release: v0.9.3

- fix: orphaned tool_use followup — ordering, empty IDs, provider_content bypass, universal repair (#220)
- feat: /retry and /rewind commands, message action controls in web UI (#221)
- docs: update tools, architecture, SDK for v0.9.2 changes

.github/renovate.json

@@ -5,55 +5,40 @@
     "helpers:pinGitHubActionDigests",
     ":separateMajorReleases"
   ],
-  "labels": [
-    "dependencies"
+  "gitIgnoredAuthors": [
+    "41898282+github-actions[bot]@users.noreply.github.com"
   ],
+  "labels": ["dependencies"],
   "prConcurrentLimit": 5,
   "prHourlyLimit": 2,
-  "schedule": [
-    "before 9am on Monday"
-  ],
+  "schedule": ["before 9am on Monday"],
   "timezone": "America/New_York",
   "lockFileMaintenance": {
     "enabled": true,
-    "schedule": [
-      "before 9am on Monday"
-    ]
+    "schedule": ["before 9am on Monday"]
   },
   "customManagers": [
     {
       "customType": "regex",
       "description": "Track vendored KaTeX version",
-      "managerFilePatterns": [
-        "/pyproject\\.toml$/"
-      ],
-      "matchStrings": [
-        "katex-(?<currentValue>[\\d.]+)/"
-      ],
+      "managerFilePatterns": ["/pyproject\\.toml$/"],
+      "matchStrings": ["katex-(?<currentValue>[\\d.]+)/"],
       "depNameTemplate": "katex",
       "datasourceTemplate": "npm"
     },
     {
       "customType": "regex",
       "description": "Track vendored Highlight.js version",
-      "managerFilePatterns": [
-        "/pyproject\\.toml$/"
-      ],
-      "matchStrings": [
-        "hljs-(?<currentValue>[\\d.]+)/"
-      ],
+      "managerFilePatterns": ["/pyproject\\.toml$/"],
+      "matchStrings": ["hljs-(?<currentValue>[\\d.]+)/"],
       "depNameTemplate": "highlight.js",
       "datasourceTemplate": "npm"
     },
     {
       "customType": "regex",
       "description": "Track vendored Mermaid version",
-      "managerFilePatterns": [
-        "/pyproject\\.toml$/"
-      ],
-      "matchStrings": [
-        "mermaid-(?<currentValue>[\\d.]+)/"
-      ],
+      "managerFilePatterns": ["/pyproject\\.toml$/"],
+      "matchStrings": ["mermaid-(?<currentValue>[\\d.]+)/"],
       "depNameTemplate": "mermaid",
       "datasourceTemplate": "npm"
     }
@@ -62,14 +47,8 @@
     {
       "description": "LLM SDKs — always review manually",
       "groupName": "LLM SDKs",
-      "matchPackageNames": [
-        "openai",
-        "anthropic",
-        "mcp"
-      ],
-      "schedule": [
-        "before 9am on Monday"
-      ],
+      "matchPackageNames": ["openai", "anthropic", "mcp"],
+      "schedule": ["before 9am on Monday"],
       "automerge": false
     },
     {
@@ -83,73 +62,38 @@
         "httpx-sse",
         "pydantic"
       ],
-      "schedule": [
-        "before 9am on Wednesday"
-      ],
+      "schedule": ["before 9am on Wednesday"],
       "automerge": true,
-      "matchUpdateTypes": [
-        "patch"
-      ]
+      "matchUpdateTypes": ["patch"]
     },
     {
       "description": "Database layer",
       "groupName": "Database",
-      "matchPackageNames": [
-        "sqlalchemy",
-        "alembic",
-        "psycopg"
-      ],
-      "schedule": [
-        "before 9am on Wednesday"
-      ],
+      "matchPackageNames": ["sqlalchemy", "alembic", "psycopg"],
+      "schedule": ["before 9am on Wednesday"],
       "automerge": true,
-      "matchUpdateTypes": [
-        "patch"
-      ]
+      "matchUpdateTypes": ["patch"]
     },
     {
       "description": "Security-critical — always review manually",
       "groupName": "Security",
-      "matchPackageNames": [
-        "PyJWT",
-        "pyjwt",
-        "bcrypt"
-      ],
+      "matchPackageNames": ["PyJWT", "pyjwt", "bcrypt"],
       "automerge": false
     },
     {
       "description": "Infrastructure dependencies",
       "groupName": "Infrastructure",
-      "matchPackageNames": [
-        "structlog",
-        "redis",
-        "croniter",
-        "discord.py"
-      ],
-      "schedule": [
-        "before 9am on the first day of the month"
-      ],
+      "matchPackageNames": ["structlog", "redis", "croniter", "discord.py"],
+      "schedule": ["before 9am on the first day of the month"],
       "automerge": true,
-      "matchUpdateTypes": [
-        "patch"
-      ]
+      "matchUpdateTypes": ["patch"]
     },
     {
-      "description": "Vendored JS — requires manual file download after merge",
+      "description": "Vendored JS — CI workflow downloads files automatically",
       "groupName": "Vendored JS",
-      "matchPackageNames": [
-        "katex",
-        "highlight.js",
-        "mermaid"
-      ],
-      "schedule": [
-        "before 9am on the first day of the month"
-      ],
-      "automerge": false,
-      "prBodyNotes": [
-        "This PR updates version references only.",
-        "After merging, run `scripts/update-vendored-js.sh <lib> <version>` to download the actual files."
-      ]
+      "matchPackageNames": ["katex", "highlight.js", "mermaid"],
+      "schedule": ["before 9am on the first day of the month"],
+      "automerge": false
     },
     {
       "description": "Dev/test tooling",
@@ -162,46 +106,29 @@
         "pytest-cov",
         "pre-commit"
       ],
-      "schedule": [
-        "before 9am on the first day of the month"
-      ],
+      "schedule": ["before 9am on the first day of the month"],
       "automerge": true,
-      "matchUpdateTypes": [
-        "patch"
-      ]
+      "matchUpdateTypes": ["patch"]
     },
     {
       "description": "Docker base images",
       "groupName": "Docker Images",
-      "matchManagers": [
-        "dockerfile",
-        "docker-compose"
-      ],
-      "schedule": [
-        "before 9am on the first day of the month"
-      ],
+      "matchManagers": ["dockerfile", "docker-compose"],
+      "schedule": ["before 9am on the first day of the month"],
       "automerge": false
     },
     {
       "description": "TypeScript SDK dev dependencies",
       "groupName": "TypeScript SDK",
-      "matchFileNames": [
-        "sdk/typescript/**"
-      ],
-      "schedule": [
-        "before 9am on the first day of the month"
-      ],
+      "matchFileNames": ["sdk/typescript/**"],
+      "schedule": ["before 9am on the first day of the month"],
       "automerge": true,
-      "matchUpdateTypes": [
-        "patch"
-      ]
+      "matchUpdateTypes": ["patch"]
     },
     {
       "description": "GitHub Actions — group all action updates",
       "groupName": "GitHub Actions",
-      "matchManagers": [
-        "github-actions"
-      ],
+      "matchManagers": ["github-actions"],
       "automerge": false
     }
   ]

.github/workflows/ci.yml

@@ -67,7 +67,7 @@ jobs:
       - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6
       - uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405 # v6
         with:
-          python-version: "3.12"
+          python-version: "3.14"
       - run: pip install -e ".[test,mq,postgres]"
       - run: pytest tests/ -m "not live" --storage-backend=postgresql -q
         env:

.github/workflows/publish.yml

@@ -19,7 +19,7 @@ jobs:
           python-version: "3.14"
       - run: pip install build
       - run: python -m build
-      - uses: pypa/gh-action-pypi-publish@release/v1
+      - uses: pypa/gh-action-pypi-publish@ed0c53931b1dc9bd32cbe73a98c7f6766f8a527e # release/v1
 
       - name: Create GitHub Release
         uses: softprops/action-gh-release@153bb8e04406b158c6c84fc1615b65b24149a1fe # v2

.github/workflows/vendor-js.yml

@@ -0,0 +1,86 @@
+name: Complete Vendored JS Updates
+
+# When Renovate bumps a vendored JS version in pyproject.toml, this
+# workflow downloads the actual files and commits them to the PR branch
+# so the PR is merge-ready without manual intervention.
+#
+# Note: the commit is made with GITHUB_TOKEN, so it won't re-trigger CI
+# automatically.  The reviewer should re-run CI once this workflow passes,
+# or Renovate's next rebase will trigger it.
+
+on:
+  pull_request:
+    paths:
+      - pyproject.toml
+  workflow_dispatch:
+    inputs:
+      pr_number:
+        description: "PR number to update"
+        required: true
+        type: number
+
+permissions:
+  contents: write
+  pull-requests: read
+
+jobs:
+  vendor-js:
+    if: github.actor == 'renovate[bot]' || github.event_name == 'workflow_dispatch'
+    runs-on: ubuntu-latest
+    steps:
+      - name: Resolve PR head ref
+        id: ref
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: |
+          if [[ "${{ github.event_name }}" == "workflow_dispatch" ]]; then
+            ref=$(gh pr view "${{ inputs.pr_number }}" --repo "${{ github.repository }}" --json headRefName -q .headRefName)
+          else
+            ref="${{ github.head_ref }}"
+          fi
+          echo "head_ref=${ref}" >> "$GITHUB_OUTPUT"
+
+      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6
+        with:
+          ref: ${{ steps.ref.outputs.head_ref }}
+
+      - name: Detect vendored JS changes
+        id: detect
+        run: |
+          updates=()
+          for lib in katex hljs mermaid; do
+            version=$(grep -oE "${lib}-[0-9.]+" pyproject.toml | head -1 | sed "s/${lib}-//")
+            [[ -z "$version" ]] && continue
+            [[ -d "turnstone/shared_static/${lib}-${version}" ]] && continue
+            updates+=("${lib}:${version}")
+          done
+          if [[ ${#updates[@]} -eq 0 ]]; then
+            echo "found=false" >> "$GITHUB_OUTPUT"
+          else
+            echo "found=true" >> "$GITHUB_OUTPUT"
+            printf '%s\n' "${updates[@]}" > /tmp/updates.txt
+            echo "Libs to update:"
+            cat /tmp/updates.txt
+          fi
+
+      - name: Download vendored files
+        if: steps.detect.outputs.found == 'true'
+        run: |
+          while IFS=: read -r lib version; do
+            echo "::group::Updating ${lib} to ${version}"
+            bash scripts/update-vendored-js.sh "$lib" "$version"
+            echo "::endgroup::"
+          done < /tmp/updates.txt
+
+      - name: Commit and push
+        if: steps.detect.outputs.found == 'true'
+        run: |
+          git config user.name "github-actions[bot]"
+          git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
+          git add -A
+          if git diff --cached --quiet; then
+            echo "No changes to commit"
+            exit 0
+          fi
+          git commit -m "chore: download vendored JS files"
+          git push

Dockerfile

@@ -8,7 +8,7 @@ FROM python:3.14-slim
 LABEL org.opencontainers.image.title="turnstone" \
       org.opencontainers.image.description="Multi-node AI orchestration platform"
 
-COPY --from=ghcr.io/astral-sh/uv:0.10.12 /uv /usr/local/bin/uv
+COPY --from=ghcr.io/astral-sh/uv:0.11.2 /uv /usr/local/bin/uv
 
 # System dependencies for psycopg (PostgreSQL client library)
 RUN apt-get update && apt-get upgrade -y && apt-get install -y --no-install-recommends libpq5 \
@@ -25,12 +25,12 @@ ENV UV_COMPILE_BYTECODE=1
 # Install dependencies first (cached layer — only re-runs when deps change)
 COPY pyproject.toml uv.lock README.md LICENSE ./
 RUN uv sync --frozen --no-install-project --no-dev \
-    --extra mq --extra console --extra sim --extra postgres --extra discord --extra anthropic
+    --extra all
 
 # Install the project itself
 COPY turnstone/ turnstone/
 RUN uv sync --frozen --no-dev \
-    --extra mq --extra console --extra sim --extra postgres --extra discord --extra anthropic
+    --extra all
 
 # Add venv to PATH so entry points are found
 ENV PATH="/app/.venv/bin:$PATH"

README.md

@@ -290,7 +290,7 @@ All entry points read `~/.config/turnstone/config.toml`. CLI flags override conf
 [api]
 base_url = "http://localhost:8000/v1"
 api_key = ""
-tavily_key = ""        # only needed for local/vLLM models without native search
+# tavily_key = ""      # only needed for local/vLLM models without native search
 
 [model]
 name = ""              # empty = auto-detect
@@ -415,6 +415,7 @@ Per-workstream metrics are labeled by `ws_id` (bounded by `[server].max_workstre
 - Redis (for message queue bridge — `pip install turnstone[mq]`)
 - Anthropic provider (optional — `pip install turnstone[anthropic]`)
 - PostgreSQL (optional, for production — `pip install turnstone[postgres]`)
+- Math sandbox packages (optional — `pip install turnstone[sandbox]` for sympy, numpy, scipy, pytest)
 - [Git LFS](https://git-lfs.com/) (for cloning — diagram PNGs are stored in LFS)
 
 ## License

deploy/docker-compose.tls.yml

@@ -0,0 +1,146 @@
+# TLS overlay — enables mTLS across the turnstone cluster.
+#
+# Usage (requires base compose.yaml with production profile):
+#   docker compose -f compose.yaml -f deploy/docker-compose.tls.yml --profile production up
+#
+# The tls-init service bootstraps a CA and issues a cert for Redis.
+# All turnstone services auto-provision their own certs via the
+# console's ACME endpoint.
+
+services:
+  # Bootstrap: create CA + Redis cert before anything starts.
+  # Runs as root to create directories in the volume, then chowns
+  # to turnstone:turnstone with restrictive perms (keys 0600).
+  tls-init:
+    build: .
+    user: root
+    command:
+      - sh
+      - -c
+      - |
+        set -e
+        turnstone-admin tls-bootstrap --out /certs --issue redis
+        chown -R turnstone:turnstone /certs
+        find /certs -type d -exec chmod 750 {} +
+        find /certs -type f -name '*key.pem' -exec chmod 600 {} +
+        find /certs -type f ! -name '*key.pem' -exec chmod 640 {} +
+    volumes:
+      - tls-certs:/certs
+    networks:
+      - turnstone-net
+    restart: "no"
+
+  # Console: runs the internal CA + ACME server
+  console:
+    depends_on:
+      tls-init:
+        condition: service_completed_successfully
+    volumes:
+      - tls-certs:/certs:ro
+    environment:
+      TURNSTONE_TLS_ENABLED: "true"
+      TURNSTONE_TLS_SANS: "console"
+      TURNSTONE_CONSOLE_URL: "http://console:8090"
+    command:
+      - turnstone-console
+      - --host=0.0.0.0
+      - --port=8090
+      - --redis-host=redis
+      - --redis-port=6379
+      - --poll-interval=${CONSOLE_POLL_INTERVAL:-10}
+      - --redis-tls
+      - --redis-tls-ca=/certs/ca.pem
+
+  # Server: auto-provisions certs via console ACME, serves HTTPS
+  server:
+    depends_on:
+      console:
+        condition: service_healthy
+    volumes:
+      - tls-certs:/certs:ro
+    environment:
+      TURNSTONE_TLS_ENABLED: "true"
+      TURNSTONE_TLS_SANS: "server"
+    # Disable healthcheck — server serves HTTPS with mTLS which the
+    # stdlib healthcheck script can't satisfy. The base compose
+    # healthcheck uses plain HTTP which won't work on an HTTPS listener.
+    # TODO: wire healthcheck with client cert from /certs volume
+    healthcheck:
+      disable: true
+
+  # Bridge: mTLS to server + Redis TLS
+  bridge:
+    depends_on:
+      console:
+        condition: service_healthy
+      server:
+        condition: service_started
+      redis:
+        condition: service_healthy
+    volumes:
+      - tls-certs:/certs:ro
+    environment:
+      TURNSTONE_TLS_ENABLED: "true"
+      TURNSTONE_TLS_SANS: "bridge"
+    command:
+      - turnstone-bridge
+      - --server-url=http://server:8080
+      - --redis-host=redis
+      - --redis-port=6379
+      - --heartbeat-ttl=${HEARTBEAT_TTL:-60}
+      - --approval-timeout=${APPROVAL_TIMEOUT:-3600}
+      - --redis-tls
+      - --redis-tls-ca=/certs/ca.pem
+
+  # Channel: Redis TLS
+  channel:
+    depends_on:
+      console:
+        condition: service_healthy
+      redis:
+        condition: service_healthy
+    volumes:
+      - tls-certs:/certs:ro
+    environment:
+      TURNSTONE_TLS_ENABLED: "true"
+      TURNSTONE_TLS_SANS: "channel"
+    command:
+      - sh
+      - -c
+      - >-
+        turnstone-channel
+        --redis-host=redis
+        --redis-port=6379
+        --redis-tls
+        --redis-tls-ca=/certs/ca.pem
+        --http-host=0.0.0.0
+        $${TURNSTONE_DISCORD_GUILD:+--discord-guild $$TURNSTONE_DISCORD_GUILD}
+
+  # Redis: TLS with certs from bootstrap
+  redis:
+    depends_on:
+      tls-init:
+        condition: service_completed_successfully
+    volumes:
+      - tls-certs:/certs:ro
+    command:
+      - sh
+      - -c
+      - |
+        ARGS="--tls-port 6379 --port 0 \
+          --tls-cert-file /certs/certs/redis/cert.pem \
+          --tls-key-file /certs/certs/redis/key.pem \
+          --tls-ca-cert-file /certs/ca.pem \
+          --tls-auth-clients no"
+        if [ -n "$$REDIS_PASSWORD" ]; then
+          ARGS="$$ARGS --requirepass $$REDIS_PASSWORD"
+        fi
+        exec redis-server $$ARGS
+    healthcheck:
+      test: ["CMD-SHELL", "if [ -n \"$$REDIS_PASSWORD\" ]; then redis-cli --tls --cacert /certs/ca.pem -a $$REDIS_PASSWORD ping; else redis-cli --tls --cacert /certs/ca.pem ping; fi"]
+      interval: 5s
+      timeout: 3s
+      retries: 5
+
+volumes:
+  tls-certs:

docs/api-reference.md

@@ -386,10 +386,10 @@ Each item in `items` (shared by `tool_info` and `approve_request`):
 {"type": "tool_output_chunk", "call_id": "call_abc123", "chunk": "Building project...\n"}
 ```
 
-**`tool_result`** -- final output from a completed tool execution. The `call_id` matches the corresponding `tool_info`/`approve_request` item and any preceding `tool_output_chunk` events. For bash tools, this arrives after all streaming chunks and includes both stdout and stderr.
+**`tool_result`** -- final output from a completed tool execution. The `call_id` matches the corresponding `tool_info`/`approve_request` item and any preceding `tool_output_chunk` events. For bash tools, this arrives after all streaming chunks and includes both stdout and stderr. The `is_error` field is `true` when the tool execution failed (e.g. bash exit code >= 2 or signal, file not found, timeout). Exit code 1 is ambiguous (e.g. `grep` no-match) and is not flagged. User denials are tracked separately via a `denied` flag. Clients should use `is_error` instead of text-prefix heuristics.
 
 ```json
-{"type": "tool_result", "call_id": "call_abc123", "name": "bash", "output": "file1.py\nfile2.py\n"}
+{"type": "tool_result", "call_id": "call_abc123", "name": "bash", "output": "file1.py\nfile2.py\n", "is_error": false}
 ```
 
 **`status`** -- token usage statistics, sent after each model turn.
@@ -452,9 +452,12 @@ after `/clear` or `/new` commands).
 {"type": "clear_ui"}
 ```
 
-**`cancelled`** -- the generation was cancelled by the user (via the Stop
-button or `POST /v1/api/cancel`). The client should finalize any in-progress
-assistant message with whatever partial content was streamed.
+**`cancelled`** -- a cancel request was acknowledged (via the Stop button or
+`POST /v1/api/cancel`). This signals that cancellation is in progress, not
+that it is complete. The worker thread may still be finishing — wait for
+`stream_end` before transitioning to a ready state. The client should clear
+any in-progress assistant rendering but not re-enable the send button until
+`stream_end` arrives.
 
 ```json
 {"type": "cancelled"}
@@ -793,23 +796,35 @@ containing the resumed session's messages.
 
 Cancels the active generation in a workstream. Sets a cooperative cancellation
 flag that is checked at multiple points in the generation loop (per streaming
-chunk, before tool execution, inside bash commands). The session transitions to
-`idle` state and preserves any partial content already streamed.
+chunk, before tool execution, inside bash commands). Also closes the underlying
+HTTP stream to the LLM provider, unblocking any pending read immediately.
+The session transitions to `idle` state and preserves any partial content
+already streamed.
 
 If the workstream is waiting for tool approval or plan review, the pending
 prompt is automatically denied/rejected to unblock the worker thread.
 
 Calling this endpoint when the workstream is already idle is a harmless no-op.
 
+**Force cancel:** When `force` is `true`, the server abandons the stuck worker
+thread immediately and transitions the workstream to `idle`. The abandoned
+thread continues to wind down in the background (killing any running
+subprocesses and exiting at the next cancellation checkpoint). During this
+wind-down it may emit a final `stream_end` event which the server suppresses
+for the orphaned thread. Use force cancel when cooperative cancel has not
+resolved within a few seconds — the web UI offers this as a "Force Stop"
+button automatically.
+
 **Request body:**
 
 ```json
-{"ws_id": "abc123"}
+{"ws_id": "abc123", "force": false}
 ```
 
 | Field  | Type   | Required | Description          |
 |--------|--------|----------|----------------------|
 | `ws_id`| string | yes      | Target workstream ID |
+| `force`| bool   | no       | Abandon stuck worker immediately (default: `false`) |
 
 **Response:**
 

docs/architecture.md

@@ -91,7 +91,7 @@ turnstone/
     _config.py        Base ChannelConfig dataclass
     discord/          Discord adapter (bot, cog, views, streaming, config)
   shared_static/      Shared design system (base.css, auth.js, theme.js, toast.js, utils.js, kb.js)
-    katex-0.16.40/    Vendored KaTeX math rendering library (MIT, woff2 fonts)
+    katex-0.16.44/    Vendored KaTeX math rendering library (MIT, woff2 fonts)
   ui/
     colors.py         ANSI color constants with NO_COLOR support
     markdown.py       Streaming terminal markdown renderer (line-buffered)
@@ -242,7 +242,7 @@ class SessionUI(Protocol):
     def on_content_token(self, text: str) -> None: ...
     def on_stream_end(self) -> None: ...
     def approve_tools(self, items: list[dict]) -> tuple[bool, str | None]: ...
-    def on_tool_result(self, call_id: str, name: str, output: str) -> None: ...
+    def on_tool_result(self, call_id: str, name: str, output: str, *, is_error: bool = False) -> None: ...
     def on_tool_output_chunk(self, call_id: str, chunk: str) -> None: ...
     def on_status(self, usage: dict, context_window: int, effort: str) -> None: ...
     def on_plan_review(self, content: str) -> str: ...
@@ -259,7 +259,7 @@ class SessionUI(Protocol):
 | Class | Module | Notes |
 |-------|--------|-------|
 | `TerminalUI` | `turnstone.cli` | ANSI colors, `MarkdownRenderer`, `Spinner`, readline-based `input()` for approval |
-| `WebUI` | `turnstone.server` | SSE event queue per workstream, `threading.Event` for blocking on approval/plan |
+| `WebUI` | `turnstone.server` | SSE event queue per workstream + global broadcast, `threading.Event` for blocking on approval/plan. `on_state_change` sends to both per-workstream and global SSE (the browser UI uses per-workstream `state_change` events to manage busy/idle transitions; `stream_end` only finalizes markdown rendering). |
 | `NullUI` | `turnstone.eval` | Discards all output; `approve_tools` always returns `(True, None)` |
 
 ### WorkstreamTerminalUI

docs/console.md

@@ -365,7 +365,7 @@ SSE streams (`/v1/api/events`, `/v1/api/events/global`) are proxied as raw byte
 
 ### Authentication
 
-The proxy forwards the user's JWT to upstream server nodes — it extracts the token from the incoming request's cookie (or `Authorization` header) and adds it as a `Bearer` header on the proxied request.  Since all services share the same `TURNSTONE_JWT_SECRET`, the user's JWT is valid on every node without re-authentication.  The console's own auth middleware also checks proxy routes — `POST` requests to proxy write endpoints (`/v1/api/send`, `/v1/api/approve`, etc.) require `write` scope, preventing read-only tokens from escalating via proxy.  The static `--auth-token` / `proxy_auth_token` is used as a fallback when no user JWT is present.
+The proxy mints a short-lived (5-minute) JWT per request carrying the real user's `user_id`, `scopes`, and `permissions` with `aud: turnstone-server`.  The user's console JWT (`aud: turnstone-console`) cannot be forwarded directly — it would be rejected by the server's audience validation — so the console re-signs a new server-audience JWT from the validated `AuthResult`.  This preserves audit attribution (the upstream server sees the real user, not a service identity) and enforces scope narrowing as defense in depth (a read-only console user's proxied request carries only `read` scope).  The JWT `src` claim is set to `"console-proxy"` for audit traceability.  When no user context is available (auth disabled), the proxy falls back to a `ServiceTokenManager` with service identity `console-proxy`.  The static `--auth-token` / `proxy_auth_token` is used as a final fallback.
 
 ---
 

docs/diagrams/03-core-engine-classes.puml

@@ -12,7 +12,7 @@ interface "SessionUI" as SessionUI <<Protocol>> {
     + on_content_token(text: str)
     + on_stream_end()
     + approve_tools(items: list) → (bool, str|None)
-    + on_tool_result(call_id: str, name: str, output: str)
+    + on_tool_result(call_id: str, name: str, output: str, *, is_error: bool = False)
     + on_tool_output_chunk(call_id: str, chunk: str)
     + on_status(usage: dict, ctx_window: int, effort: str)
     + on_plan_review(content: str) → str

docs/diagrams/04-conversation-turn.puml

@@ -133,7 +133,8 @@ group loop [while tool_calls present]
         note right of TP
           bash: on_tool_output_chunk(call_id, line)
           called per stdout line,
-          then on_tool_result(call_id, name, output).
+          then on_tool_result(call_id, name, output, is_error).
+          is_error=True when execution failed.
           call_id routes chunks/results to correct
           tool div during parallel execution.
           Other tools: on_tool_result() only.

docs/diagrams/05-tool-pipeline.puml

@@ -127,7 +127,7 @@ partition "Phase 3: Execute" #E3F2FD {
     :_truncate_output() on each result\n(max context_window × chars_per_token × 0.5 chars\ndefault: ~context_window × 2 chars);
 
     :bash: ui.on_tool_output_chunk(call_id, line) per stdout line;
-    :ui.on_tool_result(call_id, name, output) for each;
+    :ui.on_tool_result(call_id, name, output, is_error) for each;
 
     if (plan tool was executed?) then (yes)
         :ui.on_plan_review(output);

docs/diagrams/06-mq-protocol.puml

@@ -61,6 +61,7 @@ package "Inbound Messages (Client → Bridge)" as InPkg #FFF3E0 {
         + target_node: str = ""
         + initial_message: str = ""
         + skill: str = ""
+        + user_id: str = ""
     }
 
     class CloseWorkstreamMessage {
@@ -146,6 +147,7 @@ package "Outbound Events (Bridge → Client)" as OutPkg #E3F2FD {
             + call_id: str
             + name: str
             + output: str
+            + is_error: bool
         }
         class PlanReviewEvent {
             type = "plan_review"

docs/diagrams/09-workstream-states.puml

@@ -40,12 +40,23 @@ running --> error : Exception during\ntool execution
 
 error --> thinking : New send() call\n_emit_state("thinking")
 
-thinking --> idle : cancel() called\n_emit_state("idle")
+thinking --> idle : cancel() called\nstream aborted\n_emit_state("idle")
 
 running --> idle : cancel() called\n_emit_state("idle")
 
 attention --> idle : cancel() unblocks\napproval/plan wait\n_emit_state("idle")
 
+note left of idle
+  **Cancel escalation:**
+  1. **Cooperative**: cancel() sets event + closes
+     SDK stream → worker exits at next checkpoint
+  2. **Force**: force=true abandons the worker
+     thread, emits stream_end immediately.
+     Orphaned thread still kills subprocesses
+     but skips message mutations (generation
+     counter prevents stale writes).
+end note
+
 note right of thinking
   **Emitted via:**
   session._emit_state(state)

docs/diagrams/11-console-data-flow.puml

@@ -154,7 +154,7 @@ activate Server #FFECB3
 Server -> CC : _pick_best_node() or\nget_node_detail(node_id)
 CC --> Server : node validated
 
-Server -> Server : Build CreateWorkstreamMessage\n{target_node:"nodeA", name:"new-task"}
+Server -> Server : Build CreateWorkstreamMessage\n{target_node:"nodeA", name:"new-task",\nuser_id: from auth_result}
 
 Server -> Redis : RPUSH turnstone:inbound:nodeA\n(directed queue)
 Server --> Browser : {status:"ok", correlation_id:"abc",\ntarget_node:"nodeA"}
@@ -163,7 +163,8 @@ deactivate Server
 note right of Redis
   Bridge on Node-A picks up the
   message from its directed queue,
-  POSTs to /v1/api/workstreams/new,
+  POSTs to /v1/api/workstreams/new
+  (forwarding user_id in payload),
   registers ownership, publishes
   ws_created to cluster channel.
 end note

docs/diagrams/15-auth-architecture.puml

@@ -176,4 +176,15 @@ note bottom of SH
   Both share JWT signing secret
 end note
 
+note left of JWT
+  **Console Proxy Token Minting**
+  When proxying requests to server nodes:
+  1. Console AuthMiddleware validates user JWT (aud: turnstone-console)
+  2. Proxy mints new JWT (aud: turnstone-server)
+     with real user_id, scopes, permissions
+  3. src: "console-proxy" for audit traceability
+  4. 5-minute expiry (fresh per request)
+  5. Fallback: ServiceTokenManager if no user context
+end note
+
 @enduml

docs/diagrams/png/03-core-engine-classes.png

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0e605963c649574c7bf987b2d338257c78bc1fc68b524ac8276bb25365035e06
-size 594096
+oid sha256:6471e611beebf647f3a191eb16588571a404cc52a43067883a2b6f06dd936376
+size 594676

docs/diagrams/png/04-conversation-turn.png

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:da9d32000e3d92d92ce621661ced60f276f9b5be652f5ed6123b400505415f4a
-size 319702
+oid sha256:3aa8d972bba40d78152f9f0c762b9f5ec616d8052c45fa52b7dd1c679ed81d61
+size 325245

docs/diagrams/png/05-tool-pipeline.png

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:43844b07d36beb04db871f6795a3f3be17852a6a484fdc0ea207403bd7f512a6
-size 274286
+oid sha256:72b3932ce99a860f5069544cd8423d3cdae3a51f6566db262b19ced19c780eb0
+size 274374

docs/diagrams/png/06-mq-protocol.png

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2636e2d4d2f84f26f93de6e984b780f6ad5bf8d3618a0202ea0a2ee80859c5b5
-size 312409
+oid sha256:d831c5e10266f0232262b6a29b0ea8e45b3cc63df75f716a80f18462b4f85e66
+size 319125

docs/diagrams/png/09-workstream-states.png

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7896c6e041b6dbb89d034468fa980c8fe645df5eb969d45ef966ccc6399edac2
-size 200083
+oid sha256:7bf27afa267d5b8d6da38e83213ed1b8e87639d5105a0a1ccc2e5a4bf4d3b67e
+size 185282

docs/eval.md

@@ -2,7 +2,8 @@
 
 `turnstone-eval` is the evaluation and prompt optimization system for turnstone. It
 runs test cases against the LLM, scores tool call sequences against expected
-actions, and optionally uses the model to self-optimize the developer prompt.
+actions, and optionally uses a multi-agent pipeline to optimize the developer
+prompt and tool descriptions.
 
 Source: `turnstone/eval.py`
 
@@ -10,15 +11,24 @@ Source: `turnstone/eval.py`
 
 ## Overview
 
-The system works in an iterative loop:
+The system uses UCB tree search to explore prompt variants:
 
-1. Run each test case N times against the current developer prompt.
-2. Score each run by comparing the actual tool call sequence to expected actions.
-3. If not all tests pass, use the model to rewrite the prompt based on failures.
-4. Repeat until all tests pass or max iterations are reached.
+1. Maintain an **evolution tree** of prompt variants, starting from the initial prompt.
+2. Each iteration, **UCB1 selects** the most promising node to evaluate.
+3. Run each test case N times against the selected prompt.
+4. Score each run by comparing the actual tool call sequence to expected actions.
+5. If not all tests pass, run a **three-phase optimization pipeline**:
+   - Phase 1: Analyst diagnoses semantic failure patterns
+   - Phase 2: Tool optimizer adjusts tool descriptions (when `--optimize-tools`)
+   - Phase 3: Prompt optimizer proposes a child variant (when not `--optimize-tools`)
+6. Add the child to the tree and repeat until all tests pass or max iterations reached.
 
-When optimization is disabled (`--no-optimize`), only step 1 and 2 execute
-(a single iteration).
+This approach (inspired by [Learning to Self-Evolve](https://arxiv.org/abs/2603.18620))
+prevents irrecoverable collapse from bad edits — UCB naturally backtracks to
+high-scoring ancestors instead of following a linear chain.
+
+When optimization is disabled (`--no-optimize`), only steps 2-4 execute
+(a single iteration evaluating the root node).
 
 ---
 
@@ -65,6 +75,7 @@ Test suites are JSON files with this structure:
 | `match_mode`       | no       | `"ordered_subset"` | How to match actual vs expected actions (see Scoring). |
 | `max_turns`        | no       | `10`               | Maximum conversation turns before stopping. |
 | `n_runs`           | no       | suite default or 3 | Per-case override for number of runs. |
+| `holdout`          | no       | `false`            | If `true`, this case is evaluated but excluded from optimizer feedback. Used to measure progress without overfitting. |
 
 ### Expected Action Specs
 
@@ -135,6 +146,7 @@ deterministic, non-interactive execution suitable for automated testing.
 | Stdout          | Normal                  | Suppressed during execution |
 | Tool logging    | Display only            | Structured `tool_call_log`  |
 | System prompt   | Built-in developer prompt | Overridable via constructor |
+| Cancellation    | N/A                     | `_cancelled` event for timeout cleanup |
 
 ### NullUI
 
@@ -156,20 +168,34 @@ def send_headless(
 Runs a complete multi-turn conversation:
 
 1. Appends the user message.
-2. Calls the model API (non-streaming).
-3. If tool calls are returned, executes them (with stdout suppressed) and
+2. Checks `_cancelled` event — stops if set (timeout cleanup).
+3. Calls the model API (non-streaming).
+4. If tool calls are returned, executes them (with stdout suppressed) and
    logs each call to `self.tool_call_log`.
-4. Repeats up to `max_turns` or until the model responds without tool calls.
-5. Returns the tool call log: list of dicts with keys `tool`, `args`,
+5. Repeats up to `max_turns` or until the model responds without tool calls.
+6. Returns the tool call log: list of dicts with keys `tool`, `args`,
    `result` (truncated to 500 chars), and `turn`.
 
 Parallel tool calls are capped at 10 per turn to prevent degenerate repetition.
 
+### Timeout and Cancellation
+
+Each test runs in a `ThreadPoolExecutor(max_workers=1)` with a per-test
+timeout (`--test-timeout`). Each attempt gets its own `OpenAI` client with
+a matching httpx read timeout. On timeout, three layers of defense prevent
+zombie connections:
+
+1. **httpx timeout**: Per-request read timeout aborts the HTTP call and
+   releases the server slot.
+2. **`_cancelled` event**: Prevents the orphan thread from starting new turns.
+3. **`run_client.close()`**: Closes the connection pool to abort any
+   in-flight request.
+
 ### Retry Logic
 
 `send_headless()` is called inside `_run_single_test()` with retry logic:
 3 attempts with exponential backoff (sleep `2^attempt` seconds) on any
-exception. This prevents transient API errors from poisoning eval scores.
+exception. `TimeoutError` is re-raised immediately (no retry).
 
 ---
 
@@ -180,68 +206,175 @@ Each test case runs in isolation:
 1. A fresh temp directory is created.
 2. Setup files are written to the temp directory.
 3. The working directory is changed to the temp directory.
-4. A new `HeadlessSession` is created with the current developer prompt.
-5. `send_headless()` runs the user prompt through the conversation loop.
-6. The tool log is scored against expected actions.
-7. The temp directory is cleaned up.
+4. A per-attempt `OpenAI` client is created with httpx timeout matching `--test-timeout`.
+5. A new `HeadlessSession` is created with the current developer prompt.
+6. `send_headless()` runs the user prompt through the conversation loop.
+7. The tool log is scored against expected actions.
+8. The temp directory is cleaned up.
 
 The memory database is also isolated per test (an ephemeral SQLite database
 in the temp directory) so tests do not pollute each other or the user's
 real memory store.
 
+### Parallel Execution
+
+With `--parallel N` (N > 1), tests run in a `ProcessPoolExecutor` with N
+workers. Each subprocess creates its own `OpenAI` client. This is suitable
+for remote API endpoints but will overwhelm local inference servers. The
+default (`--parallel 1`) runs tests serially.
+
 ---
 
-## Optimization Loop
+## Model Roles
 
-`run_optimization()` is the main entry point for iterative prompt optimization.
+The eval pipeline uses up to five separate model roles, each independently
+configurable. All roles inherit from the test model by default, with a
+cascade chain:
+
+```
+test model (--base-url, --model)
+  └─ optimizer (--optimizer-*)
+       ├─ observer (--observer-*)
+       ├─ analyst (--analyst-*)
+       ├─ diversifier (--diversifier-*)
+       └─ tool optimizer (--tool-optimizer-*)
+```
+
+| Role | Purpose | When it runs |
+|------|---------|--------------|
+| **Test** | The model being evaluated | Every iteration |
+| **Analyst** | Diagnoses semantic failure patterns with tool use | When pass rate < 100% |
+| **Optimizer** | Rewrites the developer prompt | Every iteration (unless `--optimize-tools`) |
+| **Tool optimizer** | Rewrites tool descriptions | When `--optimize-tools` is set |
+| **Observer** | Tunes the optimizer's strategy | Every 3 iterations |
+| **Diversifier** | Generates prompt paraphrases | Once before the loop (when `--diversify N`) |
+
+Typical setup: local model for test, Opus for analyst, Sonnet for
+optimizer/observer/diversifier.
+
+---
+
+## Optimization Pipeline
 
 ### Flow
 
 ```
 for iteration in 0..max_iterations:
-    1. Run all test cases n_runs times with current prompt
-    2. Score and aggregate results
-    3. Save intermediate results to JSON
-    4. If all tests pass -> stop
-    5. Every 3 iterations (at iteration 2, 5, 8, ...):
-       -> Observer reviews optimizer strategy
-       -> Reset prompt to best-performing iteration
-    6. Propose new prompt via optimizer model call
-    7. If prompt unchanged -> stop
-    8. Continue with new prompt
+    1. UCB select → pick the most promising tree node
+    2. Run all test cases n_runs times with selected node's prompt
+    3. Update node score (rolling mean) and visit count
+    4. Save intermediate results + tree state to JSON
+    5. If all tests pass → stop
+    6. Phase 1: Analyst diagnoses semantic failure patterns
+    7. Phase 2 (--optimize-tools only): Tool optimizer adjusts descriptions
+    8. Phase 3 (default only): Prompt optimizer proposes new prompt
+    9. Every 3 iterations: Observer tunes the optimizer's strategy
+   10. Add child node to tree (if prompt or tools changed)
 ```
 
-### Prompt Proposal (`_propose_prompt_modification`)
+### Phase 1: Analyst (`_run_analyst`)
 
-Uses the model to rewrite the developer prompt based on test results:
+A multi-turn agent with `math` (Python) and `bash` tools for computing
+statistics. It receives per-case results with failure classifications and
+produces a structured diagnosis:
 
-- **Input**: Current prompt, test case definitions, per-case results with
-  actual vs expected tool sequences, and a history of the last 3 iterations.
-- **Optimizer system prompt** (`OPTIMIZER_SYSTEM`): Instructs the model to
-  act as a text rewriter. Key guidance includes:
-  - Address critical failure modes (text-only responses, write_file vs edit_file,
-    unnecessary search before create, missing plan calls).
-  - Preserve phrasing that drives 100% pass rate on passing tests.
-  - Use direct imperative style with concrete tool call examples.
-  - Stay within 130% of original prompt length.
-- **Output**: The rewritten prompt text (stripped of reasoning tags and code fences).
+- **Failure patterns**: Shared root causes across failing cases
+- **Success/failure contrast**: What distinguishes passing from failing cases
+- **Consistency signals**: Systematic (0%), flaky (1-79%), marginal (80-99%)
+- **Recommended fixes**: Priority-ordered patterns/examples to add or adjust
 
-### Observer System (`_observe_and_update_optimizer`)
+The analyst is instructed to frame fixes as patterns and examples, not
+imperative rules — this feeds cleaner signal to the optimizer.
 
-Every 3 iterations, a meta-level "observer" reviews the optimizer's strategy:
+In `--optimize-tools` mode, the analyst receives the current tool descriptions
+(with any overrides applied) and focuses on tool confusion and description
+issues rather than system prompt patterns.
 
-- Analyzes the iteration history: score trends, regressions, prompt length changes,
+### Phase 2: Tool Optimizer (`_propose_tool_overrides`)
+
+Runs when `--optimize-tools` is set. Receives the current tool descriptions,
+confusion failures (where the model picked the wrong tool), and the analyst's
+diagnosis. Returns a JSON override dict that modifies tool descriptions.
+
+Overrides are validated against known tool names — only `description` and
+`parameters` changes are accepted (no tool renaming at eval time).
+
+After each iteration, changed descriptions are logged as old → new diffs
+for easy visual inspection.
+
+### Phase 3: Prompt Optimizer (`_propose_prompt_modification`)
+
+Skipped in `--optimize-tools` mode. Receives the current prompt, test
+results with per-case pass rates and deltas from the parent node, and the
+analyst's diagnosis. Returns a rewritten prompt.
+
+The optimizer is instructed to prefer patterns over rules — concrete tool
+chain examples teach better than imperative directives like "ALWAYS" or
+"NEVER." If the current prompt contains rule-heavy language, the optimizer
+is guided to replace it with examples.
+
+### Two Optimization Surfaces
+
+The system supports alternating between two optimization surfaces:
+
+1. **System prompt optimization** (default): Freeze tool descriptions,
+   optimize the developer prompt. Run until scores plateau.
+2. **Tool description optimization** (`--optimize-tools`): Freeze the system
+   prompt, optimize tool descriptions only. Run until scores plateau.
+
+Each surface lifts the floor for the other — tool description improvements
+may unlock system prompt gains that weren't reachable before, and vice versa.
+
+### Observer (`_observe_and_update_optimizer`)
+
+Every 3 iterations, a meta-level observer reviews the optimizer's strategy:
+
+- Analyzes iteration history: score trends, regressions, prompt length changes,
   and diffs between iterations.
-- Summarizes the optimizer's behavioral patterns (list style, header usage, length).
-- Uses `OBSERVER_SYSTEM` to rewrite the optimizer's own system prompt.
+- Detects whether the optimizer is producing rule-heavy or pattern-based output.
+- Rewrites the optimizer's own system prompt to correct course.
 - Rejects degenerate outputs (over 200% of input length).
-- After updating the optimizer prompt, resets the developer prompt to the
-  best-performing iteration so far.
 
-This two-level optimization (optimizer + observer) helps the system escape
-local minima and adjust its rewriting strategy.
+### Prompt Diversification
 
-### Result Persistence
+When `--diversify N` is set, the diversifier generates N paraphrased variants
+of each test case's user prompt before the optimization loop. Each run cycles
+through variants (round-robin), testing robustness across phrasings.
+
+Variants can be cached back to the test suite JSON with `--save-variants`,
+and auto-loaded on subsequent runs even without `--diversify`.
+
+---
+
+## Evolution Tree
+
+The optimization maintains a tree of prompt variants (`EvolutionNode`), where
+each node stores its prompt text, tool overrides, aggregated score, and visit
+count. The root node (ID 0) contains the initial prompt.
+
+**UCB1 selection**: Each iteration picks the node with the highest Upper
+Confidence Bound score: `R_bar + C * sqrt(ln(N) / v)`, where `R_bar` is the
+node's mean score, `N` is total visits across all nodes, `v` is the node's
+visit count, and `C` is the exploration constant (`--explore-constant`,
+default sqrt(2)). Unvisited nodes are always selected first.
+
+### Holdout Cases
+
+Test cases with `"holdout": true` are evaluated every iteration but excluded
+from the optimizer's feedback. This prevents the optimizer from overfitting
+to specific test cases. Node scores are computed from holdout cases only
+(when present). If fewer than 2 non-holdout cases remain, holdout is disabled.
+
+### Improvement-Based Feedback
+
+The optimizer sees delta scores (`delta=+20%`) alongside absolute pass rates,
+showing how each case improved relative to the parent node's evaluation. This
+provides a cleaner signal than absolute scores alone — the optimizer can
+distinguish beneficial edits from harmful ones regardless of starting point.
+
+---
+
+## Result Persistence
 
 After each iteration, results are written to the output JSON file. The
 structure is:
@@ -251,9 +384,15 @@ structure is:
   "meta": {
     "model": "model-name",
     "base_url": "http://localhost:8000/v1",
+    "optimizer_model": "claude-opus-4-6",
+    "observer_model": "claude-opus-4-6",
     "started": "2025-01-01T00:00:00",
     "test_suite": "tests.json",
-    "n_runs_default": 3
+    "n_runs_default": 3,
+    "explore_constant": 1.414,
+    "holdout_ids": [],
+    "diversify": 10,
+    "prompt_variants": {"case_id": ["variant1", "variant2"]}
   },
   "iterations": [
     {
@@ -261,7 +400,11 @@ structure is:
       "prompt": "the developer prompt used",
       "prompt_diff": null,
       "optimizer_system": "the optimizer system prompt",
+      "analyst": "analyst diagnosis output",
+      "tool_overrides": {"bash": {"description": "..."}},
       "timestamp": "2025-01-01T00:01:00",
+      "tree_node_id": 0,
+      "tree_child_id": 1,
       "cases": {
         "test_name": {
           "runs": [
@@ -287,9 +430,21 @@ structure is:
         "overall_pass_rate": 0.8,
         "overall_avg_score": 0.87,
         "json_dumps": 0,
-        "per_case_pass_rates": {"test_name": 1.0, ...}
+        "per_case_pass_rates": {"test_name": 1.0}
       }
     }
+  ],
+  "tree": [
+    {
+      "node_id": 0,
+      "parent_id": null,
+      "prompt": "initial prompt",
+      "tool_overrides": {},
+      "score": 0.85,
+      "visit_count": 3,
+      "children": [1, 2],
+      "iteration": 0
+    }
   ]
 }
 ```
@@ -306,26 +461,57 @@ turnstone-eval tests.json                          # evaluate + optimize
 turnstone-eval tests.json --no-optimize            # evaluate only (single iteration)
 turnstone-eval tests.json --n-runs 5 --max-iter 10 # more thorough evaluation
 turnstone-eval tests.json --prompt custom.txt      # start from a custom prompt
+turnstone-eval tests.json --optimize-tools         # optimize tool descriptions only
+turnstone-eval tests.json --diversify 10           # test with prompt variants
 turnstone-eval tests.json -v                       # verbose per-turn logging
 ```
 
+### Multi-model setup (local test model, cloud optimizer)
+
+```
+turnstone-eval tests.json \
+  --base-url http://localhost:8000/v1 \
+  --optimizer-base-url https://api.anthropic.com \
+  --optimizer-model claude-sonnet-4-6 \
+  --analyst-model claude-opus-4-6
+```
+
 ### All Options
 
-| Flag                | Default                       | Description |
-|---------------------|-------------------------------|-------------|
-| `test_file`         | (positional, required)        | Path to test cases JSON file. |
-| `--base-url`        | `http://localhost:8000/v1`    | API base URL. |
-| `--model`           | auto-detect                   | Model name. Auto-detected from the API if not specified. |
-| `--prompt`          | turnstone built-in prompt         | Path to initial prompt text file. |
-| `--n-runs`          | from tests.json or 3          | Number of runs per test case. |
-| `--max-iter`        | 5                             | Maximum optimization iterations. |
-| `--no-optimize`     | false                         | Run evaluation only (sets max-iter to 1). |
-| `--temperature`     | 0.7                           | Sampling temperature. |
-| `--max-tokens`      | 32768                         | Max completion tokens. |
-| `--reasoning-effort` | `medium`                     | Reasoning effort: `low`, `medium`, or `high`. |
-| `--context-window`  | 131072                        | Context window size. |
-| `--output`          | `eval_results.json`           | Output results file path. |
-| `-v`, `--verbose`   | false                         | Show detailed per-turn logging (API calls, tool args, results). |
+| Flag                    | Default                    | Description |
+|-------------------------|----------------------------|-------------|
+| `test_file`             | (positional, required)     | Path to test cases JSON file. |
+| `--base-url`            | `http://localhost:8000/v1` | API base URL for the test model. |
+| `--model`               | auto-detect                | Model name. Auto-detected from the API if not specified. |
+| `--prompt`              | turnstone built-in prompt  | Path to initial prompt text file. |
+| `--n-runs`              | from tests.json or 3       | Number of runs per test case. |
+| `--max-iter`            | 5                          | Maximum optimization iterations. |
+| `--no-optimize`         | false                      | Run evaluation only (sets max-iter to 1). |
+| `--temperature`         | 0.7                        | Sampling temperature. |
+| `--max-tokens`          | 32768                      | Max completion tokens. |
+| `--reasoning-effort`    | `medium`                   | Reasoning effort: `low`, `medium`, or `high`. |
+| `--context-window`      | 131072                     | Context window size. |
+| `--output`              | `eval_results.json`        | Output results file path. |
+| `-v`, `--verbose`       | false                      | Show detailed per-turn logging. |
+| `--explore-constant`    | 1.414 (sqrt(2))            | UCB exploration constant C. |
+| `--test-timeout`        | 300                        | Per-test timeout in seconds. |
+| `--suite-timeout`       | 0 (unlimited)              | Total suite timeout in seconds. |
+| `--no-fast-fail`        | false                      | Disable early termination on all-zero initial runs. |
+| `--parallel`            | 1 (serial)                 | Parallel workers (0=auto, N=use N workers). |
+| `--optimizer-model`     | same as `--model`          | Model for prompt optimization. |
+| `--optimizer-base-url`  | same as `--base-url`       | Base URL for optimizer model. |
+| `--observer-model`      | same as optimizer           | Model for meta-optimization (observer). |
+| `--observer-base-url`   | same as optimizer           | Base URL for observer model. |
+| `--analyst-model`       | same as optimizer           | Model for failure analysis. |
+| `--analyst-base-url`    | same as optimizer           | Base URL for analyst model. |
+| `--diversify`           | 0 (disabled)               | Generate N prompt variants per test case. |
+| `--diversifier-model`   | same as optimizer           | Model for prompt diversification. |
+| `--diversifier-base-url`| same as optimizer           | Base URL for diversifier model. |
+| `--save-variants`       | false                      | Save generated variants back to test suite JSON. |
+| `--optimize-tools`      | false                      | Optimize tool descriptions only (freeze system prompt). |
+| `--tool-optimizer-model` | same as optimizer          | Model for tool description optimization. |
+| `--tool-optimizer-base-url` | same as optimizer      | Base URL for tool optimizer model. |
+| `--save-tools`          | false                      | Write optimized tool descriptions back to `turnstone/tools/*.json`. |
 
 ### Precedence for n_runs
 

docs/judge.md

@@ -338,7 +338,7 @@ from the output before it enters the conversation.
 | Priority | Category | Risk | Examples |
 |----------|----------|------|----------|
 | 1 | Prompt injection | high | Override phrases, role injection (`{"role":"system"}`), instruction override markers |
-| 2 | Credential leakage | high | API keys, private key blocks, connection strings, `.env` format secrets |
+| 2 | Credential leakage | high | API keys, private key blocks, connection strings, `.env` format secrets, JSON secrets (`"api_key": "..."`, `"password": "..."`, etc.) |
 | 3 | Encoded payloads | medium | Script data URIs, hex shellcode sequences |
 | 4 | Adversarial URLs | medium | Cloud metadata endpoints, credential-bearing query parameters |
 | 5 | System info disclosure | low | Private IP addresses, sensitive file paths |

docs/sdk.md

@@ -75,7 +75,7 @@ Both `TurnstoneServer` (sync) and `AsyncTurnstoneServer` (async) expose:
 | | `approve(*, ws_id, approved, feedback, always)` | `StatusResponse` |
 | | `plan_feedback(*, ws_id, feedback)` | `StatusResponse` |
 | | `command(*, ws_id, command)` | `StatusResponse` |
-| | `cancel(ws_id)` | `StatusResponse` |
+| | `cancel(ws_id, *, force=False)` | `StatusResponse` |
 | **Streaming** | `stream_events(ws_id)` | `Iterator[ServerEvent]` |
 | | `stream_global_events()` | `Iterator[ServerEvent]` |
 | **High-level** | `send_and_wait(message, ws_id, *, timeout, on_event)` | `TurnResult` |
@@ -127,7 +127,7 @@ SSE events are deserialized into typed dataclasses. Use `event.type` to discrimi
 | `reasoning` | `ReasoningEvent` | `text` |
 | `tool_info` | `ToolInfoEvent` | `items` |
 | `approve_request` | `ApproveRequestEvent` | `items` |
-| `tool_result` | `ToolResultEvent` | `call_id`, `name`, `output` |
+| `tool_result` | `ToolResultEvent` | `call_id`, `name`, `output`, `is_error` |
 | `tool_output_chunk` | `ToolOutputChunkEvent` | `call_id`, `chunk` |
 | `status` | `StatusEvent` | `prompt_tokens`, `total_tokens`, `pct`, `effort`, `cache_creation_tokens`, `cache_read_tokens` |
 | `plan_review` | `PlanReviewEvent` | `content` |

docs/security.md

@@ -457,14 +457,30 @@ without any database.
 ### Proxy auth forwarding
 
 When the console proxies requests to server nodes (via `/node/{id}/...`
-routes), it uses a dedicated **service proxy token** with
-`aud: turnstone-server` and `write` scope.  The user's console JWT
-(which has `aud: turnstone-console`) is **not** forwarded — it would be
-rejected by the server's audience validation.
+routes), it mints a **short-lived user-scoped JWT** with
+`aud: turnstone-server` carrying the real user's `user_id`, `scopes`,
+and `permissions`.  The user's console JWT (which has
+`aud: turnstone-console`) is **not** forwarded directly — it would be
+rejected by the server's audience validation.  Instead, the console
+re-signs a new JWT targeted at the server audience.
 
-The proxy token is managed by a `ServiceTokenManager` that auto-rotates
-1-hour JWTs, refreshing at 80% of lifetime.  If `--auth-token` is
-provided, that static token is used instead.
+Each proxied request gets a fresh JWT (5-minute expiry).  This ensures:
+
+- **Audit attribution** — the upstream server records the real user in
+  `ctx_user_id` and audit events, not a generic service identity.
+- **Scope narrowing** — a read-only console user's proxied request
+  carries only `read` scope, not the full `{read, write, approve}` set.
+  The server enforces this as defense in depth.
+- **Permission forwarding** — granular RBAC permissions from the
+  console JWT are carried through to the server.
+
+The JWT `src` claim is set to `"console-proxy"`, allowing servers to
+distinguish proxied requests from direct logins in audit logs.
+
+When no user context is available (auth disabled, or internal requests),
+the proxy falls back to a `ServiceTokenManager` with service identity
+`console-proxy` and full scopes.  If `--auth-token` is provided, that
+static token is used as a final fallback.
 
 ### Service-to-service authentication
 
@@ -475,7 +491,7 @@ auto-rotating JWTs when communicating with server nodes:
 |---------|----------|-------|----------|---------|
 | Bridge | `bridge` | `approve` | `turnstone-server` | Tool approval proxy, message relay |
 | Console collector | `console-collector` | `read` | `turnstone-server` | Node health polling |
-| Console proxy | `console-proxy` | `write` | `turnstone-server` | Proxied API calls |
+| Console proxy (fallback) | `console-proxy` | `approve` | `turnstone-server` | Proxied API calls when no user context |
 | Channel notify | `system` | `write` | `turnstone-channel` | Notification delivery to channel gateway |
 
 Service tokens use 1-hour expiry with automatic refresh via
@@ -483,6 +499,17 @@ Service tokens use 1-hour expiry with automatic refresh via
 httpx event hooks to ensure rotated tokens are picked up on SSE
 reconnects.
 
+### User identity in MQ-dispatched workstreams
+
+When the console creates a workstream via MQ (the normal path), the
+authenticated user's `user_id` is embedded in the
+`CreateWorkstreamMessage`.  The bridge forwards this `user_id` in the
+HTTP payload when calling the server's `POST /v1/api/workstreams/new`.
+The server accepts a `user_id` from the request body **only when the
+caller is a trusted service** — identified by `token_source` matching
+`bridge`, `console-proxy`, or `console`.  Regular API callers cannot
+override `user_id`; the server always uses their JWT identity.
+
 Note that the channel gateway uses a distinct JWT audience
 (`turnstone-channel`) from the server (`turnstone-server`) and console
 (`turnstone-console`).  A server-scoped JWT cannot authenticate to the

docs/tls.md

@@ -0,0 +1,221 @@
+# TLS / mTLS
+
+Turnstone supports end-to-end transport encryption with mutual TLS (mTLS) for
+inter-service communication, powered by [lacme](https://pypi.org/project/lacme/).
+
+---
+
+## Quick Start (Docker Compose)
+
+```bash
+docker compose -f compose.yaml -f deploy/docker-compose.tls.yml up
+```
+
+This:
+1. Bootstraps an internal CA and issues certs for Redis/PostgreSQL
+2. Starts the console with TLS enabled (internal CA + ACME server)
+3. Server nodes auto-provision certs via the console's ACME endpoint
+4. All inter-service communication uses mTLS
+
+---
+
+## Architecture
+
+```
+Console (CA + ACME Server)
+  +-- CertificateAuthority (owns root key, signs certs)
+  +-- ACMEResponder (mounted at /acme, RFC 8555)
+  +-- GET /acme/ca.pem (root cert for node bootstrapping)
+                  |
+                  | ACME protocol (auto-approve, no challenge validation)
+      +-----------+-----------+
+      |           |           |
+  Server(s)    Bridge    Channel GW
+  (auto-cert    (mTLS     (mTLS
+   + renewal)   client)    client)
+```
+
+**Two cert paths on the console:**
+- **Internal cert** (mTLS): Always from the internal CA. Used for cluster
+  service mesh communication.
+- **Frontend cert** (HTTPS): From an external ACME CA (e.g. Let's Encrypt)
+  if `tls.acme_directory` is set, otherwise self-issued from the internal CA.
+
+---
+
+## Configuration
+
+### Settings (ConfigStore / Admin Settings tab)
+
+| Setting | Default | Description |
+|---------|---------|-------------|
+| `tls.enabled` | `false` | Master switch for internal mTLS |
+| `tls.acme_directory` | `""` | External ACME CA URL for console frontend cert |
+
+### Bootstrap Config (config.toml)
+
+These are needed before storage is available:
+
+```toml
+[redis]
+tls = false
+tls_ca = ""      # path to CA cert
+tls_cert = ""    # path to client cert
+tls_key = ""     # path to client key
+
+[database]
+sslmode = "prefer"   # disable, allow, prefer, require, verify-full
+sslrootcert = ""     # path to CA cert
+sslcert = ""         # path to client cert
+sslkey = ""          # path to client key
+```
+
+### Hardcoded Defaults
+
+| Parameter | Value | Notes |
+|-----------|-------|-------|
+| CA common name | "Turnstone CA" | |
+| CA validity | 10 years | |
+| Cert validity | 48 hours | Short-lived, auto-renewed |
+| Renewal interval | 24 hours | Half of validity |
+| ACME auto-approve | true | Internal network, no challenge validation |
+
+---
+
+## CLI
+
+### Offline Bootstrap
+
+Create a CA and infrastructure certs without a running console:
+
+```bash
+# Bootstrap CA + Redis + PostgreSQL certs
+turnstone-admin tls-bootstrap --out /certs --issue redis --issue postgres
+
+# Output:
+#   /certs/ca.pem              (CA root certificate)
+#   /certs/certs/redis/        (Redis cert + key)
+#   /certs/certs/postgres/     (PostgreSQL cert + key)
+```
+
+The output directory is chmod 0700 (contains the CA private key).
+
+### Online Cert Issuance
+
+Request certs from a running console's ACME endpoint:
+
+```bash
+# Download CA root cert (TOFU — verify fingerprint)
+turnstone-admin tls-ca-cert --out ca.pem --console-url http://console:8080
+
+# Request a cert for a domain
+turnstone-admin tls-issue worker-1.internal --out /certs --console-url http://console:8080
+
+# List issued certs
+turnstone-admin tls-list --console-url http://console:8080 --auth-token $TOKEN
+```
+
+### Console URL Discovery
+
+If `--console-url` is not provided, the CLI discovers it from the `services`
+table in the shared database. The console registers itself on startup.
+
+---
+
+## Admin UI
+
+The **TLS** tab in the console admin panel (System group) shows:
+- CA status (common name, certificate count)
+- Certificate table (domain, SANs, issued, expires)
+- Force-renew and delete actions per certificate
+
+---
+
+## SDK
+
+### Python
+
+```python
+from turnstone.sdk import TurnstoneServer
+
+client = TurnstoneServer(
+    base_url="https://server:8080",
+    token="tok_xxx",
+    ca_cert="/path/to/ca.pem",
+    client_cert="/path/to/cert.pem",
+    client_key="/path/to/key.pem",
+)
+```
+
+### TypeScript
+
+```typescript
+import { TurnstoneServer } from "@turnstone/sdk";
+import { Agent } from "undici";
+import * as fs from "fs";
+
+const agent = new Agent({
+  connect: {
+    ca: fs.readFileSync("/path/to/ca.pem"),
+    cert: fs.readFileSync("/path/to/cert.pem"),
+    key: fs.readFileSync("/path/to/key.pem"),
+  },
+});
+
+const client = new TurnstoneServer({
+  baseUrl: "https://server:8080",
+  token: "tok_xxx",
+  // Node.js 18+ uses undici under the hood
+  fetch: (url, init) =>
+    fetch(url, { ...init, dispatcher: agent } as RequestInit),
+});
+```
+
+---
+
+## How It Works
+
+### Node Bootstrap Flow
+
+1. Node starts, connects to shared database (plain connection)
+2. Discovers console URL from `services` table
+3. Fetches CA root cert from `http://console/acme/ca.pem` (plain HTTP, TOFU)
+4. Requests service cert via ACME protocol (plain HTTP, JWS-signed)
+5. Starts auto-renewal (24h interval, re-issues before expiry)
+6. All subsequent inter-service communication uses mTLS
+
+### Console Startup Flow
+
+1. Read `tls.enabled` from ConfigStore
+2. Initialize CA (load from DB or generate new root key)
+3. Mount ACME responder at `/acme` (serves `/ca.pem` natively)
+4. Issue console certs (internal + optional frontend)
+5. Start CA-direct auto-renewal (no network, signs directly)
+6. Register console URL in services table with heartbeat
+
+---
+
+## Troubleshooting
+
+### Cert expired / mTLS connection refused
+
+Certs are valid for 48 hours. If auto-renewal stopped (e.g. console was down),
+restart the service to re-request a cert.
+
+### "No console service found"
+
+The console registers itself in the `services` table on startup. If the console
+hasn't started or the registration expired (1 hour TTL), nodes can't discover
+it. Use `--console-url` explicitly.
+
+### Let's Encrypt for console frontend
+
+Set `tls.acme_directory` to `https://acme-v02.api.letsencrypt.org/directory`
+in the admin Settings tab. The console will request a publicly trusted cert
+for its HTTPS endpoint. Internal mTLS still uses the private CA.
+
+### Verifying the cert chain
+
+```bash
+openssl s_client -connect server:8080 -CAfile ca.pem
+```

docs/tools.md

@@ -102,10 +102,15 @@ Each item's `execute` callable is invoked:
 - Errored or denied items return their error/denial message without executing.
 - The `bash` tool streams stdout incrementally: each line calls
   `ui.on_tool_output_chunk(call_id, line)` as it is produced, then the final
-  combined output (stdout + stderr) is delivered via `ui.on_tool_result(call_id, name, output)`.
+  combined output (stdout + stderr) is delivered via
+  `ui.on_tool_result(call_id, name, output, is_error=...)`.
   The `call_id` links `tool_info`/`approve_request` items to their streaming chunks and
   final result, enabling correct routing when multiple bash tools run in parallel.
-  Other tools deliver results atomically via `ui.on_tool_result(call_id, name, output)` only.
+  The `is_error` flag is `True` when the tool execution failed (e.g. bash exit code >= 2
+  or signal, file not found, timeout). Exit code 1 is ambiguous and not flagged; user
+  denials are tracked separately. This removes the need for text-prefix heuristics.
+  Other tools deliver results atomically via
+  `ui.on_tool_result(call_id, name, output, is_error=...)` only.
 - Special post-execution gate for `plan`: the plan output is shown to the user
   for review, and the user can reject or annotate it.
 
@@ -184,7 +189,8 @@ Execute a bash command and return stdout + stderr.
 |-----------|--------|----------|-------------|
 | `command` | string | yes      | The bash command to execute. |
 
-- **What it does**: Runs the command in a subprocess with a configurable timeout. Commands are sanitized and checked against a blocklist (e.g. `rm -rf /`).
+- **What it does**: Runs the command in a subprocess with a configurable timeout (default 120s). Commands are sanitized and checked against a blocklist (e.g. `rm -rf /`). Environment variables containing secrets are scrubbed (`*_KEY`, `*_SECRET`, `*_TOKEN`, etc.).
+- **Output format**: Stdout is returned directly. Stderr lines are prefixed with `[stderr]` so the model can distinguish them. When the command itself redirects stderr to stdout (`2>&1`), no prefix is added. Output exceeding 256KB is truncated (head + tail preserved, middle replaced with a truncation notice).
 - **Auto-approve**: No -- requires user confirmation.
 - **Agent availability**: `task_agent` only (not available to plan sub-agents).
 
@@ -225,16 +231,20 @@ Write content to a file, creating it if needed.
 
 ### edit_file
 
-Replace an exact string in a file with new content.
+Replace exact strings in a file, or apply multiple replacements atomically.
 
 | Parameter    | Type    | Required | Description |
 |--------------|---------|----------|-------------|
 | `path`       | string  | yes      | Absolute or relative file path. |
-| `old_string` | string  | yes      | The exact text to find and replace. |
-| `new_string` | string  | yes      | The replacement text. |
+| `old_string` | string  | no*      | The exact text to find and replace. |
+| `new_string` | string  | no*      | The replacement text. |
 | `near_line`  | integer | no       | Disambiguate when `old_string` matches multiple locations. |
+| `edits`      | array   | no*      | Multiple replacements to apply atomically (see below). |
+
+\* Provide either `old_string`+`new_string` (single edit) or `edits` array (batch), not both.
 
 - **What it does**: Finds `old_string` in the file and replaces it with `new_string`. Fails if the string is not found or matches multiple locations (unless `near_line` is provided to pick the nearest match). Requires a prior `read_file` call on the same path.
+- **Batch mode**: The `edits` array accepts multiple `{old_string, new_string, near_line?}` entries applied atomically. All edits are validated before any are applied. Overlapping edits (two entries targeting the same text region) are rejected. Edits are applied in reverse file-position order so character offsets stay stable.
 - **Auto-approve**: No -- requires user confirmation.
 - **Agent availability**: `task_agent` only.
 
@@ -265,8 +275,9 @@ Execute Python code for math and computation in a sandbox.
 |-----------|--------|----------|-------------|
 | `code`    | string | yes      | Python code to execute. Must use `print()` for output. |
 
-- **What it does**: Runs Python code in a sandboxed environment with pre-imported libraries: `sympy`, `numpy`, `scipy`, `math`, `fractions`, `itertools`, `functools`, `collections`, `decimal`, `operator`, `random`, `re`, `string`. Common sympy names (`symbols`, `solve`, `simplify`, `sqrt`, `Matrix`, etc.) are pre-imported.
-- **Auto-approve**: No -- requires user confirmation.
+- **What it does**: Runs Python code in a sandboxed environment with pre-imported libraries: `sympy`, `numpy`, `scipy`, `math`, `fractions`, `itertools`, `functools`, `collections`, `decimal`, `operator`, `random`, `re`, `string`. Common sympy names (`symbols`, `solve`, `simplify`, `sqrt`, `Matrix`, etc.) are pre-imported. `pytest` is also available for import.
+- **Installation**: `sympy`, `numpy`, `scipy`, and `pytest` require the `[sandbox]` extras group: `pip install turnstone[sandbox]` (included in `[all]`).
+- **Auto-approve**: Yes.
 - **Agent availability**: `agent` and `task_agent`.
 
 ---

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "turnstone"
-version = "0.8.5"
+version = "0.9.3"
 description = "Multi-node AI orchestration platform with tool use, agent routing, and cluster simulation."
 readme = "README.md"
 license = "BUSL-1.1"
@@ -51,9 +51,11 @@ console = ["redis>=7.2", "croniter>=3.0"]
 sim = ["redis>=7.2"]
 anthropic = ["anthropic>=0.39"]
 postgres = ["psycopg[binary]>=3.2"]
-ddg = ["duckduckgo-search>=8.0"]
+ddg = ["ddgs>=9.0"]
 discord = ["discord.py>=2.4", "redis>=7.2"]
-all = ["turnstone[mq,console,sim,anthropic,postgres,discord,ddg]"]
+tls = ["lacme>=1.0.4"]
+sandbox = ["sympy>=1.13", "numpy>=2.0", "scipy>=1.14", "pytest>=9.0"]
+all = ["turnstone[mq,console,sim,anthropic,postgres,discord,ddg,tls,sandbox]"]
 
 [project.scripts]
 turnstone = "turnstone.cli:main"
@@ -78,7 +80,7 @@ include = [
   "turnstone/console/static/*.js",
   "turnstone/shared_static/*.css",
   "turnstone/shared_static/*.js",
-  "turnstone/shared_static/katex-0.16.40/**/*",
+  "turnstone/shared_static/katex-0.16.44/**/*",
   "turnstone/shared_static/hljs-11.11.1/**/*",
   "turnstone/shared_static/mermaid-11.13.0/**/*",
   "turnstone/sdk/py.typed",
@@ -165,6 +167,14 @@ ignore_missing_imports = true
 module = ["frontmatter", "frontmatter.*"]
 ignore_missing_imports = true
 
+[[tool.mypy.overrides]]
+module = ["ddgs", "ddgs.*"]
+ignore_missing_imports = true
+
+[[tool.mypy.overrides]]
+module = ["lacme", "lacme.*"]
+ignore_missing_imports = true
+
 [[tool.mypy.overrides]]
 module = ["turnstone.channels.discord.*"]
 disallow_subclassing_any = false

sdk/typescript/openapi-console.json

@@ -2,7 +2,7 @@
   "openapi": "3.1.0",
   "info": {
     "title": "turnstone Console API",
-    "version": "0.8.4",
+    "version": "0.9.2",
     "description": "Cluster-wide visibility and control across all turnstone nodes."
   },
   "paths": {
@@ -2103,6 +2103,27 @@
         }
       }
     },
+    "/v1/api/models": {
+      "get": {
+        "summary": "List enabled model aliases for workstream creation",
+        "operationId": "v1_api_models_get",
+        "tags": [
+          "Models"
+        ],
+        "responses": {
+          "200": {
+            "description": "Success",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ListAvailableModelsResponse"
+                }
+              }
+            }
+          }
+        }
+      }
+    },
     "/v1/api/skills": {
       "get": {
         "summary": "List available skills (summary)",
@@ -3002,7 +3023,7 @@
             "content": {
               "application/json": {
                 "schema": {
-                  "$ref": "#/components/schemas/StatusResponse"
+                  "$ref": "#/components/schemas/DeleteSettingResponse"
                 }
               }
             }
@@ -3453,6 +3474,473 @@
         }
       }
     },
+    "/v1/api/admin/model-definitions": {
+      "get": {
+        "summary": "List model definitions with live status from cluster nodes",
+        "operationId": "v1_api_admin_model-definitions_get",
+        "tags": [
+          "Admin"
+        ],
+        "responses": {
+          "200": {
+            "description": "Success",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ListModelDefinitionsResponse"
+                }
+              }
+            }
+          }
+        }
+      },
+      "post": {
+        "summary": "Create a model definition",
+        "operationId": "v1_api_admin_model-definitions_post",
+        "tags": [
+          "Admin"
+        ],
+        "requestBody": {
+          "required": true,
+          "content": {
+            "application/json": {
+              "schema": {
+                "$ref": "#/components/schemas/CreateModelDefinitionRequest"
+              }
+            }
+          }
+        },
+        "responses": {
+          "200": {
+            "description": "Success",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ModelDefinitionInfo"
+                }
+              }
+            }
+          },
+          "400": {
+            "description": "Error 400",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ErrorResponse"
+                }
+              }
+            }
+          },
+          "409": {
+            "description": "Error 409",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ErrorResponse"
+                }
+              }
+            }
+          }
+        }
+      }
+    },
+    "/v1/api/admin/model-definitions/reload": {
+      "post": {
+        "summary": "Tell all nodes to re-read model definitions from DB and rebuild registry",
+        "operationId": "v1_api_admin_model-definitions_reload_post",
+        "tags": [
+          "Admin"
+        ],
+        "responses": {
+          "200": {
+            "description": "Success",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ModelReloadResponse"
+                }
+              }
+            }
+          }
+        }
+      }
+    },
+    "/v1/api/admin/model-definitions/{definition_id}": {
+      "get": {
+        "summary": "Get a single model definition",
+        "operationId": "v1_api_admin_model-definitions_{definition_id}_get",
+        "tags": [
+          "Admin"
+        ],
+        "parameters": [
+          {
+            "name": "definition_id",
+            "in": "path",
+            "required": true,
+            "schema": {
+              "type": "string"
+            }
+          }
+        ],
+        "responses": {
+          "200": {
+            "description": "Success",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ModelDefinitionInfo"
+                }
+              }
+            }
+          },
+          "404": {
+            "description": "Error 404",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ErrorResponse"
+                }
+              }
+            }
+          }
+        }
+      },
+      "put": {
+        "summary": "Update a model definition",
+        "operationId": "v1_api_admin_model-definitions_{definition_id}_put",
+        "tags": [
+          "Admin"
+        ],
+        "parameters": [
+          {
+            "name": "definition_id",
+            "in": "path",
+            "required": true,
+            "schema": {
+              "type": "string"
+            }
+          }
+        ],
+        "requestBody": {
+          "required": true,
+          "content": {
+            "application/json": {
+              "schema": {
+                "$ref": "#/components/schemas/UpdateModelDefinitionRequest"
+              }
+            }
+          }
+        },
+        "responses": {
+          "200": {
+            "description": "Success",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ModelDefinitionInfo"
+                }
+              }
+            }
+          },
+          "400": {
+            "description": "Error 400",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ErrorResponse"
+                }
+              }
+            }
+          },
+          "404": {
+            "description": "Error 404",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ErrorResponse"
+                }
+              }
+            }
+          },
+          "409": {
+            "description": "Error 409",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ErrorResponse"
+                }
+              }
+            }
+          }
+        }
+      },
+      "delete": {
+        "summary": "Delete a model definition",
+        "operationId": "v1_api_admin_model-definitions_{definition_id}_delete",
+        "tags": [
+          "Admin"
+        ],
+        "parameters": [
+          {
+            "name": "definition_id",
+            "in": "path",
+            "required": true,
+            "schema": {
+              "type": "string"
+            }
+          }
+        ],
+        "responses": {
+          "200": {
+            "description": "Success"
+          },
+          "404": {
+            "description": "Error 404",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ErrorResponse"
+                }
+              }
+            }
+          }
+        }
+      }
+    },
+    "/v1/api/admin/model-definitions/detect": {
+      "post": {
+        "summary": "Probe a model endpoint: verify reachability, list models, detect context window and server type",
+        "operationId": "v1_api_admin_model-definitions_detect_post",
+        "tags": [
+          "Admin"
+        ],
+        "requestBody": {
+          "required": true,
+          "content": {
+            "application/json": {
+              "schema": {
+                "$ref": "#/components/schemas/DetectModelRequest"
+              }
+            }
+          }
+        },
+        "responses": {
+          "200": {
+            "description": "Success",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/DetectModelResponse"
+                }
+              }
+            }
+          },
+          "400": {
+            "description": "Error 400",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ErrorResponse"
+                }
+              }
+            }
+          }
+        }
+      }
+    },
+    "/v1/api/admin/model-capabilities": {
+      "get": {
+        "summary": "Look up static capabilities for a known model",
+        "operationId": "v1_api_admin_model-capabilities_get",
+        "tags": [
+          "Admin"
+        ],
+        "parameters": [
+          {
+            "name": "provider",
+            "in": "query",
+            "required": true,
+            "schema": {
+              "type": "string"
+            },
+            "description": "Provider name"
+          },
+          {
+            "name": "model",
+            "in": "query",
+            "required": true,
+            "schema": {
+              "type": "string"
+            },
+            "description": "Model ID to look up"
+          }
+        ],
+        "responses": {
+          "200": {
+            "description": "Success",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ModelCapabilitiesResponse"
+                }
+              }
+            }
+          }
+        }
+      }
+    },
+    "/v1/api/admin/model-capabilities/known": {
+      "get": {
+        "summary": "List known model name prefixes for a provider",
+        "operationId": "v1_api_admin_model-capabilities_known_get",
+        "tags": [
+          "Admin"
+        ],
+        "parameters": [
+          {
+            "name": "provider",
+            "in": "query",
+            "required": true,
+            "schema": {
+              "type": "string"
+            },
+            "description": "Provider name"
+          }
+        ],
+        "responses": {
+          "200": {
+            "description": "Success",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/KnownModelsResponse"
+                }
+              }
+            }
+          }
+        }
+      }
+    },
+    "/v1/api/admin/tls/ca": {
+      "get": {
+        "summary": "CA status: initialization state, CN, cert count, cert inventory",
+        "operationId": "v1_api_admin_tls_ca_get",
+        "tags": [
+          "Admin"
+        ],
+        "responses": {
+          "200": {
+            "description": "Success"
+          }
+        }
+      }
+    },
+    "/v1/api/admin/tls/ca.pem": {
+      "get": {
+        "summary": "Download CA root certificate (PEM format)",
+        "operationId": "v1_api_admin_tls_ca.pem_get",
+        "tags": [
+          "Admin"
+        ],
+        "responses": {
+          "200": {
+            "description": "Success"
+          }
+        }
+      }
+    },
+    "/v1/api/admin/tls/certs": {
+      "get": {
+        "summary": "List all issued TLS certificates",
+        "operationId": "v1_api_admin_tls_certs_get",
+        "tags": [
+          "Admin"
+        ],
+        "responses": {
+          "200": {
+            "description": "Success"
+          }
+        }
+      }
+    },
+    "/v1/api/admin/tls/certs/{domain}/renew": {
+      "post": {
+        "summary": "Force-renew a certificate by domain",
+        "operationId": "v1_api_admin_tls_certs_{domain}_renew_post",
+        "tags": [
+          "Admin"
+        ],
+        "parameters": [
+          {
+            "name": "domain",
+            "in": "path",
+            "required": true,
+            "schema": {
+              "type": "string"
+            }
+          }
+        ],
+        "responses": {
+          "200": {
+            "description": "Success"
+          },
+          "404": {
+            "description": "Error 404",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ErrorResponse"
+                }
+              }
+            }
+          },
+          "500": {
+            "description": "Error 500",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ErrorResponse"
+                }
+              }
+            }
+          }
+        }
+      }
+    },
+    "/v1/api/admin/tls/certs/{domain}": {
+      "delete": {
+        "summary": "Delete a certificate by domain",
+        "operationId": "v1_api_admin_tls_certs_{domain}_delete",
+        "tags": [
+          "Admin"
+        ],
+        "parameters": [
+          {
+            "name": "domain",
+            "in": "path",
+            "required": true,
+            "schema": {
+              "type": "string"
+            }
+          }
+        ],
+        "responses": {
+          "200": {
+            "description": "Success"
+          },
+          "404": {
+            "description": "Error 404",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ErrorResponse"
+                }
+              }
+            }
+          }
+        }
+      }
+    },
     "/health": {
       "get": {
         "summary": "Console health check",
@@ -3507,6 +3995,34 @@
         "title": "StatusResponse",
         "type": "object"
       },
+      "DeleteSettingResponse": {
+        "description": "DELETE /v1/api/admin/settings/{key} response.",
+        "properties": {
+          "status": {
+            "default": "ok",
+            "examples": [
+              "ok"
+            ],
+            "title": "Status",
+            "type": "string"
+          },
+          "key": {
+            "description": "Dotted setting key that was reset",
+            "title": "Key",
+            "type": "string"
+          },
+          "default": {
+            "description": "Registry default value the setting reverted to",
+            "title": "Default"
+          }
+        },
+        "required": [
+          "key",
+          "default"
+        ],
+        "title": "DeleteSettingResponse",
+        "type": "object"
+      },
       "AuthLoginRequest": {
         "description": "POST /v1/api/auth/login request body.\n\nEither username+password or token must be provided.",
         "properties": {
@@ -6360,6 +6876,443 @@
         "title": "McpReloadResponse",
         "type": "object"
       },
+      "ModelDefinitionInfo": {
+        "properties": {
+          "definition_id": {
+            "title": "Definition Id",
+            "type": "string"
+          },
+          "alias": {
+            "title": "Alias",
+            "type": "string"
+          },
+          "model": {
+            "title": "Model",
+            "type": "string"
+          },
+          "provider": {
+            "default": "openai",
+            "title": "Provider",
+            "type": "string"
+          },
+          "base_url": {
+            "default": "",
+            "title": "Base Url",
+            "type": "string"
+          },
+          "api_key": {
+            "default": "",
+            "title": "Api Key",
+            "type": "string"
+          },
+          "context_window": {
+            "default": 32768,
+            "title": "Context Window",
+            "type": "integer"
+          },
+          "capabilities": {
+            "default": "{}",
+            "title": "Capabilities",
+            "type": "string"
+          },
+          "enabled": {
+            "default": true,
+            "title": "Enabled",
+            "type": "boolean"
+          },
+          "source": {
+            "default": "",
+            "title": "Source",
+            "type": "string"
+          },
+          "created_by": {
+            "default": "",
+            "title": "Created By",
+            "type": "string"
+          },
+          "created": {
+            "default": "",
+            "title": "Created",
+            "type": "string"
+          },
+          "updated": {
+            "default": "",
+            "title": "Updated",
+            "type": "string"
+          }
+        },
+        "required": [
+          "definition_id",
+          "alias",
+          "model"
+        ],
+        "title": "ModelDefinitionInfo",
+        "type": "object"
+      },
+      "CreateModelDefinitionRequest": {
+        "properties": {
+          "alias": {
+            "title": "Alias",
+            "type": "string"
+          },
+          "model": {
+            "title": "Model",
+            "type": "string"
+          },
+          "provider": {
+            "default": "openai",
+            "title": "Provider",
+            "type": "string"
+          },
+          "base_url": {
+            "default": "",
+            "title": "Base Url",
+            "type": "string"
+          },
+          "api_key": {
+            "default": "",
+            "title": "Api Key",
+            "type": "string"
+          },
+          "context_window": {
+            "default": 32768,
+            "title": "Context Window",
+            "type": "integer"
+          },
+          "capabilities": {
+            "additionalProperties": true,
+            "title": "Capabilities",
+            "type": "object"
+          },
+          "enabled": {
+            "default": true,
+            "title": "Enabled",
+            "type": "boolean"
+          }
+        },
+        "required": [
+          "alias",
+          "model"
+        ],
+        "title": "CreateModelDefinitionRequest",
+        "type": "object"
+      },
+      "UpdateModelDefinitionRequest": {
+        "properties": {
+          "alias": {
+            "anyOf": [
+              {
+                "type": "string"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "default": null,
+            "title": "Alias"
+          },
+          "model": {
+            "anyOf": [
+              {
+                "type": "string"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "default": null,
+            "title": "Model"
+          },
+          "provider": {
+            "anyOf": [
+              {
+                "type": "string"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "default": null,
+            "title": "Provider"
+          },
+          "base_url": {
+            "anyOf": [
+              {
+                "type": "string"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "default": null,
+            "title": "Base Url"
+          },
+          "api_key": {
+            "anyOf": [
+              {
+                "type": "string"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "default": null,
+            "title": "Api Key"
+          },
+          "context_window": {
+            "anyOf": [
+              {
+                "type": "integer"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "default": null,
+            "title": "Context Window"
+          },
+          "capabilities": {
+            "anyOf": [
+              {
+                "additionalProperties": true,
+                "type": "object"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "default": null,
+            "title": "Capabilities"
+          },
+          "enabled": {
+            "anyOf": [
+              {
+                "type": "boolean"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "default": null,
+            "title": "Enabled"
+          }
+        },
+        "title": "UpdateModelDefinitionRequest",
+        "type": "object"
+      },
+      "ListModelDefinitionsResponse": {
+        "properties": {
+          "models": {
+            "items": {
+              "$ref": "#/components/schemas/ModelDefinitionInfo"
+            },
+            "title": "Models",
+            "type": "array"
+          }
+        },
+        "required": [
+          "models"
+        ],
+        "title": "ListModelDefinitionsResponse",
+        "type": "object"
+      },
+      "ModelReloadResponse": {
+        "properties": {
+          "status": {
+            "default": "ok",
+            "title": "Status",
+            "type": "string"
+          },
+          "results": {
+            "additionalProperties": true,
+            "title": "Results",
+            "type": "object"
+          }
+        },
+        "title": "ModelReloadResponse",
+        "type": "object"
+      },
+      "DetectModelRequest": {
+        "properties": {
+          "provider": {
+            "default": "openai",
+            "title": "Provider",
+            "type": "string"
+          },
+          "base_url": {
+            "default": "",
+            "title": "Base Url",
+            "type": "string"
+          },
+          "api_key": {
+            "default": "",
+            "title": "Api Key",
+            "type": "string"
+          },
+          "model": {
+            "default": "",
+            "title": "Model",
+            "type": "string"
+          },
+          "definition_id": {
+            "default": "",
+            "title": "Definition Id",
+            "type": "string"
+          }
+        },
+        "title": "DetectModelRequest",
+        "type": "object"
+      },
+      "DetectModelResponse": {
+        "properties": {
+          "reachable": {
+            "default": false,
+            "title": "Reachable",
+            "type": "boolean"
+          },
+          "model_found": {
+            "anyOf": [
+              {
+                "type": "boolean"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "default": null,
+            "title": "Model Found"
+          },
+          "available_models": {
+            "items": {
+              "type": "string"
+            },
+            "title": "Available Models",
+            "type": "array"
+          },
+          "context_window": {
+            "anyOf": [
+              {
+                "type": "integer"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "default": null,
+            "title": "Context Window"
+          },
+          "server_type": {
+            "anyOf": [
+              {
+                "type": "string"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "default": null,
+            "title": "Server Type"
+          },
+          "error": {
+            "anyOf": [
+              {
+                "type": "string"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "default": null,
+            "title": "Error"
+          }
+        },
+        "title": "DetectModelResponse",
+        "type": "object"
+      },
+      "ModelCapabilitiesResponse": {
+        "properties": {
+          "model": {
+            "title": "Model",
+            "type": "string"
+          },
+          "provider": {
+            "title": "Provider",
+            "type": "string"
+          },
+          "known": {
+            "default": false,
+            "title": "Known",
+            "type": "boolean"
+          },
+          "capabilities": {
+            "additionalProperties": true,
+            "title": "Capabilities",
+            "type": "object"
+          }
+        },
+        "required": [
+          "model",
+          "provider"
+        ],
+        "title": "ModelCapabilitiesResponse",
+        "type": "object"
+      },
+      "KnownModelsResponse": {
+        "properties": {
+          "provider": {
+            "title": "Provider",
+            "type": "string"
+          },
+          "models": {
+            "items": {
+              "type": "string"
+            },
+            "title": "Models",
+            "type": "array"
+          }
+        },
+        "required": [
+          "provider"
+        ],
+        "title": "KnownModelsResponse",
+        "type": "object"
+      },
+      "AvailableModelInfo": {
+        "properties": {
+          "alias": {
+            "title": "Alias",
+            "type": "string"
+          },
+          "model": {
+            "title": "Model",
+            "type": "string"
+          },
+          "provider": {
+            "title": "Provider",
+            "type": "string"
+          }
+        },
+        "required": [
+          "alias",
+          "model",
+          "provider"
+        ],
+        "title": "AvailableModelInfo",
+        "type": "object"
+      },
+      "ListAvailableModelsResponse": {
+        "properties": {
+          "models": {
+            "items": {
+              "$ref": "#/components/schemas/AvailableModelInfo"
+            },
+            "title": "Models",
+            "type": "array"
+          }
+        },
+        "title": "ListAvailableModelsResponse",
+        "type": "object"
+      },
       "RegistrySearchResponse": {
         "properties": {
           "servers": {
@@ -7630,4 +8583,4 @@
       }
     }
   }
-}
+}

sdk/typescript/openapi-server.json

@@ -2,7 +2,7 @@
   "openapi": "3.1.0",
   "info": {
     "title": "turnstone Server API",
-    "version": "0.8.4",
+    "version": "0.9.2",
     "description": "Single-node workstream management, chat interaction, and real-time streaming."
   },
   "paths": {
@@ -458,6 +458,27 @@
         }
       }
     },
+    "/v1/api/models": {
+      "get": {
+        "summary": "List available model aliases",
+        "operationId": "v1_api_models_get",
+        "tags": [
+          "Models"
+        ],
+        "responses": {
+          "200": {
+            "description": "Success",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/ListAvailableModelsResponse"
+                }
+              }
+            }
+          }
+        }
+      }
+    },
     "/v1/api/auth/login": {
       "post": {
         "summary": "Authenticate with a token",
@@ -1223,6 +1244,12 @@
             "description": "Target workstream ID",
             "title": "Ws Id",
             "type": "string"
+          },
+          "force": {
+            "default": false,
+            "description": "Force cancel: abandon the stuck worker thread immediately. Use when cooperative cancel has not resolved within a few seconds.",
+            "title": "Force",
+            "type": "boolean"
           }
         },
         "required": [
@@ -1977,7 +2004,43 @@
         ],
         "title": "ListSkillSummaryResponse",
         "type": "object"
+      },
+      "AvailableModelInfo": {
+        "properties": {
+          "alias": {
+            "title": "Alias",
+            "type": "string"
+          },
+          "model": {
+            "title": "Model",
+            "type": "string"
+          },
+          "provider": {
+            "title": "Provider",
+            "type": "string"
+          }
+        },
+        "required": [
+          "alias",
+          "model",
+          "provider"
+        ],
+        "title": "AvailableModelInfo",
+        "type": "object"
+      },
+      "ListAvailableModelsResponse": {
+        "properties": {
+          "models": {
+            "items": {
+              "$ref": "#/components/schemas/AvailableModelInfo"
+            },
+            "title": "Models",
+            "type": "array"
+          }
+        },
+        "title": "ListAvailableModelsResponse",
+        "type": "object"
       }
     }
   }
-}
+}

sdk/typescript/package-lock.json

@@ -20,6 +20,7 @@
       "dev": true,
       "license": "MIT",
       "optional": true,
+      "peer": true,
       "dependencies": {
         "@emnapi/wasi-threads": "1.2.0",
         "tslib": "^2.4.0"
@@ -32,6 +33,7 @@
       "dev": true,
       "license": "MIT",
       "optional": true,
+      "peer": true,
       "dependencies": {
         "tslib": "^2.4.0"
       }
@@ -43,6 +45,7 @@
       "dev": true,
       "license": "MIT",
       "optional": true,
+      "peer": true,
       "dependencies": {
         "tslib": "^2.4.0"
       }
@@ -55,26 +58,28 @@
       "license": "MIT"
     },
     "node_modules/@napi-rs/wasm-runtime": {
-      "version": "1.1.1",
-      "resolved": "https://registry.npmjs.org/@napi-rs/wasm-runtime/-/wasm-runtime-1.1.1.tgz",
-      "integrity": "sha512-p64ah1M1ld8xjWv3qbvFwHiFVWrq1yFvV4f7w+mzaqiR4IlSgkqhcRdHwsGgomwzBH51sRY4NEowLxnaBjcW/A==",
+      "version": "1.1.2",
+      "resolved": "https://registry.npmjs.org/@napi-rs/wasm-runtime/-/wasm-runtime-1.1.2.tgz",
+      "integrity": "sha512-sNXv5oLJ7ob93xkZ1XnxisYhGYXfaG9f65/ZgYuAu3qt7b3NadcOEhLvx28hv31PgX8SZJRYrAIPQilQmFpLVw==",
       "dev": true,
       "license": "MIT",
       "optional": true,
       "dependencies": {
-        "@emnapi/core": "^1.7.1",
-        "@emnapi/runtime": "^1.7.1",
         "@tybys/wasm-util": "^0.10.1"
       },
       "funding": {
         "type": "github",
         "url": "https://github.com/sponsors/Brooooooklyn"
+      },
+      "peerDependencies": {
+        "@emnapi/core": "^1.7.1",
+        "@emnapi/runtime": "^1.7.1"
       }
     },
     "node_modules/@oxc-project/types": {
-      "version": "0.120.0",
-      "resolved": "https://registry.npmjs.org/@oxc-project/types/-/types-0.120.0.tgz",
-      "integrity": "sha512-k1YNu55DuvAip/MGE1FTsIuU3FUCn6v/ujG9V7Nq5Df/kX2CWb13hhwD0lmJGMGqE+bE1MXvv9SZVnMzEXlWcg==",
+      "version": "0.122.0",
+      "resolved": "https://registry.npmjs.org/@oxc-project/types/-/types-0.122.0.tgz",
+      "integrity": "sha512-oLAl5kBpV4w69UtFZ9xqcmTi+GENWOcPF7FCrczTiBbmC0ibXxCwyvZGbO39rCVEuLGAZM84DH0pUIyyv/YJzA==",
       "dev": true,
       "license": "MIT",
       "funding": {
@@ -82,9 +87,9 @@
       }
     },
     "node_modules/@rolldown/binding-android-arm64": {
-      "version": "1.0.0-rc.10",
-      "resolved": "https://registry.npmjs.org/@rolldown/binding-android-arm64/-/binding-android-arm64-1.0.0-rc.10.tgz",
-      "integrity": "sha512-jOHxwXhxmFKuXztiu1ORieJeTbx5vrTkcOkkkn2d35726+iwhrY1w/+nYY/AGgF12thg33qC3R1LMBF5tHTZHg==",
+      "version": "1.0.0-rc.12",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-android-arm64/-/binding-android-arm64-1.0.0-rc.12.tgz",
+      "integrity": "sha512-pv1y2Fv0JybcykuiiD3qBOBdz6RteYojRFY1d+b95WVuzx211CRh+ytI/+9iVyWQ6koTh5dawe4S/yRfOFjgaA==",
       "cpu": [
         "arm64"
       ],
@@ -99,9 +104,9 @@
       }
     },
     "node_modules/@rolldown/binding-darwin-arm64": {
-      "version": "1.0.0-rc.10",
-      "resolved": "https://registry.npmjs.org/@rolldown/binding-darwin-arm64/-/binding-darwin-arm64-1.0.0-rc.10.tgz",
-      "integrity": "sha512-gED05Teg/vtTZbIJBc4VNMAxAFDUPkuO/rAIyyxZjTj1a1/s6z5TII/5yMGZ0uLRCifEtwUQn8OlYzuYc0m70w==",
+      "version": "1.0.0-rc.12",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-darwin-arm64/-/binding-darwin-arm64-1.0.0-rc.12.tgz",
+      "integrity": "sha512-cFYr6zTG/3PXXF3pUO+umXxt1wkRK/0AYT8lDwuqvRC+LuKYWSAQAQZjCWDQpAH172ZV6ieYrNnFzVVcnSflAg==",
       "cpu": [
         "arm64"
       ],
@@ -116,9 +121,9 @@
       }
     },
     "node_modules/@rolldown/binding-darwin-x64": {
-      "version": "1.0.0-rc.10",
-      "resolved": "https://registry.npmjs.org/@rolldown/binding-darwin-x64/-/binding-darwin-x64-1.0.0-rc.10.tgz",
-      "integrity": "sha512-rI15NcM1mA48lqrIxVkHfAqcyFLcQwyXWThy+BQ5+mkKKPvSO26ir+ZDp36AgYoYVkqvMcdS8zOE6SeBsR9e8A==",
+      "version": "1.0.0-rc.12",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-darwin-x64/-/binding-darwin-x64-1.0.0-rc.12.tgz",
+      "integrity": "sha512-ZCsYknnHzeXYps0lGBz8JrF37GpE9bFVefrlmDrAQhOEi4IOIlcoU1+FwHEtyXGx2VkYAvhu7dyBf75EJQffBw==",
       "cpu": [
         "x64"
       ],
@@ -133,9 +138,9 @@
       }
     },
     "node_modules/@rolldown/binding-freebsd-x64": {
-      "version": "1.0.0-rc.10",
-      "resolved": "https://registry.npmjs.org/@rolldown/binding-freebsd-x64/-/binding-freebsd-x64-1.0.0-rc.10.tgz",
-      "integrity": "sha512-XZRXHdTa+4ME1MuDVp021+doQ+z6Ei4CCFmNc5/sKbqb8YmkiJdj8QKlV3rCI0AJtAeSB5n0WGPuJWNL9p/L2w==",
+      "version": "1.0.0-rc.12",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-freebsd-x64/-/binding-freebsd-x64-1.0.0-rc.12.tgz",
+      "integrity": "sha512-dMLeprcVsyJsKolRXyoTH3NL6qtsT0Y2xeuEA8WQJquWFXkEC4bcu1rLZZSnZRMtAqwtrF/Ib9Ddtpa/Gkge9Q==",
       "cpu": [
         "x64"
       ],
@@ -150,9 +155,9 @@
       }
     },
     "node_modules/@rolldown/binding-linux-arm-gnueabihf": {
-      "version": "1.0.0-rc.10",
-      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-arm-gnueabihf/-/binding-linux-arm-gnueabihf-1.0.0-rc.10.tgz",
-      "integrity": "sha512-R0SQMRluISSLzFE20sPWYHVmJdDQnRyc/FzSCN72BqQmh2SOZUFG+N3/vBZpR4C6WpEUVYJLrYUXaj43sJsNLA==",
+      "version": "1.0.0-rc.12",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-arm-gnueabihf/-/binding-linux-arm-gnueabihf-1.0.0-rc.12.tgz",
+      "integrity": "sha512-YqWjAgGC/9M1lz3GR1r1rP79nMgo3mQiiA+Hfo+pvKFK1fAJ1bCi0ZQVh8noOqNacuY1qIcfyVfP6HoyBRZ85Q==",
       "cpu": [
         "arm"
       ],
@@ -167,13 +172,16 @@
       }
     },
     "node_modules/@rolldown/binding-linux-arm64-gnu": {
-      "version": "1.0.0-rc.10",
-      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-arm64-gnu/-/binding-linux-arm64-gnu-1.0.0-rc.10.tgz",
-      "integrity": "sha512-Y1reMrV/o+cwpduYhJuOE3OMKx32RMYCidf14y+HssARRmhDuWXJ4yVguDg2R/8SyyGNo+auzz64LnPK9Hq6jg==",
+      "version": "1.0.0-rc.12",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-arm64-gnu/-/binding-linux-arm64-gnu-1.0.0-rc.12.tgz",
+      "integrity": "sha512-/I5AS4cIroLpslsmzXfwbe5OmWvSsrFuEw3mwvbQ1kDxJ822hFHIx+vsN/TAzNVyepI/j/GSzrtCIwQPeKCLIg==",
       "cpu": [
         "arm64"
       ],
       "dev": true,
+      "libc": [
+        "glibc"
+      ],
       "license": "MIT",
       "optional": true,
       "os": [
@@ -184,13 +192,16 @@
       }
     },
     "node_modules/@rolldown/binding-linux-arm64-musl": {
-      "version": "1.0.0-rc.10",
-      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-arm64-musl/-/binding-linux-arm64-musl-1.0.0-rc.10.tgz",
-      "integrity": "sha512-vELN+HNb2IzuzSBUOD4NHmP9yrGwl1DVM29wlQvx1OLSclL0NgVWnVDKl/8tEks79EFek/kebQKnNJkIAA4W2g==",
+      "version": "1.0.0-rc.12",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-arm64-musl/-/binding-linux-arm64-musl-1.0.0-rc.12.tgz",
+      "integrity": "sha512-V6/wZztnBqlx5hJQqNWwFdxIKN0m38p8Jas+VoSfgH54HSj9tKTt1dZvG6JRHcjh6D7TvrJPWFGaY9UBVOaWPw==",
       "cpu": [
         "arm64"
       ],
       "dev": true,
+      "libc": [
+        "musl"
+      ],
       "license": "MIT",
       "optional": true,
       "os": [
@@ -201,13 +212,16 @@
       }
     },
     "node_modules/@rolldown/binding-linux-ppc64-gnu": {
-      "version": "1.0.0-rc.10",
-      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-ppc64-gnu/-/binding-linux-ppc64-gnu-1.0.0-rc.10.tgz",
-      "integrity": "sha512-ZqrufYTgzxbHwpqOjzSsb0UV/aV2TFIY5rP8HdsiPTv/CuAgCRjM6s9cYFwQ4CNH+hf9Y4erHW1GjZuZ7WoI7w==",
+      "version": "1.0.0-rc.12",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-ppc64-gnu/-/binding-linux-ppc64-gnu-1.0.0-rc.12.tgz",
+      "integrity": "sha512-AP3E9BpcUYliZCxa3w5Kwj9OtEVDYK6sVoUzy4vTOJsjPOgdaJZKFmN4oOlX0Wp0RPV2ETfmIra9x1xuayFB7g==",
       "cpu": [
         "ppc64"
       ],
       "dev": true,
+      "libc": [
+        "glibc"
+      ],
       "license": "MIT",
       "optional": true,
       "os": [
@@ -218,13 +232,16 @@
       }
     },
     "node_modules/@rolldown/binding-linux-s390x-gnu": {
-      "version": "1.0.0-rc.10",
-      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-s390x-gnu/-/binding-linux-s390x-gnu-1.0.0-rc.10.tgz",
-      "integrity": "sha512-gSlmVS1FZJSRicA6IyjoRoKAFK7IIHBs7xJuHRSmjImqk3mPPWbR7RhbnfH2G6bcmMEllCt2vQ/7u9e6bBnByg==",
+      "version": "1.0.0-rc.12",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-s390x-gnu/-/binding-linux-s390x-gnu-1.0.0-rc.12.tgz",
+      "integrity": "sha512-nWwpvUSPkoFmZo0kQazZYOrT7J5DGOJ/+QHHzjvNlooDZED8oH82Yg67HvehPPLAg5fUff7TfWFHQS8IV1n3og==",
       "cpu": [
         "s390x"
       ],
       "dev": true,
+      "libc": [
+        "glibc"
+      ],
       "license": "MIT",
       "optional": true,
       "os": [
@@ -235,13 +252,16 @@
       }
     },
     "node_modules/@rolldown/binding-linux-x64-gnu": {
-      "version": "1.0.0-rc.10",
-      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-x64-gnu/-/binding-linux-x64-gnu-1.0.0-rc.10.tgz",
-      "integrity": "sha512-eOCKUpluKgfObT2pHjztnaWEIbUabWzk3qPZ5PuacuPmr4+JtQG4k2vGTY0H15edaTnicgU428XW/IH6AimcQw==",
+      "version": "1.0.0-rc.12",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-x64-gnu/-/binding-linux-x64-gnu-1.0.0-rc.12.tgz",
+      "integrity": "sha512-RNrafz5bcwRy+O9e6P8Z/OCAJW/A+qtBczIqVYwTs14pf4iV1/+eKEjdOUta93q2TsT/FI0XYDP3TCky38LMAg==",
       "cpu": [
         "x64"
       ],
       "dev": true,
+      "libc": [
+        "glibc"
+      ],
       "license": "MIT",
       "optional": true,
       "os": [
@@ -252,13 +272,16 @@
       }
     },
     "node_modules/@rolldown/binding-linux-x64-musl": {
-      "version": "1.0.0-rc.10",
-      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-x64-musl/-/binding-linux-x64-musl-1.0.0-rc.10.tgz",
-      "integrity": "sha512-Xdf2jQbfQowJnLcgYfD/m0Uu0Qj5OdxKallD78/IPPfzaiaI4KRAwZzHcKQ4ig1gtg1SuzC7jovNiM2TzQsBXA==",
+      "version": "1.0.0-rc.12",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-x64-musl/-/binding-linux-x64-musl-1.0.0-rc.12.tgz",
+      "integrity": "sha512-Jpw/0iwoKWx3LJ2rc1yjFrj+T7iHZn2JDg1Yny1ma0luviFS4mhAIcd1LFNxK3EYu3DHWCps0ydXQ5i/rrJ2ig==",
       "cpu": [
         "x64"
       ],
       "dev": true,
+      "libc": [
+        "musl"
+      ],
       "license": "MIT",
       "optional": true,
       "os": [
@@ -269,9 +292,9 @@
       }
     },
     "node_modules/@rolldown/binding-openharmony-arm64": {
-      "version": "1.0.0-rc.10",
-      "resolved": "https://registry.npmjs.org/@rolldown/binding-openharmony-arm64/-/binding-openharmony-arm64-1.0.0-rc.10.tgz",
-      "integrity": "sha512-o1hYe8hLi1EY6jgPFyxQgQ1wcycX+qz8eEbVmot2hFkgUzPxy9+kF0u0NIQBeDq+Mko47AkaFFaChcvZa9UX9Q==",
+      "version": "1.0.0-rc.12",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-openharmony-arm64/-/binding-openharmony-arm64-1.0.0-rc.12.tgz",
+      "integrity": "sha512-vRugONE4yMfVn0+7lUKdKvN4D5YusEiPilaoO2sgUWpCvrncvWgPMzK00ZFFJuiPgLwgFNP5eSiUlv2tfc+lpA==",
       "cpu": [
         "arm64"
       ],
@@ -286,9 +309,9 @@
       }
     },
     "node_modules/@rolldown/binding-wasm32-wasi": {
-      "version": "1.0.0-rc.10",
-      "resolved": "https://registry.npmjs.org/@rolldown/binding-wasm32-wasi/-/binding-wasm32-wasi-1.0.0-rc.10.tgz",
-      "integrity": "sha512-Ugv9o7qYJudqQO5Y5y2N2SOo6S4WiqiNOpuQyoPInnhVzCY+wi/GHltcLHypG9DEUYMB0iTB/huJrpadiAcNcA==",
+      "version": "1.0.0-rc.12",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-wasm32-wasi/-/binding-wasm32-wasi-1.0.0-rc.12.tgz",
+      "integrity": "sha512-ykGiLr/6kkiHc0XnBfmFJuCjr5ZYKKofkx+chJWDjitX+KsJuAmrzWhwyOMSHzPhzOHOy7u9HlFoa5MoAOJ/Zg==",
       "cpu": [
         "wasm32"
       ],
@@ -303,9 +326,9 @@
       }
     },
     "node_modules/@rolldown/binding-win32-arm64-msvc": {
-      "version": "1.0.0-rc.10",
-      "resolved": "https://registry.npmjs.org/@rolldown/binding-win32-arm64-msvc/-/binding-win32-arm64-msvc-1.0.0-rc.10.tgz",
-      "integrity": "sha512-7UODQb4fQUNT/vmgDZBl3XOBAIOutP5R3O/rkxg0aLfEGQ4opbCgU5vOw/scPe4xOqBwL9fw7/RP1vAMZ6QlAQ==",
+      "version": "1.0.0-rc.12",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-win32-arm64-msvc/-/binding-win32-arm64-msvc-1.0.0-rc.12.tgz",
+      "integrity": "sha512-5eOND4duWkwx1AzCxadcOrNeighiLwMInEADT0YM7xeEOOFcovWZCq8dadXgcRHSf3Ulh1kFo/qvzoFiCLOL1Q==",
       "cpu": [
         "arm64"
       ],
@@ -320,9 +343,9 @@
       }
     },
     "node_modules/@rolldown/binding-win32-x64-msvc": {
-      "version": "1.0.0-rc.10",
-      "resolved": "https://registry.npmjs.org/@rolldown/binding-win32-x64-msvc/-/binding-win32-x64-msvc-1.0.0-rc.10.tgz",
-      "integrity": "sha512-PYxKHMVHOb5NJuDL53vBUl1VwUjymDcYI6rzpIni0C9+9mTiJedvUxSk7/RPp7OOAm3v+EjgMu9bIy3N6b408w==",
+      "version": "1.0.0-rc.12",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-win32-x64-msvc/-/binding-win32-x64-msvc-1.0.0-rc.12.tgz",
+      "integrity": "sha512-PyqoipaswDLAZtot351MLhrlrh6lcZPo2LSYE+VDxbVk24LVKAGOuE4hb8xZQmrPAuEtTZW8E6D2zc5EUZX4Lw==",
       "cpu": [
         "x64"
       ],
@@ -337,9 +360,9 @@
       }
     },
     "node_modules/@rolldown/pluginutils": {
-      "version": "1.0.0-rc.10",
-      "resolved": "https://registry.npmjs.org/@rolldown/pluginutils/-/pluginutils-1.0.0-rc.10.tgz",
-      "integrity": "sha512-UkVDEFk1w3mveXeKgaTuYfKWtPbvgck1dT8TUG3bnccrH0XtLTuAyfCoks4Q/M5ZGToSVJTIQYCzy2g/atAOeg==",
+      "version": "1.0.0-rc.12",
+      "resolved": "https://registry.npmjs.org/@rolldown/pluginutils/-/pluginutils-1.0.0-rc.12.tgz",
+      "integrity": "sha512-HHMwmarRKvoFsJorqYlFeFRzXZqCt2ETQlEDOb9aqssrnVBB1/+xgTGtuTrIk5vzLNX1MjMtTf7W9z3tsSbrxw==",
       "dev": true,
       "license": "MIT"
     },
@@ -387,31 +410,31 @@
       "license": "MIT"
     },
     "node_modules/@vitest/expect": {
-      "version": "4.1.1",
-      "resolved": "https://registry.npmjs.org/@vitest/expect/-/expect-4.1.1.tgz",
-      "integrity": "sha512-xAV0fqBTk44Rn6SjJReEQkHP3RrqbJo6JQ4zZ7/uVOiJZRarBtblzrOfFIZeYUrukp2YD6snZG6IBqhOoHTm+A==",
+      "version": "4.1.2",
+      "resolved": "https://registry.npmjs.org/@vitest/expect/-/expect-4.1.2.tgz",
+      "integrity": "sha512-gbu+7B0YgUJ2nkdsRJrFFW6X7NTP44WlhiclHniUhxADQJH5Szt9mZ9hWnJPJ8YwOK5zUOSSlSvyzRf0u1DSBQ==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
         "@standard-schema/spec": "^1.1.0",
         "@types/chai": "^5.2.2",
-        "@vitest/spy": "4.1.1",
-        "@vitest/utils": "4.1.1",
+        "@vitest/spy": "4.1.2",
+        "@vitest/utils": "4.1.2",
         "chai": "^6.2.2",
-        "tinyrainbow": "^3.0.3"
+        "tinyrainbow": "^3.1.0"
       },
       "funding": {
         "url": "https://opencollective.com/vitest"
       }
     },
     "node_modules/@vitest/mocker": {
-      "version": "4.1.1",
-      "resolved": "https://registry.npmjs.org/@vitest/mocker/-/mocker-4.1.1.tgz",
-      "integrity": "sha512-h3BOylsfsCLPeceuCPAAJ+BvNwSENgJa4hXoXu4im0bs9Lyp4URc4JYK4pWLZ4pG/UQn7AT92K6IByi6rE6g3A==",
+      "version": "4.1.2",
+      "resolved": "https://registry.npmjs.org/@vitest/mocker/-/mocker-4.1.2.tgz",
+      "integrity": "sha512-Ize4iQtEALHDttPRCmN+FKqOl2vxTiNUhzobQFFt/BM1lRUTG7zRCLOykG/6Vo4E4hnUdfVLo5/eqKPukcWW7Q==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
-        "@vitest/spy": "4.1.1",
+        "@vitest/spy": "4.1.2",
         "estree-walker": "^3.0.3",
         "magic-string": "^0.30.21"
       },
@@ -432,26 +455,26 @@
       }
     },
     "node_modules/@vitest/pretty-format": {
-      "version": "4.1.1",
-      "resolved": "https://registry.npmjs.org/@vitest/pretty-format/-/pretty-format-4.1.1.tgz",
-      "integrity": "sha512-GM+TEQN5WhOygr1lp7skeVjdLPqqWMHsfzXrcHAqZJi/lIVh63H0kaRCY8MDhNWikx19zBUK8ceaLB7X5AH9NQ==",
+      "version": "4.1.2",
+      "resolved": "https://registry.npmjs.org/@vitest/pretty-format/-/pretty-format-4.1.2.tgz",
+      "integrity": "sha512-dwQga8aejqeuB+TvXCMzSQemvV9hNEtDDpgUKDzOmNQayl2OG241PSWeJwKRH3CiC+sESrmoFd49rfnq7T4RnA==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
-        "tinyrainbow": "^3.0.3"
+        "tinyrainbow": "^3.1.0"
       },
       "funding": {
         "url": "https://opencollective.com/vitest"
       }
     },
     "node_modules/@vitest/runner": {
-      "version": "4.1.1",
-      "resolved": "https://registry.npmjs.org/@vitest/runner/-/runner-4.1.1.tgz",
-      "integrity": "sha512-f7+FPy75vN91QGWsITueq0gedwUZy1fLtHOCMeQpjs8jTekAHeKP80zfDEnhrleviLHzVSDXIWuCIOFn3D3f8A==",
+      "version": "4.1.2",
+      "resolved": "https://registry.npmjs.org/@vitest/runner/-/runner-4.1.2.tgz",
+      "integrity": "sha512-Gr+FQan34CdiYAwpGJmQG8PgkyFVmARK8/xSijia3eTFgVfpcpztWLuP6FttGNfPLJhaZVP/euvujeNYar36OQ==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
-        "@vitest/utils": "4.1.1",
+        "@vitest/utils": "4.1.2",
         "pathe": "^2.0.3"
       },
       "funding": {
@@ -459,14 +482,14 @@
       }
     },
     "node_modules/@vitest/snapshot": {
-      "version": "4.1.1",
-      "resolved": "https://registry.npmjs.org/@vitest/snapshot/-/snapshot-4.1.1.tgz",
-      "integrity": "sha512-kMVSgcegWV2FibXEx9p9WIKgje58lcTbXgnJixfcg15iK8nzCXhmalL0ZLtTWLW9PH1+1NEDShiFFedB3tEgWg==",
+      "version": "4.1.2",
+      "resolved": "https://registry.npmjs.org/@vitest/snapshot/-/snapshot-4.1.2.tgz",
+      "integrity": "sha512-g7yfUmxYS4mNxk31qbOYsSt2F4m1E02LFqO53Xpzg3zKMhLAPZAjjfyl9e6z7HrW6LvUdTwAQR3HHfLjpko16A==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
-        "@vitest/pretty-format": "4.1.1",
-        "@vitest/utils": "4.1.1",
+        "@vitest/pretty-format": "4.1.2",
+        "@vitest/utils": "4.1.2",
         "magic-string": "^0.30.21",
         "pathe": "^2.0.3"
       },
@@ -475,9 +498,9 @@
       }
     },
     "node_modules/@vitest/spy": {
-      "version": "4.1.1",
-      "resolved": "https://registry.npmjs.org/@vitest/spy/-/spy-4.1.1.tgz",
-      "integrity": "sha512-6Ti/KT5OVaiupdIZEuZN7l3CZcR0cxnxt70Z0//3CtwgObwA6jZhmVBA3yrXSVN3gmwjgd7oDNLlsXz526gpRA==",
+      "version": "4.1.2",
+      "resolved": "https://registry.npmjs.org/@vitest/spy/-/spy-4.1.2.tgz",
+      "integrity": "sha512-DU4fBnbVCJGNBwVA6xSToNXrkZNSiw59H8tcuUspVMsBDBST4nfvsPsEHDHGtWRRnqBERBQu7TrTKskmjqTXKA==",
       "dev": true,
       "license": "MIT",
       "funding": {
@@ -485,15 +508,15 @@
       }
     },
     "node_modules/@vitest/utils": {
-      "version": "4.1.1",
-      "resolved": "https://registry.npmjs.org/@vitest/utils/-/utils-4.1.1.tgz",
-      "integrity": "sha512-cNxAlaB3sHoCdL6pj6yyUXv9Gry1NHNg0kFTXdvSIZXLHsqKH7chiWOkwJ5s5+d/oMwcoG9T0bKU38JZWKusrQ==",
+      "version": "4.1.2",
+      "resolved": "https://registry.npmjs.org/@vitest/utils/-/utils-4.1.2.tgz",
+      "integrity": "sha512-xw2/TiX82lQHA06cgbqRKFb5lCAy3axQ4H4SoUFhUsg+wztiet+co86IAMDtF6Vm1hc7J6j09oh/rgDn+JdKIQ==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
-        "@vitest/pretty-format": "4.1.1",
+        "@vitest/pretty-format": "4.1.2",
         "convert-source-map": "^2.0.0",
-        "tinyrainbow": "^3.0.3"
+        "tinyrainbow": "^3.1.0"
       },
       "funding": {
         "url": "https://opencollective.com/vitest"
@@ -739,6 +762,9 @@
         "arm64"
       ],
       "dev": true,
+      "libc": [
+        "glibc"
+      ],
       "license": "MPL-2.0",
       "optional": true,
       "os": [
@@ -760,6 +786,9 @@
         "arm64"
       ],
       "dev": true,
+      "libc": [
+        "musl"
+      ],
       "license": "MPL-2.0",
       "optional": true,
       "os": [
@@ -781,6 +810,9 @@
         "x64"
       ],
       "dev": true,
+      "libc": [
+        "glibc"
+      ],
       "license": "MPL-2.0",
       "optional": true,
       "os": [
@@ -802,6 +834,9 @@
         "x64"
       ],
       "dev": true,
+      "libc": [
+        "musl"
+      ],
       "license": "MPL-2.0",
       "optional": true,
       "os": [
@@ -912,9 +947,9 @@
       "license": "ISC"
     },
     "node_modules/picomatch": {
-      "version": "4.0.3",
-      "resolved": "https://registry.npmjs.org/picomatch/-/picomatch-4.0.3.tgz",
-      "integrity": "sha512-5gTmgEY/sqK6gFXLIsQNH19lWb4ebPDLA4SdLP7dsWkIXHWlG66oPuVvXSGFPppYZz8ZDZq0dYYrbHfBCVUb1Q==",
+      "version": "4.0.4",
+      "resolved": "https://registry.npmjs.org/picomatch/-/picomatch-4.0.4.tgz",
+      "integrity": "sha512-QP88BAKvMam/3NxH6vj2o21R6MjxZUAd6nlwAS/pnGvN9IVLocLHxGYIzFhg6fUQ+5th6P4dv4eW9jX3DSIj7A==",
       "dev": true,
       "license": "MIT",
       "engines": {
@@ -954,14 +989,14 @@
       }
     },
     "node_modules/rolldown": {
-      "version": "1.0.0-rc.10",
-      "resolved": "https://registry.npmjs.org/rolldown/-/rolldown-1.0.0-rc.10.tgz",
-      "integrity": "sha512-q7j6vvarRFmKpgJUT8HCAUljkgzEp4LAhPlJUvQhA5LA1SUL36s5QCysMutErzL3EbNOZOkoziSx9iZC4FddKA==",
+      "version": "1.0.0-rc.12",
+      "resolved": "https://registry.npmjs.org/rolldown/-/rolldown-1.0.0-rc.12.tgz",
+      "integrity": "sha512-yP4USLIMYrwpPHEFB5JGH1uxhcslv6/hL0OyvTuY+3qlOSJvZ7ntYnoWpehBxufkgN0cvXxppuTu5hHa/zPh+A==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
-        "@oxc-project/types": "=0.120.0",
-        "@rolldown/pluginutils": "1.0.0-rc.10"
+        "@oxc-project/types": "=0.122.0",
+        "@rolldown/pluginutils": "1.0.0-rc.12"
       },
       "bin": {
         "rolldown": "bin/cli.mjs"
@@ -970,21 +1005,21 @@
         "node": "^20.19.0 || >=22.12.0"
       },
       "optionalDependencies": {
-        "@rolldown/binding-android-arm64": "1.0.0-rc.10",
-        "@rolldown/binding-darwin-arm64": "1.0.0-rc.10",
-        "@rolldown/binding-darwin-x64": "1.0.0-rc.10",
-        "@rolldown/binding-freebsd-x64": "1.0.0-rc.10",
-        "@rolldown/binding-linux-arm-gnueabihf": "1.0.0-rc.10",
-        "@rolldown/binding-linux-arm64-gnu": "1.0.0-rc.10",
-        "@rolldown/binding-linux-arm64-musl": "1.0.0-rc.10",
-        "@rolldown/binding-linux-ppc64-gnu": "1.0.0-rc.10",
-        "@rolldown/binding-linux-s390x-gnu": "1.0.0-rc.10",
-        "@rolldown/binding-linux-x64-gnu": "1.0.0-rc.10",
-        "@rolldown/binding-linux-x64-musl": "1.0.0-rc.10",
-        "@rolldown/binding-openharmony-arm64": "1.0.0-rc.10",
-        "@rolldown/binding-wasm32-wasi": "1.0.0-rc.10",
-        "@rolldown/binding-win32-arm64-msvc": "1.0.0-rc.10",
-        "@rolldown/binding-win32-x64-msvc": "1.0.0-rc.10"
+        "@rolldown/binding-android-arm64": "1.0.0-rc.12",
+        "@rolldown/binding-darwin-arm64": "1.0.0-rc.12",
+        "@rolldown/binding-darwin-x64": "1.0.0-rc.12",
+        "@rolldown/binding-freebsd-x64": "1.0.0-rc.12",
+        "@rolldown/binding-linux-arm-gnueabihf": "1.0.0-rc.12",
+        "@rolldown/binding-linux-arm64-gnu": "1.0.0-rc.12",
+        "@rolldown/binding-linux-arm64-musl": "1.0.0-rc.12",
+        "@rolldown/binding-linux-ppc64-gnu": "1.0.0-rc.12",
+        "@rolldown/binding-linux-s390x-gnu": "1.0.0-rc.12",
+        "@rolldown/binding-linux-x64-gnu": "1.0.0-rc.12",
+        "@rolldown/binding-linux-x64-musl": "1.0.0-rc.12",
+        "@rolldown/binding-openharmony-arm64": "1.0.0-rc.12",
+        "@rolldown/binding-wasm32-wasi": "1.0.0-rc.12",
+        "@rolldown/binding-win32-arm64-msvc": "1.0.0-rc.12",
+        "@rolldown/binding-win32-x64-msvc": "1.0.0-rc.12"
       }
     },
     "node_modules/siginfo": {
@@ -1085,16 +1120,16 @@
       }
     },
     "node_modules/vite": {
-      "version": "8.0.1",
-      "resolved": "https://registry.npmjs.org/vite/-/vite-8.0.1.tgz",
-      "integrity": "sha512-wt+Z2qIhfFt85uiyRt5LPU4oVEJBXj8hZNWKeqFG4gRG/0RaRGJ7njQCwzFVjO+v4+Ipmf5CY7VdmZRAYYBPHw==",
+      "version": "8.0.3",
+      "resolved": "https://registry.npmjs.org/vite/-/vite-8.0.3.tgz",
+      "integrity": "sha512-B9ifbFudT1TFhfltfaIPgjo9Z3mDynBTJSUYxTjOQruf/zHH+ezCQKcoqO+h7a9Pw9Nm/OtlXAiGT1axBgwqrQ==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
         "lightningcss": "^1.32.0",
-        "picomatch": "^4.0.3",
+        "picomatch": "^4.0.4",
         "postcss": "^8.5.8",
-        "rolldown": "1.0.0-rc.10",
+        "rolldown": "1.0.0-rc.12",
         "tinyglobby": "^0.2.15"
       },
       "bin": {
@@ -1163,19 +1198,19 @@
       }
     },
     "node_modules/vitest": {
-      "version": "4.1.1",
-      "resolved": "https://registry.npmjs.org/vitest/-/vitest-4.1.1.tgz",
-      "integrity": "sha512-yF+o4POL41rpAzj5KVILUxm1GCjKnELvaqmU9TLLUbMfDzuN0UpUR9uaDs+mCtjPe+uYPksXDRLQGGPvj1cTmA==",
+      "version": "4.1.2",
+      "resolved": "https://registry.npmjs.org/vitest/-/vitest-4.1.2.tgz",
+      "integrity": "sha512-xjR1dMTVHlFLh98JE3i/f/WePqJsah4A0FK9cc8Ehp9Udk0AZk6ccpIZhh1qJ/yxVWRZ+Q54ocnD8TXmkhspGg==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
-        "@vitest/expect": "4.1.1",
-        "@vitest/mocker": "4.1.1",
-        "@vitest/pretty-format": "4.1.1",
-        "@vitest/runner": "4.1.1",
-        "@vitest/snapshot": "4.1.1",
-        "@vitest/spy": "4.1.1",
-        "@vitest/utils": "4.1.1",
+        "@vitest/expect": "4.1.2",
+        "@vitest/mocker": "4.1.2",
+        "@vitest/pretty-format": "4.1.2",
+        "@vitest/runner": "4.1.2",
+        "@vitest/snapshot": "4.1.2",
+        "@vitest/spy": "4.1.2",
+        "@vitest/utils": "4.1.2",
         "es-module-lexer": "^2.0.0",
         "expect-type": "^1.3.0",
         "magic-string": "^0.30.21",
@@ -1186,7 +1221,7 @@
         "tinybench": "^2.9.0",
         "tinyexec": "^1.0.2",
         "tinyglobby": "^0.2.15",
-        "tinyrainbow": "^3.0.3",
+        "tinyrainbow": "^3.1.0",
         "vite": "^6.0.0 || ^7.0.0 || ^8.0.0",
         "why-is-node-running": "^2.3.0"
       },
@@ -1203,10 +1238,10 @@
         "@edge-runtime/vm": "*",
         "@opentelemetry/api": "^1.9.0",
         "@types/node": "^20.0.0 || ^22.0.0 || >=24.0.0",
-        "@vitest/browser-playwright": "4.1.1",
-        "@vitest/browser-preview": "4.1.1",
-        "@vitest/browser-webdriverio": "4.1.1",
-        "@vitest/ui": "4.1.1",
+        "@vitest/browser-playwright": "4.1.2",
+        "@vitest/browser-preview": "4.1.2",
+        "@vitest/browser-webdriverio": "4.1.2",
+        "@vitest/ui": "4.1.2",
         "happy-dom": "*",
         "jsdom": "*",
         "vite": "^6.0.0 || ^7.0.0 || ^8.0.0"

sdk/typescript/src/base.ts

@@ -1,6 +1,15 @@
 import { TurnstoneAPIError } from "./errors.js";
 import { parseSSEStream } from "./sse.js";
 
+export interface TlsOptions {
+  /** Path to CA certificate PEM file (Node.js only). */
+  caCert?: string;
+  /** Path to client certificate PEM file for mTLS (Node.js only). */
+  clientCert?: string;
+  /** Path to client key PEM file for mTLS (Node.js only). */
+  clientKey?: string;
+}
+
 export interface ClientOptions {
   /** Server base URL (e.g. "http://localhost:8080"). */
   baseUrl: string;
@@ -8,6 +17,13 @@ export interface ClientOptions {
   token?: string;
   /** Custom fetch implementation (defaults to globalThis.fetch). */
   fetch?: typeof globalThis.fetch;
+  /**
+   * TLS certificate paths for documentation and tooling.
+   * The SDK does not read these directly — pass a custom `fetch`
+   * configured with your runtime's TLS agent (e.g. Node.js https.Agent).
+   * See docs/tls.md for examples.
+   */
+  tls?: TlsOptions;
 }
 
 export interface RequestOptions {

sdk/typescript/src/console.ts

@@ -44,6 +44,7 @@ import type {
   OrgInfo,
   RoleInfo,
   ScheduleInfo,
+  DeleteSettingResponse,
   SettingInfo,
   StatusResponse,
   ToolPolicyInfo,
@@ -394,7 +395,10 @@ export class TurnstoneConsole extends BaseClient {
     });
   }
 
-  async deleteSetting(key: string, nodeId?: string): Promise<StatusResponse> {
+  async deleteSetting(
+    key: string,
+    nodeId?: string,
+  ): Promise<DeleteSettingResponse> {
     const params: Record<string, string> = {};
     if (nodeId) params.node_id = nodeId;
     return this.request("DELETE", `/v1/api/admin/settings/${key}`, {

sdk/typescript/src/events.ts

@@ -38,6 +38,11 @@ export interface StreamEndEvent {
   type: "stream_end";
 }
 
+export interface StateChangeEvent {
+  type: "state_change";
+  state: "idle" | "thinking" | "running" | "attention" | "error";
+}
+
 export interface ToolInfoEvent {
   type: "tool_info";
   items: Array<Record<string, unknown>>;
@@ -59,6 +64,7 @@ export interface ToolResultEvent {
   call_id: string;
   name: string;
   output: string;
+  is_error?: boolean;
 }
 
 export interface ToolOutputChunkEvent {
@@ -149,6 +155,7 @@ export type ServerEvent =
   | ContentEvent
   | ReasoningEvent
   | StreamEndEvent
+  | StateChangeEvent
   | ToolInfoEvent
   | ApproveRequestEvent
   | ApprovalResolvedEvent
@@ -246,6 +253,10 @@ export function isStreamEndEvent(e: ServerEvent): e is StreamEndEvent {
   return e.type === "stream_end";
 }
 
+export function isStateChangeEvent(e: ServerEvent): e is StateChangeEvent {
+  return e.type === "state_change";
+}
+
 export function isToolResultEvent(e: ServerEvent): e is ToolResultEvent {
   return e.type === "tool_result";
 }

sdk/typescript/src/index.ts

@@ -19,7 +19,7 @@
 // Clients
 export { TurnstoneServer } from "./server.js";
 export { TurnstoneConsole } from "./console.js";
-export type { ClientOptions } from "./base.js";
+export type { ClientOptions, TlsOptions } from "./base.js";
 
 // Errors
 export { TurnstoneAPIError } from "./errors.js";
@@ -35,6 +35,7 @@ export type {
   ContentEvent,
   ReasoningEvent,
   StreamEndEvent,
+  StateChangeEvent,
   ToolInfoEvent,
   ApproveRequestEvent,
   ApprovalResolvedEvent,
@@ -65,6 +66,7 @@ export {
   isReasoningEvent,
   isErrorEvent,
   isStreamEndEvent,
+  isStateChangeEvent,
   isToolResultEvent,
   isWsStateEvent,
   isApproveRequestEvent,
@@ -98,6 +100,7 @@ export type {
   AuthLoginResponse,
   AuthStatusResponse,
   AuthSetupResponse,
+  DeleteSettingResponse,
   StatusResponse,
   ErrorResponse,
   ClusterOverviewResponse,

sdk/typescript/src/server.ts

@@ -93,10 +93,13 @@ export class TurnstoneServer extends BaseClient {
     });
   }
 
-  async cancel(wsId: string): Promise<StatusResponse> {
-    return this.request("POST", "/v1/api/cancel", {
-      json: { ws_id: wsId },
-    });
+  async cancel(
+    wsId: string,
+    opts?: { force?: boolean },
+  ): Promise<StatusResponse> {
+    const body: Record<string, unknown> = { ws_id: wsId };
+    if (opts?.force) body.force = true;
+    return this.request("POST", "/v1/api/cancel", { json: body });
   }
 
   // -- Streaming ------------------------------------------------------------

sdk/typescript/src/types.ts

@@ -10,6 +10,12 @@ export interface StatusResponse {
   status: string;
 }
 
+export interface DeleteSettingResponse {
+  status: string;
+  key: string;
+  default: unknown;
+}
+
 export interface AuthLoginRequest {
   token: string;
 }

tests.json

@@ -1,5 +1,5 @@
 {
-  "description": "turnstone behavior tests — tool selection, sequencing, and multi-step reasoning",
+  "description": "turnstone behavior tests \u2014 tool selection, sequencing, and multi-step reasoning",
   "defaults": {
     "n_runs": 5,
     "max_turns": 15
@@ -8,35 +8,91 @@
     {
       "id": "read-before-edit",
       "description": "Must read_file before edit_file on the same path",
-      "user_prompt": "Fix the typo in config.py — change 'recieve' to 'receive'",
+      "user_prompt": "Fix the typo in config.py \u2014 change 'recieve' to 'receive'",
       "setup": {
         "files": {
           "config.py": "# Config module\ndef recieve_data(source):\n    \"\"\"Recieve data from source.\"\"\"\n    return source.read()\n"
         }
       },
       "expected_actions": [
-        { "tool": "read_file", "args": { "path": "config.py" } },
-        { "tool": "edit_file", "args": { "path": "config.py" } }
+        {
+          "tool": "read_file",
+          "args": {
+            "path": "config.py"
+          }
+        },
+        {
+          "tool": "edit_file",
+          "args": {
+            "path": "config.py"
+          }
+        }
       ],
-      "match_mode": "ordered_subset"
+      "match_mode": "ordered_subset",
+      "user_prompts": [
+        "Fix the typo in config.py \u2014 change 'recieve' to 'receive'",
+        "In config.py, correct the misspelling of 'recieve' to 'receive'",
+        "Please update config.py by replacing 'recieve' with the correct spelling 'receive'",
+        "There's a typo in config.py: 'recieve' should be 'receive'. Please fix it.",
+        "Could you change 'recieve' to 'receive' in config.py?",
+        "Go ahead and fix 'recieve' \u2192 'receive' in config.py",
+        "I need the word 'recieve' corrected to 'receive' in the file config.py",
+        "config.py has a spelling error \u2014 'recieve' needs to be changed to 'receive'",
+        "Kindly rectify the typographical error in config.py, replacing 'recieve' with 'receive'",
+        "Hey, swap 'recieve' for 'receive' in config.py"
+      ]
     },
     {
       "id": "write-file-not-bash",
       "description": "Use write_file for file creation, not bash echo/cat",
       "user_prompt": "Create a file called hello.py that prints hello world",
       "expected_actions": [
-        { "tool": "write_file", "args_pattern": { "path": "hello\\.py" } }
+        {
+          "tool": "write_file",
+          "args_pattern": {
+            "path": "hello\\.py"
+          }
+        }
       ],
-      "match_mode": "subset"
+      "match_mode": "subset",
+      "user_prompts": [
+        "Create a file called hello.py that prints hello world",
+        "Make a hello.py file that outputs hello world",
+        "Please write a Python file named hello.py which prints hello world",
+        "I need a file called hello.py that prints hello world",
+        "Could you create hello.py with code that prints hello world?",
+        "Write hello.py \u2014 it should print hello world",
+        "Generate a hello.py file that outputs \"hello world\"",
+        "I'd like you to create a file named hello.py that prints hello world",
+        "Set up a file called hello.py to print hello world",
+        "Kindly produce a hello.py file whose purpose is to print hello world"
+      ]
     },
     {
       "id": "bash-for-commands",
       "description": "Use bash for running system commands",
       "user_prompt": "What Python version is installed?",
       "expected_actions": [
-        { "tool": "bash", "args_pattern": { "command": "python" } }
+        {
+          "tool": "bash",
+          "args_pattern": {
+            "command": "python"
+          }
+        }
       ],
-      "match_mode": "subset"
+      "match_mode": "subset",
+      "user_prompts": [
+        "What Python version is installed?",
+        "Check which version of Python is currently installed",
+        "Can you tell me the installed Python version?",
+        "python --version please",
+        "I need to know what version of Python is on this system",
+        "Which Python version do we have?",
+        "Could you look up the Python version that's installed here?",
+        "Determine the currently installed Python version",
+        "What's the Python version on this machine?",
+        "Please check the Python version"
+      ]
     },
     {
       "id": "search-for-patterns",
@@ -49,13 +105,30 @@
         }
       },
       "expected_actions": [
-        { "tool": "search", "args_pattern": { "query": "test_" } }
+        {
+          "tool": "search",
+          "args_pattern": {
+            "query": "test_"
+          }
+        }
       ],
-      "match_mode": "subset"
+      "match_mode": "subset",
+      "user_prompts": [
+        "Find all functions that start with 'test_' in the project",
+        "List every function in the project whose name begins with 'test_'",
+        "I need to locate all functions prefixed with 'test_' across the project",
+        "Could you search the project for any functions starting with 'test_'?",
+        "Show me all the test_ prefixed functions in this project",
+        "Hunt down every function that has a 'test_' prefix in the codebase",
+        "I'm looking for all functions named test_* throughout the project",
+        "Search the entire project for functions whose names start with test_",
+        "What functions beginning with 'test_' exist in this project?",
+        "Please identify all functions with the 'test_' prefix in the project files"
+      ]
     },
     {
       "id": "multi-file-edit",
-      "description": "Read and edit multiple files — must read before editing each, and edit both",
+      "description": "Read and edit multiple files \u2014 must read before editing each, and edit both",
       "user_prompt": "Change the default port from 8000 to 9000 in both server.py and config.py",
       "setup": {
         "files": {
@@ -64,12 +137,32 @@
         }
       },
       "expected_actions": [
-        { "tool": "read_file" },
-        { "tool": "read_file" },
-        { "tool": "edit_file" },
-        { "tool": "edit_file" }
+        {
+          "tool": "read_file"
+        },
+        {
+          "tool": "read_file"
+        },
+        {
+          "tool": "edit_file"
+        },
+        {
+          "tool": "edit_file"
+        }
       ],
-      "match_mode": "ordered_subset"
+      "match_mode": "ordered_subset",
+      "user_prompts": [
+        "Change the default port from 8000 to 9000 in both server.py and config.py",
+        "Update the default port to 9000 instead of 8000 in server.py and config.py",
+        "Could you modify the port number from 8000 to 9000 in both config.py and server.py?",
+        "Please replace port 8000 with 9000 in server.py and config.py",
+        "I need the default port switched from 8000 to 9000 in both server.py and config.py",
+        "In server.py and config.py, the default port should be changed from 8000 to 9000",
+        "Swap out port 8000 for 9000 in config.py and server.py",
+        "Would you mind updating the default port value from 8000 to 9000 across both server.py and config.py?",
+        "The default port in server.py and config.py needs to be 9000 instead of 8000 \u2014 please make that change",
+        "Go ahead and change 8000 to 9000 for the default port in both server.py and config.py"
+      ]
     },
     {
       "id": "search-then-edit",
@@ -83,51 +176,147 @@
         }
       },
       "expected_actions": [
-        { "tool": "search", "args_pattern": { "query": "MAX_RETRIES" } },
-        { "tool": "read_file" },
-        { "tool": "edit_file", "args_pattern": { "old_string": "3" } }
+        {
+          "tool": "search",
+          "args_pattern": {
+            "query": "MAX_RETRIES"
+          }
+        },
+        {
+          "tool": "read_file"
+        },
+        {
+          "tool": "edit_file",
+          "args_pattern": {
+            "old_string": "3"
+          }
+        }
       ],
-      "match_mode": "ordered_subset"
+      "match_mode": "ordered_subset",
+      "user_prompts": [
+        "Find where MAX_RETRIES is defined and change it from 3 to 5",
+        "Locate the definition of MAX_RETRIES and update its value from 3 to 5",
+        "Could you search for where MAX_RETRIES is defined and modify it from 3 to 5?",
+        "I need MAX_RETRIES changed from 3 to 5 \u2014 find where it's defined and update it",
+        "Please find the MAX_RETRIES definition and bump it from 3 to 5",
+        "Hunt down MAX_RETRIES in the codebase and change its value from 3 to 5",
+        "Where is MAX_RETRIES set to 3? Change it to 5.",
+        "Search the code for the MAX_RETRIES definition and alter it from 3 to 5",
+        "I'd like you to locate MAX_RETRIES (currently 3) and set it to 5 instead",
+        "Go find MAX_RETRIES and switch it from 3 to 5"
+      ]
     },
     {
       "id": "bash-git-log",
       "description": "Use bash for git commands, not other tools",
       "user_prompt": "Show me the git log for the last 5 commits",
       "expected_actions": [
-        { "tool": "bash", "args_pattern": { "command": "git\\s+log" } }
+        {
+          "tool": "bash",
+          "args_pattern": {
+            "command": "git\\s+log"
+          }
+        }
       ],
-      "match_mode": "subset"
+      "match_mode": "subset",
+      "user_prompts": [
+        "Show me the git log for the last 5 commits",
+        "Display the 5 most recent git commits",
+        "Can you pull up the git log limited to the last five commits?",
+        "I need to see the git log showing only the previous 5 commits",
+        "git log for the 5 latest commits, please",
+        "Would you mind showing me the last five entries in the git log?",
+        "Print out the most recent 5 commits from the git log",
+        "I'd like to review the git log \u2014 just the last 5 commits",
+        "Show the recent 5 commit history using git log",
+        "Could you display the git commit history for the past five commits?"
+      ]
     },
     {
       "id": "write-then-run",
       "description": "Create a script and run it to verify it works",
       "user_prompt": "Create a Python script called fib.py that prints the first 10 Fibonacci numbers, then run it to verify",
       "expected_actions": [
-        { "tool": "write_file", "args_pattern": { "path": "fib\\.py" } },
-        { "tool": "bash", "args_pattern": { "command": "python" } }
+        {
+          "tool": "write_file",
+          "args_pattern": {
+            "path": "fib\\.py"
+          }
+        },
+        {
+          "tool": "bash",
+          "args_pattern": {
+            "command": "python"
+          }
+        }
       ],
-      "match_mode": "ordered_subset"
+      "match_mode": "ordered_subset",
+      "user_prompts": [
+        "Create a Python script called fib.py that prints the first 10 Fibonacci numbers, then run it to verify",
+        "Write a Python file named fib.py that outputs the first 10 Fibonacci numbers, and execute it to confirm it works",
+        "Please make fib.py \u2013 a Python script printing the first ten Fibonacci numbers \u2013 then run it to check the output",
+        "I need a Python script fib.py that prints the first 10 Fibonacci numbers. Execute it afterwards to verify correctness.",
+        "Could you create fib.py to display the first 10 Fibonacci numbers in Python, and then run it to make sure it works?",
+        "Draft a script called fib.py in Python that outputs the first ten Fibonacci numbers, then execute it to validate",
+        "Hey, write me a fib.py that prints the first 10 Fibonacci numbers and run it so we can see it works",
+        "Generate a Python program fib.py which prints the initial 10 Fibonacci numbers, and verify by running it",
+        "Kindly produce a Python script named fib.py to print the first 10 Fibonacci numbers, then execute the script to confirm its output",
+        "Make a file fib.py containing Python code to print the first 10 Fibonacci numbers. Then run it to verify."
+      ]
     },
     {
       "id": "no-bash-for-file-write",
-      "description": "Should NOT use bash (echo/cat/heredoc) to create files — only write_file",
+      "description": "Should NOT use bash (echo/cat/heredoc) to create files \u2014 only write_file",
       "user_prompt": "Create a new file called README.md with a title and description of this project",
       "expected_actions": [
-        { "tool": "write_file", "args_pattern": { "path": "README" } }
+        {
+          "tool": "write_file",
+          "args_pattern": {
+            "path": "README"
+          }
+        }
       ],
-      "match_mode": "subset"
+      "match_mode": "subset",
+      "user_prompts": [
+        "Create a new file called README.md with a title and description of this project",
+        "Make a README.md file that includes a project title and description",
+        "I need a README.md created with a title and a brief description of the project",
+        "Please generate a README.md file containing the project's title and description",
+        "Could you set up a README.md with a title and project description?",
+        "Write a README.md that has a title and describes this project",
+        "Go ahead and create README.md \u2014 it should have a title and a description of the project",
+        "I'd like you to produce a new README.md file featuring a project title and description",
+        "Kindly establish a README.md file incorporating both a title and a description for this project",
+        "Spin up a README.md with a project title and description in it"
+      ]
     },
     {
-      "id": "plan-before-refactor",
-      "description": "Use the plan tool before a large refactoring task",
-      "user_prompt": "I need to refactor this codebase to separate the database layer from the API layer. Use the plan tool to think through the approach before making any changes.",
+      "id": "plan-when-asked",
+      "description": "Call the plan tool when the user asks to plan",
+      "user_prompt": "Plan how to add user authentication to this app.",
       "setup": {
         "files": {
-          "app.py": "import sqlite3\nfrom flask import Flask, jsonify\n\napp = Flask(__name__)\nDB = 'data.db'\n\ndef get_db():\n    return sqlite3.connect(DB)\n\n@app.route('/users')\ndef list_users():\n    db = get_db()\n    users = db.execute('SELECT * FROM users').fetchall()\n    db.close()\n    return jsonify(users)\n\n@app.route('/users/<int:uid>')\ndef get_user(uid):\n    db = get_db()\n    user = db.execute('SELECT * FROM users WHERE id=?', (uid,)).fetchone()\n    db.close()\n    return jsonify(user)\n\nif __name__ == '__main__':\n    app.run(port=8000)\n"
+          "app.py": "from flask import Flask, jsonify\n\napp = Flask(__name__)\n\n@app.route('/users')\ndef list_users():\n    return jsonify([{'id': 1, 'name': 'Alice'}])\n\nif __name__ == '__main__':\n    app.run(port=8000)\n"
         }
       },
-      "expected_actions": [{ "tool": "create_plan" }],
-      "match_mode": "subset"
+      "expected_actions": [
+        {
+          "tool": "plan_agent"
+        }
+      ],
+      "match_mode": "subset",
+      "user_prompts": [
+        "Plan how to add user authentication to this app.",
+        "Make a plan for adding pagination to the API endpoints.",
+        "Plan out how to add error handling to this application.",
+        "I need a plan for adding logging to this codebase.",
+        "Plan the approach for adding unit tests to this app.",
+        "How would you approach adding user authentication to this app? Lay out a plan.",
+        "I'd like you to outline a strategy for implementing user authentication in this application.",
+        "Could you come up with a plan for integrating user authentication into this app?",
+        "Think through the steps needed to add user auth to this app and present a plan.",
+        "Draft a plan for incorporating user authentication functionality into this application."
+      ]
     },
     {
       "id": "edit-not-rewrite",
@@ -139,10 +328,32 @@
         }
       },
       "expected_actions": [
-        { "tool": "read_file", "args": { "path": "utils.py" } },
-        { "tool": "edit_file", "args": { "path": "utils.py" } }
+        {
+          "tool": "read_file",
+          "args": {
+            "path": "utils.py"
+          }
+        },
+        {
+          "tool": "edit_file",
+          "args": {
+            "path": "utils.py"
+          }
+        }
       ],
-      "match_mode": "ordered_subset"
+      "match_mode": "ordered_subset",
+      "user_prompts": [
+        "Add a docstring to the process_data function in utils.py",
+        "Please add a docstring to the process_data function in utils.py",
+        "Could you write a docstring for process_data in utils.py?",
+        "Insert a docstring into the process_data function found in utils.py",
+        "I need a docstring added to process_data in utils.py",
+        "Put a docstring on the process_data function in utils.py",
+        "The process_data function in utils.py is missing a docstring \u2014 please add one",
+        "Would you mind adding a docstring to process_data in utils.py?",
+        "In utils.py, the process_data function needs a docstring",
+        "Add documentation via a docstring to the process_data function within utils.py"
+      ]
     },
     {
       "id": "bash-run-tests",
@@ -154,45 +365,130 @@
         }
       },
       "expected_actions": [
-        { "tool": "bash", "args_pattern": { "command": "pytest|python.*test" } }
+        {
+          "tool": "bash",
+          "args_pattern": {
+            "command": "pytest|python.*test"
+          }
+        }
       ],
-      "match_mode": "subset"
+      "match_mode": "subset",
+      "user_prompts": [
+        "Run the tests",
+        "Execute the test suite",
+        "Please go ahead and run the tests",
+        "Could you run the tests for me?",
+        "I need the tests to be run",
+        "Kick off the tests",
+        "Let's run the tests",
+        "Fire up the tests",
+        "Go ahead and execute the tests",
+        "I'd like you to run the tests"
+      ]
     },
     {
       "id": "web-fetch-url",
       "description": "Use web_fetch when asked to retrieve content from a URL",
       "user_prompt": "Fetch the contents of https://example.com and summarize what's on the page",
       "expected_actions": [
-        { "tool": "web_fetch", "args_pattern": { "url": "example\\.com" } }
+        {
+          "tool": "web_fetch",
+          "args_pattern": {
+            "url": "example\\.com"
+          }
+        }
       ],
-      "match_mode": "subset"
+      "match_mode": "subset",
+      "user_prompts": [
+        "Fetch the contents of https://example.com and summarize what's on the page",
+        "Go to https://example.com and give me a summary of what you find there",
+        "Could you pull up https://example.com and tell me what the page is about?",
+        "Retrieve the content from https://example.com, then provide a summary of it",
+        "I need you to grab https://example.com and summarize its contents for me",
+        "Please access https://example.com and give me an overview of the page",
+        "What's on https://example.com? Fetch it and summarize for me.",
+        "Download the page at https://example.com and provide a brief summary",
+        "I'd like a summary of whatever is at https://example.com \u2014 please fetch it first",
+        "Hit https://example.com and let me know what's there in summary form"
+      ]
     },
     {
       "id": "man-page-lookup",
       "description": "Use man tool to look up command documentation",
       "user_prompt": "Look up the man page for tar and tell me what the --xattrs flag does",
       "expected_actions": [
-        { "tool": "man", "args_pattern": { "page": "tar" } }
+        {
+          "tool": "man",
+          "args_pattern": {
+            "page": "tar"
+          }
+        }
       ],
-      "match_mode": "subset"
+      "match_mode": "subset",
+      "user_prompts": [
+        "Look up the man page for tar and tell me what the --xattrs flag does",
+        "What does the --xattrs flag do in tar? Check the man page for me.",
+        "Could you pull up the man page for tar and explain the --xattrs option?",
+        "I need to know what --xattrs does in tar \u2014 can you check the man page?",
+        "Check tar's man page and let me know the purpose of the --xattrs flag.",
+        "Please consult the tar man page and describe what the --xattrs flag is for.",
+        "Hey, look at the tar man page real quick \u2014 what's --xattrs do?",
+        "I'd like you to read the tar man page and summarize the --xattrs option for me.",
+        "Would you mind checking the man page for tar to find out what --xattrs means?",
+        "Look into the tar manual and explain the --xattrs flag to me."
+      ]
     },
     {
       "id": "math-calculation",
       "description": "Use the math tool for precise calculations, not bash or mental math",
       "user_prompt": "What is 2^64 - 1? Use the math tool to calculate it precisely.",
       "expected_actions": [
-        { "tool": "math", "args_pattern": { "code": "2.*64" } }
+        {
+          "tool": "math",
+          "args_pattern": {
+            "code": "2.*64"
+          }
+        }
       ],
-      "match_mode": "subset"
+      "match_mode": "subset",
+      "user_prompts": [
+        "What is 2^64 - 1? Use the math tool to calculate it precisely.",
+        "Calculate 2^64 - 1 for me using the math tool, please.",
+        "I need the exact value of 2^64 - 1. Please use the math tool.",
+        "Could you use the math tool to compute 2^64 minus 1 precisely?",
+        "Use the math tool to tell me what 2^64 - 1 equals.",
+        "I'm curious: what's 2^64 - 1? Compute it with the math tool.",
+        "Please precisely determine 2^64 - 1 via the math tool.",
+        "Mind using the math tool to figure out 2^64 - 1 exactly?",
+        "I'd like to know the precise result of 2^64 - 1 \u2014 use the math tool for this.",
+        "Leverage the math tool to give me an exact answer for 2^64 - 1."
+      ]
     },
     {
       "id": "web-search-query",
       "description": "Use web_search for general knowledge lookups, not web_fetch",
       "user_prompt": "Search the web for the current population of Tokyo",
       "expected_actions": [
-        { "tool": "web_search", "args_pattern": { "query": "Tokyo" } }
+        {
+          "tool": "web_search",
+          "args_pattern": {
+            "query": "Tokyo"
+          }
+        }
       ],
-      "match_mode": "subset"
+      "match_mode": "subset",
+      "user_prompts": [
+        "Search the web for the current population of Tokyo",
+        "What's Tokyo's current population? Look it up on the web.",
+        "Could you do a web search to find out how many people currently live in Tokyo?",
+        "Please search online for Tokyo's present-day population.",
+        "I need you to look up the current population of Tokyo on the web.",
+        "Find me Tokyo's current population via a web search.",
+        "Web search: what is the current population of Tokyo?",
+        "I'd like to know Tokyo's current population\u2014can you search the web for that?",
+        "Look up how many people live in Tokyo right now using a web search.",
+        "Do a web search for the population of Tokyo as of now."
+      ]
     }
   ]
 }

tests/test_auth.py

@@ -1237,6 +1237,52 @@ class TestJWTAudienceIssuer:
         result = validate_jwt(token, self.SECRET, audience="")
         assert result is not None
 
+    def test_create_jwt_expiry_seconds(self):
+        import jwt as pyjwt
+
+        from turnstone.core.auth import create_jwt
+
+        token = create_jwt("user1", frozenset({"read"}), "test", self.SECRET, expiry_seconds=300)
+        payload = pyjwt.decode(
+            token, self.SECRET, algorithms=["HS256"], options={"verify_aud": False}
+        )
+        assert payload["exp"] - payload["iat"] == 300
+
+    def test_create_jwt_expiry_seconds_overrides_hours(self):
+        import jwt as pyjwt
+
+        from turnstone.core.auth import create_jwt
+
+        token = create_jwt(
+            "user1",
+            frozenset({"read"}),
+            "test",
+            self.SECRET,
+            expiry_hours=24,
+            expiry_seconds=60,
+        )
+        payload = pyjwt.decode(
+            token, self.SECRET, algorithms=["HS256"], options={"verify_aud": False}
+        )
+        # expiry_seconds takes precedence over expiry_hours
+        assert payload["exp"] - payload["iat"] == 60
+
+    def test_create_jwt_expiry_seconds_rejects_zero(self):
+        import pytest
+
+        from turnstone.core.auth import create_jwt
+
+        with pytest.raises(ValueError, match="expiry_seconds must be positive"):
+            create_jwt("user1", frozenset({"read"}), "test", self.SECRET, expiry_seconds=0)
+
+    def test_create_jwt_expiry_seconds_rejects_negative(self):
+        import pytest
+
+        from turnstone.core.auth import create_jwt
+
+        with pytest.raises(ValueError, match="expiry_seconds must be positive"):
+            create_jwt("user1", frozenset({"read"}), "test", self.SECRET, expiry_seconds=-1)
+
 
 class TestServiceTokenManager:
     SECRET = "test-secret-that-is-at-least-32-chars"

tests/test_bridge_stress.py

@@ -39,7 +39,13 @@ def _make_bridge(**overrides) -> Bridge:
         approval_timeout=1,
     )
     defaults.update(overrides)
-    return Bridge(**defaults)
+    bridge = Bridge(**defaults)
+    # Replace real httpx client with a mock so daemon threads spawned by
+    # _handle_approval / _handle_plan_review don't make real HTTP calls
+    # after the test's patch context exits.
+    bridge._http.close()
+    bridge._http = MagicMock()
+    return bridge
 
 
 def _approval_items(tool_name: str = "bash") -> list[dict]:

tests/test_cancel.py

@@ -1,5 +1,6 @@
 """Tests for generation cancellation (cooperative cancel via threading.Event)."""
 
+import contextlib
 import threading
 import time
 from dataclasses import dataclass, field
@@ -7,7 +8,7 @@ from unittest.mock import MagicMock, patch
 
 import pytest
 
-from turnstone.core.session import ChatSession, GenerationCancelled
+from turnstone.core.session import ChatSession, GenerationCancelled, _CancelRef
 
 
 class NullUI:
@@ -36,7 +37,7 @@ class NullUI:
     def approve_tools(self, items):
         return True, None
 
-    def on_tool_result(self, call_id, name, output):
+    def on_tool_result(self, call_id, name, output, **kwargs):
         pass
 
     def on_tool_output_chunk(self, call_id, chunk):
@@ -407,3 +408,402 @@ class TestStreamFlushBeforeToolCalls:
         stream_end_idx = next(i for i, e in enumerate(events) if e[0] == "stream_end")
         late_content = [e for e in events[stream_end_idx + 1 :] if e[0] == "content"]
         assert late_content == [], f"Content after stream_end: {late_content}"
+
+
+class TestStreamAbort:
+    """Tests for cancel() closing the underlying SDK stream."""
+
+    def test_cancel_closes_cancel_stream(self, tmp_db):
+        """cancel() calls .close() on the stored SDK stream handle."""
+        session = _make_session()
+        mock_stream = MagicMock()
+        session._cancel_stream = mock_stream
+        session.cancel()
+        mock_stream.close.assert_called_once()
+        assert session._cancel_event.is_set()
+
+    def test_cancel_without_stream_is_safe(self, tmp_db):
+        """cancel() with no active stream just sets the event."""
+        session = _make_session()
+        assert session._cancel_stream is None
+        session.cancel()  # Should not raise
+        assert session._cancel_event.is_set()
+
+    def test_cancel_stream_close_error_suppressed(self, tmp_db):
+        """Errors from stream.close() are suppressed."""
+        session = _make_session()
+        mock_stream = MagicMock()
+        mock_stream.close.side_effect = RuntimeError("already closed")
+        session._cancel_stream = mock_stream
+        session.cancel()  # Should not raise
+        assert session._cancel_event.is_set()
+
+    def test_cancel_ref_populated_after_first_chunk(self, tmp_db):
+        """_cancel_ref is populated by the provider after the first chunk
+        arrives (lazy generator evaluation)."""
+        ui = NullUI()
+        session = _make_session(ui=ui)
+
+        @dataclass
+        class FakeChunk:
+            content_delta: str = ""
+            reasoning_delta: str = ""
+            tool_call_deltas: list = field(default_factory=list)
+            usage: None = None
+            finish_reason: str = "stop"
+            info_delta: str = ""
+            provider_blocks: list = field(default_factory=list)
+
+        sdk_stream = MagicMock()
+
+        def fake_provider_stream():
+            # Simulate provider appending to cancel_ref before first yield
+            session._cancel_ref.append(sdk_stream)
+            yield FakeChunk(content_delta="hi", finish_reason="stop")
+
+        with (
+            patch.object(
+                session,
+                "_create_stream_with_retry",
+                return_value=fake_provider_stream(),
+            ),
+            patch.object(session, "_full_messages", return_value=[]),
+        ):
+            session.send("test")
+
+        # After stream completes, cancel_stream should be cleared
+        assert session._cancel_stream is None
+        assert len(session._cancel_ref) == 0
+
+    def test_transport_error_during_cancel_becomes_generation_cancelled(self, tmp_db):
+        """When cancel() closes the stream, the resulting transport error
+        is converted to GenerationCancelled."""
+        ui = NullUI()
+        session = _make_session(ui=ui)
+
+        @dataclass
+        class FakeChunk:
+            content_delta: str = ""
+            reasoning_delta: str = ""
+            tool_call_deltas: list = field(default_factory=list)
+            usage: None = None
+            finish_reason: str = ""
+            info_delta: str = ""
+            provider_blocks: list = field(default_factory=list)
+
+        def stream_that_errors():
+            yield FakeChunk(content_delta="Hello")
+            session._cancel_event.set()
+            raise ConnectionError("stream closed")
+
+        with (
+            patch.object(
+                session,
+                "_create_stream_with_retry",
+                return_value=stream_that_errors(),
+            ),
+            patch.object(session, "_full_messages", return_value=[]),
+        ):
+            session.send("test")
+
+        # Should complete as cancelled, not error
+        assert "idle" in ui.states
+        assert any("cancelled" in i.lower() for i in ui.infos)
+        # Partial content preserved
+        assistant_msgs = [m for m in session.messages if m["role"] == "assistant"]
+        assert len(assistant_msgs) == 1
+        assert assistant_msgs[0]["content"] == "Hello"
+
+    def test_non_cancel_exception_not_swallowed(self, tmp_db):
+        """Exceptions during streaming that aren't caused by cancel
+        should propagate normally."""
+        ui = NullUI()
+        session = _make_session(ui=ui)
+
+        @dataclass
+        class FakeChunk:
+            content_delta: str = ""
+            reasoning_delta: str = ""
+            tool_call_deltas: list = field(default_factory=list)
+            usage: None = None
+            finish_reason: str = ""
+            info_delta: str = ""
+            provider_blocks: list = field(default_factory=list)
+
+        def stream_that_errors():
+            yield FakeChunk(content_delta="Hello")
+            raise ValueError("unexpected error")
+
+        with (
+            patch.object(
+                session,
+                "_create_stream_with_retry",
+                return_value=stream_that_errors(),
+            ),
+            patch.object(session, "_full_messages", return_value=[]),
+            pytest.raises(ValueError, match="unexpected error"),
+        ):
+            session.send("test")
+
+    def test_check_cancelled_between_retries(self, tmp_db):
+        """_try_stream checks for cancellation between retry attempts."""
+        session = _make_session()
+        session.cancel()
+
+        with pytest.raises(GenerationCancelled):
+            session._try_stream(
+                client=MagicMock(),
+                model="test",
+                msgs=[],
+            )
+
+
+class TestCancelRef:
+    """Tests for the _CancelRef list proxy."""
+
+    def test_append_sets_cancel_stream(self, tmp_db):
+        """Appending a stream handle to _CancelRef sets _cancel_stream eagerly."""
+        session = _make_session()
+        mock_stream = MagicMock()
+        assert session._cancel_stream is None
+
+        session._cancel_ref.append(mock_stream)
+
+        assert session._cancel_stream is mock_stream
+
+    def test_append_closes_stream_when_already_cancelled(self, tmp_db):
+        """If cancel is already set when a stream is appended, it is closed immediately."""
+        session = _make_session()
+        session.cancel()  # Set cancel event before stream is created
+
+        mock_stream = MagicMock()
+        session._cancel_ref.append(mock_stream)
+
+        mock_stream.close.assert_called_once()
+
+    def test_append_does_not_close_stream_when_not_cancelled(self, tmp_db):
+        """Stream is not closed if cancel hasn't been requested."""
+        session = _make_session()
+        mock_stream = MagicMock()
+
+        session._cancel_ref.append(mock_stream)
+
+        mock_stream.close.assert_not_called()
+        assert session._cancel_stream is mock_stream
+
+    def test_append_close_error_suppressed(self, tmp_db):
+        """Errors from stream.close() during eager close are suppressed."""
+        session = _make_session()
+        session.cancel()
+
+        mock_stream = MagicMock()
+        mock_stream.close.side_effect = RuntimeError("already closed")
+
+        session._cancel_ref.append(mock_stream)  # Should not raise
+
+    def test_cancel_ref_is_cancel_ref_instance(self, tmp_db):
+        """ChatSession._cancel_ref is a _CancelRef instance."""
+        session = _make_session()
+        assert isinstance(session._cancel_ref, _CancelRef)
+
+    def test_cancel_ref_cleared_after_stream_ends(self, tmp_db):
+        """_cancel_ref is cleared in the send() finally block after streaming."""
+        ui = NullUI()
+        session = _make_session(ui=ui)
+        mock_stream = MagicMock()
+        session._cancel_ref.append(mock_stream)
+        assert len(session._cancel_ref) == 1
+
+        @dataclass
+        class FakeChunk:
+            content_delta: str = ""
+            reasoning_delta: str = ""
+            tool_call_deltas: list = field(default_factory=list)
+            usage: None = None
+            finish_reason: str = "stop"
+            info_delta: str = ""
+            provider_blocks: list = field(default_factory=list)
+
+        with (
+            patch.object(
+                session,
+                "_create_stream_with_retry",
+                return_value=iter([FakeChunk(content_delta="hi", finish_reason="stop")]),
+            ),
+            patch.object(session, "_full_messages", return_value=[]),
+        ):
+            session.send("test")
+
+        # After send() completes, _cancel_ref is cleared in the finally block
+        assert len(session._cancel_ref) == 0
+
+
+class TestForceCancelGeneration:
+    """Tests for per-generation tracking that prevents orphaned-thread side-effects."""
+
+    def test_check_cancelled_raises_for_orphaned_generation(self, tmp_db):
+        """_check_cancelled raises GenerationCancelled when my_generation is stale."""
+        session = _make_session()
+        session._generation = 2  # Simulate two generations having run
+
+        with pytest.raises(GenerationCancelled):
+            session._check_cancelled(my_generation=1)  # Generation 1 is orphaned
+
+    def test_check_cancelled_ok_for_current_generation(self, tmp_db):
+        """_check_cancelled does not raise when my_generation matches current."""
+        session = _make_session()
+        session._generation = 3
+        session._check_cancelled(my_generation=3)  # Should not raise
+
+    def test_force_cancel_orphaned_thread_does_not_mutate_messages(self, tmp_db):
+        """An abandoned generation (force-cancel) cannot append to session.messages."""
+        ui = NullUI()
+        session = _make_session(ui=ui)
+
+        # We can't trivially test the full threading scenario in a unit test,
+        # so directly verify that _check_cancelled raises when my_generation
+        # is stale, which is what guards _stream_response against orphaned
+        # (force-cancelled) threads continuing to mutate messages.
+        session._generation = 5
+        with pytest.raises(GenerationCancelled):
+            session._check_cancelled(my_generation=4)  # orphaned generation
+
+    def test_new_cancel_event_per_generation_in_send(self, tmp_db):
+        """send() replaces _cancel_event with a fresh Event each generation."""
+        ui = NullUI()
+        session = _make_session(ui=ui)
+
+        @dataclass
+        class FakeChunk:
+            content_delta: str = ""
+            reasoning_delta: str = ""
+            tool_call_deltas: list = field(default_factory=list)
+            usage: None = None
+            finish_reason: str = "stop"
+            info_delta: str = ""
+            provider_blocks: list = field(default_factory=list)
+
+        original_event = session._cancel_event
+
+        with (
+            patch.object(
+                session,
+                "_create_stream_with_retry",
+                return_value=iter([FakeChunk(content_delta="hi", finish_reason="stop")]),
+            ),
+            patch.object(session, "_full_messages", return_value=[]),
+        ):
+            session.send("test")
+
+        # After send() completes, _cancel_event should be a NEW Event
+        # (not the same object as before the call).
+        assert session._cancel_event is not original_event
+        assert not session._cancel_event.is_set()
+
+
+class TestForceCancelThreaded:
+    """Force cancel with actual threads — verifies orphaned thread behavior."""
+
+    def test_force_cancel_orphan_does_not_mutate_messages(self, tmp_db):
+        """After force cancel + new send(), the orphaned thread must not
+        append stale content to session.messages."""
+        ui = NullUI()
+        session = _make_session(ui=ui)
+
+        @dataclass
+        class FakeChunk:
+            content_delta: str = ""
+            reasoning_delta: str = ""
+            tool_call_deltas: list = field(default_factory=list)
+            usage: None = None
+            finish_reason: str = ""
+            info_delta: str = ""
+            provider_blocks: list = field(default_factory=list)
+
+        barrier = threading.Event()
+        old_done = threading.Event()
+
+        def slow_stream():
+            yield FakeChunk(content_delta="Old content")
+            barrier.set()  # signal: first chunk delivered
+            time.sleep(2)  # simulate stuck stream
+            yield FakeChunk(content_delta=" more", finish_reason="stop")
+
+        # Start generation 1 (will get stuck)
+        with (
+            patch.object(session, "_create_stream_with_retry", return_value=slow_stream()),
+            patch.object(session, "_full_messages", return_value=[]),
+        ):
+
+            def run_old():
+                with contextlib.suppress(Exception):
+                    session.send("old message")
+                old_done.set()
+
+            t1 = threading.Thread(target=run_old, daemon=True)
+            t1.start()
+            assert barrier.wait(timeout=5), "stream did not start"
+
+        # Force cancel: simulate what the server does
+        session.cancel()
+        # Increment generation as new send() would
+        session._generation += 1
+        session._cancel_event = threading.Event()
+
+        # Wait for old thread to notice generation mismatch and exit
+        assert old_done.wait(timeout=10), "orphaned thread did not exit"
+
+        # The orphaned thread should NOT have appended its content
+        assistant_msgs = [m for m in session.messages if m["role"] == "assistant"]
+        # May have partial content from before cancel, but NOT the full
+        # "Old content more" that would appear without the generation guard
+        for msg in assistant_msgs:
+            assert "more" not in msg.get("content", "")
+
+    def test_force_cancel_then_new_send_succeeds(self, tmp_db):
+        """A new send() after force cancel works cleanly."""
+        ui = NullUI()
+        session = _make_session(ui=ui)
+
+        @dataclass
+        class FakeChunk:
+            content_delta: str = ""
+            reasoning_delta: str = ""
+            tool_call_deltas: list = field(default_factory=list)
+            usage: None = None
+            finish_reason: str = "stop"
+            info_delta: str = ""
+            provider_blocks: list = field(default_factory=list)
+
+        barrier = threading.Event()
+
+        def stuck_stream():
+            yield FakeChunk(content_delta="stuck")
+            barrier.set()
+            time.sleep(2)
+            yield FakeChunk(content_delta=" end", finish_reason="stop")
+
+        # Start stuck generation
+        with (
+            patch.object(session, "_create_stream_with_retry", return_value=stuck_stream()),
+            patch.object(session, "_full_messages", return_value=[]),
+        ):
+            t = threading.Thread(target=lambda: session.send("old"), daemon=True)
+            t.start()
+            assert barrier.wait(timeout=5), "stream did not start"
+
+        # Force cancel
+        session.cancel()
+
+        # New generation should work
+        fresh_stream = iter([FakeChunk(content_delta="Fresh response")])
+        with (
+            patch.object(session, "_create_stream_with_retry", return_value=fresh_stream),
+            patch.object(session, "_full_messages", return_value=[]),
+        ):
+            session.send("new message")
+
+        # The new generation should have completed successfully
+        assert "idle" in ui.states
+        assistant_msgs = [m for m in session.messages if m["role"] == "assistant"]
+        assert any("Fresh response" in m.get("content", "") for m in assistant_msgs)

tests/test_channel_discord.py

@@ -390,6 +390,7 @@ class TestApprovalVerdictDisplay:
         bot.config.streaming_edit_interval = 1.5
         bot.config.auto_approve = False
         bot.config.auto_approve_tools = []
+        bot.storage = None
         bot._streaming = {}
         bot._pending_approval_msgs = {}
         bot._notify_reply_channels = {}

tests/test_console.py

@@ -1699,6 +1699,236 @@ class TestSSEProxy:
         asyncio.run(_run())
 
 
+# ---------------------------------------------------------------------------
+# Proxy auth header propagation
+# ---------------------------------------------------------------------------
+
+
+class TestProxyAuthHeaders:
+    """Verify _proxy_auth_headers mints user-scoped JWTs for proxy requests."""
+
+    SECRET = "test-secret-that-is-at-least-32-chars"
+
+    def _make_request(
+        self, *, auth_result=None, jwt_secret="", proxy_token_mgr=None, proxy_auth_token=""
+    ):
+        """Build a minimal fake request for _proxy_auth_headers."""
+
+        class _State:
+            pass
+
+        class _AppState:
+            pass
+
+        class _App:
+            state = _AppState()
+
+        class _Request:
+            state = _State()
+            app = _App()
+
+        req = _Request()
+        req.state.auth_result = auth_result
+        req.app.state.jwt_secret = jwt_secret
+        req.app.state.proxy_token_mgr = proxy_token_mgr
+        req.app.state.proxy_auth_token = proxy_auth_token
+        return req
+
+    def test_mints_user_jwt(self):
+        """Real user auth_result → JWT with correct sub, scopes, src, aud, permissions."""
+        import jwt as pyjwt
+
+        from turnstone.console.server import _proxy_auth_headers
+        from turnstone.core.auth import JWT_AUD_SERVER, AuthResult
+
+        auth = AuthResult(
+            user_id="alice",
+            scopes=frozenset({"read", "write"}),
+            token_source="jwt",
+            permissions=frozenset({"admin.users"}),
+        )
+        req = self._make_request(auth_result=auth, jwt_secret=self.SECRET)
+        headers = _proxy_auth_headers(req)
+
+        assert "Authorization" in headers
+        token = headers["Authorization"].removeprefix("Bearer ")
+        payload = pyjwt.decode(token, self.SECRET, algorithms=["HS256"], audience=JWT_AUD_SERVER)
+        assert payload["sub"] == "alice"
+        assert set(payload["scopes"].split(",")) == {"read", "write"}
+        assert payload["src"] == "console-proxy"
+        assert payload["aud"] == JWT_AUD_SERVER
+        assert payload["permissions"] == "admin.users"
+
+    def test_narrows_scopes(self):
+        """Read-only user → JWT carries only read scope, not full {read,write,approve}."""
+        import jwt as pyjwt
+
+        from turnstone.console.server import _proxy_auth_headers
+        from turnstone.core.auth import JWT_AUD_SERVER, AuthResult
+
+        auth = AuthResult(
+            user_id="viewer",
+            scopes=frozenset({"read"}),
+            token_source="jwt",
+        )
+        req = self._make_request(auth_result=auth, jwt_secret=self.SECRET)
+        headers = _proxy_auth_headers(req)
+
+        token = headers["Authorization"].removeprefix("Bearer ")
+        payload = pyjwt.decode(token, self.SECRET, algorithms=["HS256"], audience=JWT_AUD_SERVER)
+        assert payload["scopes"] == "read"
+
+    def test_short_expiry(self):
+        """Minted JWT expires in 300 seconds, not hours."""
+        import jwt as pyjwt
+
+        from turnstone.console.server import _proxy_auth_headers
+        from turnstone.core.auth import AuthResult
+
+        auth = AuthResult(
+            user_id="alice",
+            scopes=frozenset({"read"}),
+            token_source="jwt",
+        )
+        req = self._make_request(auth_result=auth, jwt_secret=self.SECRET)
+        headers = _proxy_auth_headers(req)
+
+        token = headers["Authorization"].removeprefix("Bearer ")
+        payload = pyjwt.decode(
+            token, self.SECRET, algorithms=["HS256"], options={"verify_aud": False}
+        )
+        assert payload["exp"] - payload["iat"] == 300
+
+    def test_fallback_no_user(self):
+        """No auth_result → falls back to ServiceTokenManager."""
+        from turnstone.console.server import _proxy_auth_headers
+        from turnstone.core.auth import ServiceTokenManager
+
+        mgr = ServiceTokenManager(
+            user_id="console-proxy",
+            scopes=frozenset({"read", "write", "approve"}),
+            source="console",
+            secret=self.SECRET,
+        )
+        req = self._make_request(proxy_token_mgr=mgr)
+        headers = _proxy_auth_headers(req)
+
+        assert "Authorization" in headers
+        assert headers["Authorization"] == f"Bearer {mgr.token}"
+
+    def test_fallback_no_secret(self):
+        """auth_result present but empty jwt_secret → falls back to ServiceTokenManager."""
+        from turnstone.console.server import _proxy_auth_headers
+        from turnstone.core.auth import AuthResult, ServiceTokenManager
+
+        auth = AuthResult(
+            user_id="alice",
+            scopes=frozenset({"read"}),
+            token_source="jwt",
+        )
+        mgr = ServiceTokenManager(
+            user_id="console-proxy",
+            scopes=frozenset({"read", "write", "approve"}),
+            source="console",
+            secret=self.SECRET,
+        )
+        req = self._make_request(auth_result=auth, jwt_secret="", proxy_token_mgr=mgr)
+        headers = _proxy_auth_headers(req)
+
+        # Should use ServiceTokenManager, not mint a user JWT
+        assert headers["Authorization"] == f"Bearer {mgr.token}"
+
+    def test_fallback_static_token(self):
+        """No auth_result, no ServiceTokenManager → uses static proxy_auth_token."""
+        from turnstone.console.server import _proxy_auth_headers
+
+        req = self._make_request(proxy_auth_token="static-tok-123")
+        headers = _proxy_auth_headers(req)
+
+        assert headers == {"Authorization": "Bearer static-tok-123"}
+
+
+# ---------------------------------------------------------------------------
+# Server: trusted user_id forwarding on create_workstream
+# ---------------------------------------------------------------------------
+
+
+class TestCreateWorkstreamUserIdTrust:
+    """Verify that only trusted service tokens can forward user_id in create_workstream."""
+
+    def _extract_uid(self, body: dict, auth_result) -> str:
+        """Replicate the trust check from server.py:create_workstream."""
+        auth = auth_result
+        uid: str = getattr(auth, "user_id", "") or ""
+        trusted_sources = {"bridge", "console"}
+        if (
+            body.get("user_id")
+            and isinstance(body["user_id"], str)
+            and auth is not None
+            and auth.token_source in trusted_sources
+        ):
+            uid = body["user_id"]
+        return uid
+
+    def test_bridge_can_forward_user_id(self):
+        from turnstone.core.auth import AuthResult
+
+        auth = AuthResult(
+            user_id="bridge",
+            scopes=frozenset({"approve"}),
+            token_source="bridge",
+        )
+        uid = self._extract_uid({"user_id": "real-user-abc"}, auth)
+        assert uid == "real-user-abc"
+
+    def test_console_service_can_forward_user_id(self):
+        from turnstone.core.auth import AuthResult
+
+        auth = AuthResult(
+            user_id="console",
+            scopes=frozenset({"approve"}),
+            token_source="console",
+        )
+        uid = self._extract_uid({"user_id": "real-user-abc"}, auth)
+        assert uid == "real-user-abc"
+
+    def test_console_proxy_user_cannot_override_user_id(self):
+        """End-user tokens via console-proxy must NOT override user_id."""
+        from turnstone.core.auth import AuthResult
+
+        auth = AuthResult(
+            user_id="real-user-abc",
+            scopes=frozenset({"read", "write"}),
+            token_source="console-proxy",
+        )
+        uid = self._extract_uid({"user_id": "impersonated-user"}, auth)
+        # Should use JWT identity, NOT the body override
+        assert uid == "real-user-abc"
+
+    def test_direct_user_cannot_override_user_id(self):
+        """Direct JWT login must NOT override user_id."""
+        from turnstone.core.auth import AuthResult
+
+        auth = AuthResult(
+            user_id="real-user-abc",
+            scopes=frozenset({"read", "write"}),
+            token_source="password",
+        )
+        uid = self._extract_uid({"user_id": "impersonated-user"}, auth)
+        assert uid == "real-user-abc"
+
+    def test_no_body_user_id_uses_jwt(self):
+        from turnstone.core.auth import AuthResult
+
+        auth = AuthResult(
+            user_id="bridge",
+            scopes=frozenset({"approve"}),
+            token_source="bridge",
+        )
+        uid = self._extract_uid({"name": "test-ws"}, auth)
+        assert uid == "bridge"
+
+
 # ---------------------------------------------------------------------------
 # Collector — MCP aggregation in get_overview()
 # ---------------------------------------------------------------------------

tests/test_edit_file.py

@@ -0,0 +1,460 @@
+"""Tests for edit_file tool — single edit and batch edit modes."""
+
+from __future__ import annotations
+
+import os
+from unittest.mock import MagicMock
+
+import pytest
+
+from turnstone.core.session import ChatSession
+
+
+@pytest.fixture
+def session(tmp_db, mock_openai_client):
+    """Create a ChatSession wired to a temp database."""
+    return ChatSession(
+        client=mock_openai_client,
+        model="test-model",
+        ui=MagicMock(),
+        instructions=None,
+        temperature=0.5,
+        max_tokens=1000,
+        tool_timeout=10,
+    )
+
+
+@pytest.fixture
+def sample_file(tmp_path):
+    """Create a sample file and return its path."""
+    p = tmp_path / "test.py"
+    p.write_text("line1\nline2\nline3\nline4\nline5\n")
+    return str(p)
+
+
+def _mark_read(session: ChatSession, path: str) -> None:
+    """Simulate a prior read_file so the edit guard passes."""
+    resolved = os.path.realpath(os.path.expanduser(path))
+    session._read_files.add(resolved)
+
+
+# ── Single edit (backward compat) ────────────────────────────────────
+
+
+class TestSingleEdit:
+    def test_basic_replace(self, session, sample_file):
+        _mark_read(session, sample_file)
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": sample_file,
+                "old_string": "line2",
+                "new_string": "replaced",
+            },
+        )
+        assert result["needs_approval"]
+        assert result["func_name"] == "edit_file"
+
+        call_id, msg = session._exec_edit_file(result)
+        assert call_id == "c1"
+        assert "applied 1 edit" in msg
+        with open(sample_file) as f:
+            assert f.read() == "line1\nreplaced\nline3\nline4\nline5\n"
+
+    def test_missing_path(self, session):
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "old_string": "a",
+                "new_string": "b",
+            },
+        )
+        assert result.get("error")
+        assert "missing path" in result["error"]
+
+    def test_missing_old_string(self, session, sample_file):
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": sample_file,
+                "new_string": "b",
+            },
+        )
+        assert result.get("error")
+        assert "old_string" in result["error"]
+
+    def test_identical_strings(self, session, sample_file):
+        _mark_read(session, sample_file)
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": sample_file,
+                "old_string": "line1",
+                "new_string": "line1",
+            },
+        )
+        assert result.get("error")
+        assert "identical" in result["error"]
+
+    def test_old_string_not_found(self, session, sample_file):
+        _mark_read(session, sample_file)
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": sample_file,
+                "old_string": "nonexistent",
+                "new_string": "replaced",
+            },
+        )
+        assert result.get("error")
+        assert "not found" in result["error"]
+
+    def test_must_read_first(self, session, sample_file):
+        # Don't call _mark_read — should fail
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": sample_file,
+                "old_string": "line1",
+                "new_string": "replaced",
+            },
+        )
+        assert result.get("error")
+        assert "must read_file" in result["error"]
+
+    def test_multiple_occurrences_without_near_line(self, session, tmp_path):
+        p = tmp_path / "dup.txt"
+        p.write_text("foo\nbar\nfoo\n")
+        path = str(p)
+        _mark_read(session, path)
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": path,
+                "old_string": "foo",
+                "new_string": "baz",
+            },
+        )
+        assert result.get("error")
+        assert "found 2 times" in result["error"]
+
+    def test_near_line_disambiguates(self, session, tmp_path):
+        p = tmp_path / "dup.txt"
+        p.write_text("foo\nbar\nfoo\n")
+        path = str(p)
+        _mark_read(session, path)
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": path,
+                "old_string": "foo",
+                "new_string": "baz",
+                "near_line": 3,
+            },
+        )
+        assert result["needs_approval"]
+
+        call_id, msg = session._exec_edit_file(result)
+        assert "applied 1 edit" in msg
+        with open(path) as f:
+            assert f.read() == "foo\nbar\nbaz\n"
+
+    def test_deletion(self, session, sample_file):
+        _mark_read(session, sample_file)
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": sample_file,
+                "old_string": "line3\n",
+                "new_string": "",
+            },
+        )
+        assert result["needs_approval"]
+        assert "deletion" in result["preview"]
+
+        call_id, msg = session._exec_edit_file(result)
+        with open(sample_file) as f:
+            assert f.read() == "line1\nline2\nline4\nline5\n"
+
+
+# ── Batch edits ──────────────────────────────────────────────────────
+
+
+class TestBatchEdit:
+    def test_two_edits_applied_atomically(self, session, sample_file):
+        _mark_read(session, sample_file)
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": sample_file,
+                "edits": [
+                    {"old_string": "line1", "new_string": "first"},
+                    {"old_string": "line5", "new_string": "last"},
+                ],
+            },
+        )
+        assert result["needs_approval"]
+        assert "2 edits" in result["header"]
+
+        call_id, msg = session._exec_edit_file(result)
+        assert "applied 2 edits" in msg
+        with open(sample_file) as f:
+            assert f.read() == "first\nline2\nline3\nline4\nlast\n"
+
+    def test_three_edits_middle_of_file(self, session, sample_file):
+        _mark_read(session, sample_file)
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": sample_file,
+                "edits": [
+                    {"old_string": "line2", "new_string": "second"},
+                    {"old_string": "line3", "new_string": "third"},
+                    {"old_string": "line4", "new_string": "fourth"},
+                ],
+            },
+        )
+        assert result["needs_approval"]
+
+        call_id, msg = session._exec_edit_file(result)
+        assert "applied 3 edits" in msg
+        with open(sample_file) as f:
+            assert f.read() == "line1\nsecond\nthird\nfourth\nline5\n"
+
+    def test_overlapping_edits_rejected(self, session, tmp_path):
+        p = tmp_path / "overlap.txt"
+        p.write_text("abcdefgh\n")
+        path = str(p)
+        _mark_read(session, path)
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": path,
+                "edits": [
+                    {"old_string": "abcdef", "new_string": "XXX"},
+                    {"old_string": "defgh", "new_string": "YYY"},
+                ],
+            },
+        )
+        assert result["needs_approval"]
+
+        call_id, msg = session._exec_edit_file(result)
+        assert "overlap" in msg.lower()
+        # File should be untouched
+        with open(path) as f:
+            assert f.read() == "abcdefgh\n"
+
+    def test_batch_edit_not_found(self, session, sample_file):
+        _mark_read(session, sample_file)
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": sample_file,
+                "edits": [
+                    {"old_string": "line1", "new_string": "first"},
+                    {"old_string": "nonexistent", "new_string": "oops"},
+                ],
+            },
+        )
+        assert result.get("error")
+        assert "edits[1]" in result["error"]
+        assert "not found" in result["error"]
+
+    def test_batch_edit_missing_old_string(self, session, sample_file):
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": sample_file,
+                "edits": [
+                    {"old_string": "line1", "new_string": "first"},
+                    {"new_string": "oops"},
+                ],
+            },
+        )
+        assert result.get("error")
+        assert "edits[1]" in result["error"]
+        assert "old_string" in result["error"]
+
+    def test_batch_edit_identical_strings(self, session, sample_file):
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": sample_file,
+                "edits": [
+                    {"old_string": "line1", "new_string": "line1"},
+                ],
+            },
+        )
+        assert result.get("error")
+        assert "identical" in result["error"]
+
+    def test_batch_with_near_line(self, session, tmp_path):
+        p = tmp_path / "dup.txt"
+        p.write_text("foo\nbar\nfoo\nbaz\n")
+        path = str(p)
+        _mark_read(session, path)
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": path,
+                "edits": [
+                    {"old_string": "foo", "new_string": "first_foo", "near_line": 1},
+                    {"old_string": "foo", "new_string": "second_foo", "near_line": 3},
+                ],
+            },
+        )
+        assert result["needs_approval"]
+
+        call_id, msg = session._exec_edit_file(result)
+        assert "applied 2 edits" in msg
+        with open(path) as f:
+            assert f.read() == "first_foo\nbar\nsecond_foo\nbaz\n"
+
+    def test_batch_with_deletion(self, session, sample_file):
+        _mark_read(session, sample_file)
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": sample_file,
+                "edits": [
+                    {"old_string": "line2\n", "new_string": ""},
+                    {"old_string": "line4\n", "new_string": ""},
+                ],
+            },
+        )
+        assert result["needs_approval"]
+
+        call_id, msg = session._exec_edit_file(result)
+        with open(sample_file) as f:
+            assert f.read() == "line1\nline3\nline5\n"
+
+    def test_single_item_edits_array(self, session, sample_file):
+        """An edits array with one item should work like a single edit."""
+        _mark_read(session, sample_file)
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": sample_file,
+                "edits": [
+                    {"old_string": "line3", "new_string": "middle"},
+                ],
+            },
+        )
+        assert result["needs_approval"]
+        # Single edit — no "(N edits)" count in header
+        assert "edits)" not in result["header"]
+
+        call_id, msg = session._exec_edit_file(result)
+        assert "applied 1 edit" in msg
+        with open(sample_file) as f:
+            assert f.read() == "line1\nline2\nmiddle\nline4\nline5\n"
+
+
+# ── Mutual exclusivity ──────────────────────────────────────────────
+
+
+class TestMutualExclusivity:
+    def test_both_single_and_batch_rejected(self, session, sample_file):
+        _mark_read(session, sample_file)
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": sample_file,
+                "old_string": "line1",
+                "new_string": "replaced",
+                "edits": [
+                    {"old_string": "line2", "new_string": "also_replaced"},
+                ],
+            },
+        )
+        assert result.get("error")
+        assert "not both" in result["error"]
+
+    def test_neither_single_nor_batch(self, session, sample_file):
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": sample_file,
+            },
+        )
+        assert result.get("error")
+        assert "old_string" in result["error"]
+
+    def test_empty_edits_array_falls_through_to_single(self, session, sample_file):
+        """An empty edits array should be treated as no batch."""
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": sample_file,
+                "edits": [],
+            },
+        )
+        # Falls through to single-edit path, which requires old_string
+        assert result.get("error")
+        assert "old_string" in result["error"]
+
+
+# ── TOCTOU edge cases ───────────────────────────────────────────────
+
+
+class TestExecEdgeCases:
+    def test_file_changed_between_prepare_and_exec(self, session, sample_file):
+        _mark_read(session, sample_file)
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": sample_file,
+                "old_string": "line2",
+                "new_string": "replaced",
+            },
+        )
+        assert result["needs_approval"]
+
+        # Modify the file after prepare
+        with open(sample_file, "w") as f:
+            f.write("completely different content\n")
+
+        call_id, msg = session._exec_edit_file(result)
+        assert "no longer found" in msg
+
+    def test_file_deleted_between_prepare_and_exec(self, session, sample_file):
+        _mark_read(session, sample_file)
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": sample_file,
+                "old_string": "line2",
+                "new_string": "replaced",
+            },
+        )
+        assert result["needs_approval"]
+
+        os.unlink(sample_file)
+
+        call_id, msg = session._exec_edit_file(result)
+        assert "Error" in msg
+
+    def test_batch_file_changed_partial_match(self, session, sample_file):
+        """If file changes so one edit fails, none should be applied."""
+        _mark_read(session, sample_file)
+        result = session._prepare_edit_file(
+            "c1",
+            {
+                "path": sample_file,
+                "edits": [
+                    {"old_string": "line1", "new_string": "first"},
+                    {"old_string": "line5", "new_string": "last"},
+                ],
+            },
+        )
+        assert result["needs_approval"]
+
+        # Remove line5 between prepare and exec
+        with open(sample_file, "w") as f:
+            f.write("line1\nline2\nline3\nline4\n")
+
+        call_id, msg = session._exec_edit_file(result)
+        assert "no longer found" in msg
+        # line1 should NOT have been edited (atomic failure)
+        with open(sample_file) as f:
+            assert "line1" in f.read()

tests/test_judge.py

@@ -4,6 +4,7 @@ from __future__ import annotations
 
 import json
 import time
+from concurrent.futures import ThreadPoolExecutor
 from pathlib import Path
 from typing import Any
 from unittest.mock import MagicMock
@@ -179,10 +180,13 @@ class TestErrorHandling:
         provider = _make_mock_provider(side_effect=RuntimeError("API error"))
         judge = _make_judge(provider)
 
-        result = judge._evaluate_single(
-            _make_item(),
-            [{"role": "user", "content": "test"}],
-        )
+        with ThreadPoolExecutor(max_workers=1) as pool:
+            result = judge._evaluate_single(
+                _make_item(),
+                [{"role": "user", "content": "test"}],
+                cancel_event=None,
+                executor=pool,
+            )
         assert result is None
 
     def test_provider_error_heuristic_still_returned(self):
@@ -211,10 +215,13 @@ class TestErrorHandling:
         result_mock.content = ""
 
         judge = _make_judge(provider)
-        result = judge._evaluate_single(
-            _make_item(),
-            [{"role": "user", "content": "test"}],
-        )
+        with ThreadPoolExecutor(max_workers=1) as pool:
+            result = judge._evaluate_single(
+                _make_item(),
+                [{"role": "user", "content": "test"}],
+                cancel_event=None,
+                executor=pool,
+            )
         assert result is None
 
 
@@ -254,10 +261,13 @@ class TestMultiTurnToolUse:
         provider.create_completion.side_effect = [turn1, turn2]
 
         judge = _make_judge(provider)
-        verdict = judge._evaluate_single(
-            _make_item(),
-            [{"role": "user", "content": "test"}],
-        )
+        with ThreadPoolExecutor(max_workers=1) as pool:
+            verdict = judge._evaluate_single(
+                _make_item(),
+                [{"role": "user", "content": "test"}],
+                cancel_event=None,
+                executor=pool,
+            )
         assert verdict is not None
         assert verdict.tier == "llm"
         assert provider.create_completion.call_count == 2
@@ -299,10 +309,13 @@ class TestMultiTurnToolUse:
         ]
 
         judge = _make_judge(provider)
-        judge._evaluate_single(
-            _make_item(),
-            [{"role": "user", "content": "test"}],
-        )
+        with ThreadPoolExecutor(max_workers=1) as pool:
+            judge._evaluate_single(
+                _make_item(),
+                [{"role": "user", "content": "test"}],
+                cancel_event=None,
+                executor=pool,
+            )
         # Should have called create_completion exactly _JUDGE_MAX_TURNS times
         assert provider.create_completion.call_count == 5
 

tests/test_load_skill.py

@@ -54,6 +54,7 @@ def _make_session(skills: list[dict[str, Any]] | None = None):
     session._notify_on_complete = "{}"
     session.messages = []
     session._config = {}
+    session._tool_error_flags = {}
 
     # Stub set_skill to just record the call
     session._set_skill_called: list[str | None] = []
@@ -423,6 +424,7 @@ class TestSkillCatalogDisclosure:
         session._tool_search = None
         session._mcp_client = None
         session._notify_on_complete = "{}"
+        session._tool_error_flags = {}
 
         # Memory stubs
         session._memory_config = MagicMock()

tests/test_mcp_admin_api.py

@@ -573,7 +573,9 @@ def _fake_request(*nodes: dict[str, Any], proxy_client: Any = None) -> MagicMock
     collector.get_nodes.return_value = (list(nodes), len(nodes))
     collector.get_all_nodes.side_effect = lambda: collector.get_nodes.return_value[0]
     req = MagicMock()
+    req.state.auth_result = None
     req.app.state.collector = collector
+    req.app.state.jwt_secret = ""
     req.app.state.proxy_client = proxy_client or AsyncMock()
     req.app.state.proxy_token_mgr = None
     req.app.state.proxy_auth_token = "tok"

tests/test_mcp_client.py

@@ -401,6 +401,64 @@ class TestSessionIntegration:
         prepared = session._prepare_tool(tc)
         assert "error" in prepared
         assert "Unknown tool" in prepared["error"]
+        # Error lists available tools so the model can self-correct
+        assert "bash" in prepared["error"]
+        # Surfaces warning to user
+        session.ui.on_error.assert_called_once()
+        assert "nonexistent" in session.ui.on_error.call_args[0][0]
+
+    def test_prepare_tool_strips_whitespace_from_name(self, tmp_db):
+        """Local models may produce tool names with leading/trailing whitespace."""
+        session = self._make_session(mcp_client=None)
+        tc = {
+            "id": "call_strip",
+            "function": {"name": "  bash\n", "arguments": '{"command": "echo hi"}'},
+        }
+        prepared = session._prepare_tool(tc)
+        assert prepared["func_name"] == "bash"
+        assert "error" not in prepared
+
+    def test_prepare_tool_malformed_json_surfaces_error(self, tmp_db):
+        """Malformed JSON args should surface a warning to the user and
+        give the model a hint about expected format."""
+        session = self._make_session(mcp_client=None)
+        tc = {
+            "id": "call_bad",
+            "function": {"name": "bash", "arguments": "{command: echo hi}"},
+        }
+        prepared = session._prepare_tool(tc)
+        assert "error" in prepared
+        assert "JSON parse error" in prepared["error"]
+        assert "command" in prepared["error"]  # hint about expected key
+        assert "Please retry" in prepared["error"]
+        # User-facing warning
+        session.ui.on_error.assert_called_once()
+        assert "Malformed tool call" in session.ui.on_error.call_args[0][0]
+
+    def test_ensure_tool_call_ids_dict(self, tmp_db):
+        """_ensure_tool_call_ids fills empty IDs on streaming-style dict."""
+        from turnstone.core.session import ChatSession
+
+        tool_calls_acc = {
+            0: {"id": "", "function": {"name": "bash", "arguments": "{}"}},
+            1: {"id": "", "function": {"name": "read_file", "arguments": "{}"}},
+        }
+        ChatSession._ensure_tool_call_ids(tool_calls_acc)
+        ids = [tc["id"] for tc in tool_calls_acc.values()]
+        assert all(id_.startswith("call_") for id_ in ids)
+        assert len(set(ids)) == 2  # unique
+
+    def test_ensure_tool_call_ids_list(self, tmp_db):
+        """_ensure_tool_call_ids fills empty IDs on list (agent path)."""
+        from turnstone.core.session import ChatSession
+
+        tool_calls = [
+            {"id": None, "function": {"name": "bash", "arguments": "{}"}},
+            {"id": "call_existing", "function": {"name": "bash", "arguments": "{}"}},
+        ]
+        ChatSession._ensure_tool_call_ids(tool_calls)
+        assert tool_calls[0]["id"].startswith("call_")
+        assert tool_calls[1]["id"] == "call_existing"  # preserved
 
     def test_mcp_command_no_client(self, tmp_db):
         session = self._make_session(mcp_client=None)
@@ -1368,3 +1426,130 @@ class TestShutdownCleanup:
         assert mgr.get_prompts() == []
         assert mgr._resource_map == {}
         assert mgr._prompt_map == {}
+
+
+# ---------------------------------------------------------------------------
+# TCP probe and unreachable server handling
+# ---------------------------------------------------------------------------
+
+
+class TestTCPProbe:
+    """MCPClientManager._tcp_probe should fail fast on unreachable servers."""
+
+    def test_tcp_probe_unreachable_raises_connection_error(self):
+        """Unreachable host raises ConnectionError, not TimeoutError."""
+        mgr = MCPClientManager({})
+
+        async def _run():
+            with pytest.raises(ConnectionError, match="unreachable"):
+                await mgr._tcp_probe("test-server", "http://127.0.0.1:1")
+
+        asyncio.run(_run())
+
+    def test_tcp_probe_parses_url_correctly(self):
+        """Port and host are extracted from the URL."""
+        mgr = MCPClientManager({})
+
+        async def _run():
+            # Non-routable port — should fail with ConnectionError
+            with pytest.raises(ConnectionError):
+                await mgr._tcp_probe("srv", "https://127.0.0.1:1/mcp")
+
+        asyncio.run(_run())
+
+    def test_tcp_probe_default_port_http(self):
+        """Default port 80 used for http:// URLs without explicit port."""
+        mgr = MCPClientManager({})
+
+        async def _run():
+            # Will fail (nothing on port 80), but should not crash on parsing
+            with pytest.raises(ConnectionError):
+                await mgr._tcp_probe("srv", "http://127.0.0.1")
+
+        asyncio.run(_run())
+
+    def test_tcp_probe_dns_failure(self):
+        """Unresolvable hostname raises ConnectionError."""
+        mgr = MCPClientManager({})
+
+        async def _run():
+            with pytest.raises(ConnectionError):
+                await mgr._tcp_probe("srv", "http://this.host.does.not.exist.invalid:8080/mcp")
+
+        asyncio.run(_run())
+
+
+class TestConnectOneUnreachable:
+    """_connect_one should handle unreachable HTTP servers gracefully."""
+
+    def test_unreachable_http_server_raises_connection_error(self):
+        """Unreachable HTTP MCP server raises ConnectionError without spinning."""
+        mgr = MCPClientManager({})
+        mgr._loop = asyncio.new_event_loop()
+
+        async def _run():
+            with pytest.raises(ConnectionError, match="unreachable"):
+                await mgr._connect_one(
+                    "bad-server",
+                    {
+                        "type": "http",
+                        "url": "http://127.0.0.1:1/mcp",
+                    },
+                )
+
+        mgr._loop.run_until_complete(_run())
+        mgr._loop.close()
+
+        # Server should NOT be in sessions (connection failed)
+        assert "bad-server" not in mgr._sessions
+
+    def test_connect_all_continues_after_unreachable_server(self):
+        """_connect_all logs error and continues to next server."""
+        mgr = MCPClientManager(
+            {
+                "bad": {"type": "http", "url": "http://127.0.0.1:1/mcp"},
+            }
+        )
+
+        loop = asyncio.new_event_loop()
+        loop.run_until_complete(mgr._connect_all())
+        loop.close()
+
+        assert "bad" not in mgr._sessions
+        assert "bad" in mgr._last_error
+
+
+class TestSafeCloseStack:
+    """_safe_close_stack should suppress errors from broken anyio scopes."""
+
+    def test_suppresses_runtime_error(self):
+        """RuntimeError from broken cancel scope is suppressed."""
+
+        async def _run():
+            stack = AsyncExitStack()
+            await stack.__aenter__()
+
+            # Simulate a broken close that raises RuntimeError
+            async def _broken_close():
+                raise RuntimeError("Attempted to exit cancel scope in a different task")
+
+            stack.aclose = _broken_close
+            # Should not raise
+            await MCPClientManager._safe_close_stack(stack)
+
+        asyncio.run(_run())
+
+    def test_suppresses_cancelled_error(self):
+        """CancelledError during close is suppressed."""
+
+        async def _run():
+            stack = AsyncExitStack()
+            await stack.__aenter__()
+
+            async def _cancel_close():
+                raise asyncio.CancelledError()
+
+            stack.aclose = _cancel_close
+            await MCPClientManager._safe_close_stack(stack)
+
+        asyncio.run(_run())

tests/test_metacognition.py

@@ -4,6 +4,7 @@ from turnstone.core.metacognition import (
     NUDGE_COMPLETION,
     NUDGE_CORRECTION,
     NUDGE_DENIAL,
+    NUDGE_REPEAT,
     NUDGE_RESUME,
     NUDGE_START,
     NUDGE_TOOL_ERROR,
@@ -288,3 +289,22 @@ class TestToolErrorNudge:
     def test_not_with_zero_memories(self):
         state: dict[str, float] = {}
         assert should_nudge("tool_error", state, message_count=5, memory_count=0) is False
+
+
+class TestRepeatNudge:
+    def test_format(self):
+        assert format_nudge("repeat") == NUDGE_REPEAT
+
+    def test_fires(self):
+        state: dict[str, float] = {}
+        assert should_nudge("repeat", state, message_count=5) is True
+
+    def test_cooldown(self):
+        state: dict[str, float] = {}
+        assert should_nudge("repeat", state, message_count=5) is True
+        assert should_nudge("repeat", state, message_count=6) is False
+
+    def test_no_memory_requirement(self):
+        """Repeat nudge should fire even with zero memories."""
+        state: dict[str, float] = {}
+        assert should_nudge("repeat", state, message_count=5, memory_count=0) is True

tests/test_model_definition_storage.py

@@ -0,0 +1,163 @@
+"""Tests for model definition storage CRUD operations."""
+
+from __future__ import annotations
+
+import uuid
+from typing import TYPE_CHECKING
+
+if TYPE_CHECKING:
+    from turnstone.core.storage._sqlite import SQLiteBackend
+
+
+def _make_id() -> str:
+    return uuid.uuid4().hex
+
+
+class TestModelDefinitionStorage:
+    def test_create_and_get(self, db: SQLiteBackend) -> None:
+        did = _make_id()
+        db.create_model_definition(
+            definition_id=did,
+            alias="test-model",
+            model="gpt-5",
+            provider="openai",
+            base_url="https://api.openai.com/v1",
+            api_key="sk-test",
+            context_window=128000,
+        )
+        m = db.get_model_definition(did)
+        assert m is not None
+        assert m["alias"] == "test-model"
+        assert m["model"] == "gpt-5"
+        assert m["provider"] == "openai"
+        assert m["base_url"] == "https://api.openai.com/v1"
+        assert m["api_key"] == "sk-test"
+        assert m["context_window"] == 128000
+        assert m["capabilities"] == "{}"
+        assert m["enabled"] is True
+
+    def test_get_by_alias(self, db: SQLiteBackend) -> None:
+        did = _make_id()
+        db.create_model_definition(definition_id=did, alias="by-alias", model="gpt-5")
+        m = db.get_model_definition_by_alias("by-alias")
+        assert m is not None
+        assert m["definition_id"] == did
+
+    def test_get_by_alias_not_found(self, db: SQLiteBackend) -> None:
+        assert db.get_model_definition_by_alias("nope") is None
+
+    def test_get_not_found(self, db: SQLiteBackend) -> None:
+        assert db.get_model_definition("nonexistent") is None
+
+    def test_list_empty(self, db: SQLiteBackend) -> None:
+        assert db.list_model_definitions() == []
+
+    def test_list_all(self, db: SQLiteBackend) -> None:
+        db.create_model_definition(definition_id=_make_id(), alias="alpha", model="gpt-5")
+        db.create_model_definition(
+            definition_id=_make_id(), alias="beta", model="claude-opus-4-6", provider="anthropic"
+        )
+        models = db.list_model_definitions()
+        assert len(models) == 2
+        assert models[0]["alias"] == "alpha"  # ordered by alias
+        assert models[1]["alias"] == "beta"
+
+    def test_list_enabled_only(self, db: SQLiteBackend) -> None:
+        db.create_model_definition(
+            definition_id=_make_id(), alias="enabled-model", model="gpt-5", enabled=True
+        )
+        db.create_model_definition(
+            definition_id=_make_id(), alias="disabled-model", model="gpt-5", enabled=False
+        )
+        enabled = db.list_model_definitions(enabled_only=True)
+        assert len(enabled) == 1
+        assert enabled[0]["alias"] == "enabled-model"
+
+    def test_update_basic_fields(self, db: SQLiteBackend) -> None:
+        did = _make_id()
+        db.create_model_definition(
+            definition_id=did, alias="orig", model="gpt-5", base_url="http://old"
+        )
+        ok = db.update_model_definition(did, alias="renamed", base_url="http://new")
+        assert ok is True
+        m = db.get_model_definition(did)
+        assert m is not None
+        assert m["alias"] == "renamed"
+        assert m["base_url"] == "http://new"
+
+    def test_update_boolean_conversion(self, db: SQLiteBackend) -> None:
+        did = _make_id()
+        db.create_model_definition(definition_id=did, alias="booltest", model="gpt-5")
+        db.update_model_definition(did, enabled=False)
+        m = db.get_model_definition(did)
+        assert m is not None
+        assert m["enabled"] is False
+
+    def test_update_not_found(self, db: SQLiteBackend) -> None:
+        ok = db.update_model_definition("nonexistent", alias="x")
+        assert ok is False
+
+    def test_update_ignores_disallowed_fields(self, db: SQLiteBackend) -> None:
+        did = _make_id()
+        db.create_model_definition(
+            definition_id=did, alias="guard", model="gpt-5", created_by="admin"
+        )
+        original = db.get_model_definition(did)
+        assert original is not None
+        original_created = original["created"]
+        # created_by and created are not in the mutable allowlist
+        db.update_model_definition(did, created_by="evil", created="2000-01-01T00:00:00")
+        m = db.get_model_definition(did)
+        assert m is not None
+        assert m["created_by"] == "admin"  # unchanged
+        assert m["created"] == original_created  # unchanged
+
+    def test_delete(self, db: SQLiteBackend) -> None:
+        did = _make_id()
+        db.create_model_definition(definition_id=did, alias="delme", model="gpt-5")
+        ok = db.delete_model_definition(did)
+        assert ok is True
+        assert db.get_model_definition(did) is None
+
+    def test_delete_not_found(self, db: SQLiteBackend) -> None:
+        ok = db.delete_model_definition("nonexistent")
+        assert ok is False
+
+    def test_create_duplicate_alias(self, db: SQLiteBackend) -> None:
+        db.create_model_definition(definition_id=_make_id(), alias="unique", model="gpt-5")
+        # Second create with same alias but different ID should be no-op (OR IGNORE)
+        did2 = _make_id()
+        db.create_model_definition(definition_id=did2, alias="unique", model="gpt-5")
+        assert db.get_model_definition(did2) is None
+
+    def test_create_idempotent_same_id(self, db: SQLiteBackend) -> None:
+        did = _make_id()
+        db.create_model_definition(definition_id=did, alias="idem", model="gpt-5")
+        db.create_model_definition(definition_id=did, alias="idem", model="gpt-5-mini")
+        m = db.get_model_definition(did)
+        assert m is not None
+        assert m["model"] == "gpt-5"  # original preserved
+
+    def test_capabilities_json(self, db: SQLiteBackend) -> None:
+        did = _make_id()
+        caps = '{"supports_vision": true, "supports_web_search": false}'
+        db.create_model_definition(
+            definition_id=did, alias="caps-test", model="gpt-5", capabilities=caps
+        )
+        m = db.get_model_definition(did)
+        assert m is not None
+        assert m["capabilities"] == caps
+
+    def test_defaults(self, db: SQLiteBackend) -> None:
+        """Verify default values for optional fields."""
+        did = _make_id()
+        db.create_model_definition(definition_id=did, alias="defaults", model="gpt-5")
+        m = db.get_model_definition(did)
+        assert m is not None
+        assert m["provider"] == "openai"
+        assert m["base_url"] == ""
+        assert m["api_key"] == ""
+        assert m["context_window"] == 32768
+        assert m["capabilities"] == "{}"
+        assert m["enabled"] is True
+        assert m["created_by"] == ""

tests/test_model_probe.py

@@ -0,0 +1,235 @@
+"""Tests for probe_model_endpoint() and lookup_model_capabilities()."""
+
+from __future__ import annotations
+
+from typing import Any
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from turnstone.core.model_registry import probe_model_endpoint
+from turnstone.core.providers import list_known_models, lookup_model_capabilities
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+
+def _mock_model(
+    model_id: str,
+    *,
+    owned_by: str = "test",
+    meta: dict[str, Any] | None = None,
+) -> MagicMock:
+    m = MagicMock()
+    m.id = model_id
+    dumped: dict[str, Any] = {"owned_by": owned_by}
+    if meta is not None:
+        dumped["meta"] = meta
+    m.model_dump.return_value = dumped
+    return m
+
+
+def _mock_client(*models: MagicMock) -> MagicMock:
+    fast = MagicMock()
+    fast.models.list.return_value = MagicMock(data=list(models))
+    client = MagicMock()
+    client.with_options.return_value = fast
+    return client
+
+
+# ---------------------------------------------------------------------------
+# probe_model_endpoint
+# ---------------------------------------------------------------------------
+
+
+class TestProbeModelEndpoint:
+    @patch("turnstone.core.providers.create_client")
+    def test_probe_success(self, mock_cc: MagicMock) -> None:
+        m1 = _mock_model("model-a")
+        m2 = _mock_model("model-b")
+        mock_cc.return_value = _mock_client(m1, m2)
+
+        result = probe_model_endpoint("openai", "http://localhost:8000/v1", "key")
+        assert result["reachable"] is True
+        assert result["available_models"] == ["model-a", "model-b"]
+        assert result["error"] is None
+
+    @patch("turnstone.core.providers.create_client")
+    def test_target_found(self, mock_cc: MagicMock) -> None:
+        m1 = _mock_model("gpt-5")
+        mock_cc.return_value = _mock_client(m1)
+
+        result = probe_model_endpoint(
+            "openai", "http://localhost:8000/v1", "key", target_model="gpt-5"
+        )
+        assert result["model_found"] is True
+
+    @patch("turnstone.core.providers.create_client")
+    def test_target_not_found(self, mock_cc: MagicMock) -> None:
+        m1 = _mock_model("model-a")
+        mock_cc.return_value = _mock_client(m1)
+
+        result = probe_model_endpoint(
+            "openai", "http://localhost:8000/v1", "key", target_model="gpt-5"
+        )
+        assert result["model_found"] is False
+        assert result["available_models"] == ["model-a"]
+
+    @patch("turnstone.core.providers.create_client")
+    def test_no_target_model_found_is_none(self, mock_cc: MagicMock) -> None:
+        m1 = _mock_model("model-a")
+        mock_cc.return_value = _mock_client(m1)
+
+        result = probe_model_endpoint("openai", "http://localhost:8000/v1", "key")
+        assert result["model_found"] is None
+
+    @patch("turnstone.core.providers.create_client")
+    def test_context_window_llama_cpp(self, mock_cc: MagicMock) -> None:
+        m = _mock_model("qwen-32b", meta={"n_ctx_train": 131072})
+        mock_cc.return_value = _mock_client(m)
+
+        result = probe_model_endpoint("openai", "http://localhost:8000/v1", "key")
+        assert result["context_window"] == 131072
+        assert result["server_type"] == "llama.cpp"
+
+    @patch("turnstone.core.providers.create_client")
+    def test_server_type_openai(self, mock_cc: MagicMock) -> None:
+        m = _mock_model("gpt-5")
+        mock_cc.return_value = _mock_client(m)
+
+        result = probe_model_endpoint("openai", "https://api.openai.com/v1", "sk-test")
+        assert result["server_type"] == "openai"
+
+    @patch("turnstone.core.providers.create_client")
+    def test_server_type_sglang(self, mock_cc: MagicMock) -> None:
+        m = _mock_model("meta-llama/Llama-3", owned_by="sglang")
+        mock_cc.return_value = _mock_client(m)
+
+        result = probe_model_endpoint("openai", "http://localhost:30000/v1", "key")
+        assert result["server_type"] == "sglang"
+
+    @patch("turnstone.core.providers.create_client")
+    def test_server_type_vllm(self, mock_cc: MagicMock) -> None:
+        m = _mock_model("org/model-name")
+        mock_cc.return_value = _mock_client(m)
+
+        result = probe_model_endpoint("openai", "http://localhost:8000/v1", "key")
+        assert result["server_type"] == "vllm"
+
+    @patch("turnstone.core.providers.create_client")
+    def test_server_type_generic(self, mock_cc: MagicMock) -> None:
+        m = _mock_model("my-model")
+        mock_cc.return_value = _mock_client(m)
+
+        result = probe_model_endpoint("openai", "http://localhost:8000/v1", "key")
+        assert result["server_type"] == "openai-compatible"
+
+    @patch("turnstone.core.providers.create_client")
+    def test_anthropic_provider(self, mock_cc: MagicMock) -> None:
+        m = _mock_model("claude-sonnet-4-6")
+        mock_cc.return_value = _mock_client(m)
+
+        result = probe_model_endpoint(
+            "anthropic",
+            "https://api.anthropic.com",
+            "sk-ant-test",
+            target_model="claude-sonnet-4-6",
+        )
+        assert result["reachable"] is True
+        assert result["server_type"] == "anthropic"
+        assert result["context_window"] == 200000
+
+    @patch("turnstone.core.providers.create_client")
+    def test_connection_failure(self, mock_cc: MagicMock) -> None:
+        mock_cc.side_effect = OSError("Connection refused")
+
+        result = probe_model_endpoint("openai", "http://bad:1234/v1", "key")
+        assert result["reachable"] is False
+        assert "Connection refused" in (result["error"] or "")
+
+    @patch("turnstone.core.providers.create_client")
+    def test_empty_model_list(self, mock_cc: MagicMock) -> None:
+        mock_cc.return_value = _mock_client()  # no models
+
+        result = probe_model_endpoint("openai", "http://localhost:8000/v1", "key")
+        assert result["reachable"] is True
+        assert result["available_models"] == []
+        assert "No models found" in (result["error"] or "")
+
+    @patch("turnstone.core.providers.create_client")
+    def test_context_window_openai_static_table(self, mock_cc: MagicMock) -> None:
+        """When base_url is api.openai.com and model is known, use static table."""
+        m = _mock_model("gpt-5")
+        mock_cc.return_value = _mock_client(m)
+
+        result = probe_model_endpoint(
+            "openai", "https://api.openai.com/v1", "sk-test", target_model="gpt-5"
+        )
+        assert result["context_window"] == 400000
+
+
+# ---------------------------------------------------------------------------
+# lookup_model_capabilities
+# ---------------------------------------------------------------------------
+
+
+class TestLookupModelCapabilities:
+    def test_known_openai_model(self) -> None:
+        caps = lookup_model_capabilities("openai", "gpt-5")
+        assert caps is not None
+        assert caps["context_window"] == 400000
+        assert caps["supports_temperature"] is False
+
+    def test_known_anthropic_model(self) -> None:
+        caps = lookup_model_capabilities("anthropic", "claude-opus-4-6")
+        assert caps is not None
+        assert caps["context_window"] == 200000
+        assert caps["thinking_mode"] == "adaptive"
+
+    def test_unknown_model_returns_none(self) -> None:
+        caps = lookup_model_capabilities("openai", "totally-unknown-model")
+        assert caps is None
+
+    def test_tuples_converted_to_lists(self) -> None:
+        caps = lookup_model_capabilities("openai", "gpt-5")
+        assert caps is not None
+        for val in caps.values():
+            assert not isinstance(val, tuple), f"Found tuple: {val}"
+
+    def test_reasoning_effort_values_are_list(self) -> None:
+        caps = lookup_model_capabilities("openai", "gpt-5")
+        assert caps is not None
+        assert isinstance(caps["reasoning_effort_values"], list)
+        assert "medium" in caps["reasoning_effort_values"]
+
+    def test_openai_compatible_returns_none(self) -> None:
+        caps = lookup_model_capabilities("openai-compatible", "my-local-model")
+        assert caps is None
+
+    def test_invalid_provider_raises(self) -> None:
+        with pytest.raises(ValueError, match="Unknown provider"):
+            lookup_model_capabilities("bad-provider", "gpt-5")
+
+
+# ---------------------------------------------------------------------------
+# list_known_models
+# ---------------------------------------------------------------------------
+
+
+class TestListKnownModels:
+    def test_openai_models(self) -> None:
+        models = list_known_models("openai")
+        assert "gpt-5" in models
+        assert isinstance(models, list)
+        assert models == sorted(models)
+
+    def test_anthropic_models(self) -> None:
+        models = list_known_models("anthropic")
+        assert "claude-opus-4-6" in models
+
+    def test_openai_compatible_returns_empty(self) -> None:
+        assert list_known_models("openai-compatible") == []
+
+    def test_unknown_provider_returns_empty(self) -> None:
+        assert list_known_models("bad-provider") == []

tests/test_model_registry.py

@@ -10,6 +10,7 @@ import pytest
 from turnstone.core.model_registry import (
     ModelConfig,
     ModelRegistry,
+    _resolve_env_vars,
     detect_model,
     load_model_registry,
 )
@@ -319,6 +320,280 @@ class TestLoadModelRegistry:
         assert alt_cfg.api_key == "my-key"
 
 
+# ---------------------------------------------------------------------------
+# load_model_registry with DB storage
+# ---------------------------------------------------------------------------
+
+
+class _MockStorage:
+    """Minimal storage mock returning canned model definitions."""
+
+    def __init__(self, rows: list[dict[str, Any]] | None = None) -> None:
+        self._rows = rows or []
+        self.calls: list[str] = []
+
+    def list_model_definitions(self, enabled_only: bool = False) -> list[dict[str, Any]]:
+        self.calls.append("list_model_definitions")
+        if enabled_only:
+            return [r for r in self._rows if r.get("enabled", True)]
+        return list(self._rows)
+
+
+class TestLoadModelRegistryWithDB:
+    def test_db_models_loaded(self) -> None:
+        """DB model definitions are loaded into the registry."""
+        storage = _MockStorage(
+            [
+                {
+                    "alias": "cloud-gpt",
+                    "model": "gpt-5",
+                    "provider": "openai",
+                    "base_url": "https://api.openai.com/v1",
+                    "api_key": "sk-db",
+                    "context_window": 128000,
+                    "capabilities": "{}",
+                    "enabled": True,
+                }
+            ]
+        )
+        with patch("turnstone.core.model_registry.load_config", return_value={}):
+            reg = load_model_registry("http://x/v1", "x", "x", storage=storage)
+        assert reg.has_alias("cloud-gpt")
+        cfg = reg.get_config("cloud-gpt")
+        assert cfg.model == "gpt-5"
+        assert cfg.source == "db"
+
+    def test_config_overrides_db(self) -> None:
+        """Config.toml entry overrides DB entry with same alias."""
+        storage = _MockStorage(
+            [
+                {
+                    "alias": "shared",
+                    "model": "db-model",
+                    "provider": "openai",
+                    "base_url": "http://db/v1",
+                    "api_key": "sk-db",
+                    "context_window": 32768,
+                    "capabilities": "{}",
+                    "enabled": True,
+                }
+            ]
+        )
+        fake_cfg: dict[str, Any] = {
+            "models": {
+                "shared": {
+                    "model": "config-model",
+                    "base_url": "http://config/v1",
+                },
+            },
+        }
+        with patch("turnstone.core.model_registry.load_config", return_value=fake_cfg):
+            reg = load_model_registry("http://x/v1", "x", "x", storage=storage)
+        cfg = reg.get_config("shared")
+        assert cfg.model == "config-model"
+        assert cfg.source == "config"
+
+    def test_db_only_models_coexist(self) -> None:
+        """DB models coexist alongside config.toml models."""
+        storage = _MockStorage(
+            [
+                {
+                    "alias": "db-only",
+                    "model": "db-model",
+                    "provider": "anthropic",
+                    "base_url": "",
+                    "api_key": "sk-db",
+                    "context_window": 200000,
+                    "capabilities": "{}",
+                    "enabled": True,
+                }
+            ]
+        )
+        fake_cfg: dict[str, Any] = {
+            "models": {
+                "config-only": {"model": "config-model"},
+            },
+        }
+        with patch("turnstone.core.model_registry.load_config", return_value=fake_cfg):
+            reg = load_model_registry("http://x/v1", "x", "x", storage=storage)
+        assert reg.has_alias("db-only")
+        assert reg.has_alias("config-only")
+        assert reg.has_alias("default")
+        assert reg.get_config("db-only").source == "db"
+        assert reg.get_config("config-only").source == "config"
+
+    def test_source_field_set(self) -> None:
+        """Source field correctly distinguishes origin."""
+        storage = _MockStorage(
+            [
+                {
+                    "alias": "from-db",
+                    "model": "m",
+                    "provider": "openai",
+                    "base_url": "",
+                    "api_key": "",
+                    "context_window": 32768,
+                    "capabilities": "{}",
+                    "enabled": True,
+                }
+            ]
+        )
+        with patch("turnstone.core.model_registry.load_config", return_value={}):
+            reg = load_model_registry("http://x/v1", "x", "x", storage=storage)
+        assert reg.get_config("from-db").source == "db"
+        assert reg.get_config("default").source == ""
+
+    def test_disabled_db_models_excluded(self) -> None:
+        """Disabled DB models are not loaded."""
+        storage = _MockStorage(
+            [
+                {
+                    "alias": "disabled",
+                    "model": "m",
+                    "provider": "openai",
+                    "base_url": "",
+                    "api_key": "",
+                    "context_window": 32768,
+                    "capabilities": "{}",
+                    "enabled": False,
+                }
+            ]
+        )
+        with patch("turnstone.core.model_registry.load_config", return_value={}):
+            reg = load_model_registry("http://x/v1", "x", "x", storage=storage)
+        assert not reg.has_alias("disabled")
+
+    def test_db_capabilities_parsed(self) -> None:
+        """JSON capabilities from DB are parsed into dict."""
+        storage = _MockStorage(
+            [
+                {
+                    "alias": "caps-model",
+                    "model": "m",
+                    "provider": "openai",
+                    "base_url": "",
+                    "api_key": "",
+                    "context_window": 32768,
+                    "capabilities": '{"supports_vision": true}',
+                    "enabled": True,
+                }
+            ]
+        )
+        with patch("turnstone.core.model_registry.load_config", return_value={}):
+            reg = load_model_registry("http://x/v1", "x", "x", storage=storage)
+        assert reg.get_config("caps-model").capabilities == {"supports_vision": True}
+
+    def test_db_default_alias_not_clobbered(self) -> None:
+        """DB model with alias='default' is not overwritten by CLI args."""
+        storage = _MockStorage(
+            [
+                {
+                    "alias": "default",
+                    "model": "db-default-model",
+                    "provider": "openai",
+                    "base_url": "http://db/v1",
+                    "api_key": "sk-db",
+                    "context_window": 128000,
+                    "capabilities": "{}",
+                    "enabled": True,
+                }
+            ]
+        )
+        with patch("turnstone.core.model_registry.load_config", return_value={}):
+            reg = load_model_registry("http://cli/v1", "cli-key", "cli-model", storage=storage)
+        cfg = reg.get_config("default")
+        assert cfg.model == "db-default-model"
+        assert cfg.source == "db"
+
+    def test_no_db_writes(self) -> None:
+        """Config.toml models are NOT written to storage."""
+        storage = _MockStorage()
+        fake_cfg: dict[str, Any] = {
+            "models": {"local": {"model": "llama"}},
+        }
+        with patch("turnstone.core.model_registry.load_config", return_value=fake_cfg):
+            load_model_registry("http://x/v1", "x", "x", storage=storage)
+        # Only list_model_definitions should be called, no create
+        assert storage.calls == ["list_model_definitions"]
+
+    def test_storage_failure_graceful(self) -> None:
+        """Storage errors don't prevent registry creation."""
+        storage = MagicMock()
+        storage.list_model_definitions.side_effect = RuntimeError("db down")
+        with patch("turnstone.core.model_registry.load_config", return_value={}):
+            reg = load_model_registry("http://x/v1", "x", "x", storage=storage)
+        assert reg.has_alias("default")
+
+
+# ---------------------------------------------------------------------------
+# _resolve_env_vars
+# ---------------------------------------------------------------------------
+
+
+class TestResolveEnvVars:
+    def test_expand_single(self) -> None:
+        with patch.dict("os.environ", {"MY_KEY": "secret123"}):
+            assert _resolve_env_vars("sk-${MY_KEY}") == "sk-secret123"
+
+    def test_expand_multiple(self) -> None:
+        with patch.dict("os.environ", {"A": "1", "B": "2"}):
+            assert _resolve_env_vars("${A}-${B}") == "1-2"
+
+    def test_missing_var_empty(self) -> None:
+        with patch.dict("os.environ", {}, clear=True):
+            assert _resolve_env_vars("${MISSING}") == ""
+
+    def test_no_vars(self) -> None:
+        assert _resolve_env_vars("plain-key") == "plain-key"
+
+    def test_empty_string(self) -> None:
+        assert _resolve_env_vars("") == ""
+
+
+# ---------------------------------------------------------------------------
+# ModelRegistry.reload
+# ---------------------------------------------------------------------------
+
+
+class TestRegistryReload:
+    def test_reload_replaces_models(self) -> None:
+        models_a = {"a": ModelConfig("a", "x", "x", "m1")}
+        reg = ModelRegistry(models=models_a, default="a")
+        assert reg.has_alias("a")
+
+        models_b = {"b": ModelConfig("b", "y", "y", "m2")}
+        reg.reload(models_b, "b")
+        assert not reg.has_alias("a")
+        assert reg.has_alias("b")
+        assert reg.default == "b"
+
+    def test_reload_clears_clients(self) -> None:
+        models = {"a": ModelConfig("a", "http://x/v1", "key", "m")}
+        reg = ModelRegistry(models=models, default="a")
+        # Force client creation
+        reg.get_client("a")
+        assert "a" in reg._clients
+
+        # Reload with same models — clients should be cleared
+        reg.reload(dict(models), "a")
+        assert "a" not in reg._clients
+
+    def test_reload_validates_default(self) -> None:
+        models_a = {"a": ModelConfig("a", "x", "x", "m")}
+        reg = ModelRegistry(models=models_a, default="a")
+        with pytest.raises(ValueError, match="Default model"):
+            reg.reload(models_a, "nonexistent")
+        # Registry should be unchanged after failed reload
+        assert reg.has_alias("a")
+        assert reg.default == "a"
+
+    def test_reload_validates_empty(self) -> None:
+        models_a = {"a": ModelConfig("a", "x", "x", "m")}
+        reg = ModelRegistry(models=models_a, default="a")
+        with pytest.raises(ValueError, match="at least one"):
+            reg.reload({}, "a")
+
+
 # ---------------------------------------------------------------------------
 # Session integration
 # ---------------------------------------------------------------------------
@@ -339,7 +614,7 @@ class _FakeUI:
     def approve_tools(self, items: list[dict[str, Any]]) -> tuple[bool, str | None]:
         return True, None
 
-    def on_tool_result(self, call_id: str, name: str, output: str) -> None: ...
+    def on_tool_result(self, call_id: str, name: str, output: str, **kwargs: Any) -> None: ...
     def on_tool_output_chunk(self, call_id: str, chunk: str) -> None: ...
     def on_status(self, usage: dict[str, Any], context_window: int, effort: str) -> None: ...
     def on_plan_review(self, content: str) -> str:

tests/test_prompt_templates_runtime.py

@@ -29,7 +29,7 @@ class NullUI:
     def approve_tools(self, items):
         return True, None
 
-    def on_tool_result(self, call_id, name, output):
+    def on_tool_result(self, call_id, name, output, **kwargs):
         pass
 
     def on_tool_output_chunk(self, call_id, chunk):

tests/test_providers.py

@@ -94,6 +94,7 @@ def _anthropic_event(
         delta.type = kwargs.get("delta_type", "text_delta")
         delta.text = kwargs.get("text", "")
         delta.thinking = kwargs.get("thinking", "")
+        delta.signature = kwargs.get("signature", "")
         delta.partial_json = kwargs.get("partial_json", "")
         event.delta = delta
         event.index = kwargs.get("index", 0)
@@ -140,6 +141,34 @@ class TestOpenAIProvider:
     def test_provider_name(self) -> None:
         assert self.provider.provider_name == "openai"
 
+    # -- _sanitize_messages ---------------------------------------------------
+
+    def test_sanitize_messages_none_content_no_tool_calls(self) -> None:
+        msgs = [{"role": "assistant", "content": None}]
+        assert self.provider._sanitize_messages(msgs) == [{"role": "assistant", "content": ""}]
+
+    def test_sanitize_messages_none_content_with_tool_calls(self) -> None:
+        msgs = [{"role": "assistant", "content": None, "tool_calls": [{"id": "1"}]}]
+        result = self.provider._sanitize_messages(msgs)
+        assert result[0]["content"] is None
+        assert result[0]["tool_calls"] == [{"id": "1"}]
+
+    def test_sanitize_messages_empty_string_passthrough(self) -> None:
+        msgs = [{"role": "assistant", "content": ""}]
+        assert self.provider._sanitize_messages(msgs) == msgs
+
+    def test_sanitize_messages_non_assistant_unchanged(self) -> None:
+        msgs = [{"role": "user", "content": None}]
+        result = self.provider._sanitize_messages(msgs)
+        assert result[0]["content"] is None
+
+    def test_sanitize_messages_does_not_mutate_original(self) -> None:
+        original = {"role": "assistant", "content": None}
+        self.provider._sanitize_messages([original])
+        assert original["content"] is None
+
+    # -- convert_tools --------------------------------------------------------
+
     def test_convert_tools_passthrough(self) -> None:
         tools = [
             {
@@ -479,10 +508,11 @@ class TestAnthropicProvider:
                         },
                     }
                 ],
-            }
+            },
+            {"role": "tool", "tool_call_id": "call_1", "content": "file contents"},
         ]
         _, converted = self.provider._convert_messages(messages)
-        assert len(converted) == 1
+        assert len(converted) == 2
         blocks = converted[0]["content"]
         assert len(blocks) == 2
         assert blocks[0] == {"type": "text", "text": "Let me check that."}
@@ -490,6 +520,8 @@ class TestAnthropicProvider:
         assert blocks[1]["id"] == "call_1"
         assert blocks[1]["name"] == "read_file"
         assert blocks[1]["input"] == {"path": "foo.py"}
+        # Tool result in user message
+        assert converted[1]["role"] == "user"
 
     def test_message_conversion_tool_results(self) -> None:
         messages = [
@@ -598,7 +630,11 @@ class TestAnthropicProvider:
         response.usage.output_tokens = 5
 
         client = MagicMock()
-        client.messages.create.return_value = response
+        stream_ctx = MagicMock()
+        stream_ctx.__enter__ = MagicMock(return_value=stream_ctx)
+        stream_ctx.__exit__ = MagicMock(return_value=False)
+        stream_ctx.get_final_message.return_value = response
+        client.messages.stream.return_value = stream_ctx
 
         result = self.provider.create_completion(
             client=client,
@@ -630,7 +666,11 @@ class TestAnthropicProvider:
         response.usage.output_tokens = 20
 
         client = MagicMock()
-        client.messages.create.return_value = response
+        stream_ctx = MagicMock()
+        stream_ctx.__enter__ = MagicMock(return_value=stream_ctx)
+        stream_ctx.__exit__ = MagicMock(return_value=False)
+        stream_ctx.get_final_message.return_value = response
+        client.messages.stream.return_value = stream_ctx
 
         result = self.provider.create_completion(
             client=client,
@@ -661,7 +701,11 @@ class TestAnthropicProvider:
         response.usage.output_tokens = 50
 
         client = MagicMock()
-        client.messages.create.return_value = response
+        stream_ctx = MagicMock()
+        stream_ctx.__enter__ = MagicMock(return_value=stream_ctx)
+        stream_ctx.__exit__ = MagicMock(return_value=False)
+        stream_ctx.get_final_message.return_value = response
+        client.messages.stream.return_value = stream_ctx
 
         result = self.provider.create_completion(
             client=client,
@@ -1141,6 +1185,189 @@ class TestOpenAIParameterGating:
         assert kwargs["reasoning_effort"] == "medium"  # fell back from unsupported "low"
 
 
+class TestAnthropicOrphanedToolUse:
+    """Verify _convert_messages synthesizes tool_results for orphaned tool_use."""
+
+    def setup_method(self) -> None:
+        from turnstone.core.providers._anthropic import AnthropicProvider
+
+        self.provider = AnthropicProvider()
+
+    def test_orphaned_tool_use_gets_synthetic_result(self) -> None:
+        """Assistant has tool_calls but next message is user (no tool results)."""
+        messages = [
+            {"role": "user", "content": "do something"},
+            {
+                "role": "assistant",
+                "content": "I'll run that.",
+                "tool_calls": [
+                    {
+                        "id": "call_abc",
+                        "function": {"name": "bash", "arguments": '{"command": "ls"}'},
+                    }
+                ],
+            },
+            {"role": "user", "content": "never mind, do something else"},
+        ]
+        _, converted = self.provider._convert_messages(messages)
+        # Should have: user, assistant(tool_use), user(synthetic tool_result), user
+        # After _merge_consecutive, the two user messages may merge.
+        # Find the synthetic tool_result
+        tool_results = []
+        for msg in converted:
+            if msg["role"] == "user" and isinstance(msg["content"], list):
+                for block in msg["content"]:
+                    if isinstance(block, dict) and block.get("type") == "tool_result":
+                        tool_results.append(block)
+        assert len(tool_results) == 1
+        assert tool_results[0]["tool_use_id"] == "call_abc"
+        assert tool_results[0]["is_error"] is True
+        assert "cancelled" in tool_results[0]["content"].lower()
+
+    def test_multiple_orphaned_tool_calls(self) -> None:
+        """Assistant has 3 tool_calls, none have results."""
+        messages = [
+            {"role": "user", "content": "do three things"},
+            {
+                "role": "assistant",
+                "content": "",
+                "tool_calls": [
+                    {"id": "c1", "function": {"name": "bash", "arguments": "{}"}},
+                    {"id": "c2", "function": {"name": "read_file", "arguments": "{}"}},
+                    {"id": "c3", "function": {"name": "write_file", "arguments": "{}"}},
+                ],
+            },
+            {"role": "user", "content": "skip all that"},
+        ]
+        _, converted = self.provider._convert_messages(messages)
+        tool_results = []
+        for msg in converted:
+            if msg["role"] == "user" and isinstance(msg["content"], list):
+                for block in msg["content"]:
+                    if isinstance(block, dict) and block.get("type") == "tool_result":
+                        tool_results.append(block)
+        assert len(tool_results) == 3
+        result_ids = {r["tool_use_id"] for r in tool_results}
+        assert result_ids == {"c1", "c2", "c3"}
+
+    def test_partial_results_only_orphans_synthesized(self) -> None:
+        """2 tool_calls, only 1 has a result — synthesize for the missing one."""
+        messages = [
+            {"role": "user", "content": "do two things"},
+            {
+                "role": "assistant",
+                "content": "",
+                "tool_calls": [
+                    {"id": "c1", "function": {"name": "bash", "arguments": "{}"}},
+                    {"id": "c2", "function": {"name": "write_file", "arguments": "{}"}},
+                ],
+            },
+            {"role": "tool", "tool_call_id": "c1", "content": "file1.txt"},
+            {"role": "user", "content": "skip the write"},
+        ]
+        _, converted = self.provider._convert_messages(messages)
+        # c1 should have a real result, c2 should have a synthetic one
+        tool_results = []
+        for msg in converted:
+            if msg["role"] == "user" and isinstance(msg["content"], list):
+                for block in msg["content"]:
+                    if isinstance(block, dict) and block.get("type") == "tool_result":
+                        tool_results.append(block)
+        # Real result should come before synthetic (ordering matters for Anthropic)
+        assert len(tool_results) == 2
+        assert tool_results[0]["tool_use_id"] == "c1"
+        assert tool_results[0]["content"] == "file1.txt"  # real result
+        assert tool_results[0].get("is_error") is not True
+        assert tool_results[1]["tool_use_id"] == "c2"
+        assert tool_results[1]["is_error"] is True  # synthetic
+
+    def test_complete_results_no_synthesis(self) -> None:
+        """All tool_calls have results — no synthesis needed."""
+        messages = [
+            {"role": "user", "content": "do it"},
+            {
+                "role": "assistant",
+                "content": "",
+                "tool_calls": [
+                    {"id": "c1", "function": {"name": "bash", "arguments": "{}"}},
+                ],
+            },
+            {"role": "tool", "tool_call_id": "c1", "content": "done"},
+            {"role": "user", "content": "thanks"},
+        ]
+        _, converted = self.provider._convert_messages(messages)
+        # No synthetic results — only the real one (no is_error flag)
+        tool_results = []
+        for msg in converted:
+            if msg["role"] == "user" and isinstance(msg["content"], list):
+                for block in msg["content"]:
+                    if isinstance(block, dict) and block.get("type") == "tool_result":
+                        tool_results.append(block)
+        assert len(tool_results) == 1
+        assert tool_results[0]["tool_use_id"] == "c1"
+        assert tool_results[0].get("is_error") is not True
+
+    def test_trailing_orphan(self) -> None:
+        """Orphaned tool_use at end of conversation (no following messages)."""
+        messages = [
+            {"role": "user", "content": "do it"},
+            {
+                "role": "assistant",
+                "content": "Running...",
+                "tool_calls": [
+                    {"id": "c1", "function": {"name": "bash", "arguments": "{}"}},
+                ],
+            },
+        ]
+        _, converted = self.provider._convert_messages(messages)
+        tool_results = []
+        for msg in converted:
+            if msg["role"] == "user" and isinstance(msg["content"], list):
+                for block in msg["content"]:
+                    if isinstance(block, dict) and block.get("type") == "tool_result":
+                        tool_results.append(block)
+        assert len(tool_results) == 1
+        assert tool_results[0]["tool_use_id"] == "c1"
+        assert tool_results[0]["is_error"] is True
+
+    def test_provider_content_orphan(self) -> None:
+        """Orphaned tool_use inside _provider_content (Anthropic raw blocks)."""
+        messages = [
+            {"role": "user", "content": "run something"},
+            {
+                "role": "assistant",
+                "content": "Running...",
+                "_provider_content": [
+                    {"type": "text", "text": "Running..."},
+                    {
+                        "type": "tool_use",
+                        "id": "toolu_abc",
+                        "name": "bash",
+                        "input": {"command": "sleep 30"},
+                    },
+                ],
+                "tool_calls": [
+                    {
+                        "id": "toolu_abc",
+                        "function": {"name": "bash", "arguments": '{"command": "sleep 30"}'},
+                    },
+                ],
+            },
+            {"role": "user", "content": "never mind"},
+        ]
+        _, converted = self.provider._convert_messages(messages)
+        # Should synthesize a tool_result for the orphaned tool_use in provider_content
+        tool_results = []
+        for msg in converted:
+            if msg["role"] == "user" and isinstance(msg["content"], list):
+                for block in msg["content"]:
+                    if isinstance(block, dict) and block.get("type") == "tool_result":
+                        tool_results.append(block)
+        assert len(tool_results) == 1
+        assert tool_results[0]["tool_use_id"] == "toolu_abc"
+        assert tool_results[0]["is_error"] is True
+
+
 class TestAnthropicReasoningNone:
     """Verify 'none' effort disables thinking for manual-thinking models."""
 
@@ -1383,7 +1610,11 @@ class TestAnthropicWebSearch:
         response.usage.output_tokens = 50
 
         client = MagicMock()
-        client.messages.create.return_value = response
+        stream_ctx = MagicMock()
+        stream_ctx.__enter__ = MagicMock(return_value=stream_ctx)
+        stream_ctx.__exit__ = MagicMock(return_value=False)
+        stream_ctx.get_final_message.return_value = response
+        client.messages.stream.return_value = stream_ctx
 
         with patch("turnstone.core.providers._anthropic._ensure_anthropic"):
             result = self.provider.create_completion(
@@ -1934,6 +2165,104 @@ class TestAnthropicProviderBlocks:
         assert blocks[2]["type"] == "web_search_tool_result"
         assert blocks[2]["encrypted_content"] == "enc_data"
 
+    def test_streaming_thinking_block_captures_signature(self) -> None:
+        """Streaming thinking block accumulates signature from signature_delta events."""
+        thinking_block = MagicMock()
+        thinking_block.type = "thinking"
+        thinking_block.model_dump.return_value = {
+            "type": "thinking",
+            "thinking": "",
+            "signature": "",
+        }
+
+        text_block = MagicMock()
+        text_block.type = "text"
+        text_block.model_dump.return_value = {"type": "text", "text": ""}
+
+        events = [
+            MagicMock(type="content_block_start", index=0, content_block=thinking_block),
+            _anthropic_event(
+                "content_block_delta", delta_type="thinking_delta", thinking="step 1", index=0
+            ),
+            _anthropic_event(
+                "content_block_delta", delta_type="thinking_delta", thinking=" step 2", index=0
+            ),
+            _anthropic_event(
+                "content_block_delta",
+                delta_type="signature_delta",
+                signature="sig_part1",
+                index=0,
+            ),
+            _anthropic_event(
+                "content_block_delta",
+                delta_type="signature_delta",
+                signature="sig_part2",
+                index=0,
+            ),
+            _anthropic_event("content_block_stop", index=0),
+            MagicMock(type="content_block_start", index=1, content_block=text_block),
+            _anthropic_event("content_block_delta", delta_type="text_delta", text="Hello", index=1),
+            _anthropic_event("content_block_stop", index=1),
+            _anthropic_event("message_delta", stop_reason="end_turn", usage_output_tokens=50),
+        ]
+
+        chunks = list(self.provider._iter_anthropic_stream(iter(events)))
+        final_chunks = [c for c in chunks if c.provider_blocks]
+        assert len(final_chunks) == 1
+        blocks = final_chunks[0].provider_blocks
+        assert blocks[0]["type"] == "thinking"
+        assert blocks[0]["thinking"] == "step 1 step 2"
+        assert blocks[0]["signature"] == "sig_part1sig_part2"
+
+    def test_thinking_block_multiturn_roundtrip(self) -> None:
+        """Thinking block with signature survives _convert_messages round-trip."""
+        provider_content = [
+            {
+                "type": "thinking",
+                "thinking": "Let me reason...",
+                "signature": "ErUBCkYIAxgCIkD_valid_sig",
+            },
+            {"type": "text", "text": "Here is my answer."},
+        ]
+        messages = [
+            {"role": "user", "content": "Question"},
+            {
+                "role": "assistant",
+                "content": "Here is my answer.",
+                "_provider_content": provider_content,
+            },
+            {"role": "user", "content": "Follow up"},
+        ]
+        _, converted = self.provider._convert_messages(messages)
+        assistant_msg = converted[1]
+        assert assistant_msg["content"] is provider_content
+        assert assistant_msg["content"][0]["signature"] == "ErUBCkYIAxgCIkD_valid_sig"
+        assert assistant_msg["content"][0]["type"] == "thinking"
+
+    def test_block_to_dict_preserves_thinking_signature(self) -> None:
+        """_block_to_dict preserves signature on thinking blocks."""
+        from turnstone.core.providers._anthropic import _block_to_dict
+
+        class FakeThinkingBlock:
+            def model_dump(self, **kwargs: Any) -> dict[str, Any]:
+                return {
+                    "type": "thinking",
+                    "thinking": "reasoning...",
+                    "signature": "abc123sig",
+                }
+
+        result = _block_to_dict(FakeThinkingBlock())
+        assert result["signature"] == "abc123sig"
+
+        # Also test fallback path (no model_dump)
+        class FallbackBlock:
+            type = "thinking"
+            thinking = "reasoning..."
+            signature = "abc123sig"
+
+        result2 = _block_to_dict(FallbackBlock())
+        assert result2["signature"] == "abc123sig"
+
 
 # ---------------------------------------------------------------------------
 # Tool search tests
@@ -2307,7 +2636,11 @@ class TestAnthropicPromptCaching:
         response.usage = usage
 
         client = MagicMock()
-        client.messages.create.return_value = response
+        stream_ctx = MagicMock()
+        stream_ctx.__enter__ = MagicMock(return_value=stream_ctx)
+        stream_ctx.__exit__ = MagicMock(return_value=False)
+        stream_ctx.get_final_message.return_value = response
+        client.messages.stream.return_value = stream_ctx
 
         result = self.provider.create_completion(
             client=client,

tests/test_reconstruct_messages.py

@@ -9,13 +9,12 @@ def _row(
     role,
     content=None,
     tool_name=None,
-    tool_args=None,
     tc_id=None,
     pdata=None,
     tool_calls=None,
 ):
-    """Build a 7-element conversation row tuple (post-migration 013 format)."""
-    return (role, content, tool_name, tool_args, tc_id, pdata, tool_calls)
+    """Build a 6-element conversation row tuple (post-migration 027 format)."""
+    return (role, content, tool_name, tc_id, pdata, tool_calls)
 
 
 class TestAssistantWithToolCalls:
@@ -227,3 +226,91 @@ class TestEdgeCases:
         assert len(msgs) == 2
         assert msgs[0]["role"] == "user"
         assert msgs[1]["role"] == "assistant"
+
+
+class TestMidConversationOrphanRepair:
+    """Mid-conversation orphaned tool_calls get synthetic tool results."""
+
+    def test_all_orphaned_mid_conversation(self):
+        """Assistant has 2 tool_calls, no tool results, then user message."""
+        tc = json.dumps(
+            [
+                {"id": "c1", "function": {"name": "bash", "arguments": "{}"}},
+                {"id": "c2", "function": {"name": "write_file", "arguments": "{}"}},
+            ]
+        )
+        rows = [
+            _row("user", "do stuff"),
+            _row("assistant", "Running...", tool_calls=tc),
+            _row("user", "never mind"),
+        ]
+        msgs = reconstruct_messages(rows, "ws1")
+        # Should have: user, assistant, tool(c1), tool(c2), user
+        assert len(msgs) == 5
+        assert msgs[2]["role"] == "tool"
+        assert msgs[2]["tool_call_id"] == "c1"
+        assert msgs[2]["is_error"] is True
+        assert msgs[3]["role"] == "tool"
+        assert msgs[3]["tool_call_id"] == "c2"
+        assert msgs[4]["role"] == "user"
+
+    def test_partial_results_mid_conversation(self):
+        """2 tool_calls, 1 result present, 1 missing — synthesize only the missing one."""
+        tc = json.dumps(
+            [
+                {"id": "c1", "function": {"name": "bash", "arguments": "{}"}},
+                {"id": "c2", "function": {"name": "write_file", "arguments": "{}"}},
+            ]
+        )
+        rows = [
+            _row("user", "do stuff"),
+            _row("assistant", "", tool_calls=tc),
+            _row("tool", "file1.txt", tool_name="bash", tc_id="c1"),
+            _row("user", "skip the write"),
+        ]
+        msgs = reconstruct_messages(rows, "ws1")
+        # Should have: user, assistant, tool(c1 real), tool(c2 synthetic), user
+        assert len(msgs) == 5
+        assert msgs[2]["role"] == "tool"
+        assert msgs[2]["tool_call_id"] == "c1"
+        assert msgs[2]["content"] == "file1.txt"
+        assert msgs[2].get("is_error") is not True
+        assert msgs[3]["role"] == "tool"
+        assert msgs[3]["tool_call_id"] == "c2"
+        assert msgs[3]["is_error"] is True
+        assert msgs[4]["role"] == "user"
+
+    def test_complete_results_no_synthesis(self):
+        """All tool_calls have results — no synthesis needed."""
+        tc = json.dumps(
+            [
+                {"id": "c1", "function": {"name": "bash", "arguments": "{}"}},
+            ]
+        )
+        rows = [
+            _row("user", "do it"),
+            _row("assistant", "", tool_calls=tc),
+            _row("tool", "done", tool_name="bash", tc_id="c1"),
+            _row("user", "thanks"),
+        ]
+        msgs = reconstruct_messages(rows, "ws1")
+        assert len(msgs) == 4
+        tool_msgs = [m for m in msgs if m["role"] == "tool"]
+        assert len(tool_msgs) == 1
+        assert tool_msgs[0].get("is_error") is not True
+
+    def test_trailing_orphan_stripped_not_synthesized(self):
+        """Trailing orphan is handled by the existing strip repair, not synthesis."""
+        tc = json.dumps(
+            [
+                {"id": "c1", "function": {"name": "bash", "arguments": "{}"}},
+            ]
+        )
+        rows = [
+            _row("user", "do it"),
+            _row("assistant", "Running...", tool_calls=tc),
+        ]
+        msgs = reconstruct_messages(rows, "ws1")
+        # Trailing strip removes the assistant message entirely
+        assert len(msgs) == 1
+        assert msgs[0]["role"] == "user"

tests/test_rewind_retry.py

@@ -0,0 +1,387 @@
+"""Tests for conversation rewind and retry functionality."""
+
+from __future__ import annotations
+
+from unittest.mock import MagicMock
+
+from turnstone.core.session import ChatSession
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+
+class NullUI:
+    """UI adapter that discards all output."""
+
+    def on_thinking_start(self):
+        pass
+
+    def on_thinking_stop(self):
+        pass
+
+    def on_reasoning_token(self, text):
+        pass
+
+    def on_content_token(self, text):
+        pass
+
+    def on_stream_end(self):
+        pass
+
+    def approve_tools(self, items):
+        return True, None
+
+    def on_tool_result(self, call_id, name, output, **kwargs):
+        pass
+
+    def on_tool_output_chunk(self, call_id, chunk):
+        pass
+
+    def on_status(self, usage, context_window, effort):
+        pass
+
+    def on_plan_review(self, content):
+        return ""
+
+    def on_info(self, message):
+        pass
+
+    def on_error(self, message):
+        pass
+
+    def on_state_change(self, state):
+        pass
+
+    def on_rename(self, name):
+        pass
+
+    def on_output_warning(self, call_id, assessment):
+        pass
+
+
+def _make_session(tmp_db) -> ChatSession:
+    return ChatSession(
+        client=MagicMock(),
+        model="test-model",
+        ui=NullUI(),
+        instructions="",
+        temperature=0.5,
+        max_tokens=4096,
+        tool_timeout=30,
+    )
+
+
+def _populate_simple(session: ChatSession) -> None:
+    """Populate with 2 simple turns (no tool calls)."""
+    session.messages = [
+        {"role": "user", "content": "Hello"},
+        {"role": "assistant", "content": "Hi there!"},
+        {"role": "user", "content": "How are you?"},
+        {"role": "assistant", "content": "I'm fine."},
+    ]
+    session._msg_tokens = [10, 20, 10, 20]
+
+
+def _populate_with_tools(session: ChatSession) -> None:
+    """Populate with 2 turns, first has tool calls."""
+    session.messages = [
+        {"role": "user", "content": "Write a test"},
+        {
+            "role": "assistant",
+            "content": "",
+            "tool_calls": [
+                {"id": "tc1", "function": {"name": "bash", "arguments": '{"cmd":"echo hi"}'}}
+            ],
+        },
+        {"role": "tool", "tool_call_id": "tc1", "content": "hi"},
+        {"role": "assistant", "content": "Done."},
+        {"role": "user", "content": "Fix the import"},
+        {"role": "assistant", "content": "Fixed."},
+    ]
+    session._msg_tokens = [10, 20, 10, 20, 10, 20]
+
+
+# ---------------------------------------------------------------------------
+# _find_turn_boundaries
+# ---------------------------------------------------------------------------
+
+
+class TestFindTurnBoundaries:
+    def test_empty_messages(self, tmp_db):
+        session = _make_session(tmp_db)
+        assert session._find_turn_boundaries() == []
+
+    def test_single_turn(self, tmp_db):
+        session = _make_session(tmp_db)
+        session.messages = [
+            {"role": "user", "content": "Hello"},
+            {"role": "assistant", "content": "Hi!"},
+        ]
+        assert session._find_turn_boundaries() == [0]
+
+    def test_multi_turn(self, tmp_db):
+        session = _make_session(tmp_db)
+        _populate_simple(session)
+        assert session._find_turn_boundaries() == [0, 2]
+
+    def test_with_tool_calls(self, tmp_db):
+        session = _make_session(tmp_db)
+        _populate_with_tools(session)
+        assert session._find_turn_boundaries() == [0, 4]
+
+
+# ---------------------------------------------------------------------------
+# rewind
+# ---------------------------------------------------------------------------
+
+
+class TestRewind:
+    def test_rewind_zero(self, tmp_db):
+        session = _make_session(tmp_db)
+        _populate_simple(session)
+        assert session.rewind(0) == 0
+        assert len(session.messages) == 4
+
+    def test_rewind_one_turn(self, tmp_db):
+        session = _make_session(tmp_db)
+        _populate_simple(session)
+        removed = session.rewind(1)
+        assert removed == 2  # user + assistant
+        assert len(session.messages) == 2
+        assert session.messages[0]["content"] == "Hello"
+        assert session.messages[1]["content"] == "Hi there!"
+        assert len(session._msg_tokens) == 2
+
+    def test_rewind_all_turns(self, tmp_db):
+        session = _make_session(tmp_db)
+        _populate_simple(session)
+        removed = session.rewind(2)
+        assert removed == 4
+        assert len(session.messages) == 0
+        assert len(session._msg_tokens) == 0
+
+    def test_rewind_clamped(self, tmp_db):
+        """Rewinding more turns than exist should clamp to available."""
+        session = _make_session(tmp_db)
+        _populate_simple(session)
+        removed = session.rewind(999)
+        assert removed == 4
+        assert len(session.messages) == 0
+
+    def test_rewind_empty(self, tmp_db):
+        session = _make_session(tmp_db)
+        assert session.rewind(1) == 0
+
+    def test_rewind_with_tools(self, tmp_db):
+        """Rewinding 1 turn on a multi-sub-turn conversation."""
+        session = _make_session(tmp_db)
+        _populate_with_tools(session)
+        removed = session.rewind(1)
+        assert removed == 2  # user "Fix the import" + assistant "Fixed."
+        assert len(session.messages) == 4
+        assert session.messages[-1]["content"] == "Done."
+
+    def test_rewind_tokens_sync(self, tmp_db):
+        """_msg_tokens stays in sync with messages."""
+        session = _make_session(tmp_db)
+        _populate_simple(session)
+        session.rewind(1)
+        assert len(session._msg_tokens) == len(session.messages)
+
+
+# ---------------------------------------------------------------------------
+# retry
+# ---------------------------------------------------------------------------
+
+
+class TestRetry:
+    def test_retry_returns_user_message(self, tmp_db):
+        session = _make_session(tmp_db)
+        _populate_simple(session)
+        msg = session.retry()
+        assert msg == "How are you?"
+        # Only Turn 1 remains, without the second user message
+        assert len(session.messages) == 2
+        assert session.messages[-1]["content"] == "Hi there!"
+
+    def test_retry_empty(self, tmp_db):
+        session = _make_session(tmp_db)
+        assert session.retry() is None
+
+    def test_retry_with_tools(self, tmp_db):
+        session = _make_session(tmp_db)
+        _populate_with_tools(session)
+        msg = session.retry()
+        assert msg == "Fix the import"
+        # Only Turn 1 remains (user + assistant w/tools + tool result + assistant)
+        assert len(session.messages) == 4
+
+    def test_retry_sets_pending(self, tmp_db):
+        """handle_command for /retry should set _pending_retry."""
+        session = _make_session(tmp_db)
+        _populate_simple(session)
+        session.handle_command("/retry")
+        assert session._pending_retry == "How are you?"
+
+    def test_retry_tokens_sync(self, tmp_db):
+        session = _make_session(tmp_db)
+        _populate_simple(session)
+        session.retry()
+        assert len(session._msg_tokens) == len(session.messages)
+
+    def test_retry_multipart_content_returns_none(self, tmp_db):
+        """retry() should refuse multipart (vision/image) messages."""
+        session = _make_session(tmp_db)
+        session.messages = [
+            {"role": "user", "content": [{"type": "text", "text": "describe this"}]},
+            {"role": "assistant", "content": "It's an image."},
+        ]
+        session._msg_tokens = [10, 20]
+        assert session.retry() is None
+        # Messages should be unchanged
+        assert len(session.messages) == 2
+
+    def test_retry_none_content_returns_none(self, tmp_db):
+        """retry() should handle content=None gracefully."""
+        session = _make_session(tmp_db)
+        session.messages = [
+            {"role": "user", "content": None},
+            {"role": "assistant", "content": "Ok."},
+        ]
+        session._msg_tokens = [10, 20]
+        assert session.retry() is None
+
+
+# ---------------------------------------------------------------------------
+# handle_command integration
+# ---------------------------------------------------------------------------
+
+
+class TestHandleCommand:
+    def test_rewind_command(self, tmp_db):
+        session = _make_session(tmp_db)
+        _populate_simple(session)
+        session.handle_command("/rewind 1")
+        assert len(session.messages) == 2
+
+    def test_rewind_no_arg(self, tmp_db):
+        session = _make_session(tmp_db)
+        ui = session.ui
+        ui.on_info = MagicMock()
+        session.handle_command("/rewind")
+        ui.on_info.assert_called_once()
+        assert "Usage" in ui.on_info.call_args[0][0]
+
+    def test_rewind_invalid_arg(self, tmp_db):
+        session = _make_session(tmp_db)
+        ui = session.ui
+        ui.on_info = MagicMock()
+        session.handle_command("/rewind abc")
+        ui.on_info.assert_called_once()
+        assert "integer" in ui.on_info.call_args[0][0]
+
+    def test_retry_nothing_to_retry(self, tmp_db):
+        session = _make_session(tmp_db)
+        ui = session.ui
+        ui.on_info = MagicMock()
+        session.handle_command("/retry")
+        ui.on_info.assert_called_once()
+        assert "Nothing" in ui.on_info.call_args[0][0]
+
+
+# ---------------------------------------------------------------------------
+# Storage integration — delete_messages_after
+# ---------------------------------------------------------------------------
+
+
+class TestDeleteMessagesAfter:
+    def test_delete_truncates_db(self, tmp_db):
+        from turnstone.core.memory import (
+            delete_messages_after,
+            load_messages,
+            register_workstream,
+            save_message,
+        )
+
+        ws_id = "test-ws-delete"
+        register_workstream(ws_id)
+        save_message(ws_id, "user", "Hello")
+        save_message(ws_id, "assistant", "Hi!")
+        save_message(ws_id, "user", "Bye")
+        save_message(ws_id, "assistant", "Goodbye!")
+
+        deleted = delete_messages_after(ws_id, 2)
+        assert deleted == 2
+
+        msgs = load_messages(ws_id)
+        assert len(msgs) == 2
+        assert msgs[0]["content"] == "Hello"
+        assert msgs[1]["content"] == "Hi!"
+
+    def test_delete_nothing(self, tmp_db):
+        from turnstone.core.memory import (
+            delete_messages_after,
+            register_workstream,
+            save_message,
+        )
+
+        ws_id = "test-ws-noop"
+        register_workstream(ws_id)
+        save_message(ws_id, "user", "Hello")
+
+        deleted = delete_messages_after(ws_id, 10)
+        assert deleted == 0
+
+    def test_delete_all(self, tmp_db):
+        from turnstone.core.memory import (
+            delete_messages_after,
+            load_messages,
+            register_workstream,
+            save_message,
+        )
+
+        ws_id = "test-ws-all"
+        register_workstream(ws_id)
+        save_message(ws_id, "user", "Hello")
+        save_message(ws_id, "assistant", "Hi!")
+
+        deleted = delete_messages_after(ws_id, 0)
+        assert deleted == 2
+        assert load_messages(ws_id) == []
+
+
+# ---------------------------------------------------------------------------
+# End-to-end: rewind + DB sync
+# ---------------------------------------------------------------------------
+
+
+class TestRewindDBSync:
+    def test_rewind_persists_to_db(self, tmp_db):
+        from turnstone.core.memory import load_messages, register_workstream, save_message
+
+        session = _make_session(tmp_db)
+        ws_id = session.ws_id
+        register_workstream(ws_id)
+
+        # Persist messages to DB and set in-memory state
+        save_message(ws_id, "user", "Hello")
+        save_message(ws_id, "assistant", "Hi!")
+        save_message(ws_id, "user", "Bye")
+        save_message(ws_id, "assistant", "Goodbye!")
+
+        session.messages = [
+            {"role": "user", "content": "Hello"},
+            {"role": "assistant", "content": "Hi!"},
+            {"role": "user", "content": "Bye"},
+            {"role": "assistant", "content": "Goodbye!"},
+        ]
+        session._msg_tokens = [5, 5, 5, 5]
+
+        session.rewind(1)
+
+        # Verify DB matches in-memory state
+        db_msgs = load_messages(ws_id)
+        assert len(db_msgs) == 2
+        assert db_msgs[0]["content"] == "Hello"
+        assert db_msgs[1]["content"] == "Hi!"

tests/test_server_live.py

@@ -88,7 +88,7 @@ class RecordingUI:
     def approve_tools(self, items):
         return True, None  # auto-approve everything
 
-    def on_tool_result(self, call_id, name, output):
+    def on_tool_result(self, call_id, name, output, **kwargs):
         self.tool_results.append((call_id, name, output))
 
     def on_tool_output_chunk(self, call_id, chunk):

tests/test_session.py

@@ -28,7 +28,7 @@ class NullUI:
     def approve_tools(self, items):
         return True, None
 
-    def on_tool_result(self, call_id, name, output):
+    def on_tool_result(self, call_id, name, output, **kwargs):
         pass
 
     def on_tool_output_chunk(self, call_id, chunk):
@@ -214,7 +214,7 @@ class TestPlanExec:
                         "id": tc_id,
                         "type": "function",
                         "function": {
-                            "name": "create_plan",
+                            "name": "plan_agent",
                             "arguments": json.dumps({"goal": prior_prompt}),
                         },
                     }
@@ -250,7 +250,7 @@ class TestPlanExec:
             m for m in messages if m["role"] == "assistant" and m.get("tool_calls")
         ]
         assert len(assistant_with_tc) == 1
-        assert assistant_with_tc[0]["tool_calls"][0]["function"]["name"] == "create_plan"
+        assert assistant_with_tc[0]["tool_calls"][0]["function"]["name"] == "plan_agent"
 
         # The real tool result is forwarded with its original content
         tool_msgs = [m for m in messages if m["role"] == "tool"]
@@ -463,7 +463,7 @@ class TestPlanRefinement:
         with patch.object(session, "_refine_plan", side_effect=fake_refine):
             items = [
                 {
-                    "func_name": "create_plan",
+                    "func_name": "plan_agent",
                     "call_id": "c1",
                     "prompt": "add auth",
                 }
@@ -576,7 +576,7 @@ class TestPlanRefinement:
         msgs = captured["messages"]
         assert msgs[0]["role"] == "system"
         assert msgs[1]["role"] == "assistant"
-        assert msgs[1]["tool_calls"][0]["function"]["name"] == "create_plan"
+        assert msgs[1]["tool_calls"][0]["function"]["name"] == "plan_agent"
         assert msgs[2]["role"] == "tool"
         assert msgs[2]["content"] == self.GOOD_PLAN
         assert msgs[3]["role"] == "user"

tests/test_sessions.py

@@ -525,6 +525,37 @@ class TestWorkstreamConfig:
         assert session.instructions == "be concise"
         assert session.creative_mode is True
 
+    def test_resume_restores_model(self, tmp_db):
+        """ChatSession.resume() should restore the model from workstream config."""
+        client = MagicMock()
+        client.models.list.return_value.data = [MagicMock(id="test-model")]
+        ui = MagicMock()
+        ui.on_info = MagicMock()
+        ui.on_error = MagicMock()
+        ui.on_state_change = MagicMock()
+        ui.on_rename = MagicMock()
+
+        # Create a workstream that was using a specific model
+        register_workstream("model_ws")
+        save_message("model_ws", "user", "hello")
+        save_message("model_ws", "assistant", "hi")
+        save_workstream_config("model_ws", {"model": "gpt-5", "model_alias": ""})
+
+        # Resume into a session that was created with a different model
+        session = ChatSession(
+            client=client,
+            model="gpt-5-nano",
+            ui=ui,
+            instructions=None,
+            temperature=0.5,
+            max_tokens=1000,
+            tool_timeout=10,
+        )
+        assert session.model == "gpt-5-nano"
+        result = session.resume("model_ws")
+        assert result is True
+        assert session.model == "gpt-5"
+
 
 # ── Prune workstreams ─────────────────────────────────────────────────
 
@@ -699,6 +730,7 @@ class TestWebSearchGating:
         with (
             patch.object(session, "_get_capabilities", return_value=caps),
             patch("turnstone.core.session.get_tavily_key", return_value=None),
+            patch("turnstone.core.web_search._ddg_available", return_value=False),
         ):
             tools = session._get_active_tools()
 
@@ -754,3 +786,150 @@ class TestWebSearchGating:
 
         names = [t.get("function", {}).get("name") for t in tools]
         assert "web_search" in names
+
+
+class TestMCPToolGating:
+    """MCP tools should not be offered when no MCP servers provide them."""
+
+    def test_mcp_tools_filtered_without_mcp_client(self, tmp_db, mock_openai_client):
+        """read_resource and use_prompt excluded when no MCP client."""
+        session = ChatSession(
+            client=mock_openai_client,
+            model="local-model",
+            ui=MagicMock(),
+            instructions=None,
+            temperature=0.5,
+            max_tokens=1000,
+            tool_timeout=10,
+        )
+        assert session._mcp_client is None
+
+        tools = session._get_active_tools()
+        names = [t.get("function", {}).get("name") for t in tools]
+        assert "read_resource" not in names
+        assert "use_prompt" not in names
+
+    def test_read_resource_filtered_when_no_resources(self, tmp_db, mock_openai_client):
+        """read_resource excluded when MCP client has no resources."""
+        mcp_client = MagicMock()
+        mcp_client.get_tools.return_value = []
+        mcp_client.resource_count = 0
+        mcp_client.prompt_count = 2
+
+        session = ChatSession(
+            client=mock_openai_client,
+            model="local-model",
+            ui=MagicMock(),
+            instructions=None,
+            temperature=0.5,
+            max_tokens=1000,
+            tool_timeout=10,
+            mcp_client=mcp_client,
+        )
+
+        tools = session._get_active_tools()
+        names = [t.get("function", {}).get("name") for t in tools]
+        assert "read_resource" not in names
+        assert "use_prompt" in names
+
+    def test_use_prompt_filtered_when_no_prompts(self, tmp_db, mock_openai_client):
+        """use_prompt excluded when MCP client has no prompts."""
+        mcp_client = MagicMock()
+        mcp_client.get_tools.return_value = []
+        mcp_client.resource_count = 3
+        mcp_client.prompt_count = 0
+
+        session = ChatSession(
+            client=mock_openai_client,
+            model="local-model",
+            ui=MagicMock(),
+            instructions=None,
+            temperature=0.5,
+            max_tokens=1000,
+            tool_timeout=10,
+            mcp_client=mcp_client,
+        )
+
+        tools = session._get_active_tools()
+        names = [t.get("function", {}).get("name") for t in tools]
+        assert "use_prompt" not in names
+        assert "read_resource" in names
+
+    def test_mcp_tools_kept_when_servers_have_both(self, tmp_db, mock_openai_client):
+        """Both tools present when MCP client has resources and prompts."""
+        mcp_client = MagicMock()
+        mcp_client.get_tools.return_value = []
+        mcp_client.resource_count = 1
+        mcp_client.prompt_count = 1
+
+        session = ChatSession(
+            client=mock_openai_client,
+            model="local-model",
+            ui=MagicMock(),
+            instructions=None,
+            temperature=0.5,
+            max_tokens=1000,
+            tool_timeout=10,
+            mcp_client=mcp_client,
+        )
+
+        tools = session._get_active_tools()
+        names = [t.get("function", {}).get("name") for t in tools]
+        assert "read_resource" in names
+        assert "use_prompt" in names
+
+    def test_mcp_tools_filtered_with_tool_search_active(self, tmp_db, mock_openai_client):
+        """Gating applies even when tool_search is active (client-side path)."""
+        mcp_client = MagicMock()
+        mcp_client.get_tools.return_value = []
+        mcp_client.resource_count = 0
+        mcp_client.prompt_count = 0
+
+        session = ChatSession(
+            client=mock_openai_client,
+            model="local-model",
+            ui=MagicMock(),
+            instructions=None,
+            temperature=0.5,
+            max_tokens=1000,
+            tool_timeout=10,
+            mcp_client=mcp_client,
+            tool_search="on",
+        )
+        assert session._tool_search is not None
+
+        tools = session._get_active_tools()
+        names = [t.get("function", {}).get("name") for t in tools]
+        assert "read_resource" not in names
+        assert "use_prompt" not in names
+
+    def test_mcp_tools_filtered_with_native_tool_search(self, tmp_db, mock_openai_client):
+        """Gating applies when provider handles tool search natively."""
+        from unittest.mock import patch
+
+        from turnstone.core.providers._protocol import ModelCapabilities
+
+        mcp_client = MagicMock()
+        mcp_client.get_tools.return_value = []
+        mcp_client.resource_count = 0
+        mcp_client.prompt_count = 0
+
+        session = ChatSession(
+            client=mock_openai_client,
+            model="local-model",
+            ui=MagicMock(),
+            instructions=None,
+            temperature=0.5,
+            max_tokens=1000,
+            tool_timeout=10,
+            mcp_client=mcp_client,
+            tool_search="on",
+        )
+
+        caps = ModelCapabilities(supports_tool_search=True)
+        with patch.object(session, "_get_capabilities", return_value=caps):
+            tools = session._get_active_tools()
+
+        names = [t.get("function", {}).get("name") for t in tools]
+        assert "read_resource" not in names
+        assert "use_prompt" not in names

tests/test_settings_api.py

@@ -254,20 +254,46 @@ class TestSecretMasking:
         by_key = {s["key"]: s for s in r.json()["settings"]}
         assert by_key["judge.api_key"]["value"] == "***"
 
-    def test_secret_write_blocked(self, client):
-        """Secret settings cannot be modified via API."""
+    def test_secret_writable_via_api(self, client):
+        """Secret settings can be written via API (write-only pattern)."""
         r = client.put(
             "/v1/api/admin/settings/judge.api_key",
             json={"value": "sk-secret-123"},
         )
-        assert r.status_code == 403
-        assert "config.toml" in r.json()["error"]
+        assert r.status_code == 200
+        # Response value is masked even for the write confirmation
+        assert r.json()["value"] == "***"
 
-    def test_secret_shows_managed_label(self, client):
-        """Secret settings show a label instead of a value."""
+    def test_secret_sentinel_preserves_existing(self, client):
+        """Submitting '***' for a secret setting is a no-op (preserve existing)."""
+        # First write a real value
+        r1 = client.put(
+            "/v1/api/admin/settings/judge.api_key",
+            json={"value": "sk-real-key"},
+        )
+        assert r1.status_code == 200
+        # Now submit the sentinel — should return unchanged with full response shape
+        r2 = client.put(
+            "/v1/api/admin/settings/judge.api_key",
+            json={"value": "***"},
+        )
+        assert r2.status_code == 200
+        data = r2.json()
+        assert data.get("unchanged") is True
+        assert data["key"] == "judge.api_key"
+        assert data["value"] == "***"
+        assert data["type"] == "str"
+        assert data["is_secret"] is True
+
+    def test_secret_still_masked_in_list(self, client):
+        """After writing a secret, list still shows '***'."""
+        client.put(
+            "/v1/api/admin/settings/judge.api_key",
+            json={"value": "sk-written-via-api"},
+        )
         r = client.get("/v1/api/admin/settings")
         by_key = {s["key"]: s for s in r.json()["settings"]}
-        assert "managed via" in by_key["judge.api_key"]["value"]
+        assert by_key["judge.api_key"]["value"] == "***"
 
 
 # ---------------------------------------------------------------------------

tests/test_skills.py

@@ -52,7 +52,7 @@ class NullUI:
     def approve_tools(self, items):
         return True, None
 
-    def on_tool_result(self, call_id, name, output):
+    def on_tool_result(self, call_id, name, output, **kwargs):
         pass
 
     def on_tool_output_chunk(self, call_id, chunk):

tests/test_tls_admin.py

@@ -0,0 +1,234 @@
+"""Tests for TLS admin API endpoints and CLI commands."""
+
+from __future__ import annotations
+
+import pytest
+
+from turnstone.core.storage import get_storage, init_storage, reset_storage
+
+lacme = pytest.importorskip("lacme")
+
+
+@pytest.fixture(autouse=True)
+def _storage(tmp_path):
+    """Initialize ephemeral SQLite storage for each test."""
+    reset_storage()
+    db = str(tmp_path / "test.db")
+    init_storage("sqlite", path=db)
+    yield
+    reset_storage()
+
+
+@pytest.fixture
+def tls_manager():
+    """Create an initialized TLSManager."""
+    import asyncio
+
+    from turnstone.console.tls import TLSManager
+
+    mgr = TLSManager(get_storage())
+    asyncio.run(mgr.init_ca())
+    # Issue a test cert
+    asyncio.run(mgr.issue_console_certs(["test.internal", "localhost"]))
+    return mgr
+
+
+# ── Admin API endpoints ───────────────────────────────────────────────────────
+
+
+def _make_app(tls_manager):
+    """Create a minimal Starlette app with TLS endpoints."""
+    from starlette.applications import Starlette
+    from starlette.middleware import Middleware
+    from starlette.middleware.base import BaseHTTPMiddleware
+    from starlette.routing import Route
+
+    from turnstone.console.server import (
+        tls_ca_cert,
+        tls_ca_status,
+        tls_delete_cert,
+        tls_list_certs,
+        tls_renew_cert,
+    )
+    from turnstone.core.auth import AuthResult
+
+    async def _grant_access(request, call_next):  # type: ignore[no-untyped-def]
+        request.state.auth_result = AuthResult(
+            user_id="",
+            scopes=frozenset({"approve"}),
+            token_source="config",
+        )
+        return await call_next(request)
+
+    app = Starlette(
+        routes=[
+            Route("/ca", tls_ca_status),
+            Route("/ca.pem", tls_ca_cert),
+            Route("/certs", tls_list_certs),
+            Route("/certs/{domain}/renew", tls_renew_cert, methods=["POST"]),
+            Route("/certs/{domain}", tls_delete_cert, methods=["DELETE"]),
+        ],
+        middleware=[Middleware(BaseHTTPMiddleware, dispatch=_grant_access)],
+    )
+    app.state.tls_manager = tls_manager
+    return app
+
+
+def test_list_certs(tls_manager):
+    from starlette.testclient import TestClient
+
+    client = TestClient(_make_app(tls_manager))
+    resp = client.get("/certs")
+    assert resp.status_code == 200
+    data = resp.json()
+    assert len(data["certs"]) >= 1
+    assert data["certs"][0]["domain"] == "test.internal"
+
+
+def test_renew_cert(tls_manager):
+    from starlette.testclient import TestClient
+
+    client = TestClient(_make_app(tls_manager))
+    resp = client.post("/certs/test.internal/renew")
+    assert resp.status_code == 200
+    data = resp.json()
+    assert data["domain"] == "test.internal"
+
+
+def test_renew_cert_not_found(tls_manager):
+    from starlette.testclient import TestClient
+
+    client = TestClient(_make_app(tls_manager))
+    resp = client.post("/certs/nonexistent.internal/renew")
+    assert resp.status_code == 404
+
+
+def test_delete_cert(tls_manager):
+    from starlette.testclient import TestClient
+
+    client = TestClient(_make_app(tls_manager))
+    resp = client.delete("/certs/test.internal")
+    assert resp.status_code == 200
+    assert resp.json()["deleted"] == "test.internal"
+    # Verify it's gone
+    resp = client.get("/certs")
+    domains = [c["domain"] for c in resp.json()["certs"]]
+    assert "test.internal" not in domains
+
+
+def test_delete_cert_not_found(tls_manager):
+    from starlette.testclient import TestClient
+
+    client = TestClient(_make_app(tls_manager))
+    resp = client.delete("/certs/nonexistent.internal")
+    assert resp.status_code == 404
+
+
+# ── CLI bootstrap ─────────────────────────────────────────────────────────────
+
+
+def test_cli_bootstrap(tmp_path):
+    """Test offline CA bootstrap."""
+    import argparse
+
+    from turnstone.admin import _cmd_tls_bootstrap
+
+    out = tmp_path / "certs"
+    args = argparse.Namespace(out=str(out), issue=["redis.internal", "pg.internal"])
+    _cmd_tls_bootstrap(args)
+
+    assert (out / "ca.pem").exists()
+    assert b"BEGIN CERTIFICATE" in (out / "ca.pem").read_bytes()
+    # Check certs were issued
+    assert (out / "certs" / "redis.internal").exists()
+    assert (out / "certs" / "pg.internal").exists()
+
+
+def test_cli_bootstrap_no_issue(tmp_path):
+    """Bootstrap with no --issue creates CA only."""
+    import argparse
+
+    from turnstone.admin import _cmd_tls_bootstrap
+
+    out = tmp_path / "certs"
+    args = argparse.Namespace(out=str(out), issue=[])
+    _cmd_tls_bootstrap(args)
+
+    assert (out / "ca.pem").exists()
+    # No certs dir
+    certs_dir = out / "certs"
+    if certs_dir.exists():
+        assert len(list(certs_dir.iterdir())) == 0
+
+
+# ── Config parsing ────────────────────────────────────────────────────────────
+
+
+def test_redis_tls_config_map():
+    """Redis TLS keys are in the config map."""
+    from turnstone.core.config import _CONFIG_MAP
+
+    redis_map = _CONFIG_MAP["redis"]
+    assert "tls" in redis_map
+    assert "tls_ca" in redis_map
+    assert "tls_cert" in redis_map
+    assert "tls_key" in redis_map
+
+
+def test_database_ssl_config_map():
+    """Database SSL keys are in the config map."""
+    from turnstone.core.config import _CONFIG_MAP
+
+    db_map = _CONFIG_MAP["database"]
+    assert "sslmode" in db_map
+    assert "sslrootcert" in db_map
+    assert "sslcert" in db_map
+    assert "sslkey" in db_map
+
+
+# ── Auth enforcement ──────────────────────────────────────────────────────────
+
+
+def test_tls_endpoints_require_auth(tls_manager):
+    """TLS admin endpoints return 401 without auth."""
+    from starlette.applications import Starlette
+    from starlette.routing import Route
+    from starlette.testclient import TestClient
+
+    from turnstone.console.server import tls_ca_status, tls_list_certs
+
+    # No auth middleware — request.state.auth_result will be missing
+    app = Starlette(
+        routes=[
+            Route("/ca", tls_ca_status),
+            Route("/certs", tls_list_certs),
+        ]
+    )
+    app.state.tls_manager = tls_manager
+
+    client = TestClient(app)
+    resp = client.get("/ca")
+    assert resp.status_code == 401
+
+    resp = client.get("/certs")
+    assert resp.status_code == 401
+
+
+# ── SDK TLS params ────────────────────────────────────────────────────────────
+
+
+def test_sdk_client_cert_requires_both():
+    """SDK raises ValueError if only one of client_cert/client_key provided."""
+    from turnstone.sdk._base import _BaseClient
+
+    with pytest.raises(ValueError, match="Both client_cert and client_key"):
+        _BaseClient(
+            base_url="http://localhost:8080",
+            client_cert="/path/to/cert.pem",
+        )
+
+    with pytest.raises(ValueError, match="Both client_cert and client_key"):
+        _BaseClient(
+            base_url="http://localhost:8080",
+            client_key="/path/to/key.pem",
+        )

tests/test_tls_client.py

@@ -0,0 +1,98 @@
+"""Tests for TLSClient — service node certificate provisioning."""
+
+from __future__ import annotations
+
+from unittest.mock import MagicMock
+
+import pytest
+
+from turnstone.core.storage import get_storage, init_storage, reset_storage
+
+lacme = pytest.importorskip("lacme")
+
+
+@pytest.fixture(autouse=True)
+def _storage(tmp_path):
+    """Initialize ephemeral SQLite storage for each test."""
+    reset_storage()
+    db = str(tmp_path / "test.db")
+    init_storage("sqlite", path=db)
+    yield
+    reset_storage()
+
+
+# ── Console URL discovery ─────────────────────────────────────────────────────
+
+
+def test_discover_console_url():
+    """TLSClient discovers console URL from services table."""
+    from turnstone.core.tls import TLSClient
+
+    storage = get_storage()
+    storage.register_service("console", "console", "http://console:8080")
+
+    client = TLSClient(storage=storage, hostnames=["node-1"])
+    url = client._discover_console_url()
+    assert url == "http://console:8080"
+
+
+def test_discover_console_url_missing():
+    """TLSClient raises if no console registered."""
+    from turnstone.core.tls import TLSClient
+
+    client = TLSClient(storage=get_storage(), hostnames=["node-1"])
+    with pytest.raises(RuntimeError, match="No console service found"):
+        client._discover_console_url()
+
+
+def test_explicit_console_url_skips_discovery():
+    """When console_url is provided, discovery is skipped."""
+    from turnstone.core.tls import TLSClient
+
+    client = TLSClient(
+        storage=get_storage(),
+        console_url="http://explicit:9090",
+        hostnames=["node-1"],
+    )
+    assert client._console_url == "http://explicit:9090"
+
+
+# ── SSL context construction ─────────────────────────────────────────────────
+
+
+@pytest.mark.anyio
+async def test_ssl_contexts_none_before_init():
+    """SSL contexts are None before init()."""
+    from turnstone.core.tls import TLSClient
+
+    client = TLSClient(
+        storage=get_storage(),
+        console_url="http://localhost:8080",
+        hostnames=["node-1"],
+    )
+    assert client.get_server_ssl_context() is None
+    assert client.get_client_ssl_context() is None
+    assert not client.initialized
+
+
+# ── Backward compatibility ───────────────────────────────────────────────────
+
+
+def test_bridge_tls_defaults():
+    """Bridge with default TLS params works without changes."""
+    from turnstone.mq.bridge import Bridge
+
+    # Default: tls_verify=True, tls_cert=None — no mTLS
+    bridge = Bridge(server_url="http://localhost:8080")
+    assert bridge._tls_verify is True
+    assert bridge._tls_cert is None
+
+
+def test_collector_tls_defaults():
+    """Collector with default TLS params works without changes."""
+    from turnstone.console.collector import ClusterCollector
+
+    broker_mock = MagicMock()
+    collector = ClusterCollector(broker=broker_mock)
+    # Should create httpx client without errors
+    assert collector._http_client is not None

tests/test_tls_manager.py

@@ -0,0 +1,226 @@
+"""Tests for TLSManager — console CA and ACME server."""
+
+from __future__ import annotations
+
+import pytest
+
+from turnstone.core.storage import get_storage, init_storage, reset_storage
+
+lacme = pytest.importorskip("lacme")
+
+
+@pytest.fixture(autouse=True)
+def _storage(tmp_path):
+    """Initialize ephemeral SQLite storage for each test."""
+    reset_storage()
+    db = str(tmp_path / "test.db")
+    init_storage("sqlite", path=db)
+    yield
+    reset_storage()
+
+
+@pytest.fixture
+def tls_manager():
+    """Create a TLSManager backed by test storage."""
+    from turnstone.console.tls import TLSManager
+
+    return TLSManager(get_storage())
+
+
+# ── CA initialization ─────────────────────────────────────────────────────────
+
+
+@pytest.mark.anyio
+async def test_init_ca(tls_manager):
+    await tls_manager.init_ca()
+    assert tls_manager.ca_initialized
+    root_pem = tls_manager.get_root_cert_pem()
+    assert b"BEGIN CERTIFICATE" in root_pem
+
+
+@pytest.mark.anyio
+async def test_init_ca_persists(tls_manager):
+    """CA root survives re-initialization (loaded from storage)."""
+    await tls_manager.init_ca()
+    pem1 = tls_manager.get_root_cert_pem()
+
+    # Create a new manager on the same storage
+    from turnstone.console.tls import TLSManager
+
+    mgr2 = TLSManager(get_storage())
+    await mgr2.init_ca()
+    pem2 = mgr2.get_root_cert_pem()
+
+    assert pem1 == pem2  # Same CA loaded from DB
+
+
+@pytest.mark.anyio
+async def test_get_responder_before_init(tls_manager):
+    with pytest.raises(RuntimeError, match="CA not initialized"):
+        tls_manager.get_responder()
+
+
+@pytest.mark.anyio
+async def test_get_responder(tls_manager):
+    await tls_manager.init_ca()
+    responder = tls_manager.get_responder()
+    assert responder is not None
+    # Should be an ASGI app (callable)
+    assert callable(responder)
+
+
+# ── Cert issuance ─────────────────────────────────────────────────────────────
+
+
+@pytest.mark.anyio
+async def test_issue_console_certs_internal(tls_manager):
+    """Console certs issued from internal CA when no external directory."""
+    await tls_manager.init_ca()
+    await tls_manager.issue_console_certs(["console.internal", "localhost"])
+    assert tls_manager.internal_bundle is not None
+    assert tls_manager.frontend_bundle is not None
+    assert tls_manager.internal_bundle.domain == "console.internal"
+    assert b"BEGIN CERTIFICATE" in tls_manager.internal_bundle.cert_pem
+
+
+@pytest.mark.anyio
+async def test_issue_console_certs_persists(tls_manager):
+    """Certs loaded from storage on re-issue."""
+    await tls_manager.init_ca()
+    await tls_manager.issue_console_certs(["console.internal"])
+    bundle1 = tls_manager.internal_bundle
+
+    # New manager, same storage
+    from turnstone.console.tls import TLSManager
+
+    mgr2 = TLSManager(get_storage())
+    await mgr2.init_ca()
+    await mgr2.issue_console_certs(["console.internal"])
+    bundle2 = mgr2.internal_bundle
+
+    assert bundle1.cert_pem == bundle2.cert_pem
+
+
+# ── SSL contexts ──────────────────────────────────────────────────────────────
+
+
+@pytest.mark.anyio
+async def test_ssl_contexts_none_before_certs(tls_manager):
+    await tls_manager.init_ca()
+    assert tls_manager.get_server_ssl_context() is None
+    assert tls_manager.get_client_ssl_context() is None
+
+
+@pytest.mark.anyio
+async def test_ssl_contexts_after_certs(tls_manager):
+    await tls_manager.init_ca()
+    await tls_manager.issue_console_certs(["console.internal"])
+    server_ctx = tls_manager.get_server_ssl_context()
+    client_ctx = tls_manager.get_client_ssl_context()
+    assert server_ctx is not None
+    assert client_ctx is not None
+    import ssl
+
+    assert isinstance(server_ctx, ssl.SSLContext)
+    assert isinstance(client_ctx, ssl.SSLContext)
+
+
+# ── Root cert endpoint ────────────────────────────────────────────────────────
+
+
+@pytest.mark.anyio
+async def test_tls_ca_cert_endpoint(tls_manager):
+    """Test the CA cert download endpoint via test client."""
+    await tls_manager.init_ca()
+
+    from starlette.applications import Starlette
+    from starlette.middleware import Middleware
+    from starlette.routing import Route
+    from starlette.testclient import TestClient
+
+    from turnstone.console.server import tls_ca_cert, tls_ca_status
+
+    # Middleware that grants full access (config-token style: no user_id)
+    from turnstone.core.auth import AuthResult
+
+    async def _grant_access(request, call_next):  # type: ignore[no-untyped-def]
+        request.state.auth_result = AuthResult(
+            user_id="", scopes=frozenset({"approve"}), token_source="config"
+        )
+        return await call_next(request)
+
+    from starlette.middleware.base import BaseHTTPMiddleware
+
+    app = Starlette(
+        routes=[
+            Route("/ca.pem", tls_ca_cert),
+            Route("/ca", tls_ca_status),
+        ],
+        middleware=[Middleware(BaseHTTPMiddleware, dispatch=_grant_access)],
+    )
+    app.state.tls_manager = tls_manager
+
+    client = TestClient(app)
+
+    # CA cert download
+    resp = client.get("/ca.pem")
+    assert resp.status_code == 200
+    assert b"BEGIN CERTIFICATE" in resp.content
+    assert resp.headers["content-type"] == "application/x-pem-file"
+
+    # CA status
+    resp = client.get("/ca")
+    assert resp.status_code == 200
+    data = resp.json()
+    assert data["enabled"] is True
+    assert data["ca_cn"] == "Turnstone CA"
+
+
+@pytest.mark.anyio
+async def test_tls_endpoints_disabled():
+    """Endpoints return 404/disabled when TLS not enabled."""
+    from starlette.applications import Starlette
+    from starlette.middleware import Middleware
+    from starlette.middleware.base import BaseHTTPMiddleware
+    from starlette.routing import Route
+    from starlette.testclient import TestClient
+
+    from turnstone.console.server import tls_ca_cert, tls_ca_status
+    from turnstone.core.auth import AuthResult
+
+    async def _grant_access(request, call_next):  # type: ignore[no-untyped-def]
+        request.state.auth_result = AuthResult(
+            user_id="", scopes=frozenset({"approve"}), token_source="config"
+        )
+        return await call_next(request)
+
+    app = Starlette(
+        routes=[
+            Route("/ca.pem", tls_ca_cert),
+            Route("/ca", tls_ca_status),
+        ],
+        middleware=[Middleware(BaseHTTPMiddleware, dispatch=_grant_access)],
+    )
+    # No tls_manager on state
+
+    client = TestClient(app)
+
+    resp = client.get("/ca.pem")
+    assert resp.status_code == 404
+
+    resp = client.get("/ca")
+    data = resp.json()
+    assert data["enabled"] is False
+
+
+# ── Events ────────────────────────────────────────────────────────────────────
+
+
+@pytest.mark.anyio
+async def test_event_dispatcher_wired(tls_manager):
+    """Verify the event dispatcher has subscribers."""
+    assert tls_manager._event_dispatcher is not None
+    # Should have at least 4 subscriptions (issued, renewed, expiring, failed)
+    # The exact check depends on lacme's EventDispatcher internals,
+    # so just verify the dispatcher exists and the manager initializes cleanly
+    await tls_manager.init_ca()

tests/test_tls_storage.py

@@ -0,0 +1,226 @@
+"""Tests for TLS storage backend and lacme Store adapter."""
+
+from __future__ import annotations
+
+import json
+from datetime import UTC, datetime
+
+import pytest
+
+from turnstone.core.storage import get_storage, init_storage, reset_storage
+
+
+@pytest.fixture(autouse=True)
+def _storage(tmp_path):
+    """Initialize ephemeral SQLite storage for each test."""
+    reset_storage()
+    db = str(tmp_path / "test.db")
+    init_storage("sqlite", path=db)
+    yield
+    reset_storage()
+
+
+# ── Account keys ──────────────────────────────────────────────────────────────
+
+
+def test_save_and_load_account_key():
+    s = get_storage()
+    s.save_tls_account_key(
+        "default",
+        "-----BEGIN EC PRIVATE KEY-----\nfake\n-----END EC PRIVATE KEY-----",
+    )
+    result = s.load_tls_account_key("default")
+    assert result is not None
+    assert "EC PRIVATE KEY" in result
+
+
+def test_load_account_key_missing():
+    s = get_storage()
+    assert s.load_tls_account_key("nonexistent") is None
+
+
+def test_save_account_key_upsert():
+    s = get_storage()
+    s.save_tls_account_key("default", "key-v1")
+    s.save_tls_account_key("default", "key-v2")
+    assert s.load_tls_account_key("default") == "key-v2"
+
+
+# ── CA ────────────────────────────────────────────────────────────────────────
+
+
+def test_save_and_load_ca():
+    s = get_storage()
+    s.save_tls_ca("Turnstone CA", "cert-pem-data", "key-pem-data")
+    result = s.load_tls_ca("Turnstone CA")
+    assert result is not None
+    assert result["cert_pem"] == "cert-pem-data"
+    assert result["key_pem"] == "key-pem-data"
+    assert result["name"] == "Turnstone CA"
+
+
+def test_load_ca_missing():
+    s = get_storage()
+    assert s.load_tls_ca("nonexistent") is None
+
+
+def test_save_ca_upsert():
+    s = get_storage()
+    s.save_tls_ca("CA", "cert-v1", "key-v1")
+    s.save_tls_ca("CA", "cert-v2", "key-v2")
+    result = s.load_tls_ca("CA")
+    assert result["cert_pem"] == "cert-v2"
+    assert result["key_pem"] == "key-v2"
+
+
+# ── Certificates ──────────────────────────────────────────────────────────────
+
+
+def test_save_and_load_cert():
+    s = get_storage()
+    s.save_tls_cert(
+        domain="node-1.internal",
+        cert_pem="cert-data",
+        fullchain_pem="fullchain-data",
+        key_pem="key-data",
+        issued_at="2026-03-25T00:00:00",
+        expires_at="2026-03-27T00:00:00",
+        meta=json.dumps({"domains": ["node-1.internal", "10.0.1.5"]}),
+    )
+    result = s.load_tls_cert("node-1.internal")
+    assert result is not None
+    assert result["domain"] == "node-1.internal"
+    assert result["cert_pem"] == "cert-data"
+    assert result["fullchain_pem"] == "fullchain-data"
+    assert result["key_pem"] == "key-data"
+    assert result["issued_at"] == "2026-03-25T00:00:00"
+    assert result["expires_at"] == "2026-03-27T00:00:00"
+    meta = json.loads(result["meta"])
+    assert meta["domains"] == ["node-1.internal", "10.0.1.5"]
+
+
+def test_load_cert_missing():
+    s = get_storage()
+    assert s.load_tls_cert("nonexistent") is None
+
+
+def test_save_cert_upsert():
+    s = get_storage()
+    s.save_tls_cert("d", "c1", "f1", "k1", "2026-01-01", "2026-01-02")
+    s.save_tls_cert("d", "c2", "f2", "k2", "2026-02-01", "2026-02-02")
+    result = s.load_tls_cert("d")
+    assert result["cert_pem"] == "c2"
+    assert result["issued_at"] == "2026-02-01"
+
+
+def test_list_certs_empty():
+    s = get_storage()
+    assert s.list_tls_certs() == []
+
+
+def test_list_certs():
+    s = get_storage()
+    s.save_tls_cert("alpha.internal", "c", "f", "k", "2026-01-01", "2026-01-02")
+    s.save_tls_cert("beta.internal", "c", "f", "k", "2026-01-01", "2026-01-02")
+    certs = s.list_tls_certs()
+    assert len(certs) == 2
+    assert certs[0]["domain"] == "alpha.internal"  # sorted by domain
+    assert certs[1]["domain"] == "beta.internal"
+
+
+def test_delete_cert():
+    s = get_storage()
+    s.save_tls_cert("d", "c", "f", "k", "2026-01-01", "2026-01-02")
+    assert s.delete_tls_cert("d") is True
+    assert s.load_tls_cert("d") is None
+
+
+def test_delete_cert_missing():
+    s = get_storage()
+    assert s.delete_tls_cert("nonexistent") is False
+
+
+# ── StorageStore adapter ──────────────────────────────────────────────────────
+
+
+@pytest.fixture
+def store_adapter():
+    """Create a StorageStore backed by the test database."""
+    from turnstone.core.tls_store import StorageStore
+
+    return StorageStore(get_storage())
+
+
+def test_adapter_save_load_ca(store_adapter):
+    store_adapter.save_ca("test-ca", b"cert-pem", b"key-pem")
+    result = store_adapter.load_ca("test-ca")
+    assert result is not None
+    cert_pem, key_pem = result
+    assert cert_pem == b"cert-pem"
+    assert key_pem == b"key-pem"
+
+
+def test_adapter_load_ca_missing(store_adapter):
+    assert store_adapter.load_ca("missing") is None
+
+
+def test_adapter_save_load_cert(store_adapter):
+    lacme = pytest.importorskip("lacme")
+    now = datetime.now(UTC)
+    bundle = lacme.CertBundle(
+        domain="test.internal",
+        domains=("test.internal", "10.0.1.1"),
+        cert_pem=b"cert",
+        fullchain_pem=b"fullchain",
+        key_pem=b"key",
+        issued_at=now,
+        expires_at=now,
+    )
+    store_adapter.save_cert(bundle)
+    loaded = store_adapter.load_cert("test.internal")
+    assert loaded is not None
+    assert loaded.domain == "test.internal"
+    assert loaded.domains == ("test.internal", "10.0.1.1")
+    assert loaded.cert_pem == b"cert"
+    assert loaded.fullchain_pem == b"fullchain"
+    assert loaded.key_pem == b"key"
+
+
+def test_adapter_list_certs(store_adapter):
+    lacme = pytest.importorskip("lacme")
+    now = datetime.now(UTC)
+    for name in ["alpha", "beta"]:
+        bundle = lacme.CertBundle(
+            domain=f"{name}.internal",
+            domains=(f"{name}.internal",),
+            cert_pem=b"c",
+            fullchain_pem=b"f",
+            key_pem=b"k",
+            issued_at=now,
+            expires_at=now,
+        )
+        store_adapter.save_cert(bundle)
+    certs = store_adapter.list_certs()
+    assert len(certs) == 2
+    assert certs[0].domain == "alpha.internal"
+
+
+def test_adapter_load_cert_missing(store_adapter):
+    assert store_adapter.load_cert("missing") is None
+
+
+def test_adapter_account_key_roundtrip(store_adapter):
+    """Test account key save/load with real cryptography objects."""
+    pytest.importorskip("lacme")
+    from cryptography.hazmat.primitives.asymmetric import ec
+
+    key = ec.generate_private_key(ec.SECP256R1())
+    store_adapter.save_account_key(key)
+    loaded = store_adapter.load_account_key()
+    assert loaded is not None
+    # Verify it's a usable EC key
+    assert loaded.key_size == key.key_size
+
+
+def test_adapter_account_key_missing(store_adapter):
+    assert store_adapter.load_account_key() is None

tests/test_tools_schema.py

@@ -104,8 +104,8 @@ class TestToolsMetadata:
             "man": "page",
             "web_fetch": "url",
             "web_search": "query",
-            "task": "prompt",
-            "create_plan": "goal",
+            "task_agent": "prompt",
+            "plan_agent": "goal",
             "memory": "name",
             "recall": "query",
             "notify": "message",

tests/test_watch_dispatch.py

@@ -0,0 +1,217 @@
+"""Tests for _make_watch_dispatch error/cancel handling and concurrency guards."""
+
+import queue
+import threading
+import time
+
+from turnstone.core.session import GenerationCancelled
+from turnstone.core.workstream import Workstream
+from turnstone.server import _make_watch_dispatch
+
+
+class _StubSession:
+    """Minimal ChatSession stand-in with controllable send() behaviour."""
+
+    def __init__(self, *, side_effect=None):
+        self._watch_pending: queue.Queue = queue.Queue(maxsize=20)
+        self._side_effect = side_effect
+
+    def send(self, msg: str) -> None:
+        if self._side_effect is not None:
+            raise self._side_effect
+
+
+class _RecordingUI:
+    """Track calls made by the dispatch error handlers."""
+
+    def __init__(self):
+        self.errors: list[str] = []
+        self.state_changes: list[str] = []
+        self.stream_end_calls: int = 0
+
+    # -- SessionUI protocol stubs used by the dispatch code --
+
+    def on_error(self, message: str) -> None:
+        self.errors.append(message)
+
+    def on_state_change(self, state: str) -> None:
+        self.state_changes.append(state)
+
+    def on_stream_end(self) -> None:
+        self.stream_end_calls += 1
+
+
+# ── helpers ──────────────────────────────────────────────────────────────────
+
+
+def _wait_for_worker(ws: Workstream, timeout: float = 2.0) -> None:
+    """Block until the worker thread started by dispatch() finishes."""
+    t = ws.worker_thread
+    if t is not None:
+        t.join(timeout)
+        assert not t.is_alive(), "worker thread did not finish in time"
+
+
+# ── GenerationCancelled path ────────────────────────────────────────────────
+
+
+def test_cancelled_emits_stream_end_and_idle():
+    session = _StubSession(side_effect=GenerationCancelled())
+    ws = Workstream()
+    ui = _RecordingUI()
+
+    dispatch = _make_watch_dispatch(ws, session, ui)
+    dispatch("hello")
+    _wait_for_worker(ws)
+
+    assert ui.stream_end_calls == 1
+    assert ui.state_changes == ["idle"]
+    assert ui.errors == []
+
+
+# ── Generic exception path ──────────────────────────────────────────────────
+
+
+def test_exception_emits_stream_end_and_error():
+    session = _StubSession(side_effect=RuntimeError("boom"))
+    ws = Workstream()
+    ui = _RecordingUI()
+
+    dispatch = _make_watch_dispatch(ws, session, ui)
+    dispatch("hello")
+    _wait_for_worker(ws)
+
+    assert ui.stream_end_calls == 1
+    assert ui.state_changes == ["error"]
+    assert len(ui.errors) == 1
+    assert "boom" in ui.errors[0]
+
+
+# ── Worker-thread identity guard ────────────────────────────────────────────
+
+
+def test_abandoned_thread_emits_no_events():
+    """After force-cancel sets worker_thread=None, the old thread must not
+    emit stream_end or state changes."""
+    barrier = threading.Event()
+
+    class _BlockingSession(_StubSession):
+        def send(self, msg: str) -> None:
+            barrier.wait(timeout=5)
+            raise RuntimeError("late error")
+
+    session = _BlockingSession()
+    ws = Workstream()
+    ui = _RecordingUI()
+
+    dispatch = _make_watch_dispatch(ws, session, ui)
+    dispatch("hello")
+
+    # Simulate force-cancel: clear the worker_thread reference.
+    ws.worker_thread = None
+    barrier.set()
+
+    # Wait for the thread to actually complete (it's still running).
+    time.sleep(0.3)
+
+    assert ui.stream_end_calls == 0
+    assert ui.state_changes == []
+    assert ui.errors == []
+
+
+# ── Path A: busy workstream enqueue ─────────────────────────────────────────
+
+
+def test_busy_workstream_enqueues_message():
+    """When the workstream already has a live worker, dispatch enqueues."""
+    session = _StubSession()
+    ws = Workstream()
+    ui = _RecordingUI()
+
+    # Simulate a live worker thread.
+    blocker = threading.Event()
+    ws.worker_thread = threading.Thread(target=blocker.wait, args=(5,), daemon=True)
+    ws.worker_thread.start()
+
+    try:
+        dispatch = _make_watch_dispatch(ws, session, ui)
+        dispatch("queued msg")
+
+        item = session._watch_pending.get_nowait()
+        assert item == {"message": "queued msg"}
+    finally:
+        blocker.set()
+        ws.worker_thread.join(2)
+
+
+def test_busy_workstream_drops_on_full_queue():
+    """When the pending queue is full, dispatch drops the message."""
+    session = _StubSession()
+    # Fill the queue to capacity.
+    for i in range(20):
+        session._watch_pending.put_nowait({"message": f"msg{i}"})
+
+    ws = Workstream()
+    ui = _RecordingUI()
+
+    blocker = threading.Event()
+    ws.worker_thread = threading.Thread(target=blocker.wait, args=(5,), daemon=True)
+    ws.worker_thread.start()
+
+    try:
+        dispatch = _make_watch_dispatch(ws, session, ui)
+        # Should not block or raise — just log a warning and drop.
+        dispatch("overflow msg")
+
+        assert session._watch_pending.full()
+    finally:
+        blocker.set()
+        ws.worker_thread.join(2)
+
+
+# ── Lock guard ───────────────────────────────────────────────────────────────
+
+
+def test_dispatch_holds_lock_during_thread_start():
+    """Dispatch acquires ws._lock before checking/starting the worker."""
+    session = _StubSession()
+    ws = Workstream()
+    ui = _RecordingUI()
+
+    acquire_count = 0
+    inner = ws._lock
+
+    class _CountingLock:
+        def __enter__(self):
+            nonlocal acquire_count
+            acquire_count += 1
+            return inner.__enter__()
+
+        def __exit__(self, *args):
+            return inner.__exit__(*args)
+
+    ws._lock = _CountingLock()  # type: ignore[assignment]
+
+    dispatch = _make_watch_dispatch(ws, session, ui)
+    dispatch("hello")
+    _wait_for_worker(ws)
+
+    assert acquire_count >= 1
+
+
+# ── Happy path ───────────────────────────────────────────────────────────────
+
+
+def test_successful_send_no_error_events():
+    """Normal send() completion should not trigger error/cancel events."""
+    session = _StubSession()  # send() does nothing (success)
+    ws = Workstream()
+    ui = _RecordingUI()
+
+    dispatch = _make_watch_dispatch(ws, session, ui)
+    dispatch("hello")
+    _wait_for_worker(ws)
+
+    assert ui.stream_end_calls == 0
+    assert ui.state_changes == []
+    assert ui.errors == []

tests/test_web_search.py

@@ -84,7 +84,7 @@ class TestTavilyClient:
 
 class TestDuckDuckGoClient:
     def test_integration_via_mock_ddgs(self):
-        """Patch the duckduckgo_search import inside DuckDuckGoClient.search."""
+        """Patch the ddgs import inside DuckDuckGoClient.search."""
         mock_ddgs = MagicMock()
         mock_ddgs.__enter__ = MagicMock(return_value=mock_ddgs)
         mock_ddgs.__exit__ = MagicMock(return_value=False)
@@ -93,7 +93,7 @@ class TestDuckDuckGoClient:
         ]
         mock_module = MagicMock()
         mock_module.DDGS.return_value = mock_ddgs
-        with patch.dict("sys.modules", {"duckduckgo_search": mock_module}):
+        with patch.dict("sys.modules", {"ddgs": mock_module}):
             client = DuckDuckGoClient(timeout=10)
             result = client.search("test query", max_results=3)
         mock_ddgs.text.assert_called_once_with("test query", max_results=3)

tests/test_workstream.py

@@ -54,7 +54,7 @@ class FakeUI:
     def approve_tools(self, items):
         return True, None
 
-    def on_tool_result(self, call_id, name, output):
+    def on_tool_result(self, call_id, name, output, **kwargs):
         pass
 
     def on_tool_output_chunk(self, call_id, chunk):

turnstone.example.toml

@@ -0,0 +1,124 @@
+# turnstone.toml — shared bootstrap configuration
+#
+# This file is read once at startup. Values here are overridden by
+# environment variables, which are in turn overridden by CLI flags.
+#
+# All sections are optional. Missing sections use binary defaults.
+# Config file location precedence:
+#   1. --config flag
+#   2. $TURNSTONE_CONFIG env var
+#   3. ~/.config/turnstone/config.toml
+
+# --- LLM API (turnstone, node, eval) ---
+
+[api]
+# base_url = ""                # API endpoint; empty = binary default
+# api_key = ""                 # env: OPENAI_API_KEY or ANTHROPIC_API_KEY
+
+# --- Default Model (turnstone, node, eval) ---
+
+[model]
+# name = ""                    # Model ID; empty = provider default (gpt-5 / claude-sonnet-4)
+# temperature = 0.0            # 0 = provider default
+# reasoning_effort = ""        # "low", "medium", "high", "max"
+# context_window = 0           # 0 = auto-detect from provider capabilities
+# max_tokens = 0               # 0 = provider default
+
+# --- Named Models (turnstone, node, eval) ---
+# Define model aliases with per-model overrides. Useful for local model
+# servers or mixing providers. Reference by name with --model flag.
+#
+# [models.local]
+# name = "llama-3-70b"
+# provider = "openai"
+# base_url = "http://localhost:8000/v1"
+# context_window = 8192
+#
+# [models.local.capabilities]
+# supports_vision = false
+# supports_web_search = false
+#
+# [models.claude]
+# name = "claude-opus-4-6"
+# provider = "anthropic"
+
+# --- Database (turnstone, node, console) ---
+
+[database]
+# url = ""                     # postgres://user:pass@host/db or /path/to.db
+                               # env: TURNSTONE_DB_URL
+# SSL params (passed through to SQLAlchemy connection):
+# sslmode = "prefer"           # disable, allow, prefer, require, verify-ca, verify-full
+# sslrootcert = ""             # path to CA cert for verify-ca/verify-full
+# sslcert = ""                 # path to client cert (mTLS)
+# sslkey = ""                  # path to client key (mTLS)
+
+# --- Redis (bridge, console, channel) ---
+
+[redis]
+# url = ""                     # redis://host:6379/0 or rediss://host:6380/0
+                               # env: TURNSTONE_REDIS_URL
+# TLS params (passed through to Redis connection):
+# tls = false                  # enable TLS (also auto-enabled by rediss:// scheme)
+# tls_ca = ""                  # path to CA cert
+# tls_cert = ""                # path to client cert (mTLS)
+# tls_key = ""                 # path to client key (mTLS)
+
+# --- Auth (node, console) ---
+
+[auth]
+# enabled = true               # env: TURNSTONE_AUTH_ENABLED
+# jwt_secret = ""              # HS256 signing secret (min 32 bytes recommended)
+                               # env: TURNSTONE_JWT_SECRET
+# token = ""                   # Static config token for full access
+                               # env: TURNSTONE_AUTH_TOKEN
+
+# --- Logging (turnstone, node, console) ---
+
+[log]
+# level = ""                   # "debug", "info", "warn", "error"
+                               # empty = binary default (warn for CLI, info for servers)
+                               # env: TURNSTONE_LOG_LEVEL
+# json = false                 # JSON output; auto-enabled when stderr is not a TTY
+
+# --- Session (turnstone, node) ---
+
+[session]
+# instructions = ""            # Default system message
+# compact_max_tokens = 32768   # Max tokens for context compaction summary
+# auto_compact_pct = 0.8       # Trigger compaction at this % of context window
+
+# --- Tools (turnstone, node) ---
+
+[tools]
+# timeout = 120                # Tool execution timeout in seconds
+# skip_permissions = false     # Auto-approve all tool calls
+
+# --- Judge (turnstone, node) ---
+
+[judge]
+# enabled = true               # Enable intent validation
+# confidence_threshold = 0.7   # Minimum confidence for heuristic verdicts
+# output_guard = true          # Scan tool output for security signals
+# redact_secrets = true        # Redact detected credentials in output
+
+# --- Memory (turnstone, node) ---
+
+[memory]
+# relevance_k = 5              # Top-K memories for context injection
+# fetch_limit = 50             # Max memories to fetch for ranking
+# max_content = 32768          # Max memory content size in chars
+# nudge_cooldown = 300         # Min seconds between metacognitive nudges
+# nudges = true                # Enable memory nudges
+
+# --- MCP (turnstone, node) ---
+
+[mcp]
+# config_path = ""             # Path to MCP servers config file (JSON)
+# refresh_interval = 14400     # Refresh interval in seconds (default: 4h)
+
+# --- Server (node, console) ---
+
+[server]
+# max_workstreams = 50         # Maximum concurrent workstreams per node
+                               # env: TURNSTONE_MAX_WORKSTREAMS

turnstone/__init__.py

@@ -1,3 +1,3 @@
 """turnstone - Multi-node AI orchestration platform with tool use, agent routing, and cluster simulation."""
 
-__version__ = "0.8.5"
+__version__ = "0.9.3"

turnstone/admin.py

@@ -142,6 +142,189 @@ def _cmd_revoke_token(args: argparse.Namespace) -> None:
         sys.exit(1)
 
 
+# ---------------------------------------------------------------------------
+# TLS commands
+# ---------------------------------------------------------------------------
+
+
+def _cmd_tls_bootstrap(args: argparse.Namespace) -> None:
+    """Initialize CA and issue certs offline."""
+    try:
+        from lacme import CertificateAuthority, FileStore
+    except ImportError:
+        print("lacme not installed. Run: pip install turnstone[tls]", file=sys.stderr)
+        sys.exit(1)
+
+    import contextlib
+    import os
+    from pathlib import Path
+
+    out_dir = Path(args.out)
+    out_dir.mkdir(parents=True, exist_ok=True)
+    with contextlib.suppress(PermissionError):
+        os.chmod(out_dir, 0o700)  # Restrict access — contains CA private key
+
+    store = FileStore(str(out_dir))
+    ca = CertificateAuthority(store, name="turnstone")
+    ca.init(cn="Turnstone CA", validity_days=3650)
+    print(f"CA initialized in {out_dir} (permissions: 0700)")
+
+    # Write CA cert to a well-known location
+    ca_cert_path = out_dir / "ca.pem"
+    ca_cert_path.write_bytes(ca.root_cert_pem)
+    with contextlib.suppress(PermissionError):
+        os.chmod(ca_cert_path, 0o644)
+    print(f"CA cert: {ca_cert_path}")
+
+    # Issue certs for requested domains
+    for domain in args.issue:
+        bundle = ca.issue([domain], validity_hours=48)
+        store.save_cert(bundle)
+        cert_dir = out_dir / "certs" / domain
+        print(f"Issued: {domain} -> {cert_dir}")
+
+    print(f"\nBootstrap complete. {len(args.issue)} cert(s) issued.")
+    print(f"CA and certs written to: {out_dir}")
+
+
+def _cmd_tls_issue(args: argparse.Namespace) -> None:
+    """Request a cert from the console's ACME endpoint."""
+    try:
+        from lacme import SyncClient
+    except ImportError:
+        print("lacme not installed. Run: pip install turnstone[tls]", file=sys.stderr)
+        sys.exit(1)
+
+    import os
+    from pathlib import Path
+
+    console_url = args.console_url
+    if not console_url:
+        console_url = _discover_console_url()
+
+    domains = [args.domain] + args.san
+    directory_url = f"{console_url}/acme/directory"
+    print(f"Requesting cert for {domains} from {directory_url}")
+
+    client = SyncClient(
+        directory_url=directory_url,
+        allow_insecure=True,
+    )
+    bundle = client.issue(domains)
+
+    out_dir = Path(args.out)
+    out_dir.mkdir(parents=True, exist_ok=True)
+    (out_dir / "cert.pem").write_bytes(bundle.cert_pem)
+    (out_dir / "fullchain.pem").write_bytes(bundle.fullchain_pem)
+    (out_dir / "key.pem").write_bytes(bundle.key_pem)
+    os.chmod(out_dir / "key.pem", 0o600)
+
+    print(f"Certificate written to {out_dir}/")
+    print("  cert.pem      (leaf certificate)")
+    print("  fullchain.pem (cert + chain)")
+    print("  key.pem       (private key, 0600)")
+
+
+def _cmd_tls_ca_cert(args: argparse.Namespace) -> None:
+    """Download the CA root certificate from the console."""
+    import httpx
+
+    console_url = args.console_url
+    if not console_url:
+        console_url = _discover_console_url()
+
+    # Use plain HTTP for bootstrap (node may not have CA cert yet)
+    # WARNING: This is trust-on-first-use (TOFU) — verify the fingerprint
+    base = console_url.replace("https://", "http://")
+    url = f"{base}/acme/ca.pem"
+    print(f"Fetching CA cert from {url}")
+    print("WARNING: Fetching over plain HTTP — verify the fingerprint below")
+
+    resp = httpx.get(url)
+    resp.raise_for_status()
+
+    # Show fingerprint for out-of-band verification
+    import hashlib
+
+    fingerprint = hashlib.sha256(resp.content).hexdigest()
+    print(f"CA cert SHA-256: {fingerprint}")
+
+    from pathlib import Path
+
+    Path(args.out).write_bytes(resp.content)
+    print(f"CA cert written to {args.out}")
+
+
+def _cmd_tls_list(args: argparse.Namespace) -> None:
+    """List certificates from the console."""
+    import httpx
+
+    console_url = args.console_url
+    if not console_url:
+        console_url = _discover_console_url()
+
+    url = f"{console_url}/v1/api/admin/tls/certs"
+    headers = {}
+    token = getattr(args, "auth_token", "") or _get_config_token()
+    if token:
+        headers["Authorization"] = f"Bearer {token}"
+    resp = httpx.get(url, headers=headers)
+    resp.raise_for_status()
+    data = resp.json()
+
+    certs = data.get("certs", [])
+    if not certs:
+        print("No certificates issued.")
+        return
+
+    print(f"{'DOMAIN':<30s} {'ISSUED':<22s} {'EXPIRES':<22s}")
+    print("-" * 74)
+    for c in certs:
+        print(f"{c['domain']:<30s} {c['issued_at']:<22s} {c['expires_at']:<22s}")
+
+
+def _get_config_token() -> str:
+    """Try to load auth token from config.toml or environment."""
+    token = os.environ.get("TURNSTONE_AUTH_TOKEN", "")
+    if token:
+        return token
+    try:
+        from turnstone.core.config import load_config
+
+        cfg = load_config("auth")
+        return str(cfg.get("token", ""))
+    except Exception:
+        return ""
+
+
+def _discover_console_url() -> str:
+    """Discover console URL from the services table."""
+    from turnstone.core.storage import get_storage
+
+    try:
+        storage = get_storage()
+    except Exception:
+        print(
+            "No storage configured. Use --console-url or run from a "
+            "directory with a turnstone database.",
+            file=sys.stderr,
+        )
+        sys.exit(1)
+    consoles = storage.list_services("console", max_age_seconds=3600)
+    if not consoles:
+        print(
+            "No console found in services table. Use --console-url explicitly.",
+            file=sys.stderr,
+        )
+        sys.exit(1)
+    return consoles[0]["url"]
+
+
+# ---------------------------------------------------------------------------
+# Entry point
+# ---------------------------------------------------------------------------
+
+
 def main() -> None:
     """Entry point for turnstone-admin CLI."""
     parser = argparse.ArgumentParser(
@@ -171,6 +354,35 @@ def main() -> None:
     p_rt = sub.add_parser("revoke-token", help="Revoke an API token")
     p_rt.add_argument("--token-id", required=True, help="Token ID to revoke")
 
+    # TLS subcommands
+    p_bootstrap = sub.add_parser(
+        "tls-bootstrap",
+        help="Initialize CA and issue certs offline (no running console needed)",
+    )
+    p_bootstrap.add_argument("--out", required=True, help="Output directory for PEM files")
+    p_bootstrap.add_argument(
+        "--issue",
+        action="append",
+        default=[],
+        help="Domain to issue cert for (repeatable)",
+    )
+
+    p_issue = sub.add_parser("tls-issue", help="Request cert from console ACME")
+    p_issue.add_argument("domain", help="Primary domain for the certificate")
+    p_issue.add_argument("--san", action="append", default=[], help="Additional SAN (repeatable)")
+    p_issue.add_argument("--out", default=".", help="Output directory for PEM files")
+    p_issue.add_argument(
+        "--console-url", default="", help="Console URL (discovered from DB if empty)"
+    )
+
+    p_cacert = sub.add_parser("tls-ca-cert", help="Download CA root certificate")
+    p_cacert.add_argument("--out", default="ca.pem", help="Output file path")
+    p_cacert.add_argument("--console-url", default="", help="Console URL")
+
+    p_tlslist = sub.add_parser("tls-list", help="List issued certificates")
+    p_tlslist.add_argument("--console-url", default="", help="Console URL")
+    p_tlslist.add_argument("--auth-token", default="", help="Auth token for admin API")
+
     args = parser.parse_args()
     if not args.command:
         parser.print_help()
@@ -182,5 +394,9 @@ def main() -> None:
         "list-users": _cmd_list_users,
         "list-tokens": _cmd_list_tokens,
         "revoke-token": _cmd_revoke_token,
+        "tls-bootstrap": _cmd_tls_bootstrap,
+        "tls-issue": _cmd_tls_issue,
+        "tls-ca-cert": _cmd_tls_ca_cert,
+        "tls-list": _cmd_tls_list,
     }
     dispatch[args.command](args)

turnstone/api/console_schemas.py

@@ -785,3 +785,94 @@ class RegistryInstallRequest(BaseModel):
     variables: dict[str, str] = Field(default_factory=dict)
     env: dict[str, str] = Field(default_factory=dict)
     headers: dict[str, str] = Field(default_factory=dict)
+
+
+# ---------------------------------------------------------------------------
+# Admin: Model Definitions
+# ---------------------------------------------------------------------------
+
+
+class ModelDefinitionInfo(BaseModel):
+    definition_id: str
+    alias: str
+    model: str
+    provider: str = "openai"
+    base_url: str = ""
+    api_key: str = ""
+    context_window: int = 32768
+    capabilities: str = "{}"
+    enabled: bool = True
+    source: str = ""
+    created_by: str = ""
+    created: str = ""
+    updated: str = ""
+
+
+class CreateModelDefinitionRequest(BaseModel):
+    alias: str
+    model: str
+    provider: str = "openai"
+    base_url: str = ""
+    api_key: str = ""
+    context_window: int = 32768
+    capabilities: dict[str, Any] = Field(default_factory=dict)
+    enabled: bool = True
+
+
+class UpdateModelDefinitionRequest(BaseModel):
+    alias: str | None = None
+    model: str | None = None
+    provider: str | None = None
+    base_url: str | None = None
+    api_key: str | None = None
+    context_window: int | None = None
+    capabilities: dict[str, Any] | None = None
+    enabled: bool | None = None
+
+
+class ListModelDefinitionsResponse(BaseModel):
+    models: list[ModelDefinitionInfo]
+
+
+class ModelReloadResponse(BaseModel):
+    status: str = "ok"
+    results: dict[str, Any] = Field(default_factory=dict)
+
+
+class DetectModelRequest(BaseModel):
+    provider: str = "openai"
+    base_url: str = ""
+    api_key: str = ""
+    model: str = ""
+    definition_id: str = ""
+
+
+class DetectModelResponse(BaseModel):
+    reachable: bool = False
+    model_found: bool | None = None
+    available_models: list[str] = Field(default_factory=list)
+    context_window: int | None = None
+    server_type: str | None = None
+    error: str | None = None
+
+
+class ModelCapabilitiesResponse(BaseModel):
+    model: str
+    provider: str
+    known: bool = False
+    capabilities: dict[str, Any] = Field(default_factory=dict)
+
+
+class KnownModelsResponse(BaseModel):
+    provider: str
+    models: list[str] = Field(default_factory=list)
+
+
+class AvailableModelInfo(BaseModel):
+    alias: str
+    model: str
+    provider: str
+
+
+class ListAvailableModelsResponse(BaseModel):
+    models: list[AvailableModelInfo] = Field(default_factory=list)

turnstone/api/console_spec.py

@@ -11,6 +11,7 @@ from turnstone.api.console_schemas import (
     AdminMemoryInfo,
     AssignRoleRequest,
     AuditEventInfo,
+    AvailableModelInfo,
     ChannelUserInfo,
     ClusterNodesResponse,
     ClusterOverviewResponse,
@@ -21,16 +22,22 @@ from turnstone.api.console_schemas import (
     ConsoleHealthResponse,
     CreateChannelUserRequest,
     CreateMcpServerRequest,
+    CreateModelDefinitionRequest,
     CreateRoleRequest,
     CreateSkillRequest,
     CreateSkillResourceRequest,
     CreateToolPolicyRequest,
+    DetectModelRequest,
+    DetectModelResponse,
     ImportMcpConfigRequest,
     ImportMcpConfigResponse,
+    KnownModelsResponse,
     ListAdminMemoriesResponse,
     ListAuditEventsResponse,
+    ListAvailableModelsResponse,
     ListChannelUsersResponse,
     ListMcpServersResponse,
+    ListModelDefinitionsResponse,
     ListOrgsResponse,
     ListOutputAssessmentsResponse,
     ListRolesResponse,
@@ -44,6 +51,9 @@ from turnstone.api.console_schemas import (
     ListVerdictsResponse,
     McpReloadResponse,
     McpServerDetail,
+    ModelCapabilitiesResponse,
+    ModelDefinitionInfo,
+    ModelReloadResponse,
     NodeDetailResponse,
     OrgInfo,
     OutputAssessmentInfo,
@@ -60,6 +70,7 @@ from turnstone.api.console_schemas import (
     SkillVersionInfo,
     ToolPolicyInfo,
     UpdateMcpServerRequest,
+    UpdateModelDefinitionRequest,
     UpdateOrgRequest,
     UpdateRoleRequest,
     UpdateSettingRequest,
@@ -560,6 +571,14 @@ CONSOLE_ENDPOINTS: list[EndpointSpec] = [
         response_model=ListSkillVersionsResponse,
         tags=["Admin"],
     ),
+    # --- Models ---
+    EndpointSpec(
+        "/v1/api/models",
+        "GET",
+        "List enabled model aliases for workstream creation",
+        response_model=ListAvailableModelsResponse,
+        tags=["Models"],
+    ),
     # --- Skills ---
     EndpointSpec(
         "/v1/api/skills",
@@ -843,6 +862,117 @@ CONSOLE_ENDPOINTS: list[EndpointSpec] = [
         error_codes=[400],
         tags=["Admin"],
     ),
+    # --- Admin: Model Definitions ---
+    EndpointSpec(
+        "/v1/api/admin/model-definitions",
+        "GET",
+        "List model definitions with live status from cluster nodes",
+        response_model=ListModelDefinitionsResponse,
+        tags=["Admin"],
+    ),
+    EndpointSpec(
+        "/v1/api/admin/model-definitions",
+        "POST",
+        "Create a model definition",
+        request_model=CreateModelDefinitionRequest,
+        response_model=ModelDefinitionInfo,
+        error_codes=[400, 409],
+        tags=["Admin"],
+    ),
+    EndpointSpec(
+        "/v1/api/admin/model-definitions/reload",
+        "POST",
+        "Tell all nodes to re-read model definitions from DB and rebuild registry",
+        response_model=ModelReloadResponse,
+        tags=["Admin"],
+    ),
+    EndpointSpec(
+        "/v1/api/admin/model-definitions/{definition_id}",
+        "GET",
+        "Get a single model definition",
+        response_model=ModelDefinitionInfo,
+        error_codes=[404],
+        tags=["Admin"],
+    ),
+    EndpointSpec(
+        "/v1/api/admin/model-definitions/{definition_id}",
+        "PUT",
+        "Update a model definition",
+        request_model=UpdateModelDefinitionRequest,
+        response_model=ModelDefinitionInfo,
+        error_codes=[400, 404, 409],
+        tags=["Admin"],
+    ),
+    EndpointSpec(
+        "/v1/api/admin/model-definitions/{definition_id}",
+        "DELETE",
+        "Delete a model definition",
+        error_codes=[404],
+        tags=["Admin"],
+    ),
+    EndpointSpec(
+        "/v1/api/admin/model-definitions/detect",
+        "POST",
+        "Probe a model endpoint: verify reachability, list models, detect context window and server type",
+        request_model=DetectModelRequest,
+        response_model=DetectModelResponse,
+        error_codes=[400],
+        tags=["Admin"],
+    ),
+    EndpointSpec(
+        "/v1/api/admin/model-capabilities",
+        "GET",
+        "Look up static capabilities for a known model",
+        response_model=ModelCapabilitiesResponse,
+        query_params=[
+            QueryParam(name="provider", description="Provider name", required=True),
+            QueryParam(name="model", description="Model ID to look up", required=True),
+        ],
+        tags=["Admin"],
+    ),
+    EndpointSpec(
+        "/v1/api/admin/model-capabilities/known",
+        "GET",
+        "List known model name prefixes for a provider",
+        response_model=KnownModelsResponse,
+        query_params=[
+            QueryParam(name="provider", description="Provider name", required=True),
+        ],
+        tags=["Admin"],
+    ),
+    # --- Admin: TLS / ACME ---
+    EndpointSpec(
+        "/v1/api/admin/tls/ca",
+        "GET",
+        "CA status: initialization state, CN, cert count, cert inventory",
+        tags=["Admin"],
+    ),
+    EndpointSpec(
+        "/v1/api/admin/tls/ca.pem",
+        "GET",
+        "Download CA root certificate (PEM format)",
+        tags=["Admin"],
+    ),
+    EndpointSpec(
+        "/v1/api/admin/tls/certs",
+        "GET",
+        "List all issued TLS certificates",
+        tags=["Admin"],
+    ),
+    EndpointSpec(
+        "/v1/api/admin/tls/certs/{domain}/renew",
+        "POST",
+        "Force-renew a certificate by domain",
+        error_codes=[404, 500],
+        tags=["Admin"],
+    ),
+    EndpointSpec(
+        "/v1/api/admin/tls/certs/{domain}",
+        "DELETE",
+        "Delete a certificate by domain",
+        error_codes=[404],
+        tags=["Admin"],
+    ),
     # --- Observability ---
     EndpointSpec(
         "/health",
@@ -920,6 +1050,17 @@ _ALL_MODELS: list[type[BaseModel]] = [
     ImportMcpConfigRequest,
     ImportMcpConfigResponse,
     McpReloadResponse,
+    ModelDefinitionInfo,
+    CreateModelDefinitionRequest,
+    UpdateModelDefinitionRequest,
+    ListModelDefinitionsResponse,
+    ModelReloadResponse,
+    DetectModelRequest,
+    DetectModelResponse,
+    ModelCapabilitiesResponse,
+    KnownModelsResponse,
+    AvailableModelInfo,
+    ListAvailableModelsResponse,
     RegistrySearchResponse,
     RegistryInstallRequest,
     SkillDiscoverResponse,

turnstone/api/server_schemas.py

@@ -41,6 +41,11 @@ class CommandRequest(BaseModel):
 
 class CancelRequest(BaseModel):
     ws_id: str = Field(description="Target workstream ID")
+    force: bool = Field(
+        default=False,
+        description="Force cancel: abandon the stuck worker thread immediately. "
+        "Use when cooperative cancel has not resolved within a few seconds.",
+    )
 
 
 class CreateWorkstreamRequest(BaseModel):
@@ -252,3 +257,13 @@ class SkillSummary(BaseModel):
 
 class ListSkillSummaryResponse(BaseModel):
     skills: list[SkillSummary]
+
+
+class AvailableModelInfo(BaseModel):
+    alias: str
+    model: str
+    provider: str
+
+
+class ListAvailableModelsResponse(BaseModel):
+    models: list[AvailableModelInfo] = Field(default_factory=list)

turnstone/api/server_spec.py

@@ -20,6 +20,7 @@ from turnstone.api.schemas import (
 )
 from turnstone.api.server_schemas import (
     ApproveRequest,
+    AvailableModelInfo,
     CancelRequest,
     CloseWorkstreamRequest,
     CommandRequest,
@@ -27,6 +28,7 @@ from turnstone.api.server_schemas import (
     CreateWorkstreamResponse,
     DashboardResponse,
     HealthResponse,
+    ListAvailableModelsResponse,
     ListMemoriesResponse,
     ListSavedWorkstreamsResponse,
     ListSkillSummaryResponse,
@@ -155,6 +157,14 @@ SERVER_ENDPOINTS: list[EndpointSpec] = [
         response_model=ListSkillSummaryResponse,
         tags=["Skills"],
     ),
+    # --- Models ---
+    EndpointSpec(
+        "/v1/api/models",
+        "GET",
+        "List available model aliases",
+        response_model=ListAvailableModelsResponse,
+        tags=["Models"],
+    ),
     # --- Auth ---
     EndpointSpec(
         "/v1/api/auth/login",
@@ -293,6 +303,8 @@ _ALL_MODELS: list[type[BaseModel]] = [
     SearchMemoriesRequest,
     SkillSummary,
     ListSkillSummaryResponse,
+    AvailableModelInfo,
+    ListAvailableModelsResponse,
 ]
 
 

turnstone/channels/cli.py

@@ -58,6 +58,11 @@ def main() -> None:
         help="HTTP server port (default: $TURNSTONE_CHANNEL_PORT or 8091)",
     )
 
+    # -- TLS -----------------------------------------------------------------
+    parser.add_argument("--ssl-certfile", default=None, help="SSL certificate file for HTTPS")
+    parser.add_argument("--ssl-keyfile", default=None, help="SSL private key file")
+    parser.add_argument("--ssl-ca-certs", default=None, help="SSL CA certs for client verification")
+
     # -- Auth ----------------------------------------------------------------
     parser.add_argument(
         "--auth-token",
@@ -189,7 +194,8 @@ def main() -> None:
                     advertise_host = socket.gethostname()
                 else:
                     advertise_host = args.http_host
-                advertise_url = f"http://{advertise_host}:{args.http_port}"
+                scheme = "https" if args.ssl_certfile else "http"
+                advertise_url = f"{scheme}://{advertise_host}:{args.http_port}"
             service_url = advertise_url
 
             # Register in service registry
@@ -209,11 +215,25 @@ def main() -> None:
                     except Exception:
                         log.exception("channel.heartbeat_failed")
 
+            # TLS: use cert files if available (from bootstrap or TLSClient)
+            ssl_certfile = getattr(args, "ssl_certfile", None)
+            ssl_keyfile = getattr(args, "ssl_keyfile", None)
+            ssl_ca_certs = getattr(args, "ssl_ca_certs", None)
+            if bool(ssl_certfile) != bool(ssl_keyfile):
+                print(
+                    "Both --ssl-certfile and --ssl-keyfile are required for TLS",
+                    file=sys.stderr,
+                )
+                sys.exit(1)
+
             uv_config = uvicorn.Config(
                 channel_app,
                 host=args.http_host,
                 port=args.http_port,
                 log_level="warning",
+                ssl_certfile=ssl_certfile,
+                ssl_keyfile=ssl_keyfile,
+                ssl_ca_certs=ssl_ca_certs,
             )
             server = uvicorn.Server(uv_config)
 

turnstone/cli.py

@@ -63,6 +63,8 @@ SLASH_COMMANDS = [
     "/creative",
     "/debug",
     "/mcp",
+    "/retry",
+    "/rewind",
     "/help",
     "/exit",
     "/quit",
@@ -251,8 +253,18 @@ class TerminalUI(SessionUI):
                     item["denial_msg"] = denial_msg
                 return False, None
 
-    def on_tool_result(self, call_id: str, name: str, output: str) -> None:
-        pass  # Optional: display summary
+    def on_tool_result(
+        self,
+        call_id: str,
+        name: str,
+        output: str,
+        *,
+        is_error: bool = False,
+    ) -> None:
+        if is_error:
+            with self._print_lock:
+                sys.stderr.write(f"{RED}\u2717 {name}: {output}{RESET}\n")
+                sys.stderr.flush()
 
     def on_tool_output_chunk(self, call_id: str, chunk: str) -> None:
         pass  # Terminal shows spinner during tool execution
@@ -424,9 +436,16 @@ class WorkstreamTerminalUI(TerminalUI):
         else:
             self._buffer("error", message)
 
-    def on_tool_result(self, call_id: str, name: str, output: str) -> None:
+    def on_tool_result(
+        self,
+        call_id: str,
+        name: str,
+        output: str,
+        *,
+        is_error: bool = False,
+    ) -> None:
         if self.is_foreground:
-            super().on_tool_result(call_id, name, output)
+            super().on_tool_result(call_id, name, output, is_error=is_error)
 
     def on_tool_output_chunk(self, call_id: str, chunk: str) -> None:
         if self.is_foreground:
@@ -1237,6 +1256,16 @@ def main() -> None:
             should_exit = active.session.handle_command(user_input)
             if should_exit:
                 break
+            # Dispatch deferred retry (handle_command sets _pending_retry)
+            retry_msg = active.session._pending_retry
+            if retry_msg:
+                active.session._pending_retry = None
+                try:
+                    active.session.send(retry_msg)
+                except KeyboardInterrupt:
+                    print(f"\n{yellow('Interrupted.')}")
+                except Exception as e:
+                    print(f"\n{red(f'Error: {e}')}")
         else:
             try:
                 active.session.send(user_input)

turnstone/console/collector.py

@@ -60,6 +60,8 @@ class ClusterCollector:
         http_timeout: float = 30.0,
         auth_token: str = "",
         token_manager: ServiceTokenManager | None = None,
+        tls_verify: Any = True,
+        tls_cert: tuple[str, str] | None = None,
     ):
         self._broker = broker
         self._prefix = prefix
@@ -86,12 +88,33 @@ class ClusterCollector:
                 max_connections=max_poll_workers + 10,
                 max_keepalive_connections=min(max_poll_workers, 200),
             ),
+            verify=tls_verify,
+            cert=tls_cert,
         )
 
         # SSE fan-out to browser clients
         self._listeners: list[queue.Queue[dict[str, Any]]] = []
         self._listeners_lock = threading.Lock()
 
+    def upgrade_tls(self, tls_verify: Any = True, tls_cert: tuple[str, str] | None = None) -> None:
+        """Replace the httpx client with one using mTLS context."""
+        old = self._http_client
+        self._http_client = httpx.Client(
+            timeout=httpx.Timeout(
+                connect=10, read=self._http_timeout, write=5, pool=self._http_timeout
+            ),
+            limits=httpx.Limits(
+                max_connections=self._max_poll_workers + 10,
+                max_keepalive_connections=min(self._max_poll_workers, 200),
+            ),
+            verify=tls_verify,
+            cert=tls_cert,
+        )
+        # Don't close old client — concurrent _fetch_node() threads may still
+        # be using it.  It will be GC'd once all references are released, and
+        # the current client is closed in stop().
+        del old
+
     # -- lifecycle -----------------------------------------------------------
 
     def start(self) -> None:

turnstone/console/static/admin.js

@@ -64,7 +64,9 @@ function showAdmin() {
     audit: "admin.audit",
     memories: "admin.memories",
     settings: "admin.settings",
+    tls: "admin.settings",
     mcp: "admin.mcp",
+    models: "admin.models",
   };
   if (perms) {
     var permSet = perms.split(",");
@@ -192,7 +194,9 @@ function switchAdminTab(tab) {
     "usage",
     "audit",
     "memories",
+    "models",
     "settings",
+    "tls",
     "mcp",
   ];
   for (var p = 0; p < panels.length; p++) {
@@ -214,7 +218,9 @@ function switchAdminTab(tab) {
     loadGovAudit();
   }
   if (tab === "memories") loadAdminMemories();
+  if (tab === "models") loadAdminModels();
   if (tab === "settings") loadSettings();
+  if (tab === "tls") loadTlsCerts();
   if (tab === "mcp") loadAdminMcp();
 
   // Update breadcrumb with active tab label
@@ -1841,6 +1847,7 @@ function _installTrap(overlayId, boxId, trapRef) {
         else if (overlayId === "mcp-detail-overlay") hideMcpDetailModal();
         else if (overlayId === "mcp-install-overlay") hideInstallMcpModal();
         else if (overlayId === "github-import-overlay") hideGitHubImportModal();
+        else if (overlayId === "model-create-overlay") hideCreateModelModal();
       }
     };
   }
@@ -1929,6 +1936,7 @@ document.addEventListener("keydown", function (e) {
     ["mcp-import-overlay", hideImportMcpModal],
     ["mcp-create-overlay", hideCreateMcpModal],
     ["github-import-overlay", hideGitHubImportModal],
+    ["model-create-overlay", hideCreateModelModal],
   ];
   for (var gi = 0; gi < govOverlays.length; gi++) {
     var govEl = document.getElementById(govOverlays[gi][0]);
@@ -2083,6 +2091,180 @@ function _settingsSectionLabel(section) {
   return labels[section] || section;
 }
 
+// ---------------------------------------------------------------------------
+// TLS tab
+// ---------------------------------------------------------------------------
+
+function loadTlsCerts() {
+  var statusEl = document.getElementById("tls-ca-status");
+  var listEl = document.getElementById("tls-cert-list");
+  if (!statusEl || !listEl) return;
+
+  // Fetch CA status and cert list in parallel
+  Promise.all([
+    authFetch("/v1/api/admin/tls/ca").then(function (r) {
+      if (!r.ok) throw new Error("Failed");
+      return r.json();
+    }),
+    authFetch("/v1/api/admin/tls/certs").then(function (r) {
+      if (!r.ok) return { certs: [] };
+      return r.json();
+    }),
+  ])
+    .then(function (results) {
+      var data = results[0];
+      var certData = results[1];
+      while (statusEl.firstChild) statusEl.removeChild(statusEl.firstChild);
+      while (listEl.firstChild) listEl.removeChild(listEl.firstChild);
+
+      if (!data.enabled) {
+        var msg = document.createElement("div");
+        msg.className = "dashboard-empty";
+        msg.textContent =
+          "TLS is not enabled. Set tls.enabled = true in Settings.";
+        statusEl.appendChild(msg);
+        return;
+      }
+
+      // CA status bar
+      var bar = document.createElement("div");
+      bar.className = "tls-ca-bar";
+      var caLabel = document.createElement("span");
+      caLabel.textContent = "CA: " + data.ca_cn;
+      var countLabel = document.createElement("span");
+      countLabel.textContent = "Certificates: " + data.cert_count;
+      bar.appendChild(caLabel);
+      bar.appendChild(countLabel);
+      statusEl.appendChild(bar);
+
+      var certs = certData.certs || [];
+      if (certs.length === 0) {
+        var empty = document.createElement("div");
+        empty.className = "dashboard-empty";
+        empty.textContent = "No certificates issued yet.";
+        listEl.appendChild(empty);
+        return;
+      }
+
+      // Cert rows
+      certs.forEach(function (c) {
+        var row = document.createElement("div");
+        row.className = "admin-row";
+        row.setAttribute("role", "listitem");
+
+        var colDomain = document.createElement("span");
+        colDomain.className = "admin-col";
+        colDomain.textContent = c.domain;
+
+        var colSans = document.createElement("span");
+        colSans.className = "admin-col";
+        colSans.textContent = (c.domains || [c.domain]).join(", ");
+
+        var colIssued = document.createElement("span");
+        colIssued.className = "admin-col";
+        colIssued.textContent = (c.issued_at || "")
+          .slice(0, 16)
+          .replace("T", " ");
+
+        var colExpires = document.createElement("span");
+        colExpires.className = "admin-col";
+        var expires = new Date(c.expires_at);
+        var isExpired = expires < new Date();
+        colExpires.textContent =
+          (isExpired ? "EXPIRED " : "") +
+          (c.expires_at || "").slice(0, 16).replace("T", " ");
+        if (isExpired) colExpires.style.color = "var(--red)";
+
+        var colActions = document.createElement("span");
+        colActions.className = "admin-col admin-col-actions";
+        var renewBtn = document.createElement("button");
+        renewBtn.className = "admin-btn-action";
+        renewBtn.textContent = "Renew";
+        renewBtn.setAttribute(
+          "aria-label",
+          "Renew certificate for " + c.domain,
+        );
+        renewBtn.onclick = function () {
+          tlsRenewCert(c.domain);
+        };
+        var deleteBtn = document.createElement("button");
+        deleteBtn.className = "admin-btn-danger";
+        deleteBtn.textContent = "Delete";
+        deleteBtn.setAttribute(
+          "aria-label",
+          "Delete certificate for " + c.domain,
+        );
+        deleteBtn.onclick = function () {
+          tlsDeleteCert(c.domain);
+        };
+        colActions.appendChild(renewBtn);
+        colActions.appendChild(deleteBtn);
+
+        row.appendChild(colDomain);
+        row.appendChild(colSans);
+        row.appendChild(colIssued);
+        row.appendChild(colExpires);
+        row.appendChild(colActions);
+        listEl.appendChild(row);
+      });
+    })
+    .catch(function () {
+      while (statusEl.firstChild) statusEl.removeChild(statusEl.firstChild);
+      while (listEl.firstChild) listEl.removeChild(listEl.firstChild);
+      var errMsg = document.createElement("div");
+      errMsg.className = "dashboard-empty";
+      errMsg.textContent = "Failed to load TLS status";
+      statusEl.appendChild(errMsg);
+    });
+}
+
+function tlsRenewCert(domain) {
+  showConfirmModal(
+    "Renew Certificate",
+    "Force renew certificate for \u2018" + domain + "\u2019?",
+    "Renew",
+    function () {
+      authFetch(
+        "/v1/api/admin/tls/certs/" + encodeURIComponent(domain) + "/renew",
+        { method: "POST" },
+      )
+        .then(function (r) {
+          if (!r.ok) throw new Error("Renew failed");
+          showToast("Certificate renewed for " + domain);
+          loadTlsCerts();
+        })
+        .catch(function () {
+          showToast("Failed to renew certificate", "error");
+        });
+    },
+  );
+}
+
+function tlsDeleteCert(domain) {
+  showConfirmModal(
+    "Delete Certificate",
+    "Delete certificate for \u2018" + domain + "\u2019? This cannot be undone.",
+    "Delete",
+    function () {
+      authFetch("/v1/api/admin/tls/certs/" + encodeURIComponent(domain), {
+        method: "DELETE",
+      })
+        .then(function (r) {
+          if (!r.ok) throw new Error("Delete failed");
+          showToast("Certificate deleted for " + domain);
+          loadTlsCerts();
+        })
+        .catch(function () {
+          showToast("Failed to delete certificate", "error");
+        });
+    },
+  );
+}
+
+// ---------------------------------------------------------------------------
+// Settings tab
+// ---------------------------------------------------------------------------
+
 function loadSettings() {
   var el = document.getElementById("admin-settings-content");
   if (!el) return;
@@ -3848,3 +4030,563 @@ function _pollInstallStatus(serverId, serverName, attempt) {
       .catch(function () {});
   }, 3000);
 }
+
+// ---------------------------------------------------------------------------
+// Models tab
+// ---------------------------------------------------------------------------
+
+var _modelDefs = [];
+var _modelCreateTrap = null;
+var _modelCreateTrigger = null;
+
+function loadAdminModels() {
+  authFetch("/v1/api/admin/model-definitions")
+    .then(function (r) {
+      if (!r.ok) throw new Error("Failed");
+      return r.json();
+    })
+    .then(function (data) {
+      _modelDefs = data.models || [];
+      _renderModels(_modelDefs);
+    })
+    .catch(function () {
+      var el = document.getElementById("admin-models-table");
+      el.textContent = "";
+      var d = document.createElement("div");
+      d.className = "dashboard-empty";
+      d.textContent = "Failed to load models";
+      el.appendChild(d);
+    });
+}
+
+function _renderModels(items) {
+  var el = document.getElementById("admin-models-table");
+  // Clear previous content
+  el.textContent = "";
+  if (!items.length) {
+    var empty = document.createElement("div");
+    empty.className = "dashboard-empty";
+    empty.textContent = "No model definitions configured";
+    el.appendChild(empty);
+    return;
+  }
+  for (var i = 0; i < items.length; i++) {
+    var m = items[i];
+    var isConfig = m.source === "config";
+
+    // Status
+    var dotClass = m.enabled
+      ? "model-status-dot enabled"
+      : "model-status-dot disabled";
+    var rowClass = m.enabled ? "model-row-enabled" : "model-row-disabled";
+    var statusText = m.enabled ? "enabled" : "disabled";
+
+    // Context window formatting (0 = auto-detect)
+    var ctxText = m.context_window
+      ? m.context_window >= 1000
+        ? Math.round(m.context_window / 1000) + "k"
+        : String(m.context_window)
+      : "auto";
+
+    // Provider badge class
+    var providerCls =
+      m.provider === "anthropic"
+        ? "model-provider-anthropic"
+        : "model-provider-openai";
+
+    // Build row via DOM
+    var row = document.createElement("div");
+    row.className = "admin-row models-grid " + rowClass;
+    row.setAttribute("role", "listitem");
+
+    // Alias + source badge
+    var colAlias = document.createElement("span");
+    colAlias.className = "admin-col";
+    colAlias.textContent = m.alias;
+    var badge = document.createElement("span");
+    badge.className = isConfig
+      ? "scope-badge scope-config"
+      : "scope-badge scope-db";
+    badge.textContent = isConfig ? "config" : "db";
+    colAlias.appendChild(document.createTextNode(" "));
+    colAlias.appendChild(badge);
+    row.appendChild(colAlias);
+
+    // Model ID
+    var colModel = document.createElement("span");
+    colModel.className = "admin-col";
+    var code = document.createElement("code");
+    code.textContent = m.model;
+    colModel.appendChild(code);
+    row.appendChild(colModel);
+
+    // Provider
+    var colProvider = document.createElement("span");
+    colProvider.className = "admin-col";
+    var provBadge = document.createElement("span");
+    provBadge.className = "model-provider-badge " + providerCls;
+    provBadge.textContent = m.provider;
+    colProvider.appendChild(provBadge);
+    row.appendChild(colProvider);
+
+    // Context window
+    var colCtx = document.createElement("span");
+    colCtx.className = "admin-col";
+    colCtx.textContent = ctxText;
+    row.appendChild(colCtx);
+
+    // Status
+    var colStatus = document.createElement("span");
+    colStatus.className = "admin-col";
+    var dot = document.createElement("span");
+    dot.className = dotClass;
+    dot.setAttribute("aria-hidden", "true");
+    colStatus.appendChild(dot);
+    colStatus.appendChild(document.createTextNode(statusText));
+    row.appendChild(colStatus);
+
+    // Actions
+    var colActions = document.createElement("span");
+    colActions.className = "admin-col";
+    if (!isConfig) {
+      var editBtn = document.createElement("button");
+      editBtn.className = "admin-btn-action";
+      editBtn.textContent = "edit";
+      editBtn.setAttribute("data-model-edit", m.definition_id);
+      colActions.appendChild(editBtn);
+
+      var delBtn = document.createElement("button");
+      delBtn.className = "admin-btn-danger";
+      delBtn.textContent = "del";
+      delBtn.setAttribute("data-model-delete", m.definition_id);
+      delBtn.setAttribute("data-model-alias", m.alias);
+      colActions.appendChild(delBtn);
+    }
+    row.appendChild(colActions);
+
+    el.appendChild(row);
+  }
+
+  // Bind event handlers
+  el.querySelectorAll("[data-model-edit]").forEach(function (btn) {
+    btn.addEventListener("click", function () {
+      showEditModelModal(this.getAttribute("data-model-edit"));
+    });
+  });
+  el.querySelectorAll("[data-model-delete]").forEach(function (btn) {
+    btn.addEventListener("click", function () {
+      var did = this.getAttribute("data-model-delete");
+      var dalias = this.getAttribute("data-model-alias");
+      showConfirmModal(
+        "Delete Model",
+        'Delete model "' + dalias + '"?',
+        "Delete",
+        function () {
+          authFetch(
+            "/v1/api/admin/model-definitions/" + encodeURIComponent(did),
+            {
+              method: "DELETE",
+            },
+          )
+            .then(function (r) {
+              if (!r.ok) throw new Error();
+              return r.json();
+            })
+            .then(function () {
+              showToast("Model deleted");
+              _flagModelSyncPending();
+              loadAdminModels();
+            })
+            .catch(function () {
+              showToast("Failed to delete model");
+            });
+        },
+      );
+    });
+  });
+}
+
+function showCreateModelModal() {
+  _modelCreateTrigger = document.activeElement;
+  var ov = document.getElementById("model-create-overlay");
+  ov.style.display = "flex";
+  document.getElementById("model-edit-id").value = "";
+  document.getElementById("model-create-title").textContent = "Add Model";
+  document.getElementById("model-create-submit").textContent = "Create";
+  document.getElementById("model-create-error").classList.remove("is-visible");
+  document.getElementById("model-alias").value = "";
+  document.getElementById("model-name").value = "";
+  document.getElementById("model-provider").value = "openai";
+  document.getElementById("model-base-url").value = "";
+  document.getElementById("model-api-key").value = "";
+  document.getElementById("model-api-key").placeholder = "sk-...";
+  document.getElementById("model-ctx-window").value = "0";
+  document.getElementById("model-capabilities").value = "";
+  document.getElementById("model-enabled").checked = true;
+  document.getElementById("model-detect-result").style.display = "none";
+  document.getElementById("model-detect-btn").disabled = false;
+  document.getElementById("model-detect-btn").textContent = "Detect";
+  _refreshModelSuggestions();
+  document.getElementById("model-alias").focus();
+  _modelCreateTrap = _installTrap("model-create-overlay", "model-create-box");
+}
+
+function showEditModelModal(definitionId) {
+  authFetch(
+    "/v1/api/admin/model-definitions/" + encodeURIComponent(definitionId),
+  )
+    .then(function (r) {
+      if (!r.ok) throw new Error("Failed");
+      return r.json();
+    })
+    .then(function (m) {
+      showCreateModelModal();
+      document.getElementById("model-edit-id").value = definitionId;
+      document.getElementById("model-create-title").textContent = "Edit Model";
+      document.getElementById("model-create-submit").textContent = "Save";
+      document.getElementById("model-alias").value = m.alias || "";
+      document.getElementById("model-name").value = m.model || "";
+      document.getElementById("model-provider").value = m.provider || "openai";
+      document.getElementById("model-base-url").value = m.base_url || "";
+      document.getElementById("model-api-key").value = "";
+      document.getElementById("model-api-key").placeholder =
+        "\u2022\u2022\u2022 (leave blank to keep existing)";
+      document.getElementById("model-ctx-window").value =
+        m.context_window != null ? m.context_window : 0;
+      // Parse capabilities JSON for display
+      var caps = m.capabilities || "{}";
+      try {
+        caps = JSON.stringify(JSON.parse(caps), null, 2);
+      } catch (e) {
+        /* keep raw */
+      }
+      if (caps === "{}") caps = "";
+      document.getElementById("model-capabilities").value = caps;
+      document.getElementById("model-enabled").checked = m.enabled !== false;
+    })
+    .catch(function () {
+      showToast("Failed to load model details");
+    });
+}
+
+function hideCreateModelModal() {
+  document.getElementById("model-create-overlay").style.display = "none";
+  _modelCreateTrap = _removeTrap(_modelCreateTrap);
+  if (_modelCreateTrigger && _modelCreateTrigger.focus)
+    _modelCreateTrigger.focus();
+  _modelCreateTrigger = null;
+}
+
+function submitCreateModel() {
+  var alias = document.getElementById("model-alias").value.trim();
+  var modelName = document.getElementById("model-name").value.trim();
+  if (!alias) {
+    _showModelError("Alias is required");
+    return;
+  }
+  if (!modelName) {
+    _showModelError("Model ID is required");
+    return;
+  }
+  if (!/^[a-zA-Z0-9._-]+$/.test(alias)) {
+    _showModelError("Alias must be alphanumeric (with . _ -)");
+    return;
+  }
+
+  var capsText = document.getElementById("model-capabilities").value.trim();
+  var caps = {};
+  if (capsText) {
+    try {
+      caps = JSON.parse(capsText);
+    } catch (e) {
+      _showModelError("Invalid JSON in capabilities");
+      return;
+    }
+  }
+
+  var form = {
+    alias: alias,
+    model: modelName,
+    provider: document.getElementById("model-provider").value,
+    base_url: document.getElementById("model-base-url").value.trim(),
+    context_window:
+      parseInt(document.getElementById("model-ctx-window").value, 10) || 0,
+    capabilities: caps,
+    enabled: document.getElementById("model-enabled").checked,
+  };
+
+  var apiKey = document.getElementById("model-api-key").value;
+  if (apiKey) form.api_key = apiKey;
+
+  var editId = document.getElementById("model-edit-id").value;
+  var method = editId ? "PUT" : "POST";
+  var url = editId
+    ? "/v1/api/admin/model-definitions/" + encodeURIComponent(editId)
+    : "/v1/api/admin/model-definitions";
+
+  document.getElementById("model-create-submit").disabled = true;
+  authFetch(url, {
+    method: method,
+    headers: { "Content-Type": "application/json" },
+    body: JSON.stringify(form),
+  })
+    .then(function (r) {
+      if (!r.ok)
+        return r.json().then(function (d) {
+          throw new Error(d.error || "Failed");
+        });
+      return r.json();
+    })
+    .then(function () {
+      hideCreateModelModal();
+      showToast(editId ? "Model updated" : "Model created");
+      _flagModelSyncPending();
+      loadAdminModels();
+    })
+    .catch(function (e) {
+      _showModelError(e.message);
+    })
+    .finally(function () {
+      document.getElementById("model-create-submit").disabled = false;
+    });
+}
+
+function _showModelError(msg) {
+  var e = document.getElementById("model-create-error");
+  e.textContent = msg;
+  e.classList.add("is-visible");
+}
+
+function _detectResultLine(text, color) {
+  var div = document.createElement("div");
+  div.style.marginTop = "3px";
+  if (color) div.style.color = "var(--" + color + ")";
+  div.textContent = text;
+  return div;
+}
+
+function _clearDetectResult() {
+  var rd = document.getElementById("model-detect-result");
+  if (rd) {
+    rd.style.display = "none";
+    rd.textContent = "";
+    rd.style.borderColor = "";
+  }
+}
+
+function detectModel() {
+  var btn = document.getElementById("model-detect-btn");
+  var resultDiv = document.getElementById("model-detect-result");
+  btn.disabled = true;
+  btn.setAttribute("aria-busy", "true");
+  btn.textContent = "Detecting\u2026";
+  resultDiv.style.display = "none";
+  resultDiv.textContent = "";
+
+  var form = {
+    provider: document.getElementById("model-provider").value,
+    base_url: document.getElementById("model-base-url").value.trim(),
+    model: document.getElementById("model-name").value.trim(),
+  };
+  var apiKey = document.getElementById("model-api-key").value;
+  if (apiKey) form.api_key = apiKey;
+  var editId = document.getElementById("model-edit-id").value;
+  if (editId) form.definition_id = editId;
+
+  authFetch("/v1/api/admin/model-definitions/detect", {
+    method: "POST",
+    headers: { "Content-Type": "application/json" },
+    body: JSON.stringify(form),
+  })
+    .then(function (r) {
+      if (!r.ok)
+        return r.json().then(function (d) {
+          throw new Error(d.error || "Detect failed");
+        });
+      return r.json();
+    })
+    .then(function (d) {
+      resultDiv.style.display = "block";
+      resultDiv.textContent = "";
+      if (d.error && !d.reachable) {
+        resultDiv.appendChild(
+          _detectResultLine("\u2717 Failed: " + d.error, "red"),
+        );
+        resultDiv.style.borderColor = "var(--red)";
+        return;
+      }
+      var line1 = "\u2713 Connected";
+      if (d.available_models && d.available_models.length) {
+        line1 += " \u2014 " + d.available_models.length + " model(s) available";
+      }
+      resultDiv.appendChild(_detectResultLine(line1, "green"));
+
+      if (d.model_found === false) {
+        var models = d.available_models || [];
+        var msg =
+          '\u26A0 Model "' +
+          form.model +
+          '" not found in ' +
+          models.length +
+          " available model(s)";
+        if (models.length > 0) {
+          var shown = models.slice(0, 8);
+          msg += ": " + shown.join(", ");
+          if (models.length > 8)
+            msg += ", \u2026 +" + (models.length - 8) + " more";
+        }
+        resultDiv.appendChild(_detectResultLine(msg, "yellow"));
+      }
+      if (d.context_window) {
+        resultDiv.appendChild(
+          _detectResultLine(
+            "Context window: " + d.context_window.toLocaleString() + " tokens",
+          ),
+        );
+        var ctxInput = document.getElementById("model-ctx-window");
+        if (parseInt(ctxInput.value, 10) === 0) {
+          ctxInput.value = d.context_window;
+        }
+      }
+      if (d.server_type) {
+        resultDiv.appendChild(
+          _detectResultLine("Server type: " + d.server_type),
+        );
+      }
+      resultDiv.style.borderColor = "var(--green)";
+    })
+    .catch(function (e) {
+      if (e.message === "auth") return;
+      resultDiv.style.display = "block";
+      resultDiv.textContent = "";
+      resultDiv.appendChild(_detectResultLine("\u2717 " + e.message, "red"));
+      resultDiv.style.borderColor = "var(--red)";
+    })
+    .finally(function () {
+      btn.disabled = false;
+      btn.removeAttribute("aria-busy");
+      btn.textContent = "Detect";
+    });
+}
+
+/* Capability auto-fill: when the user types a known model name or
+   changes the provider, look up static capabilities and pre-fill
+   context_window and the capabilities textarea. */
+var _capsTimer = null;
+function _onModelFieldChange() {
+  clearTimeout(_capsTimer);
+  _capsTimer = setTimeout(function () {
+    var overlay = document.getElementById("model-create-overlay");
+    if (!overlay || overlay.style.display === "none") return;
+    var provider = document.getElementById("model-provider").value;
+    var modelName = document.getElementById("model-name").value.trim();
+    if (!modelName) return;
+    authFetch(
+      "/v1/api/admin/model-capabilities?provider=" +
+        encodeURIComponent(provider) +
+        "&model=" +
+        encodeURIComponent(modelName),
+    )
+      .then(function (r) {
+        return r.json();
+      })
+      .then(function (d) {
+        if (!d.known || !d.capabilities) return;
+        var ctxInput = document.getElementById("model-ctx-window");
+        if (
+          parseInt(ctxInput.value, 10) === 0 &&
+          d.capabilities.context_window
+        ) {
+          ctxInput.value = d.capabilities.context_window;
+        }
+        var capsInput = document.getElementById("model-capabilities");
+        if (!capsInput.value.trim()) {
+          var caps = Object.assign({}, d.capabilities);
+          delete caps.context_window;
+          delete caps.max_output_tokens;
+          delete caps.token_param;
+          delete caps.supports_streaming;
+          delete caps.supports_tools;
+          var text = JSON.stringify(caps, null, 2);
+          if (text !== "{}") capsInput.value = text;
+        }
+      })
+      .catch(function () {
+        /* silent */
+      });
+  }, 500);
+}
+/* Populate the model name datalist with known model prefixes for the
+   selected provider.  Called on page load and provider change. */
+function _refreshModelSuggestions() {
+  var dl = document.getElementById("model-name-suggestions");
+  if (!dl) return;
+  var provider = document.getElementById("model-provider").value;
+  authFetch(
+    "/v1/api/admin/model-capabilities/known?provider=" +
+      encodeURIComponent(provider),
+  )
+    .then(function (r) {
+      return r.json();
+    })
+    .then(function (d) {
+      dl.textContent = "";
+      (d.models || []).forEach(function (m) {
+        var opt = document.createElement("option");
+        opt.value = m;
+        dl.appendChild(opt);
+      });
+    })
+    .catch(function () {
+      dl.textContent = "";
+    });
+}
+
+/* Register listeners once at page load */
+(function () {
+  var nameEl = document.getElementById("model-name");
+  var provEl = document.getElementById("model-provider");
+  if (nameEl) nameEl.addEventListener("input", _onModelFieldChange);
+  if (provEl) {
+    provEl.addEventListener("change", _onModelFieldChange);
+    provEl.addEventListener("change", _refreshModelSuggestions);
+    provEl.addEventListener("change", _clearDetectResult);
+  }
+  /* Clear stale detect results when probe-relevant inputs change */
+  ["model-base-url", "model-api-key"].forEach(function (id) {
+    var el = document.getElementById(id);
+    if (el) el.addEventListener("input", _clearDetectResult);
+  });
+})();
+
+function _flagModelSyncPending() {
+  var btn = document.getElementById("model-sync-btn");
+  if (btn) btn.classList.add("model-sync-pending");
+}
+function _clearModelSyncPending() {
+  var btn = document.getElementById("model-sync-btn");
+  if (btn) btn.classList.remove("model-sync-pending");
+}
+
+function reloadModelNodes() {
+  var btn = document.getElementById("model-sync-btn");
+  btn.disabled = true;
+  btn.textContent = "Syncing...";
+  authFetch("/v1/api/admin/model-definitions/reload", { method: "POST" })
+    .then(function (r) {
+      if (!r.ok) throw new Error();
+      return r.json();
+    })
+    .then(function () {
+      showToast("Model reload dispatched");
+      _clearModelSyncPending();
+      loadAdminModels();
+    })
+    .catch(function () {
+      showToast("Failed to sync models");
+    })
+    .finally(function () {
+      btn.disabled = false;
+      btn.textContent = "Sync to Nodes";
+    });
+}

turnstone/console/static/app.js

@@ -1281,8 +1281,31 @@ function showNewWsModal() {
     .catch(function () {
       /* ignore — defaults still work */
     });
+  // Populate model dropdown
+  var modelSelect = document.getElementById("new-ws-model");
+  modelSelect.textContent = "";
+  var defaultOpt = document.createElement("option");
+  defaultOpt.value = "";
+  defaultOpt.textContent = "Default model";
+  modelSelect.appendChild(defaultOpt);
+  authFetch("/v1/api/models")
+    .then(function (r) {
+      return r.json();
+    })
+    .then(function (data) {
+      (data.models || []).forEach(function (m) {
+        var opt = document.createElement("option");
+        opt.value = m.alias;
+        opt.textContent =
+          m.alias === m.model ? m.alias : m.alias + " (" + m.model + ")";
+        modelSelect.appendChild(opt);
+      });
+    })
+    .catch(function () {
+      /* ignore — default model still works */
+    });
   document.getElementById("new-ws-name").value = "";
-  document.getElementById("new-ws-model").value = "";
+  modelSelect.value = "";
   var taskEl = document.getElementById("new-ws-task");
   taskEl.value = "";
   var mod =

turnstone/console/static/governance.js

@@ -1838,7 +1838,9 @@ function _renderGovAudit(events, total) {
       _relativeTime(ev.timestamp) +
       "</span>" +
       '<span class="admin-col admin-col-auser">' +
-      escapeHtml(ev.user_id ? ev.user_id.slice(0, 8) : "\u2014") +
+      escapeHtml(
+        ev.username || (ev.user_id ? ev.user_id.slice(0, 8) : "\u2014"),
+      ) +
       "</span>" +
       '<span class="admin-col admin-col-aaction"><span class="' +
       actionCls +

turnstone/console/static/index.html

@@ -109,7 +109,9 @@
         </div>
         <div class="admin-sidebar-group" data-group="system" role="group" aria-label="System">
           <div class="admin-sidebar-group-label" aria-hidden="true">System</div>
+          <button id="tab-models" class="admin-nav" data-tab="models" role="tab" aria-selected="false" aria-controls="admin-models" tabindex="-1" onclick="switchAdminTab('models')">Models</button>
           <button id="tab-settings" class="admin-nav" data-tab="settings" role="tab" aria-selected="false" aria-controls="admin-settings" tabindex="-1" onclick="switchAdminTab('settings')">Settings</button>
+          <button id="tab-tls" class="admin-nav" data-tab="tls" role="tab" aria-selected="false" aria-controls="admin-tls" tabindex="-1" onclick="switchAdminTab('tls')">TLS</button>
         </div>
       </nav>
       <div id="admin-sidebar-backdrop" class="admin-sidebar-backdrop" aria-hidden="true"></div>
@@ -393,6 +395,26 @@
       </div>
     </div>
 
+    <!-- Models Tab -->
+    <div id="admin-models" class="admin-panel" role="tabpanel" aria-labelledby="tab-models" style="display:none">
+      <div class="admin-toolbar">
+        <span class="section-header">Models</span>
+        <button id="model-sync-btn" class="admin-action-btn admin-action-btn-ghost" onclick="reloadModelNodes()" title="Push model config to all cluster nodes">Sync to Nodes</button>
+        <button class="admin-action-btn" onclick="showCreateModelModal()">+ Add Model</button>
+      </div>
+      <div class="admin-colheaders models-grid" aria-hidden="true">
+        <span class="admin-col">ALIAS</span>
+        <span class="admin-col">MODEL</span>
+        <span class="admin-col">PROVIDER</span>
+        <span class="admin-col">CTX WINDOW</span>
+        <span class="admin-col">STATUS</span>
+        <span class="admin-col">ACTIONS</span>
+      </div>
+      <div id="admin-models-table" role="list" aria-label="Model definitions" aria-live="polite">
+        <div class="dashboard-empty">Loading...</div>
+      </div>
+    </div>
+
     <!-- Settings Tab -->
     <div id="admin-settings" class="admin-panel" role="tabpanel" aria-labelledby="tab-settings" style="display:none">
       <div class="admin-toolbar">
@@ -404,6 +426,24 @@
       </div>
     </div>
 
+    <!-- TLS Tab -->
+    <div id="admin-tls" class="admin-panel" role="tabpanel" aria-labelledby="tab-tls" style="display:none">
+      <div class="admin-toolbar">
+        <span class="section-header" style="margin:0">TLS / CERTIFICATES</span>
+      </div>
+      <div id="tls-ca-status"></div>
+      <div class="admin-colheaders admin-colheaders-tls" aria-hidden="true">
+        <span class="admin-col">DOMAIN</span>
+        <span class="admin-col">SANS</span>
+        <span class="admin-col">ISSUED</span>
+        <span class="admin-col">EXPIRES</span>
+        <span class="admin-col admin-col-actions">ACTIONS</span>
+      </div>
+      <div id="tls-cert-list" role="list" aria-label="TLS certificates" aria-live="polite">
+        <div class="dashboard-empty">Loading&hellip;</div>
+      </div>
+    </div>
+
     <div id="admin-mcp" class="admin-panel" role="tabpanel" aria-labelledby="tab-mcp" style="display:none">
       <div class="admin-toolbar">
         <span class="section-header">MCP</span>
@@ -504,7 +544,9 @@ window.TURNSTONE_KB_SHORTCUTS = [
     <label for="new-ws-name">Name <span class="label-hint">optional</span></label>
     <input id="new-ws-name" type="text" placeholder="Auto-generated if empty" autocomplete="off">
     <label for="new-ws-model">Model <span class="label-hint">optional</span></label>
-    <input id="new-ws-model" type="text" placeholder="Default model" autocomplete="off">
+    <select id="new-ws-model">
+      <option value="">Default model</option>
+    </select>
     <label for="new-ws-skill">Skill <span class="label-hint">optional</span></label>
     <select id="new-ws-skill">
       <option value="">Use defaults</option>
@@ -1170,6 +1212,45 @@ window.TURNSTONE_KB_SHORTCUTS = [
   </div>
 </div>
 
+<!-- Model create/edit modal -->
+<div id="model-create-overlay" style="display:none" role="dialog" aria-modal="true" aria-labelledby="model-create-title">
+  <div id="model-create-box" class="admin-modal">
+    <h2 id="model-create-title">Add Model</h2>
+    <div id="model-create-error" role="alert" aria-live="assertive"></div>
+    <input type="hidden" id="model-edit-id" value="">
+    <label for="model-alias">Alias</label>
+    <input type="text" id="model-alias" placeholder="e.g. gpt5-prod" maxlength="64" pattern="[a-zA-Z0-9._-]+">
+    <label for="model-name">Model ID <span style="font-weight:400;text-transform:none">(type to autocomplete)</span></label>
+    <input type="text" id="model-name" placeholder="e.g. gpt-5" list="model-name-suggestions">
+    <datalist id="model-name-suggestions"></datalist>
+    <label for="model-provider">Provider</label>
+    <select id="model-provider">
+      <option value="openai">openai</option>
+      <option value="anthropic">anthropic</option>
+      <option value="openai-compatible">openai-compatible</option>
+    </select>
+    <label for="model-base-url">Base URL <span style="font-weight:400;text-transform:none">(empty = provider default)</span></label>
+    <input type="text" id="model-base-url" placeholder="https://api.openai.com/v1">
+    <label for="model-api-key">API Key <span style="font-weight:400;text-transform:none">(write-only, never displayed)</span></label>
+    <input type="password" id="model-api-key" placeholder="sk-..." autocomplete="off">
+    <label for="model-ctx-window">Context Window <span style="font-weight:400;text-transform:none">(0 = auto-detect from model)</span></label>
+    <input type="number" id="model-ctx-window" value="0" min="0">
+    <label for="model-capabilities">Capabilities <span style="font-weight:400;text-transform:none">(JSON)</span></label>
+    <textarea id="model-capabilities" rows="3" placeholder='{"supports_vision": true}' style="font-family:var(--font-mono);font-size:11px"></textarea>
+    <div style="display:flex;gap:20px;margin-top:14px">
+      <label style="margin:0;font-size:12px;color:var(--fg-dim)"><input type="checkbox" id="model-enabled" checked style="margin-right:5px">Enabled</label>
+    </div>
+    <div id="model-detect-area" style="margin-top:14px">
+      <button type="button" id="model-detect-btn" class="modal-cancel" onclick="detectModel()" style="width:auto;padding:7px 16px;font-size:12px" title="Probe endpoint to verify connectivity and discover models">Detect</button>
+      <div id="model-detect-result" role="status" aria-live="polite" style="display:none;margin-top:8px;padding:10px 12px;border-radius:6px;font-size:12px;border:1px solid var(--border);word-break:break-word"></div>
+    </div>
+    <div class="modal-buttons">
+      <button class="modal-cancel" onclick="hideCreateModelModal()">Cancel</button>
+      <button id="model-create-submit" class="modal-submit" onclick="submitCreateModel()">Create</button>
+    </div>
+  </div>
+</div>
+
 <script src="/static/admin.js"></script>
 <script src="/static/governance.js"></script>
 <script src="/static/app.js"></script>

turnstone/console/static/style.css

@@ -948,6 +948,19 @@
   grid-template-columns: 100px 1fr 100px 80px;
 }
 
+/* TLS grid: DOMAIN | SANS | ISSUED | EXPIRES | ACTIONS */
+#admin-tls .admin-colheaders,
+#admin-tls .admin-row {
+  grid-template-columns: 160px 1fr 130px 150px 120px;
+}
+.tls-ca-bar {
+  display: flex;
+  gap: 2rem;
+  margin-bottom: 0.75rem;
+  font-size: 12px;
+  color: var(--fg-dim);
+}
+
 /* Scope badges */
 .scope-badge {
   display: inline-block;
@@ -1176,7 +1189,8 @@
 .admin-modal label.admin-checkbox input:disabled { opacity: 0.4; }
 .admin-modal input::placeholder, .admin-modal textarea::placeholder { color: var(--fg-dim); opacity: 0.6; }
 .admin-modal textarea { resize: vertical; min-height: 40px; }
-.admin-modal [role="alert"] { color: var(--red); font-size: 12px; margin-bottom: 8px; display: none; }
+.admin-modal [role="alert"] { display: none; color: var(--red); font-size: 12px; margin-bottom: 8px; }
+.admin-modal [role="alert"].is-visible { display: block; }
 
 .admin-details { margin-top: 12px; border: 1px solid var(--border); border-radius: 6px; padding: 0 12px; }
 .admin-details[open] { padding-bottom: 12px; }
@@ -1356,6 +1370,7 @@ h3.skill-spec-heading { font-size: inherit; margin-block: 0; }
 }
 .modal-cancel:hover { background: var(--bg-elevated); }
 .modal-cancel:focus-visible { outline: 2px solid var(--accent); outline-offset: 2px; }
+.modal-cancel:disabled { opacity: 0.5; cursor: not-allowed; pointer-events: none; }
 .modal-submit {
   flex: 1;
   padding: 9px;
@@ -1381,7 +1396,8 @@ h3.skill-spec-heading { font-size: inherit; margin-block: 0; }
 #create-template-overlay, #edit-template-overlay,
 #memory-detail-overlay,
 #mcp-create-overlay, #mcp-import-overlay, #mcp-detail-overlay, #mcp-install-overlay,
-#github-import-overlay {
+#github-import-overlay,
+#model-create-overlay {
   position: fixed;
   inset: 0;
   background: rgba(0, 0, 0, 0.7);
@@ -2103,7 +2119,7 @@ h3.skill-spec-heading { font-size: inherit; margin-block: 0; }
 
 .admin-action-btn-ghost{background:transparent;color:var(--fg-dim);border:1px solid var(--border-strong)}
 .admin-action-btn-ghost:hover{color:var(--fg);background:var(--bg-highlight)}
-.mcp-sync-pending{color:var(--yellow)!important;border-color:var(--yellow)!important;animation:mcp-sync-pulse 2s ease-in-out infinite}
+.mcp-sync-pending,.model-sync-pending{color:var(--yellow)!important;border-color:var(--yellow)!important;animation:mcp-sync-pulse 2s ease-in-out infinite}
 @keyframes mcp-sync-pulse{0%,100%{border-color:var(--yellow)}50%{border-color:rgba(251,191,36,.3)}}
 
 /* -- MCP sub-view toggle -------------------------------------------------- */
@@ -2170,7 +2186,8 @@ h3.skill-spec-heading { font-size: inherit; margin-block: 0; }
   .mcp-reg-card{grid-template-columns:1fr;gap:8px}
   .mcp-reg-card-actions{flex-direction:row;align-items:center}
   .mcp-registry-search{flex-direction:column}
-  #admin-mcp .admin-toolbar{flex-wrap:wrap;gap:8px}
+  #admin-mcp .admin-toolbar,
+  #admin-models .admin-toolbar{flex-wrap:wrap;gap:8px}
   #mcp-servers-toolbar{display:flex;gap:6px;width:100%}
   #admin-skills .admin-toolbar{flex-wrap:wrap;gap:8px}
   #skill-installed-toolbar{display:flex;gap:6px;width:100%}
@@ -2274,6 +2291,31 @@ h3.skill-spec-heading { font-size: inherit; margin-block: 0; }
   .oidc-detail-panel { margin-left: 8px; }
 }
 
+/* -- Models grid --------------------------------------------------------- */
+.models-grid{grid-template-columns:1.2fr 1.2fr 80px 90px 80px 120px;gap:0 6px}
+@media(max-width:700px){
+  .models-grid{grid-template-columns:1fr 80px 120px}
+  .models-grid .admin-col:nth-child(2),
+  .models-grid .admin-col:nth-child(3),
+  .models-grid .admin-col:nth-child(4){display:none}
+}
+
+/* Model status indicators */
+.model-status-dot{display:inline-block;width:8px;height:8px;border-radius:50%;vertical-align:middle;margin-right:6px}
+.model-status-dot.enabled{background:var(--blue);box-shadow:0 0 6px var(--blue-glow)}
+.model-status-dot.disabled{background:var(--fg-dim);opacity:.35}
+
+.model-row-enabled{border-left:3px solid var(--blue)}
+.model-row-disabled{border-left:3px solid transparent}
+
+/* Provider badges */
+.model-provider-badge{display:inline-block;font-size:9px;font-weight:600;text-transform:uppercase;letter-spacing:.06em;padding:1px 6px;border-radius:2px;background:var(--bg-highlight);border:1px solid var(--border)}
+.model-provider-openai{color:var(--blue);border-color:rgba(56,189,248,.2)}
+.model-provider-anthropic{color:var(--magenta);border-color:rgba(192,132,252,.25)}
+
+/* Model source badge */
+.scope-db{color:var(--blue);border-color:rgba(56,189,248,.2)}
+
 /* ==========================================================================
    Reduced motion — console-specific
    ========================================================================== */
@@ -2297,5 +2339,5 @@ h3.skill-spec-heading { font-size: inherit; margin-block: 0; }
   .mcp-view-btn, .mcp-reg-card, .mcp-install-btn, .mcp-install-source-label { transition: none; }
   .mcp-registry-search input[type="search"] { transition: none; }
   .mcp-reg-card-repo { transition: none; }
-  .mcp-sync-pending { animation: none; }
+  .mcp-sync-pending, .model-sync-pending { animation: none; }
 }

turnstone/console/tls.py

@@ -0,0 +1,391 @@
+"""TLS Manager — Certificate Authority and ACME server for the console.
+
+Owns the lacme CertificateAuthority, ACMEResponder, and RenewalManager
+lifecycle. When TLS is enabled, the console acts as the cluster's internal
+CA and ACME server, issuing short-lived mTLS certificates to all services.
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Any
+
+import structlog
+
+if TYPE_CHECKING:
+    import ssl
+
+    from starlette.types import ASGIApp
+
+    from turnstone.core.config_store import ConfigStore
+    from turnstone.core.storage._protocol import StorageBackend
+
+log = structlog.get_logger(__name__)
+
+# Hardcoded defaults — no operator config needed
+_CA_CN = "Turnstone CA"
+_CA_NAME = "turnstone"  # Store key for save_ca/load_ca
+_CA_VALIDITY_DAYS = 3650  # 10 years
+_CERT_VALIDITY_HOURS = 48
+_RENEW_INTERVAL_HOURS = 24
+_RENEW_BEFORE_EXPIRY_DAYS = 1
+
+
+def _require_lacme() -> Any:
+    try:
+        import lacme
+    except ImportError:
+        raise ImportError(
+            "lacme is required for TLS support. Install with: pip install turnstone[tls]",
+        ) from None
+    return lacme
+
+
+class TLSManager:
+    """Manages the internal CA, ACME responder, and certificate lifecycle.
+
+    Typical usage::
+
+        mgr = TLSManager(storage, config_store)
+        await mgr.init_ca()
+        responder = mgr.get_responder()       # Mount at /acme
+        await mgr.issue_console_certs()        # Self-issue for this node
+        mgr.start_renewal()                    # Background auto-renewal
+    """
+
+    def __init__(
+        self,
+        storage: StorageBackend,
+        config_store: ConfigStore | None = None,
+    ) -> None:
+        lacme = _require_lacme()
+
+        from turnstone.core.tls_store import StorageStore
+
+        self._store = StorageStore(storage)
+        self._config_store = config_store
+        self._event_dispatcher = lacme.EventDispatcher()
+        self._ca: Any | None = None
+        self._responder: Any | None = None
+        self._renewal_task: Any | None = None
+        self._renewal_manager: Any | None = None
+        self._internal_bundle: Any | None = None
+        self._frontend_bundle: Any | None = None
+
+        # Wire structlog to lacme events
+        self._subscribe_events()
+
+        # Wire Prometheus metrics (if prometheus_client available)
+        try:
+            from lacme.metrics import setup_metrics
+
+            setup_metrics(self._event_dispatcher)
+        except ImportError:
+            pass  # prometheus_client or lacme.metrics missing
+        except ValueError as exc:
+            if "Duplicated timeseries" in str(exc):
+                log.debug("tls_metrics_already_registered")
+            else:
+                raise
+
+    def _subscribe_events(self) -> None:
+        """Subscribe structlog handlers to lacme lifecycle events."""
+        _require_lacme()
+        from lacme.events import (
+            CertificateExpiring,
+            CertificateIssued,
+            CertificateRenewed,
+            ChallengeFailed,
+        )
+
+        def _on_issued(event: Any) -> None:
+            if isinstance(event, CertificateIssued):
+                log.info("tls.cert.issued", domain=event.domain)
+
+        def _on_renewed(event: Any) -> None:
+            if isinstance(event, CertificateRenewed):
+                log.info("tls.cert.renewed", domain=event.domain)
+
+        def _on_expiring(event: Any) -> None:
+            if isinstance(event, CertificateExpiring):
+                log.warning("tls.cert.expiring", domain=event.domain)
+
+        def _on_failed(event: Any) -> None:
+            if isinstance(event, ChallengeFailed):
+                log.error("tls.challenge.failed", domain=getattr(event, "domain", "unknown"))
+
+        self._event_dispatcher.subscribe(_on_issued, event_type=CertificateIssued)
+        self._event_dispatcher.subscribe(_on_renewed, event_type=CertificateRenewed)
+        self._event_dispatcher.subscribe(_on_expiring, event_type=CertificateExpiring)
+        self._event_dispatcher.subscribe(_on_failed, event_type=ChallengeFailed)
+
+    # -- CA lifecycle ----------------------------------------------------------
+
+    async def init_ca(self) -> None:
+        """Initialize the internal Certificate Authority.
+
+        If a bootstrap CA exists on disk (from tls-bootstrap), imports it
+        into the database store first so the console uses the same CA that
+        signed the infrastructure certs.
+        """
+        lacme = _require_lacme()
+
+        # Import bootstrap CA from well-known volume path if not already in DB
+        self._import_bootstrap_ca()
+
+        self._ca = lacme.CertificateAuthority(
+            self._store,
+            name=_CA_NAME,
+            event_dispatcher=self._event_dispatcher,
+        )
+        self._ca.init(cn=_CA_CN, validity_days=_CA_VALIDITY_DAYS)
+        log.info("tls.ca.initialized", cn=_CA_CN)
+
+    def _import_bootstrap_ca(self) -> None:
+        """Import a bootstrap CA from /certs into the database store.
+
+        The tls-bootstrap CLI writes the CA to a FileStore at /certs.
+        On first boot, the console imports it so all services share
+        the same trust root.
+        """
+        import os
+        from pathlib import Path
+
+        # Check if bootstrap CA exists and DB CA doesn't
+        bootstrap_dir = Path(os.environ.get("TURNSTONE_TLS_BOOTSTRAP_DIR", "/certs"))
+        ca_dir = bootstrap_dir / "ca" / _CA_NAME
+        ca_cert_file = ca_dir / "cert.pem"
+        ca_key_file = ca_dir / "key.pem"
+
+        if not ca_cert_file.exists() or not ca_key_file.exists():
+            return  # No bootstrap CA found
+
+        existing = self._store.load_ca(_CA_NAME)
+        if existing is not None:
+            return  # Already imported
+
+        cert_pem = ca_cert_file.read_bytes()
+        key_pem = ca_key_file.read_bytes()
+        self._store.save_ca(_CA_NAME, cert_pem, key_pem)
+        log.info("tls.ca.imported_from_bootstrap", path=str(ca_dir))
+
+    def get_responder(self) -> ASGIApp:
+        """Return the ACME responder ASGI app for mounting."""
+        if self._ca is None:
+            raise RuntimeError("CA not initialized — call init_ca() first")
+        lacme = _require_lacme()
+        if self._responder is None:
+            self._responder = lacme.ACMEResponder(
+                ca=self._ca,
+                auto_approve=True,
+            )
+        return self._responder  # type: ignore[no-any-return]
+
+    def get_root_cert_pem(self) -> bytes:
+        """Return the CA root certificate in PEM format."""
+        if self._ca is None:
+            raise RuntimeError("CA not initialized — call init_ca() first")
+        return self._ca.root_cert_pem  # type: ignore[no-any-return]
+
+    # -- Cert issuance ---------------------------------------------------------
+
+    async def issue_console_certs(self, hostnames: list[str]) -> None:
+        """Issue certificates for the console node.
+
+        Raises ValueError if hostnames is empty.
+
+        Issues two certificates:
+        - Internal cert: always from the internal CA (for mTLS with cluster)
+        - Frontend cert: from external ACME CA if configured, else internal CA
+        """
+        if not hostnames:
+            raise ValueError("issue_console_certs requires at least one hostname")
+
+        # Internal cert — always from our own CA
+        await self._issue_internal_cert(hostnames)
+
+        # Frontend cert — external CA if configured
+        acme_directory = ""
+        if self._config_store:
+            acme_directory = self._config_store.get("tls.acme_directory") or ""
+
+        if acme_directory:
+            await self._issue_frontend_cert(hostnames, acme_directory)
+        else:
+            # Self-issue from internal CA (behind reverse proxy or internal only)
+            self._frontend_bundle = self._internal_bundle
+            log.info("tls.frontend.self_issued", hostnames=hostnames)
+
+    async def _issue_internal_cert(self, hostnames: list[str]) -> None:
+        """Issue an internal mTLS cert from the internal CA."""
+        if self._ca is None:
+            raise RuntimeError("CA not initialized")
+
+        # Check for existing cert in store (skip if expired)
+        existing = self._store.load_cert(hostnames[0])
+        if existing is not None:
+            from datetime import UTC, datetime
+
+            if existing.expires_at > datetime.now(UTC):
+                self._internal_bundle = existing
+                log.info("tls.internal.loaded", domain=hostnames[0])
+                return
+            log.info("tls.internal.expired", domain=hostnames[0])
+            self._store.delete_cert(hostnames[0])
+
+        # Issue new cert
+        bundle = self._ca.issue(
+            hostnames,
+            validity_hours=_CERT_VALIDITY_HOURS,
+        )
+        self._store.save_cert(bundle)
+        self._internal_bundle = bundle
+        log.info("tls.internal.issued", domain=hostnames[0])
+
+    async def _issue_frontend_cert(
+        self,
+        hostnames: list[str],
+        acme_directory: str,
+    ) -> None:
+        """Issue a frontend cert from an external ACME CA."""
+        lacme = _require_lacme()
+        from lacme.challenges.http01 import HTTP01Handler
+
+        handler = HTTP01Handler()
+
+        async with lacme.Client(
+            directory_url=acme_directory,
+            store=self._store,
+            challenge_handler=handler,
+            event_dispatcher=self._event_dispatcher,
+        ) as client:
+            self._frontend_bundle = await client.issue(hostnames)
+            self._store.save_cert(self._frontend_bundle)
+            log.info(
+                "tls.frontend.issued",
+                domain=hostnames[0],
+                ca=acme_directory,
+            )
+
+    # -- Auto-renewal ----------------------------------------------------------
+
+    async def start_renewal(self) -> None:
+        """Start background auto-renewal for all stored certificates.
+
+        Uses CA-direct mode (lacme 1.0.2+) — signs directly via the CA
+        without going through ACME. No loopback client, no network,
+        no startup ordering dependency.
+        """
+        if self._ca is None:
+            raise RuntimeError("CA not initialized")
+        lacme = _require_lacme()
+
+        def _on_renewed(bundle: Any) -> None:
+            # Update our cached bundles if the renewed domain matches
+            if self._internal_bundle and bundle.domain == self._internal_bundle.domain:
+                self._internal_bundle = bundle
+            if self._frontend_bundle and bundle.domain == self._frontend_bundle.domain:
+                self._frontend_bundle = bundle
+
+        self._renewal_manager = lacme.RenewalManager(
+            ca=self._ca,
+            store=self._store,
+            interval_hours=_RENEW_INTERVAL_HOURS,
+            days_before_expiry=_RENEW_BEFORE_EXPIRY_DAYS,
+            on_renewed=_on_renewed,
+            event_dispatcher=self._event_dispatcher,
+        )
+        self._renewal_task = self._renewal_manager.start()
+        log.info(
+            "tls.renewal.started",
+            interval_hours=_RENEW_INTERVAL_HOURS,
+        )
+
+    async def stop_renewal(self) -> None:
+        """Stop the background renewal task."""
+        if self._renewal_task is not None:
+            import asyncio
+            import contextlib
+
+            self._renewal_task.cancel()
+            try:
+                with contextlib.suppress(asyncio.CancelledError):
+                    await self._renewal_task
+            except Exception:
+                log.exception("tls.renewal.stop_error")
+            self._renewal_task = None
+
+    # -- SSL contexts ----------------------------------------------------------
+
+    def get_server_ssl_context(self) -> ssl.SSLContext | None:
+        """Build an SSL context for the uvicorn HTTPS listener.
+
+        Uses the frontend cert (external CA or self-issued).
+        Returns None if no certs are available.
+        """
+        if self._frontend_bundle is None:
+            return None
+        _require_lacme()
+        from lacme.mtls import server_ssl_context
+
+        return server_ssl_context(  # type: ignore[no-any-return,unused-ignore]
+            cert_pem=self._frontend_bundle.fullchain_pem,
+            key_pem=self._frontend_bundle.key_pem,
+            ca_cert_pem=self.get_root_cert_pem(),
+        )
+
+    def get_client_ssl_context(self) -> ssl.SSLContext | None:
+        """Build an mTLS client context for connecting to cluster services.
+
+        Uses the internal cert for mutual authentication.
+        Returns None if no certs are available.
+        """
+        if self._internal_bundle is None:
+            return None
+        _require_lacme()
+        from lacme.mtls import client_ssl_context
+
+        return client_ssl_context(  # type: ignore[no-any-return,unused-ignore]
+            cert_pem=self._internal_bundle.cert_pem,
+            key_pem=self._internal_bundle.key_pem,
+            ca_cert_pem=self.get_root_cert_pem(),
+        )
+
+    # -- Properties ------------------------------------------------------------
+
+    def list_certs(self) -> list[Any]:
+        """List all stored certificate bundles."""
+        return self._store.list_certs()
+
+    def renew_cert(self, domain: str) -> Any:
+        """Force-renew a certificate by domain. Returns the new bundle."""
+        if self._ca is None:
+            raise RuntimeError("CA not initialized")
+        existing = self._store.load_cert(domain)
+        if existing is None:
+            raise ValueError(f"No certificate for {domain}")
+        # Issue new cert first, then delete old (safe if issuance fails)
+        bundle = self._ca.issue(list(existing.domains))
+        self._store.delete_cert(domain)
+        self._store.save_cert(bundle)
+        # Update in-memory bundles if this is the console's own cert
+        if self._internal_bundle and bundle.domain == self._internal_bundle.domain:
+            self._internal_bundle = bundle
+        if self._frontend_bundle and bundle.domain == self._frontend_bundle.domain:
+            self._frontend_bundle = bundle
+        return bundle
+
+    def delete_cert(self, domain: str) -> bool:
+        """Delete a certificate by domain."""
+        return self._store.delete_cert(domain)
+
+    @property
+    def ca_initialized(self) -> bool:
+        return self._ca is not None
+
+    @property
+    def internal_bundle(self) -> Any | None:
+        return self._internal_bundle
+
+    @property
+    def frontend_bundle(self) -> Any | None:
+        return self._frontend_bundle

turnstone/core/auth.py

@@ -157,7 +157,7 @@ PUBLIC_PATHS: frozenset[str] = frozenset(
         "/api/auth/oidc/callback",
     }
 )
-PUBLIC_PREFIXES: tuple[str, ...] = ("/static/", "/shared/")
+PUBLIC_PREFIXES: tuple[str, ...] = ("/static/", "/shared/", "/acme/")
 
 WRITE_PATHS: frozenset[str] = frozenset(
     {
@@ -173,7 +173,12 @@ WRITE_PATHS: frozenset[str] = frozenset(
 )
 
 APPROVE_PATHS: frozenset[str] = frozenset(
-    {"/api/approve", "/api/_internal/config-reload", "/api/_internal/mcp-reload"}
+    {
+        "/api/approve",
+        "/api/_internal/config-reload",
+        "/api/_internal/mcp-reload",
+        "/api/_internal/model-reload",
+    }
 )
 ADMIN_PREFIX = "/api/admin/"
 
@@ -329,18 +334,22 @@ def create_jwt(
     expiry_hours: int = 24,
     audience: str = "",
     permissions: frozenset[str] = frozenset(),
+    expiry_seconds: int | None = None,
 ) -> str:
     """Create a signed JWT with user identity, scopes, and permissions."""
     import jwt
 
+    if expiry_seconds is not None and expiry_seconds <= 0:
+        raise ValueError("expiry_seconds must be positive")
     now = int(time.time())
+    ttl = expiry_seconds if expiry_seconds is not None else expiry_hours * 3600
     payload: dict[str, Any] = {
         "sub": user_id,
         "scopes": ",".join(sorted(scopes)),
         "src": source,
         "iss": JWT_ISSUER,
         "iat": now,
-        "exp": now + expiry_hours * 3600,
+        "exp": now + ttl,
     }
     if audience:
         payload["aud"] = audience

turnstone/core/config.py

@@ -125,6 +125,10 @@ _CONFIG_MAP: dict[str, dict[str, str]] = {
         "port": "redis_port",
         "password": "redis_password",
         "db": "redis_db",
+        "tls": "redis_tls",
+        "tls_ca": "redis_tls_ca",
+        "tls_cert": "redis_tls_cert",
+        "tls_key": "redis_tls_key",
     },
     "console": {
         "host": "host",
@@ -157,6 +161,10 @@ _CONFIG_MAP: dict[str, dict[str, str]] = {
         "url": "db_url",
         "path": "db_path",
         "pool_size": "db_pool_size",
+        "sslmode": "db_sslmode",
+        "sslrootcert": "db_sslrootcert",
+        "sslcert": "db_sslcert",
+        "sslkey": "db_sslkey",
     },
     "judge": {
         "enabled": "judge_enabled",

turnstone/core/judge.py

@@ -853,6 +853,10 @@ If you used read_file to check a target, cite what you found."""
 # ---------------------------------------------------------------------------
 
 
+class _ExecutorPoisonedError(Exception):
+    """Raised when a timeout leaves the executor's worker thread stuck."""
+
+
 class IntentJudge:
     """Session-scoped LLM judge for intent validation.
 
@@ -912,18 +916,12 @@ class IntentJudge:
             self._model = session_model
             self._judge_context_window = context_window
 
-        # Executor for timeout-guarded API calls (1 thread — judge is serial)
-        self._executor = ThreadPoolExecutor(max_workers=1, thread_name_prefix="judge-api")
-
-    def shutdown(self) -> None:
-        """Release the executor thread pool."""
-        self._executor.shutdown(wait=False, cancel_futures=True)
-
     def evaluate(
         self,
         items: list[dict[str, Any]],
         messages: list[dict[str, Any]],
         callback: Callable[[IntentVerdict], None],
+        cancel_event: threading.Event | None = None,
     ) -> list[IntentVerdict]:
         """Evaluate tool calls. Returns heuristic verdicts immediately.
 
@@ -936,6 +934,10 @@ class IntentJudge:
                 ``func_args``, ``approval_label``, ``call_id``).
             messages: Conversation history (OpenAI message format).
             callback: Called with each LLM verdict (or timeout/error fallback).
+            cancel_event: When set, the daemon judge thread abandons
+                remaining work.  Callers should set this after the user
+                has already made an approval decision so the judge does
+                not keep consuming inference resources.
 
         Returns:
             List of heuristic verdicts (one per item), available immediately.
@@ -958,7 +960,7 @@ class IntentJudge:
         # Spawn daemon thread for LLM judge
         thread = threading.Thread(
             target=self._run_judge,
-            args=(items, messages, heuristic_verdicts, callback),
+            args=(items, messages, heuristic_verdicts, callback, cancel_event),
             daemon=True,
             name="intent-judge",
         )
@@ -972,25 +974,44 @@ class IntentJudge:
         messages: list[dict[str, Any]],
         heuristic_verdicts: list[IntentVerdict],
         callback: Callable[[IntentVerdict], None],
+        cancel_event: threading.Event | None = None,
     ) -> None:
         """Daemon thread: run LLM judge for each item and invoke callback."""
-        for item, h_verdict in zip(items, heuristic_verdicts, strict=True):
-            try:
-                llm_verdict = self._evaluate_single(item, messages)
-                # Arbitrate: only callback when LLM upgrades the heuristic
-                if llm_verdict and llm_verdict.confidence > h_verdict.confidence:
-                    callback(llm_verdict)
-                # else: heuristic already delivered, no duplicate callback
-            except Exception:
-                log.exception(
-                    "Judge evaluation failed for %s",
-                    item.get("func_name", "?"),
-                )
+        # Evaluation-scoped executor — avoids sharing mutable state with
+        # other daemon threads from concurrent evaluate() calls.
+        executor = ThreadPoolExecutor(max_workers=1, thread_name_prefix="judge-api")
+        try:
+            for idx, (item, h_verdict) in enumerate(zip(items, heuristic_verdicts, strict=True)):
+                if cancel_event and cancel_event.is_set():
+                    log.debug("judge.cancelled", remaining=len(items) - idx)
+                    return
+                try:
+                    llm_verdict = self._evaluate_single(item, messages, cancel_event, executor)
+                    if cancel_event and cancel_event.is_set():
+                        return
+                    # Arbitrate: only callback when LLM upgrades the heuristic
+                    if llm_verdict and llm_verdict.confidence > h_verdict.confidence:
+                        callback(llm_verdict)
+                    # else: heuristic already delivered, no duplicate callback
+                except _ExecutorPoisonedError:
+                    # Timeout left the worker stuck — replace the executor
+                    # so subsequent items don't queue behind it.
+                    executor.shutdown(wait=False, cancel_futures=True)
+                    executor = ThreadPoolExecutor(max_workers=1, thread_name_prefix="judge-api")
+                except Exception:
+                    log.exception(
+                        "Judge evaluation failed for %s",
+                        item.get("func_name", "?"),
+                    )
+        finally:
+            executor.shutdown(wait=False, cancel_futures=True)
 
     def _evaluate_single(
         self,
         item: dict[str, Any],
         messages: list[dict[str, Any]],
+        cancel_event: threading.Event | None,
+        executor: ThreadPoolExecutor,
     ) -> IntentVerdict | None:
         """Run LLM judge for a single tool call. Returns verdict or None."""
         start = time.monotonic()
@@ -1020,6 +1041,9 @@ class IntentJudge:
         result = None  # will hold the last CompletionResult
 
         for turn in range(_JUDGE_MAX_TURNS):
+            if cancel_event and cancel_event.is_set():
+                return None
+
             turn_start = time.monotonic()
 
             is_last_turn = turn == _JUDGE_MAX_TURNS - 1
@@ -1043,7 +1067,7 @@ class IntentJudge:
             # 10 minutes — far too long for an advisory judge on local models.
             per_call_timeout = max(timeout_budget, 5.0)  # at least 5s
             try:
-                future = self._executor.submit(
+                future = executor.submit(
                     self._provider.create_completion,
                     client=self._client,
                     model=self._model,
@@ -1053,17 +1077,24 @@ class IntentJudge:
                     temperature=0.0,
                     reasoning_effort="medium",
                 )
-                result = future.result(timeout=per_call_timeout)
+                # Poll in 1s increments so we notice cancellation promptly
+                # instead of blocking for the full per_call_timeout.
+                deadline = time.monotonic() + per_call_timeout
+                while True:
+                    remaining = deadline - time.monotonic()
+                    if cancel_event and cancel_event.is_set():
+                        future.cancel()
+                        return None
+                    if remaining <= 0:
+                        raise TimeoutError
+                    try:
+                        result = future.result(timeout=min(remaining, 1.0))
+                        break
+                    except TimeoutError:
+                        pass  # loop back to check remaining/cancel
             except TimeoutError:
                 log.warning("Judge LLM call timed out on turn %d (%.0fs)", turn, per_call_timeout)
-                # Abandon the lingering API call and replace the executor so
-                # subsequent items in the batch don't queue behind it.
-                self._executor.shutdown(wait=False, cancel_futures=True)
-                self._executor = ThreadPoolExecutor(
-                    max_workers=1,
-                    thread_name_prefix="judge-api",
-                )
-                return None
+                raise _ExecutorPoisonedError from None
             except Exception:
                 log.exception("Judge LLM call failed on turn %d", turn)
                 return None

turnstone/core/mcp_client.py

@@ -176,6 +176,8 @@ class MCPClientManager:
         for name, cfg in self._server_configs.items():
             try:
                 await self._connect_one(name, cfg)
+            except asyncio.CancelledError:
+                raise  # propagate so the background task can be cleanly stopped
             except Exception as exc:
                 log.warning("Failed to connect MCP server '%s'", name, exc_info=True)
                 self._set_error(name, f"{type(exc).__name__}: {exc}")
@@ -199,6 +201,49 @@ class MCPClientManager:
             self._refresh_task = asyncio.get_running_loop().create_task(self._periodic_refresh())
 
     _CONNECT_TIMEOUT = 30  # seconds — prevents hung connections on broken remotes
+    _TCP_PROBE_TIMEOUT = 5  # seconds — fast TCP pre-flight for HTTP transports
+
+    async def _tcp_probe(self, name: str, url: str) -> None:
+        """Fast TCP connect check before entering the MCP transport context.
+
+        Fails fast when the server is unreachable, avoiding the anyio
+        cancel-scope orphan bug that causes 100% CPU spin.
+        """
+        from urllib.parse import urlparse
+
+        parsed = urlparse(url)
+        host = parsed.hostname
+        if not host:
+            raise ConnectionError(f"MCP server '{name}' has invalid URL (no hostname): {url}")
+        try:
+            port = parsed.port or (443 if parsed.scheme == "https" else 80)
+        except ValueError:
+            raise ConnectionError(f"MCP server '{name}' has invalid port in URL: {url}") from None
+        try:
+            _, writer = await asyncio.wait_for(
+                asyncio.open_connection(host, port),
+                timeout=self._TCP_PROBE_TIMEOUT,
+            )
+            writer.close()
+            await writer.wait_closed()
+        except (TimeoutError, OSError) as exc:
+            raise ConnectionError(
+                f"MCP server '{name}' unreachable at {host}:{port}: {exc}"
+            ) from None
+
+    @staticmethod
+    async def _safe_close_stack(stack: AsyncExitStack) -> None:
+        """Close an AsyncExitStack, suppressing errors from broken anyio scopes.
+
+        Called from exception handlers — must not raise, otherwise cleanup
+        errors could mask the original exception.  CancelledError is caught
+        explicitly because it is the primary failure mode (stray cancel from
+        broken anyio scope) and is BaseException, not Exception.
+        """
+        try:
+            await asyncio.wait_for(stack.aclose(), timeout=5)
+        except (Exception, asyncio.CancelledError):
+            log.debug("Error closing AsyncExitStack; ignoring", exc_info=True)
 
     async def _connect_one(self, name: str, cfg: dict[str, Any]) -> None:
         """Connect to a single MCP server and discover its tools."""
@@ -213,6 +258,13 @@ class MCPClientManager:
         transport = cfg.get("type", "stdio")
         try:
             if transport in ("http", "streamable-http") or "url" in cfg:
+                # Pre-flight TCP check: fail fast before entering the anyio
+                # task group in streamablehttp_client.  An immediate connect
+                # failure (ECONNREFUSED) inside the anyio context causes a
+                # CancelledError that escapes asyncio.wait_for and leaves
+                # orphaned cancel-scope tasks spinning at 100% CPU.
+                await self._tcp_probe(name, cfg["url"])
+
                 read, write, _ = await asyncio.wait_for(
                     stack.enter_async_context(
                         streamablehttp_client(url=cfg["url"], headers=cfg.get("headers"))
@@ -235,15 +287,25 @@ class MCPClientManager:
                     env=env,
                 )
                 read, write = await stack.enter_async_context(stdio_client(params))
+        except asyncio.CancelledError:
+            # Stray CancelledError from broken anyio cancel scope — treat as
+            # connection failure.  But if the task is genuinely being cancelled
+            # (shutdown), re-raise so we don't block teardown.
+            task = asyncio.current_task()
+            if task is not None and task.cancelling():
+                await self._safe_close_stack(stack)
+                raise
+            log.warning("MCP server '%s' connection failed (anyio cancel)", name)
+            await self._safe_close_stack(stack)
+            raise TimeoutError(f"Connection failed for '{name}'") from None
         except TimeoutError:
             log.warning(
                 "MCP server '%s' connection timed out after %ds", name, self._CONNECT_TIMEOUT
             )
-            with contextlib.suppress(Exception):
-                await stack.aclose()
+            await self._safe_close_stack(stack)
             raise TimeoutError(f"Connection timed out after {self._CONNECT_TIMEOUT}s") from None
         except Exception:
-            await stack.aclose()
+            await self._safe_close_stack(stack)
             raise
 
         # Register notification handler — dispatches tool, resource, and
@@ -274,21 +336,27 @@ class MCPClientManager:
                 ClientSession(read, write, message_handler=_on_notification)  # type: ignore[arg-type]
             )
         except Exception:
-            await stack.aclose()
+            await self._safe_close_stack(stack)
             raise
 
         self._per_server_stacks[name] = stack
         try:
             await asyncio.wait_for(session.initialize(), timeout=self._CONNECT_TIMEOUT)
+        except asyncio.CancelledError:
+            self._per_server_stacks.pop(name, None)
+            task = asyncio.current_task()
+            if task is not None and task.cancelling():
+                await self._safe_close_stack(stack)
+                raise
+            await self._safe_close_stack(stack)
+            raise TimeoutError(f"MCP handshake failed for '{name}'") from None
         except TimeoutError:
             self._per_server_stacks.pop(name, None)
-            with contextlib.suppress(Exception):
-                await stack.aclose()
+            await self._safe_close_stack(stack)
             raise TimeoutError(f"MCP handshake timed out after {self._CONNECT_TIMEOUT}s") from None
         except Exception:
             self._per_server_stacks.pop(name, None)
-            with contextlib.suppress(Exception):
-                await stack.aclose()
+            await self._safe_close_stack(stack)
             raise
         self._sessions[name] = session
 
@@ -873,8 +941,7 @@ class MCPClientManager:
 
             async def _close_all_stacks() -> None:
                 for stack in self._per_server_stacks.values():
-                    with contextlib.suppress(Exception):
-                        await stack.aclose()
+                    await self._safe_close_stack(stack)
 
             future = asyncio.run_coroutine_threadsafe(_close_all_stacks(), self._loop)
             try:
@@ -984,10 +1051,7 @@ class MCPClientManager:
                 self._sessions.pop(name, None)
                 stack = self._per_server_stacks.pop(name, None)
                 if stack is not None:
-                    try:
-                        await asyncio.wait_for(stack.aclose(), timeout=10)
-                    except (TimeoutError, Exception):
-                        log.warning("Timed out closing MCP server '%s', forcing cleanup", name)
+                    await self._safe_close_stack(stack)
                 # Clean up per-server state (on the event loop thread)
                 self._per_server_tools.pop(name, None)
                 self._per_server_resources.pop(name, None)

turnstone/core/memory.py

@@ -37,7 +37,6 @@ def save_message(
     role: str,
     content: str | None,
     tool_name: str | None = None,
-    tool_args: str | None = None,
     tool_call_id: str | None = None,
     provider_data: str | None = None,
     tool_calls: str | None = None,
@@ -49,7 +48,6 @@ def save_message(
             role,
             content,
             tool_name,
-            tool_args,
             tool_call_id,
             provider_data,
             tool_calls=tool_calls,
@@ -67,6 +65,23 @@ def load_messages(ws_id: str) -> list[dict[str, Any]]:
         return []
 
 
+def delete_messages_after(ws_id: str, keep_count: int) -> int:
+    """Delete conversation rows beyond the first *keep_count* rows.
+
+    Returns the number of rows deleted, or 0 on error.
+    """
+    try:
+        return get_storage().delete_messages_after(ws_id, keep_count)
+    except Exception:
+        log.warning(
+            "Failed to delete messages after count=%d for ws=%s",
+            keep_count,
+            ws_id,
+            exc_info=True,
+        )
+        return 0
+
+
 # -- Workstream management ----------------------------------------------------
 
 

turnstone/core/metacognition.py

@@ -50,6 +50,13 @@ NUDGE_TOOL_ERROR = (
     "previous session. Use memory(action='search') to find relevant guidance."
 )
 
+NUDGE_REPEAT = (
+    "You just called the same tool with the same arguments as a previous "
+    "call in this conversation. Repeating the exact same action will produce "
+    "the same result. Stop and reconsider your approach — try a different "
+    "tool, different arguments, or ask the user for clarification."
+)
+
 _NUDGE_MAP: dict[str, str] = {
     "correction": NUDGE_CORRECTION,
     "denial": NUDGE_DENIAL,
@@ -57,6 +64,7 @@ _NUDGE_MAP: dict[str, str] = {
     "completion": NUDGE_COMPLETION,
     "start": NUDGE_START,
     "tool_error": NUDGE_TOOL_ERROR,
+    "repeat": NUDGE_REPEAT,
 }
 
 # ---------------------------------------------------------------------------

turnstone/core/model_registry.py

@@ -34,6 +34,7 @@ class ModelConfig:
     context_window: int = 32768
     provider: str = "openai"
     capabilities: dict[str, Any] = field(default_factory=dict)
+    source: str = ""  # "config", "db", or "" (CLI default)
 
 
 # ---------------------------------------------------------------------------
@@ -81,9 +82,9 @@ class ModelRegistry:
 
     def get_client(self, alias: str) -> Any:
         """Get or lazily create an API client for *alias*. Thread-safe."""
-        if alias not in self._models:
-            raise ValueError(f"Unknown model alias: {alias}")
         with self._client_lock:
+            if alias not in self._models:
+                raise ValueError(f"Unknown model alias: {alias}")
             if alias not in self._clients:
                 cfg = self._models[alias]
                 self._clients[alias] = create_client(
@@ -93,9 +94,9 @@ class ModelRegistry:
 
     def get_provider(self, alias: str) -> LLMProvider:
         """Get the ``LLMProvider`` for *alias*. Thread-safe, cached."""
-        if alias not in self._models:
-            raise ValueError(f"Unknown model alias: {alias}")
         with self._client_lock:
+            if alias not in self._models:
+                raise ValueError(f"Unknown model alias: {alias}")
             if alias not in self._providers:
                 cfg = self._models[alias]
                 self._providers[alias] = create_provider(cfg.provider)
@@ -129,8 +130,46 @@ class ModelRegistry:
         """Number of registered models."""
         return len(self._models)
 
+    @property
+    def models(self) -> dict[str, ModelConfig]:
+        """Return a copy of the models dict (public accessor for reload)."""
+        return dict(self._models)
+
     # -- lifecycle -----------------------------------------------------------
 
+    def reload(
+        self,
+        models: dict[str, ModelConfig],
+        default: str,
+        fallback: list[str] | None = None,
+        agent_model: str | None = None,
+    ) -> None:
+        """Hot-reload all model configs. Thread-safe; clears cached clients.
+
+        Validates arguments before mutating state so a bad reload
+        does not leave the registry in an inconsistent state.
+        """
+        if not models:
+            raise ValueError("ModelRegistry requires at least one model config")
+        if default not in models:
+            raise ValueError(f"Default model '{default}' not found in registry")
+        if fallback:
+            for alias in fallback:
+                if alias not in models:
+                    raise ValueError(f"Fallback model '{alias}' not found in registry")
+        if agent_model and agent_model not in models:
+            raise ValueError(f"Agent model '{agent_model}' not found in registry")
+        with self._client_lock:
+            self._models = dict(models)
+            self.default = default
+            self.fallback = list(fallback) if fallback else []
+            self.agent_model = agent_model
+            for client in self._clients.values():
+                if hasattr(client, "close"):
+                    client.close()
+            self._clients.clear()
+            self._providers.clear()
+
     def shutdown(self) -> None:
         """Close all cached client connections."""
         with self._client_lock:
@@ -146,32 +185,82 @@ class ModelRegistry:
 # ---------------------------------------------------------------------------
 
 
+def _resolve_env_vars(value: str) -> str:
+    """Expand ``${VAR}`` patterns in *value* using environment variables.
+
+    Unresolved variables are replaced with empty strings.
+    """
+    import os
+    import re
+
+    def _replace(m: re.Match[str]) -> str:
+        return os.environ.get(m.group(1), "")
+
+    return re.sub(r"\$\{([A-Za-z_][A-Za-z0-9_]*)\}", _replace, value)
+
+
 def load_model_registry(
     base_url: str,
     api_key: str,
     model: str,
     context_window: int = 32768,
     provider: str = "openai",
+    storage: Any | None = None,
 ) -> ModelRegistry:
-    """Build a ModelRegistry from CLI args and ``config.toml``.
+    """Build a ModelRegistry from CLI args, ``config.toml``, and database.
 
-    Precedence:
+    Precedence (highest to lowest):
 
-    1. ``[models.*]`` sections in config.toml define named models.
-    2. CLI ``--base-url`` / ``--api-key`` / ``--model`` always create a
-       ``"default"`` entry (overrides any ``[models.default]`` section).
-    3. ``[model].default``, ``[model].fallback``, ``[model].agent_model``
+    1. ``[models.*]`` sections in config.toml define named models
+       (``source="config"``).  These override DB entries with the same
+       alias in-memory only — the DB rows are never modified.
+    2. Database model definitions (``source="db"``), loaded when
+       *storage* is provided.
+    3. CLI ``--base-url`` / ``--api-key`` / ``--model`` always create a
+       ``"default"`` entry.
+    4. ``[model].default``, ``[model].fallback``, ``[model].agent_model``
        control routing.
-    4. If no ``[models.*]`` sections exist, a single-entry registry is built
-       from the CLI args.
     """
+    import json as _json
+
     cfg = load_config()
     models_section: dict[str, Any] = cfg.get("models", {})
     model_section: dict[str, Any] = cfg.get("model", {})
 
     configs: dict[str, ModelConfig] = {}
 
-    # Build configs from [models.*] sections
+    # 1. Load DB model definitions (lowest priority, overridden by config.toml)
+    if storage is not None:
+        try:
+            for row in storage.list_model_definitions(enabled_only=True):
+                alias = row["alias"]
+                caps: dict[str, Any] = {}
+                if row.get("capabilities"):
+                    try:
+                        parsed = _json.loads(row["capabilities"])
+                        if isinstance(parsed, dict):
+                            caps = parsed
+                    except (_json.JSONDecodeError, TypeError):
+                        pass
+                row_provider = row.get("provider", "openai")
+                row_model = row["model"]
+                # 0 = auto-detect: inherit CLI-detected context_window,
+                # same fallback chain as config.toml models
+                row_ctx = row.get("context_window", 0) or context_window
+                configs[alias] = ModelConfig(
+                    alias=alias,
+                    base_url=_resolve_env_vars(row.get("base_url", "")),
+                    api_key=_resolve_env_vars(row.get("api_key", "")),
+                    model=row_model,
+                    context_window=row_ctx,
+                    provider=row_provider,
+                    capabilities=caps,
+                    source="db",
+                )
+        except Exception:
+            log.warning("Failed to load model definitions from storage", exc_info=True)
+
+    # 2. Build configs from [models.*] sections (overrides DB for same alias)
     for alias, entry in models_section.items():
         if not isinstance(entry, dict):
             continue
@@ -189,17 +278,20 @@ def load_model_registry(
             capabilities=entry.get("capabilities", {})
             if isinstance(entry.get("capabilities"), dict)
             else {},
+            source="config",
         )
 
-    # Ensure a "default" entry from CLI args
-    configs["default"] = ModelConfig(
-        alias="default",
-        base_url=base_url,
-        api_key=api_key,
-        model=model,
-        context_window=context_window,
-        provider=provider,
-    )
+    # 3. Ensure a "default" entry from CLI args (only if not already defined
+    # by config.toml or DB — those take precedence)
+    if "default" not in configs:
+        configs["default"] = ModelConfig(
+            alias="default",
+            base_url=base_url,
+            api_key=api_key,
+            model=model,
+            context_window=context_window,
+            provider=provider,
+        )
 
     # Determine default alias
     default_alias = model_section.get("default", "default")
@@ -342,3 +434,114 @@ def detect_model(
         log_fn(f"Warning: Could not connect to LLM backend: {e}")
         log_fn("Starting in degraded mode — requests will fail until backend is reachable.")
         return None, None
+
+
+def probe_model_endpoint(
+    provider: str,
+    base_url: str,
+    api_key: str,
+    target_model: str = "",
+) -> dict[str, Any]:
+    """Stateless probe of a model endpoint.
+
+    Creates a temporary SDK client, calls ``/v1/models``, and returns
+    reachability status, available model IDs, detected context window,
+    and server type.  Used by the admin *Detect* button — never persists
+    state or stores the API key.
+    """
+    from turnstone.core.providers import create_client
+
+    result: dict[str, Any] = {
+        "reachable": False,
+        "model_found": None,
+        "available_models": [],
+        "context_window": None,
+        "server_type": None,
+        "error": None,
+    }
+    client = None
+    try:
+        client = create_client(provider, base_url=base_url, api_key=api_key)
+        fast = client.with_options(timeout=10.0, max_retries=0)
+        models = fast.models.list()
+        if not models.data:
+            result["reachable"] = True
+            result["error"] = "No models found at endpoint"
+            return result
+
+        all_ids = [m.id for m in models.data]
+        result["reachable"] = True
+        result["available_models"] = all_ids
+
+        # Determine which model to inspect for context_window
+        if target_model:
+            result["model_found"] = target_model in all_ids
+            inspect_id = target_model if result["model_found"] else all_ids[0]
+        else:
+            inspect_id = all_ids[0]
+
+        inspect_obj = next((m for m in models.data if m.id == inspect_id), None)
+
+        # --- context window detection ---
+        if provider == "anthropic":
+            from turnstone.core.providers import lookup_model_capabilities
+
+            known = lookup_model_capabilities("anthropic", inspect_id)
+            if known is not None:
+                result["context_window"] = known["context_window"]
+            result["server_type"] = "anthropic"
+        else:
+            # OpenAI-compatible path
+            _detect_openai_compat(result, inspect_obj, inspect_id, base_url)
+    except Exception as exc:
+        err_msg = str(exc)
+        if len(err_msg) > 500:
+            err_msg = err_msg[:500] + "..."
+        result["error"] = err_msg
+    finally:
+        if client is not None and hasattr(client, "close"):
+            client.close()
+    return result
+
+
+def _detect_openai_compat(
+    result: dict[str, Any],
+    model_obj: Any,
+    model_id: str,
+    base_url: str,
+) -> None:
+    """Fill context_window and server_type for an OpenAI-compatible endpoint."""
+
+    meta: dict[str, Any] | None = None
+    owned_by: str = ""
+    if model_obj is not None:
+        dumped = model_obj.model_dump()
+        raw_meta = dumped.get("meta")
+        if isinstance(raw_meta, dict):
+            meta = raw_meta
+        owned_by = str(dumped.get("owned_by", ""))
+
+    # Context window: prefer backend metadata, fall back to static table
+    # (only for known models — the default 200k would be misleading for local servers)
+    if meta is not None:
+        n_ctx = meta.get("n_ctx_train")
+        if isinstance(n_ctx, int) and n_ctx > 0:
+            result["context_window"] = n_ctx
+    if result["context_window"] is None:
+        from turnstone.core.providers import lookup_model_capabilities
+
+        known = lookup_model_capabilities("openai", model_id)
+        if known is not None:
+            result["context_window"] = known["context_window"]
+
+    # Server type heuristics
+    if base_url and "api.openai.com" in base_url:
+        result["server_type"] = "openai"
+    elif meta is not None and "n_ctx_train" in meta:
+        result["server_type"] = "llama.cpp"
+    elif "sglang" in owned_by.lower():
+        result["server_type"] = "sglang"
+    elif "/" in (model_id or ""):
+        result["server_type"] = "vllm"
+    else:
+        result["server_type"] = "openai-compatible"

turnstone/core/output_guard.py

@@ -58,6 +58,12 @@ _RE_ENV_SECRET_KEY = re.compile(
     r"(?:^|_)(?:SECRET|TOKEN|PASSWORD|CREDENTIAL)(?:_|$)|(?:^|_)KEY(?:_|$)",
     re.IGNORECASE,
 )
+_RE_JSON_SECRET = re.compile(
+    r'"(?:api_key|apikey|api_secret|secret_key|secret|password|passwd|'
+    r"token|access_token|refresh_token|auth_token|private_key|"
+    r'client_secret|webhook_secret|signing_key|encryption_key)"\s*:\s*"([^"]{8,})"',
+    re.IGNORECASE,
+)
 
 # (pattern, redact_label) — ordered most-specific first for redaction.
 _CREDENTIAL_PATTERNS: list[tuple[re.Pattern[str], str]] = [
@@ -228,6 +234,15 @@ def _check_credentials(
         found = True
         risk = "high"
 
+    if _RE_JSON_SECRET.search(text):
+        _add_flag(flags, "credential_leak")
+        flags.append("json_secret_leak")
+        ann.append(
+            "Output contains JSON with secret-bearing keys (api_key, password, token, etc.)."
+        )
+        found = True
+        risk = "high"
+
     return risk, _redact_credentials(text) if found else None
 
 
@@ -247,6 +262,15 @@ def _redact_credentials(text: str) -> str:
         return key + "=[REDACTED:secret]" if _RE_ENV_SECRET_KEY.search(key) else m.group()
 
     result = _RE_ENV_SECRET_LINE.sub(_redact_env, result)
+
+    def _redact_json_secret(m: re.Match[str]) -> str:
+        # Positional replacement to avoid corrupting key when value == key name
+        start = m.start(1) - m.start()
+        end = m.end(1) - m.start()
+        full = m.group()
+        return full[:start] + "[REDACTED:secret]" + full[end:]
+
+    result = _RE_JSON_SECRET.sub(_redact_json_secret, result)
     return result
 
 

turnstone/core/providers/__init__.py

@@ -25,6 +25,8 @@ __all__ = [
     "UsageInfo",
     "create_client",
     "create_provider",
+    "list_known_models",
+    "lookup_model_capabilities",
 ]
 
 # Singleton instances (stateless, safe to share)
@@ -36,7 +38,7 @@ _anthropic_provider: LLMProvider | None = None
 def create_provider(provider_name: str) -> LLMProvider:
     """Return a provider adapter for the given provider name. Thread-safe."""
     global _anthropic_provider  # noqa: PLW0603
-    if provider_name == "openai":
+    if provider_name in ("openai", "openai-compatible"):
         return _openai_provider
     if provider_name == "anthropic":
         with _provider_lock:
@@ -45,15 +47,19 @@ def create_provider(provider_name: str) -> LLMProvider:
 
                 _anthropic_provider = AnthropicProvider()
             return _anthropic_provider
-    raise ValueError(f"Unknown provider: {provider_name!r}. Supported: openai, anthropic")
+    raise ValueError(
+        f"Unknown provider: {provider_name!r}. Supported: openai, anthropic, openai-compatible"
+    )
 
 
 def create_client(provider_name: str, *, base_url: str, api_key: str) -> Any:
     """Create an SDK client for the given provider."""
-    if provider_name == "openai":
+    if provider_name in ("openai", "openai-compatible"):
         from openai import OpenAI
 
-        return OpenAI(base_url=base_url, api_key=api_key)
+        if base_url:
+            return OpenAI(base_url=base_url, api_key=api_key)
+        return OpenAI(api_key=api_key)
     if provider_name == "anthropic":
         from turnstone.core.providers._anthropic import _ensure_anthropic
 
@@ -62,4 +68,42 @@ def create_client(provider_name: str, *, base_url: str, api_key: str) -> Any:
         if base_url and base_url != "https://api.anthropic.com":
             kwargs["base_url"] = base_url
         return anthropic.Anthropic(**kwargs)
-    raise ValueError(f"Unknown provider: {provider_name!r}. Supported: openai, anthropic")
+    raise ValueError(
+        f"Unknown provider: {provider_name!r}. Supported: openai, anthropic, openai-compatible"
+    )
+
+
+def lookup_model_capabilities(provider: str, model: str) -> dict[str, Any] | None:
+    """Return static capabilities for a known model, or ``None`` if unknown.
+
+    The returned dict has JSON-friendly values (tuples converted to lists).
+    Returns ``None`` for ``openai-compatible`` (no static table for local models).
+    """
+    import dataclasses
+
+    if provider == "openai-compatible":
+        return None
+    prov = create_provider(provider)
+    caps = prov.get_capabilities(model)
+    default = prov.get_capabilities("")
+    if caps is default:
+        return None
+    result = dataclasses.asdict(caps)
+    # Convert tuples to lists for JSON serialisation
+    for key, val in result.items():
+        if isinstance(val, tuple):
+            result[key] = list(val)
+    return result
+
+
+def list_known_models(provider: str) -> list[str]:
+    """Return the model name prefixes in the static capability table."""
+    if provider == "openai":
+        from turnstone.core.providers._openai import _OPENAI_CAPABILITIES
+
+        return sorted(_OPENAI_CAPABILITIES.keys())
+    if provider == "anthropic":
+        from turnstone.core.providers._anthropic import _ANTHROPIC_CAPABILITIES
+
+        return sorted(_ANTHROPIC_CAPABILITIES.keys())
+    return []

turnstone/core/providers/_anthropic.py

@@ -7,6 +7,8 @@ The ``anthropic`` SDK is imported lazily so it remains an optional dependency.
 from __future__ import annotations
 
 import json
+import logging
+import sys
 from typing import TYPE_CHECKING, Any
 
 from turnstone.core.providers._protocol import (
@@ -21,6 +23,8 @@ from turnstone.core.providers._protocol import (
 if TYPE_CHECKING:
     from collections.abc import Iterator
 
+log = logging.getLogger(__name__)
+
 
 def _ensure_anthropic() -> Any:
     """Lazy import anthropic SDK, raising helpful error if not installed."""
@@ -286,6 +290,7 @@ class AnthropicProvider:
         """
         system_parts: list[str] = []
         converted: list[dict[str, Any]] = []
+        pending_orphan_results: list[dict[str, Any]] = []
 
         i = 0
         while i < len(messages):
@@ -299,11 +304,50 @@ class AnthropicProvider:
                 continue
 
             if role == "assistant":
+                # Safety: flush any unconsumed synthetic results from a prior
+                # assistant message (should not happen with well-formed data).
+                if pending_orphan_results:
+                    converted.append({"role": "user", "content": pending_orphan_results})
+                    pending_orphan_results = []
                 # If raw provider content was preserved, pass it through verbatim
                 # so encrypted_content/encrypted_index from web search are retained
                 provider_content = msg.get("_provider_content")
                 if provider_content:
                     converted.append({"role": "assistant", "content": provider_content})
+                    # Check for orphaned tool_use in provider content too
+                    if isinstance(provider_content, list):
+                        pc_tool_ids = [
+                            b["id"]
+                            for b in provider_content
+                            if isinstance(b, dict) and b.get("type") == "tool_use" and b.get("id")
+                        ]
+                        if pc_tool_ids:
+                            j = i + 1
+                            result_ids_pc: set[str] = set()
+                            while j < len(messages) and messages[j]["role"] == "tool":
+                                tc_id = messages[j].get("tool_call_id", "")
+                                if tc_id:
+                                    result_ids_pc.add(tc_id)
+                                j += 1
+                            orphaned_pc = [uid for uid in pc_tool_ids if uid not in result_ids_pc]
+                            if orphaned_pc:
+                                log.debug(
+                                    "Synthesizing %d tool_result(s) for orphaned provider_content tool_use IDs",
+                                    len(orphaned_pc),
+                                )
+                                synthetic_pc = [
+                                    {
+                                        "type": "tool_result",
+                                        "tool_use_id": uid,
+                                        "content": "Tool execution was cancelled.",
+                                        "is_error": True,
+                                    }
+                                    for uid in orphaned_pc
+                                ]
+                                if j == i + 1:
+                                    converted.append({"role": "user", "content": synthetic_pc})
+                                else:
+                                    pending_orphan_results = synthetic_pc
                     i += 1
                     continue
 
@@ -328,6 +372,53 @@ class AnthropicProvider:
                     )
                 if content_blocks:
                     converted.append({"role": "assistant", "content": content_blocks})
+
+                # Repair orphaned tool_use blocks: if this assistant message
+                # has tool_use blocks but the next messages don't provide
+                # matching tool_results, synthesize error results.  This
+                # happens when a cancel interrupts tool execution — the
+                # assistant message is saved to DB before tools run, but
+                # GenerationCancelled prevents tool results from being created.
+                # Collect IDs in order, skip empty IDs (from malformed tool calls).
+                tool_use_ids = [
+                    b["id"] for b in content_blocks if b.get("type") == "tool_use" and b.get("id")
+                ]
+                if tool_use_ids:
+                    # Peek ahead to collect tool_result IDs
+                    j = i + 1
+                    result_ids: set[str] = set()
+                    while j < len(messages) and messages[j]["role"] == "tool":
+                        tc_id = messages[j].get("tool_call_id", "")
+                        if tc_id:
+                            result_ids.add(tc_id)
+                        j += 1
+                    orphaned = [uid for uid in tool_use_ids if uid not in result_ids]
+                    if orphaned:
+                        log.debug(
+                            "Synthesizing %d tool_result(s) for orphaned tool_use IDs",
+                            len(orphaned),
+                        )
+                        # Store for deferred injection — synthetic results are
+                        # appended after any real tool results so
+                        # _merge_consecutive produces them in tool_use order.
+                        synthetic = [
+                            {
+                                "type": "tool_result",
+                                "tool_use_id": uid,
+                                "content": "Tool execution was cancelled.",
+                                "is_error": True,
+                            }
+                            for uid in orphaned
+                        ]
+                        if j == i + 1:
+                            # No real tool messages follow — inject immediately
+                            converted.append({"role": "user", "content": synthetic})
+                        else:
+                            # Real tool messages follow — they'll be converted
+                            # next iteration.  Stash synthetic results to append
+                            # after them.
+                            pending_orphan_results = synthetic
+
                 i += 1
                 continue
 
@@ -340,14 +431,19 @@ class AnthropicProvider:
                     # Convert image_url parts to Anthropic image format
                     if isinstance(content, list):
                         content = self._convert_content_parts(content)
-                    tool_results.append(
-                        {
-                            "type": "tool_result",
-                            "tool_use_id": tool_msg.get("tool_call_id", ""),
-                            "content": content,
-                        }
-                    )
+                    result_block: dict[str, Any] = {
+                        "type": "tool_result",
+                        "tool_use_id": tool_msg.get("tool_call_id", ""),
+                        "content": content,
+                    }
+                    if tool_msg.get("is_error"):
+                        result_block["is_error"] = True
+                    tool_results.append(result_block)
                     i += 1
+                # Append any deferred synthetic results after real ones
+                if pending_orphan_results:
+                    tool_results.extend(pending_orphan_results)
+                    pending_orphan_results = []
                 converted.append({"role": "user", "content": tool_results})
                 continue
 
@@ -459,6 +555,7 @@ class AnthropicProvider:
         reasoning_effort: str = "medium",
         extra_params: dict[str, Any] | None = None,
         deferred_names: frozenset[str] | None = None,
+        cancel_ref: list[Any] | None = None,
     ) -> Iterator[StreamChunk]:
         _ensure_anthropic()
         caps = self.get_capabilities(model)
@@ -476,8 +573,25 @@ class AnthropicProvider:
             deferred_names,
         )
 
-        with client.messages.stream(**kwargs) as stream:
+        manager = client.messages.stream(**kwargs)
+        try:
+            stream = manager.__enter__()
+        except BaseException:
+            manager.__exit__(*sys.exc_info())
+            raise
+        if cancel_ref is not None:
+            cancel_ref.append(stream)
+        return self._iter_with_cleanup(stream, manager)
+
+    def _iter_with_cleanup(self, stream: Any, manager: Any) -> Iterator[StreamChunk]:
+        """Iterate the Anthropic stream, ensuring the context manager exits."""
+        try:
             yield from self._iter_anthropic_stream(stream)
+        except BaseException:
+            manager.__exit__(*sys.exc_info())
+            raise
+        else:
+            manager.__exit__(None, None, None)
 
     def _iter_anthropic_stream(self, stream: Any) -> Iterator[StreamChunk]:
         """Convert Anthropic streaming events to normalized StreamChunks."""
@@ -541,6 +655,12 @@ class AnthropicProvider:
                         raw_blocks[event.index]["thinking"] = (
                             raw_blocks[event.index].get("thinking", "") + delta.thinking
                         )
+                elif delta.type == "signature_delta":
+                    # Accumulate signature into raw block for round-trip
+                    if event.index in raw_blocks:
+                        raw_blocks[event.index]["signature"] = (
+                            raw_blocks[event.index].get("signature", "") + delta.signature
+                        )
                 elif delta.type == "input_json_delta":
                     if event.index in server_tool_blocks:
                         # Accumulate server tool input (search query)
@@ -647,7 +767,24 @@ class AnthropicProvider:
             deferred_names,
         )
 
-        response = client.messages.create(**kwargs)
+        # Use streaming internally to avoid the Anthropic SDK's 10-minute
+        # timeout on non-streaming requests.  get_final_message() returns the
+        # same Message object as messages.create() would.
+        # Mirror create_streaming's defensive __enter__/__exit__ pattern so
+        # resources are cleaned up even if __enter__ fails.
+        manager = client.messages.stream(**kwargs)
+        try:
+            stream = manager.__enter__()
+        except BaseException:
+            manager.__exit__(*sys.exc_info())
+            raise
+        try:
+            response = stream.get_final_message()
+        except BaseException:
+            manager.__exit__(*sys.exc_info())
+            raise
+        else:
+            manager.__exit__(None, None, None)
 
         # Extract content and tool_calls from content blocks.
         # Skip server-side blocks (server_tool_use, web_search_tool_result)

turnstone/core/providers/_openai.py

@@ -11,6 +11,8 @@ from typing import TYPE_CHECKING, Any
 if TYPE_CHECKING:
     from collections.abc import Iterator
 
+import structlog
+
 from turnstone.core.providers._protocol import (
     CompletionResult,
     ModelCapabilities,
@@ -20,6 +22,8 @@ from turnstone.core.providers._protocol import (
     _lookup_capabilities,
 )
 
+log = structlog.get_logger(__name__)
+
 # -- model capabilities -------------------------------------------------------
 
 _OPENAI_CAPABILITIES: dict[str, ModelCapabilities] = {
@@ -273,6 +277,29 @@ class OpenAIProvider:
                 result.append(tool)
         return result
 
+    # -- message sanitisation ------------------------------------------------
+
+    @staticmethod
+    def _sanitize_messages(
+        messages: list[dict[str, Any]],
+    ) -> list[dict[str, Any]]:
+        """Ensure assistant messages always have ``content`` or ``tool_calls``.
+
+        OpenAI-compatible APIs reject assistant messages that have neither.
+        This is a defensive catch-all; the upstream layers should already
+        guarantee well-formed messages.
+        """
+        out: list[dict[str, Any]] = []
+        for msg in messages:
+            if (
+                msg.get("role") == "assistant"
+                and msg.get("content") is None
+                and not msg.get("tool_calls")
+            ):
+                msg = {**msg, "content": ""}
+            out.append(msg)
+        return out
+
     # -- streaming -----------------------------------------------------------
 
     def create_streaming(
@@ -287,8 +314,10 @@ class OpenAIProvider:
         reasoning_effort: str = "medium",
         extra_params: dict[str, Any] | None = None,
         deferred_names: frozenset[str] | None = None,
+        cancel_ref: list[Any] | None = None,
     ) -> Iterator[StreamChunk]:
         caps = self.get_capabilities(model)
+        messages = self._sanitize_messages(messages)
         kwargs: dict[str, Any] = {
             "model": model,
             "messages": messages,
@@ -305,19 +334,34 @@ class OpenAIProvider:
         if extra_params:
             kwargs["extra_body"] = extra_params
 
+        log.debug(
+            "openai.request",
+            model=model,
+            stream=True,
+            max_tokens=max_tokens,
+            message_count=len(messages),
+            tool_count=len(tools) if tools else 0,
+        )
         stream = client.chat.completions.create(**kwargs)
-        yield from self._iter_stream(stream)
+        if cancel_ref is not None:
+            cancel_ref.append(stream)
+        return self._iter_stream(stream)
 
     def _iter_stream(self, stream: Any) -> Iterator[StreamChunk]:
         """Convert OpenAI stream chunks to normalized StreamChunks."""
         first = True
         annotations: list[Any] = []
+        content_len = 0
+        tool_call_count = 0
+        last_finish_reason: str | None = None
+        completion_tokens: int | None = None
         for chunk in stream:
             sc = StreamChunk()
 
             # Finish reason
             if chunk.choices and chunk.choices[0].finish_reason:
                 sc.finish_reason = chunk.choices[0].finish_reason
+                last_finish_reason = sc.finish_reason
 
             # Usage from final chunk
             if hasattr(chunk, "usage") and chunk.usage is not None:
@@ -325,6 +369,7 @@ class OpenAIProvider:
                 pt = getattr(u, "prompt_tokens", None)
                 ct = getattr(u, "completion_tokens", None)
                 tt = getattr(u, "total_tokens", None)
+                completion_tokens = ct
                 if pt is not None and ct is not None:
                     # Extract cached_tokens from prompt_tokens_details.
                     # OpenAI caching is automatic with no write premium, so
@@ -353,6 +398,7 @@ class OpenAIProvider:
             # Content
             if delta.content:
                 sc.content_delta = delta.content
+                content_len += len(delta.content)
 
             # Tool calls
             if delta.tool_calls:
@@ -366,6 +412,7 @@ class OpenAIProvider:
                         if tc_delta.function.arguments:
                             tcd.arguments_delta = tc_delta.function.arguments
                     sc.tool_call_deltas.append(tcd)
+                    tool_call_count += 1
 
             # Accumulate url_citation annotations from search models
             delta_anns = getattr(delta, "annotations", None)
@@ -380,6 +427,15 @@ class OpenAIProvider:
             if has_content or sc.finish_reason or sc.usage:
                 yield sc
 
+        log.debug(
+            "openai.response",
+            stream=True,
+            finish_reason=last_finish_reason,
+            content_length=content_len,
+            tool_call_deltas=tool_call_count,
+            completion_tokens=completion_tokens,
+        )
+
         # Emit accumulated citations as a final info chunk
         if annotations:
             citation_text = self._format_citations("", annotations).strip()
@@ -402,6 +458,7 @@ class OpenAIProvider:
         deferred_names: frozenset[str] | None = None,
     ) -> CompletionResult:
         caps = self.get_capabilities(model)
+        messages = self._sanitize_messages(messages)
         kwargs: dict[str, Any] = {
             "model": model,
             "messages": messages,
@@ -417,6 +474,14 @@ class OpenAIProvider:
         if extra_params:
             kwargs["extra_body"] = extra_params
 
+        log.debug(
+            "openai.request",
+            model=model,
+            stream=False,
+            max_tokens=max_tokens,
+            message_count=len(messages),
+            tool_count=len(tools) if tools else 0,
+        )
         response = client.chat.completions.create(**kwargs)
         choice = response.choices[0]
         msg = choice.message
@@ -454,12 +519,21 @@ class OpenAIProvider:
                 cache_read_tokens=cached or 0,
             )
 
-        return CompletionResult(
+        result = CompletionResult(
             content=content,
             tool_calls=tool_calls,
             finish_reason=choice.finish_reason or "stop",
             usage=usage,
         )
+        log.debug(
+            "openai.response",
+            stream=False,
+            finish_reason=result.finish_reason,
+            content_length=len(content),
+            tool_call_count=len(tool_calls) if tool_calls else 0,
+            completion_tokens=usage.completion_tokens if usage else None,
+        )
+        return result
 
     @staticmethod
     def _format_citations(content: str, annotations: list[Any]) -> str:

turnstone/core/providers/_protocol.py

@@ -125,8 +125,15 @@ class LLMProvider(Protocol):
         reasoning_effort: str = "medium",
         extra_params: dict[str, Any] | None = None,
         deferred_names: frozenset[str] | None = None,
+        cancel_ref: list[Any] | None = None,
     ) -> Iterator[StreamChunk]:
-        """Create a streaming request, yielding normalized StreamChunks."""
+        """Create a streaming request, yielding normalized StreamChunks.
+
+        If *cancel_ref* is provided the provider appends the underlying SDK
+        stream object (which has a ``.close()`` method) before yielding the
+        first chunk.  The caller can then close it from another thread to
+        abort a blocked HTTP read immediately.
+        """
         ...
 
     def create_completion(