release: v0.9.6

Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>

.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

.trivyignore

@@ -0,0 +1,4 @@
+# libexpat integer overflow — no fix available in Debian repos yet
+# https://avd.aquasec.com/nvd/cve-2026-25210
+# Review: remove this entry once a patched libexpat1 is published
+CVE-2026-25210

Dockerfile

@@ -8,10 +8,14 @@ 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:0.11.1 /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 \
+# Remove the slim image's man page exclusion so man-db has actual content
+RUN rm -f /etc/dpkg/dpkg.cfg.d/docker
+
+# System dependencies: psycopg (libpq5), developer tooling for agent workflows
+RUN apt-get update && apt-get upgrade -y && apt-get install -y --no-install-recommends \
+    libpq5 git curl jq man-db manpages procps file \
     && rm -rf /var/lib/apt/lists/*
 
 # Non-root user

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

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

@@ -3,7 +3,7 @@
 Turnstone is an AI orchestration platform with tool use, parallel workstreams, and persistent
 memory. It connects to any OpenAI-compatible API (local vLLM, OpenAI, etc.) or
 Anthropic's native Messages API via pluggable provider adapters, and gives the
-model 17 built-in tools plus external tools via MCP (Model Context Protocol) for
+model 19 built-in tools plus external tools via MCP (Model Context Protocol) for
 reading, writing, searching, planning, and executing code.
 
 The core design principle is a **UI-agnostic engine with pluggable frontends**.
@@ -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-0.16.42/    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/diagrams/02-package-structure.puml

@@ -97,7 +97,7 @@ package "turnstone/sdk/" <<Rectangle>> {
 
 ' Tool schemas
 package "turnstone/tools/" <<Rectangle>> {
-    component [*.json\n18 tool schemas] as schemas <<artifact>>
+    component [*.json\n19 tool schemas] as schemas <<artifact>>
 }
 
 ' Entry point dependencies

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

@@ -24,7 +24,7 @@ partition "Phase 1: Prepare" #E8F5E9 {
         :Dispatch to _prepare_{func_name}();
 
         note right
-            **Dispatch table (17 tools):**
+            **Dispatch table (19 built-in + tool_search):**
             ┌───────────────┬──────────────────┐
             │ Tool          │ Needs Approval?  │
             ├───────────────┼──────────────────┤
@@ -33,16 +33,19 @@ partition "Phase 1: Prepare" #E8F5E9 {
             │ write_file    │ ✓ Yes            │
             │ edit_file     │ ✓ Yes            │
             │ search        │ ✗ Auto-approve   │
+            │ diff_file     │ ✗ Auto-approve   │
             │ math          │ ✗ Auto-approve   │
             │ man           │ ✗ Auto-approve   │
             │ web_fetch     │ ✗ Auto-approve   │
             │ web_search    │ ✗ Auto-approve   │
             │ tool_search   │ ✗ Auto-approve   │
-            │ task          │ ✓ Yes            │
-            │ plan          │ ✓ Yes            │
+            │ task_agent    │ ✓ Yes            │
+            │ plan_agent    │ ✓ Yes            │
             │ memory        │ ✗ Auto-approve   │
             │ recall        │ ✗ Auto-approve   │
             │ notify        │ ✗ Auto-approve   │
+            │ watch         │ ✓ create only    │
+            │ skill         │ ✓ load only      │
             │ read_resource │ ✓ Yes            │
             │ use_prompt    │ ✓ Yes            │
             ├───────────────┼──────────────────┤
@@ -127,7 +130,7 @@ 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

@@ -147,6 +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/23-memory-architecture.puml

@@ -61,7 +61,7 @@ end
 
 Session -> Session : _init_system_messages()\nevery conversation turn
 
-Session -> Session : _get_visible_memories(\nlimit=fetch_limit)
+Session -> Session : _list_visible_memories(\nlimit=fetch_limit)
 note right
   **Scope resolution:**
   1. global scope (always)

docs/diagrams/png/02-package-structure.png

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:efcc7cbe8161a54b5ec24bdfd47e8a142f70029e6e66c707e811b99369f85ebf
-size 310079
+oid sha256:2b3ea69f852e93dc1bc7943db0c71d0cdd1afcbcf470a8737189ae56e85b3206
+size 310075

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 sha256:674712a0563f51837383184652efeb28b7bec13378be636e89d2959bfba39d1e
+size 281519

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/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/tools.md

@@ -1,6 +1,6 @@
 # Tools Reference
 
-turnstone exposes 18 built-in tools plus any number of external MCP tools to the
+turnstone exposes 19 built-in tools plus any number of external MCP tools to the
 LLM via the OpenAI function-calling interface. Built-in tools are defined as JSON
 files under `turnstone/tools/` and loaded at startup by `turnstone/core/tools.py`.
 MCP tools are discovered from configured MCP servers at startup by
@@ -46,12 +46,12 @@ schema plus turnstone-specific metadata keys:
 
 | Name                | Description |
 |---------------------|-------------|
-| `TOOLS`             | All 17 tool definitions (sent to the model). |
+| `TOOLS`             | All 19 tool definitions (sent to the model). |
 | `AGENT_TOOLS`       | Tools with `agent: true` -- available to plan sub-agents. Read-only tools. |
 | `TASK_AGENT_TOOLS`  | Tools with `task_agent: true` -- available to task sub-agents. Includes write operations. |
 | `AGENT_AUTO_TOOLS`  | Set of tool names with `auto_approve: true` -- no user confirmation needed. |
 | `TASK_AUTO_TOOLS`   | Same as `AGENT_AUTO_TOOLS` (identical filter). |
-| `BUILTIN_TOOL_NAMES`| Frozenset of all 17 built-in tool names. Used by tool search to distinguish always-on tools from deferrable MCP tools. |
+| `BUILTIN_TOOL_NAMES`| Frozenset of all 19 built-in tool names. Used by tool search to distinguish always-on tools from deferrable MCP tools. |
 | `PRIMARY_KEY_MAP`   | Dict mapping tool name to its `primary_key` parameter name. |
 
 ---
@@ -69,7 +69,7 @@ Tool execution follows a three-phase pipeline inside `ChatSession._execute_tools
 - Parses the JSON arguments (with fallback for malformed JSON).
 - If JSON parsing fails entirely, uses `PRIMARY_KEY_MAP` to map a bare string
   to the correct parameter.
-- Dispatches to the matching `_prepare_{func_name}()` handler. There are 17
+- Dispatches to the matching `_prepare_{func_name}()` handler. There are 19
   built-in tools plus `tool_search` (synthetic, client-side BM25 fallback) and
   the generic `_prepare_mcp_tool()` handler for MCP tools.
 - Validates arguments and builds a preview dict containing:
@@ -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.
 
@@ -183,8 +188,11 @@ Execute a bash command and return stdout + stderr.
 | Parameter | Type   | Required | Description |
 |-----------|--------|----------|-------------|
 | `command` | string | yes      | The bash command to execute. |
+| `timeout` | integer | no      | Timeout in seconds (1-600). Omit to use the global `tools.timeout` setting (typically 120s). |
+| `stop_on_error` | boolean | no | Enable `set -e` so the script exits on the first command failure. Default false. |
 
-- **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. 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).
 
@@ -216,8 +224,9 @@ Write content to a file, creating it if needed.
 |-----------|--------|----------|-------------|
 | `path`    | string | yes      | Absolute or relative file path. |
 | `content` | string | yes      | The full file content to write. |
+| `mode`    | string | no       | `"overwrite"` (default) replaces the file. `"append"` adds content to the end. |
 
-- **What it does**: Creates or overwrites the file at the given path. Parent directories are created as needed.
+- **What it does**: Creates or overwrites (or appends to) the file at the given path. Parent directories are created as needed.
 - **Auto-approve**: No -- requires user confirmation.
 - **Agent availability**: `task_agent` only.
 
@@ -225,21 +234,44 @@ 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). |
+| `replace_all` | boolean | no       | Replace ALL occurrences of `old_string`. Cannot combine with `near_line` or `edits`. |
 
-- **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.
+\* 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` or `replace_all` is provided). Requires a prior `read_file` or `diff_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.
+- **Replace-all mode**: When `replace_all` is true, all occurrences are replaced via `str.replace()`. The approval preview shows the occurrence count.
 - **Auto-approve**: No -- requires user confirmation.
 - **Agent availability**: `task_agent` only.
 
 ---
 
+### diff_file
+
+Show a unified diff between two files, or between a file and a provided string.
+
+| Parameter       | Type    | Required | Description |
+|-----------------|---------|----------|-------------|
+| `path_a`        | string  | yes      | Path to the first file. |
+| `path_b`        | string  | no       | Path to the second file. Mutually exclusive with `content_b`. |
+| `content_b`     | string  | no       | String content to compare against `path_a`. Mutually exclusive with `path_b`. |
+| `context_lines` | integer | no       | Number of context lines around changes (default 3, max 20). |
+
+- **What it does**: Returns unified diff output using Python's `difflib`. Binary files (containing null bytes) are rejected with a clear error. Files read through `diff_file` satisfy `edit_file`'s read guard — you can diff then edit without a separate `read_file` call. Large diffs are streamed with early cutoff at the tool truncation limit.
+- **Auto-approve**: Yes (read-only).
+- **Agent availability**: `agent` and `task_agent`.
+
+---
+
 ### search
 
 Search file contents for a regex pattern.
@@ -265,8 +297,9 @@ 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`.
 
 ---
@@ -596,7 +629,7 @@ CLI flags override the config file:
    directly.
 
 2. **Partitioning**: When active, tools are split into two sets:
-   - **Always-on** -- the 17 built-in tools (members of `BUILTIN_TOOL_NAMES`).
+   - **Always-on** -- the 19 built-in tools (members of `BUILTIN_TOOL_NAMES`).
      These are always visible to the model.
    - **Deferred** -- all MCP tools. These are not sent in the tool list unless
      the model searches for them.
@@ -639,7 +672,7 @@ MCP-compatible service.
 3. **Schema conversion**: Each MCP tool's `inputSchema` is converted to OpenAI
    function-calling format. The tool name is prefixed: `mcp__{server}__{tool}`.
 
-4. **Merging**: MCP tools are appended after the 17 built-in tools via
+4. **Merging**: MCP tools are appended after the 19 built-in tools via
    `merge_mcp_tools()`. Built-in tools appear first, giving them natural LLM priority.
    When dynamic tool search is active, MCP tools are deferred rather than directly
    visible -- the model discovers them via search as needed (see

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "turnstone"
-version = "0.8.9"
+version = "0.9.6"
 description = "Multi-node AI orchestration platform with tool use, agent routing, and cluster simulation."
 readme = "README.md"
 license = "BUSL-1.1"
@@ -54,7 +54,8 @@ postgres = ["psycopg[binary]>=3.2"]
 ddg = ["ddgs>=9.0"]
 discord = ["discord.py>=2.4", "redis>=7.2"]
 tls = ["lacme>=1.0.4"]
-all = ["turnstone[mq,console,sim,anthropic,postgres,discord,ddg,tls]"]
+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"
@@ -79,7 +80,7 @@ include = [
   "turnstone/console/static/*.js",
   "turnstone/shared_static/*.css",
   "turnstone/shared_static/*.js",
-  "turnstone/shared_static/katex-0.16.42/**/*",
+  "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",

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": [
@@ -916,6 +951,7 @@
       "resolved": "https://registry.npmjs.org/picomatch/-/picomatch-4.0.4.tgz",
       "integrity": "sha512-QP88BAKvMam/3NxH6vj2o21R6MjxZUAd6nlwAS/pnGvN9IVLocLHxGYIzFhg6fUQ+5th6P4dv4eW9jX3DSIj7A==",
       "dev": true,
+      "license": "MIT",
       "engines": {
         "node": ">=12"
       },
@@ -953,14 +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"
@@ -969,21 +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": {
@@ -1084,16 +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": {
@@ -1162,19 +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",
@@ -1185,7 +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"
       },
@@ -1202,10 +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/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 {
@@ -77,6 +83,8 @@ export interface StatusEvent {
   effort: string;
   cache_creation_tokens?: number;
   cache_read_tokens?: number;
+  tool_calls_this_turn?: number;
+  turn_count?: number;
 }
 
 export interface PlanReviewEvent {
@@ -149,6 +157,7 @@ export type ServerEvent =
   | ContentEvent
   | ReasoningEvent
   | StreamEndEvent
+  | StateChangeEvent
   | ToolInfoEvent
   | ApproveRequestEvent
   | ApprovalResolvedEvent
@@ -246,6 +255,10 @@ 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

@@ -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/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):
@@ -179,7 +180,7 @@ class TestCancelDuringToolExecution:
     """Cancel while tools are being executed."""
 
     def test_rollback_incomplete_tool_results(self, tmp_db):
-        """When cancelled during tool execution, incomplete results are rolled back."""
+        """When cancelled during tool execution, synthesized results replace missing tool outputs."""
         ui = NullUI()
         session = _make_session(ui=ui)
 
@@ -235,13 +236,15 @@ class TestCancelDuringToolExecution:
 
         # Session should be idle
         assert ui.states[-1] == "idle"
-        # No tool result messages should remain (rolled back)
-        roles = [m["role"] for m in session.messages]
-        assert "tool" not in roles
-        # The assistant message with tool_calls should also be rolled back
-        for m in session.messages:
-            if m["role"] == "assistant":
-                assert "tool_calls" not in m or not m["tool_calls"]
+        # Cancelled tool calls should have synthesized results
+        tool_msgs = [m for m in session.messages if m["role"] == "tool"]
+        assert len(tool_msgs) == 1
+        assert tool_msgs[0]["tool_call_id"] == "tc_1"
+        assert "Cancelled by user" in tool_msgs[0]["content"]
+        assert tool_msgs[0].get("is_error") is True
+        # The assistant message with tool_calls should still be present
+        assistant_msgs = [m for m in session.messages if m.get("tool_calls")]
+        assert len(assistant_msgs) == 1
 
 
 class TestCancelWhenIdle:
@@ -407,3 +410,402 @@ 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_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()
@@ -434,7 +436,7 @@ class TestSkillCatalogDisclosure:
                 "turnstone.core.session.list_skills_by_activation",
                 return_value=search_skills or [],
             ),
-            patch.object(session, "_get_visible_memories", return_value=[]),
+            patch.object(session, "_list_visible_memories", return_value=[]),
         ):
             session._init_system_messages()
 

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 +1,259 @@
+"""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,
+    **kwargs: Any,
+) -> MagicMock:
+    m = MagicMock()
+    m.id = model_id
+    dumped: dict[str, Any] = {"owned_by": owned_by}
+    if meta is not None:
+        dumped["meta"] = meta
+    if kwargs.get("max_model_len") is not None:
+        dumped["max_model_len"] = kwargs["max_model_len"]
+    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_vllm_max_model_len(self, mock_cc: MagicMock) -> None:
+        m = _mock_model("/models/nemotron", max_model_len=262144, owned_by="vllm")
+        mock_cc.return_value = _mock_client(m)
+
+        result = probe_model_endpoint("openai", "http://localhost:8000/v1", "key")
+        assert result["context_window"] == 262144
+        assert result["server_type"] == "vllm"
+
+    @patch("turnstone.core.providers.create_client")
+    def test_vllm_max_model_len_preferred_over_meta(self, mock_cc: MagicMock) -> None:
+        m = _mock_model(
+            "/models/test",
+            meta={"n_ctx_train": 8192},
+            max_model_len=131072,
+        )
+        mock_cc.return_value = _mock_client(m)
+
+        result = probe_model_endpoint("openai", "http://localhost:8000/v1", "key")
+        assert result["context_window"] == 131072
+
+    @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"] == 1000000
+
+    @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"] == 1000000
+        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:
@@ -624,3 +899,125 @@ class TestDetectModelTimeout:
 
         result = detect_model(client, provider="openai", fatal=False)
         assert result == (None, None)
+
+    def test_vllm_max_model_len_detected(self) -> None:
+        """detect_model() reads max_model_len from vLLM model objects."""
+        mock_model = MagicMock()
+        mock_model.id = "/models/nemotron"
+        mock_model.model_dump.return_value = {
+            "owned_by": "vllm",
+            "max_model_len": 262144,
+        }
+
+        fast_client = MagicMock()
+        fast_client.models.list.return_value = MagicMock(data=[mock_model])
+
+        client = MagicMock()
+        client.with_options.return_value = fast_client
+
+        model_id, ctx = detect_model(client, provider="openai")
+        assert model_id == "/models/nemotron"
+        assert ctx == 262144
+
+
+class TestExtractContextWindow:
+    def test_vllm_max_model_len(self) -> None:
+        from turnstone.core.model_registry import _extract_context_window
+
+        m = MagicMock()
+        m.id = "/models/test"
+        m.model_dump.return_value = {"max_model_len": 131072}
+        assert _extract_context_window(m, "openai") == 131072
+
+    def test_llama_cpp_meta(self) -> None:
+        from turnstone.core.model_registry import _extract_context_window
+
+        m = MagicMock()
+        m.id = "test"
+        m.model_dump.return_value = {"meta": {"n_ctx_train": 8192}}
+        assert _extract_context_window(m, "openai") == 8192
+
+    def test_vllm_preferred_over_meta(self) -> None:
+        from turnstone.core.model_registry import _extract_context_window
+
+        m = MagicMock()
+        m.id = "test"
+        m.model_dump.return_value = {"max_model_len": 262144, "meta": {"n_ctx_train": 4096}}
+        assert _extract_context_window(m, "openai") == 262144
+
+    def test_no_metadata_returns_none(self) -> None:
+        from turnstone.core.model_registry import _extract_context_window
+
+        m = MagicMock()
+        m.id = "test"
+        m.model_dump.return_value = {}
+        assert _extract_context_window(m, "openai") is None
+
+
+class TestHealthMonitorModelChange:
+    def test_model_change_fires_callback(self) -> None:
+        from turnstone.core.healthcheck import BackendHealthMonitor
+
+        changes: list[tuple[str, int | None]] = []
+
+        def on_change(model_id: str, ctx: int | None) -> None:
+            changes.append((model_id, ctx))
+
+        client = MagicMock()
+        monitor = BackendHealthMonitor(
+            client=client,
+            provider="openai",
+            initial_model="model-a",
+            on_model_changed=on_change,
+        )
+
+        # Simulate probe returning a different model
+        resp = MagicMock()
+        m = MagicMock()
+        m.id = "model-b"
+        m.model_dump.return_value = {"max_model_len": 131072}
+        resp.data = [m]
+
+        monitor._check_model_change(resp)
+        assert len(changes) == 1
+        assert changes[0] == ("model-b", 131072)
+        assert monitor._last_detected_model == "model-b"
+
+    def test_same_model_no_callback(self) -> None:
+        from turnstone.core.healthcheck import BackendHealthMonitor
+
+        changes: list[tuple[str, int | None]] = []
+
+        def on_change(model_id: str, ctx: int | None) -> None:
+            changes.append((model_id, ctx))
+
+        client = MagicMock()
+        monitor = BackendHealthMonitor(
+            client=client,
+            provider="openai",
+            initial_model="model-a",
+            on_model_changed=on_change,
+        )
+
+        resp = MagicMock()
+        m = MagicMock()
+        m.id = "model-a"
+        m.model_dump.return_value = {}
+        resp.data = [m]
+
+        monitor._check_model_change(resp)
+        assert len(changes) == 0
+
+    def test_no_callback_configured(self) -> None:
+        from turnstone.core.healthcheck import BackendHealthMonitor
+
+        client = MagicMock()
+        monitor = BackendHealthMonitor(client=client, initial_model="model-a")
+
+        resp = MagicMock()
+        m = MagicMock()
+        m.id = "model-b"
+        resp.data = [m]
+
+        # Should not raise
+        monitor._check_model_change(resp)

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,
@@ -913,7 +957,7 @@ class TestAnthropicHelpers:
 
         provider = AnthropicProvider()
         caps = provider.get_capabilities("claude-opus-4-6")
-        assert caps.context_window == 200000
+        assert caps.context_window == 1000000
         assert caps.max_output_tokens == 128000
         assert caps.thinking_mode == "adaptive"
         assert caps.supports_effort is True
@@ -922,11 +966,11 @@ class TestAnthropicHelpers:
         from turnstone.core.providers._anthropic import AnthropicProvider
 
         provider = AnthropicProvider()
-        # Prefix match: "claude-sonnet-4" matches dated variants
-        caps = provider.get_capabilities("claude-sonnet-4-20260101")
-        assert caps.context_window == 200000
+        # Prefix match: "claude-sonnet-4-6" matches dated variants
+        caps = provider.get_capabilities("claude-sonnet-4-6-20260101")
+        assert caps.context_window == 1000000
         assert caps.token_param == "max_tokens"
-        assert caps.thinking_mode == "manual"
+        assert caps.thinking_mode == "adaptive"
 
     def test_capabilities_lookup_unknown(self) -> None:
         from turnstone.core.providers._anthropic import AnthropicProvider
@@ -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."""
 
@@ -1180,7 +1407,7 @@ class TestAnthropicWebSearch:
         """All Anthropic models should support native web search."""
         caps = self.provider.get_capabilities("claude-opus-4-6")
         assert caps.supports_web_search is True
-        caps = self.provider.get_capabilities("claude-sonnet-4")
+        caps = self.provider.get_capabilities("claude-sonnet-4-6")
         assert caps.supports_web_search is True
         # Unknown models use default which also has web search
         caps = self.provider.get_capabilities("claude-unknown-99")
@@ -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):

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_structured_memory.py

@@ -3,6 +3,7 @@
 from turnstone.core.memory import (
     count_structured_memories,
     delete_structured_memory,
+    get_structured_memory_by_name,
     list_structured_memories,
     normalize_key,
     save_structured_memory,
@@ -67,6 +68,29 @@ class TestSearchStructuredMemories:
         assert any(r["name"] == "db_host" for r in results)
 
 
+class TestGetStructuredMemoryByName:
+    def test_get_existing(self, tmp_db):
+        save_structured_memory("my_mem", "full content here that is quite long")
+        mem = get_structured_memory_by_name("my_mem", "global", "")
+        assert mem is not None
+        assert mem["content"] == "full content here that is quite long"
+        assert mem["name"] == "my_mem"
+
+    def test_get_nonexistent(self, tmp_db):
+        assert get_structured_memory_by_name("nope", "global", "") is None
+
+    def test_get_wrong_scope(self, tmp_db):
+        save_structured_memory("ws_mem", "data", scope="workstream", scope_id="ws1")
+        assert get_structured_memory_by_name("ws_mem", "global", "") is None
+        assert get_structured_memory_by_name("ws_mem", "workstream", "ws1") is not None
+
+    def test_get_normalizes_key(self, tmp_db):
+        save_structured_memory("My-Key", "value")
+        mem = get_structured_memory_by_name("My-Key", "global", "")
+        assert mem is not None
+        assert mem["name"] == "my_key"
+
+
 class TestCountStructuredMemories:
     def test_count_zero(self, tmp_db):
         assert count_structured_memories() == 0
@@ -80,3 +104,109 @@ class TestCountStructuredMemories:
 class TestNormalizeKey:
     def test_basic(self):
         assert normalize_key("My-Key Name") == "my_key_name"
+
+
+class TestScopeIsolation:
+    """Verify that list/search without scope only returns visible memories.
+
+    Reproduces the cross-workstream leak: unscoped list/search must not
+    return workstream-scoped memories from other workstreams or
+    user-scoped memories from other users.
+    """
+
+    def _seed(self):
+        """Create memories across multiple scopes."""
+        save_structured_memory("global_note", "visible to all", scope="global")
+        save_structured_memory("ws1_note", "belongs to ws1", scope="workstream", scope_id="ws1")
+        save_structured_memory("ws2_note", "belongs to ws2", scope="workstream", scope_id="ws2")
+        save_structured_memory("u1_note", "belongs to user1", scope="user", scope_id="u1")
+        save_structured_memory("u2_note", "belongs to user2", scope="user", scope_id="u2")
+
+    @staticmethod
+    def _list_visible(ws_id: str, user_id: str, mem_type: str = "", limit: int = 50):
+        """Replicate the scope-filtered list logic from ChatSession."""
+        global_mems = list_structured_memories(mem_type=mem_type, scope="global", limit=limit)
+        ws_mems = list_structured_memories(
+            mem_type=mem_type, scope="workstream", scope_id=ws_id, limit=limit
+        )
+        user_mems = (
+            list_structured_memories(mem_type=mem_type, scope="user", scope_id=user_id, limit=limit)
+            if user_id
+            else []
+        )
+        combined = global_mems + ws_mems + user_mems
+        combined.sort(key=lambda m: m.get("updated", ""), reverse=True)
+        return combined[:limit]
+
+    @staticmethod
+    def _search_visible(query: str, ws_id: str, user_id: str, mem_type: str = "", limit: int = 20):
+        """Replicate the scope-filtered search logic from ChatSession."""
+        global_mems = search_structured_memories(
+            query, mem_type=mem_type, scope="global", limit=limit
+        )
+        ws_mems = search_structured_memories(
+            query, mem_type=mem_type, scope="workstream", scope_id=ws_id, limit=limit
+        )
+        user_mems = (
+            search_structured_memories(
+                query, mem_type=mem_type, scope="user", scope_id=user_id, limit=limit
+            )
+            if user_id
+            else []
+        )
+        combined = global_mems + ws_mems + user_mems
+        combined.sort(key=lambda m: m.get("updated", ""), reverse=True)
+        return combined[:limit]
+
+    def test_unscoped_list_returns_all_scopes(self, tmp_db):
+        """Demonstrate the leak: unscoped list returns everything."""
+        self._seed()
+        all_mems = list_structured_memories()
+        assert len(all_mems) == 5  # no scope filter → all memories
+
+    def test_visible_list_excludes_other_workstreams(self, tmp_db):
+        """Scope-filtered list for ws1/u1 excludes ws2 and u2 memories."""
+        self._seed()
+        visible = self._list_visible("ws1", "u1")
+        names = {m["name"] for m in visible}
+        assert "global_note" in names
+        assert "ws1_note" in names
+        assert "u1_note" in names
+        assert "ws2_note" not in names
+        assert "u2_note" not in names
+
+    def test_visible_list_no_user(self, tmp_db):
+        """Scope-filtered list with no user_id excludes all user memories."""
+        self._seed()
+        visible = self._list_visible("ws1", "")
+        names = {m["name"] for m in visible}
+        assert "global_note" in names
+        assert "ws1_note" in names
+        assert "u1_note" not in names
+        assert "u2_note" not in names
+
+    def test_visible_search_no_user(self, tmp_db):
+        """Scope-filtered search with no user_id excludes all user memories."""
+        self._seed()
+        visible = self._search_visible("belongs", "ws1", "")
+        names = {m["name"] for m in visible}
+        assert "ws1_note" in names
+        assert "u1_note" not in names
+        assert "u2_note" not in names
+
+    def test_visible_search_excludes_other_workstreams(self, tmp_db):
+        """Scope-filtered search for ws1/u1 excludes ws2 and u2 memories."""
+        self._seed()
+        visible = self._search_visible("belongs", "ws1", "u1")
+        names = {m["name"] for m in visible}
+        assert "ws1_note" in names
+        assert "u1_note" in names
+        assert "ws2_note" not in names
+        assert "u2_note" not in names
+
+    def test_explicit_scope_still_works(self, tmp_db):
+        """Explicit scope filter continues to work as before."""
+        self._seed()
+        ws2_only = list_structured_memories(scope="workstream", scope_id="ws2")
+        assert len(ws2_only) == 1
+        assert ws2_only[0]["name"] == "ws2_note"

tests/test_tools_schema.py

@@ -72,18 +72,19 @@ class TestToolsMetadata:
     """Validate the metadata extracted from JSON files."""
 
     def test_tool_count(self):
-        assert len(TOOLS) == 18
+        assert len(TOOLS) == 19
 
     def test_agent_tools_count(self):
-        assert len(AGENT_TOOLS) == 9
+        assert len(AGENT_TOOLS) == 10
 
     def test_task_agent_tools_count(self):
-        assert len(TASK_AGENT_TOOLS) == 12
+        assert len(TASK_AGENT_TOOLS) == 13
 
     def test_auto_approve_sets_match(self):
         expected = {
             "read_file",
             "search",
+            "diff_file",
             "math",
             "man",
             "web_fetch",
@@ -113,6 +114,7 @@ class TestToolsMetadata:
             "read_resource": "uri",
             "use_prompt": "name",
             "skill": "name",
+            "diff_file": "path_a",
         }
         assert expected == PRIMARY_KEY_MAP
 

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_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/__init__.py

@@ -1,3 +1,3 @@
 """turnstone - Multi-node AI orchestration platform with tool use, agent routing, and cluster simulation."""
 
-__version__ = "0.8.9"
+__version__ = "0.9.6"

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 +862,84 @@ 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",
@@ -953,6 +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/discord/cog.py

@@ -183,16 +183,21 @@ class MessageCog:
                 auto_archive_duration=self.ts.config.thread_auto_archive,  # type: ignore[arg-type]
             )
 
-            # Create workstream with the initial message.
+            # Create workstream WITHOUT initial_message — subscribe to events
+            # first, then send the message.  Sending initial_message through
+            # the bridge races with subscription: Redis pub/sub is fire-and-
+            # forget, so response events published before subscribe completes
+            # are silently dropped.
             ws_id, _is_new = await self.ts.router.get_or_create_workstream(
                 channel_type="discord",
                 channel_id=str(thread.id),
                 name=thread_name,
                 model=self.ts.config.model,
-                initial_message=content,
+                initial_message="",
             )
 
             await self.ts.subscribe_ws(ws_id, thread)
+            await self.ts.router.send_message(ws_id, content)
             log.info(
                 "discord.workstream_created",
                 ws_id=ws_id,
@@ -362,10 +367,11 @@ class MessageCog:
             channel_id=str(thread.id),
             name=thread_name,
             model=self.ts.config.model,
-            initial_message=message,
+            initial_message="",
         )
 
         await self.ts.subscribe_ws(ws_id, thread)
+        await self.ts.router.send_message(ws_id, message)
 
         await interaction.followup.send(
             f"Workstream started in {thread.mention}",

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/server.py

@@ -365,6 +365,25 @@ async def oidc_callback(request: Request) -> Response:
     return await handle_oidc_callback(request, JWT_AUD_CONSOLE)
 
 
+# ---------------------------------------------------------------------------
+# Route handlers — available models (lightweight, no admin permission)
+# ---------------------------------------------------------------------------
+
+
+async def list_available_models(request: Request) -> JSONResponse:
+    """GET /v1/api/models — enabled model aliases for workstream creation."""
+    from turnstone.core.web_helpers import require_storage_or_503
+
+    storage, err = require_storage_or_503(request)
+    if err:
+        return err
+
+    rows = storage.list_model_definitions(enabled_only=True)
+    # Only expose alias/model/provider — rows also contain api_key, base_url, etc.
+    models = [{"alias": r["alias"], "model": r["model"], "provider": r["provider"]} for r in rows]
+    return JSONResponse({"models": models})
+
+
 # ---------------------------------------------------------------------------
 # Route handlers — workstream creation
 # ---------------------------------------------------------------------------
@@ -1748,9 +1767,11 @@ _VALID_PERMISSIONS = frozenset(
         "admin.memories",
         "admin.settings",
         "admin.mcp",
+        "admin.models",
         "tools.approve",
         "workstreams.create",
         "workstreams.close",
+        "conversation.modify",
     }
 )
 
@@ -3656,7 +3677,6 @@ async def admin_list_settings(request: Request) -> JSONResponse:
     if err:
         return err
 
-    reveal = request.query_params.get("reveal") == "true"
     stored = {r["key"]: r for r in storage.list_system_settings() if r.get("node_id", "") == ""}
 
     settings: list[dict[str, Any]] = []
@@ -3669,7 +3689,7 @@ async def admin_list_settings(request: Request) -> JSONResponse:
                 val = row["value"]
             info = {
                 "key": key,
-                "value": "***" if defn.is_secret and not reveal else val,
+                "value": "***" if defn.is_secret else val,
                 "source": "storage",
                 "type": defn.type,
                 "description": defn.description,
@@ -3683,7 +3703,7 @@ async def admin_list_settings(request: Request) -> JSONResponse:
         else:
             info = {
                 "key": key,
-                "value": "(managed via config file / env)" if defn.is_secret else defn.default,
+                "value": "***" if defn.is_secret else defn.default,
                 "source": "default",
                 "type": defn.type,
                 "description": defn.description,
@@ -3758,18 +3778,31 @@ async def admin_update_setting(request: Request) -> JSONResponse:
     except ValueError:
         return JSONResponse({"error": f"Unknown setting: {key}"}, status_code=400)
 
-    if defn.is_secret:
-        return JSONResponse(
-            {
-                "error": "Secret settings cannot be modified via API — use config.toml or environment variables"
-            },
-            status_code=403,
-        )
-
     if "value" not in body:
         return JSONResponse({"error": "value is required"}, status_code=400)
 
     raw_value = body.get("value")
+
+    # Secret sentinel: "***" means "keep existing value"
+    if defn.is_secret and raw_value == "***":
+        existing = storage.get_system_setting(key)
+        return JSONResponse(
+            {
+                "key": key,
+                "value": "***",
+                "source": "storage" if existing else "default",
+                "type": defn.type,
+                "description": defn.description,
+                "section": defn.section,
+                "is_secret": True,
+                "node_id": existing.get("node_id", "") if existing else "",
+                "changed_by": existing.get("changed_by", "") if existing else "",
+                "updated": existing.get("updated", "") if existing else "",
+                "restart_required": defn.restart_required,
+                "unchanged": True,
+            }
+        )
+
     try:
         typed_value = validate_value(key, raw_value)
     except ValueError as e:
@@ -4663,6 +4696,484 @@ async def admin_import_mcp_config(request: Request) -> JSONResponse:
     return JSONResponse({"imported": imported, "skipped": skipped, "errors": errors})
 
 
+# ---------------------------------------------------------------------------
+# Admin: Model Definitions
+# ---------------------------------------------------------------------------
+
+_MODEL_ALIAS_RE = re.compile(r"^[a-zA-Z0-9._-]+$")
+_MODEL_PROVIDERS = frozenset({"openai", "anthropic", "openai-compatible"})
+
+
+def _mask_model_secrets(model: dict[str, Any]) -> dict[str, Any]:
+    """Replace api_key with '***' (unconditional, write-only)."""
+    m = dict(model)
+    if m.get("api_key"):
+        m["api_key"] = "***"
+    return m
+
+
+async def _collect_model_status(
+    request: Request,
+) -> dict[str, dict[str, dict[str, Any]]]:
+    """Query all nodes for model status. Returns {node_id: {alias: info}}."""
+    collector: ClusterCollector = request.app.state.collector
+    nodes = collector.get_all_nodes()
+    client: httpx.AsyncClient = request.app.state.proxy_client
+    headers = _proxy_auth_headers(request)
+    sem = asyncio.Semaphore(_get_fan_out_limit(request))
+
+    async def _fetch(node: dict[str, Any]) -> tuple[str, dict[str, dict[str, Any]] | None]:
+        node_id = node.get("node_id", "")
+        url = node.get("server_url", "")
+        if not url:
+            return node_id, None
+        async with sem:
+            try:
+                resp = await client.get(
+                    f"{url.rstrip('/')}/v1/api/_internal/model-status",
+                    headers=headers,
+                    timeout=10,
+                )
+                if resp.status_code == 200:
+                    return node_id, resp.json().get("models", {})
+            except Exception:
+                log.debug("Failed to fetch model status from node %s", node_id, exc_info=True)
+        return node_id, None
+
+    results = await asyncio.gather(*[_fetch(n) for n in nodes])
+    return {nid: models for nid, models in results if models is not None}
+
+
+async def _notify_nodes_model_reload(request: Request) -> dict[str, Any]:
+    """Tell all nodes to re-read model definitions from DB and rebuild registry."""
+    collector: ClusterCollector = request.app.state.collector
+    nodes = collector.get_all_nodes()
+    client: httpx.AsyncClient = request.app.state.proxy_client
+    headers = _proxy_auth_headers(request)
+    sem = asyncio.Semaphore(_get_fan_out_limit(request))
+
+    async def _notify(node: dict[str, Any]) -> tuple[str, Any]:
+        node_id = node.get("node_id", "")
+        url = node.get("server_url", "")
+        if not url:
+            return node_id, None
+        async with sem:
+            try:
+                resp = await client.post(
+                    f"{url.rstrip('/')}/v1/api/_internal/model-reload",
+                    headers=headers,
+                    timeout=30,
+                )
+                return node_id, resp.json()
+            except Exception as exc:
+                log.debug("Failed to notify node %s for model reload", node_id, exc_info=True)
+                return node_id, {"error": str(exc)}
+
+    results = await asyncio.gather(*[_notify(n) for n in nodes])
+    return {nid: data for nid, data in results if data is not None}
+
+
+async def admin_list_model_definitions(request: Request) -> JSONResponse:
+    """GET /v1/api/admin/model-definitions — list all model definitions."""
+    from turnstone.core.auth import require_permission
+    from turnstone.core.web_helpers import require_storage_or_503
+
+    storage, err = require_storage_or_503(request)
+    if err:
+        return err
+    err = require_permission(request, "admin.models")
+    if err:
+        return err
+
+    db_models = storage.list_model_definitions()
+
+    # Collect live status from all nodes
+    node_statuses = await _collect_model_status(request)
+
+    db_aliases: set[str] = set()
+    result = []
+    for m in db_models:
+        db_aliases.add(m["alias"])
+        m["source"] = "db"
+        result.append(_mask_model_secrets(m))
+
+    # Merge config-sourced models visible on nodes but not in DB
+    config_aliases: set[str] = set()
+    for node_models in node_statuses.values():
+        for alias in node_models:
+            if alias not in db_aliases:
+                config_aliases.add(alias)
+    for alias in sorted(config_aliases):
+        # Build a synthetic read-only entry from node-reported data
+        model_name = ""
+        provider = "openai"
+        context_window = 0
+        for node_models in node_statuses.values():
+            nm = node_models.get(alias)
+            if nm:
+                model_name = nm.get("model", "")
+                provider = nm.get("provider", "openai")
+                context_window = nm.get("context_window", 0)
+                break
+        result.append(
+            {
+                "definition_id": "",
+                "alias": alias,
+                "model": model_name,
+                "provider": provider,
+                "base_url": "",
+                "api_key": "",
+                "context_window": context_window,
+                "capabilities": "{}",
+                "enabled": True,
+                "source": "config",
+                "created_by": "",
+                "created": "",
+                "updated": "",
+            }
+        )
+
+    return JSONResponse({"models": result})
+
+
+async def admin_create_model_definition(request: Request) -> JSONResponse:
+    """POST /v1/api/admin/model-definitions — create a model definition."""
+    import uuid
+
+    from turnstone.core.audit import record_audit
+    from turnstone.core.auth import require_permission
+    from turnstone.core.web_helpers import read_json_or_400, require_storage_or_503
+
+    storage, err = require_storage_or_503(request)
+    if err:
+        return err
+    err = require_permission(request, "admin.models")
+    if err:
+        return err
+
+    body = await read_json_or_400(request)
+    if isinstance(body, JSONResponse):
+        return body
+
+    alias = str(body.get("alias", "")).strip()[:64]
+    model_name = str(body.get("model", "")).strip()[:128]
+    if not alias:
+        return JSONResponse({"error": "alias is required"}, status_code=400)
+    if not model_name:
+        return JSONResponse({"error": "model is required"}, status_code=400)
+    if not _MODEL_ALIAS_RE.match(alias):
+        return JSONResponse(
+            {"error": "alias must match [a-zA-Z0-9._-]+"},
+            status_code=400,
+        )
+
+    # Check alias uniqueness
+    if storage.get_model_definition_by_alias(alias):
+        return JSONResponse(
+            {"error": f"Model alias '{alias}' already exists"},
+            status_code=409,
+        )
+
+    definition_id = uuid.uuid4().hex
+    audit_uid, ip = _audit_context(request)
+
+    provider = str(body.get("provider", "openai")).strip()
+    if provider not in _MODEL_PROVIDERS:
+        return JSONResponse(
+            {"error": f"Unknown provider: {provider!r}"},
+            status_code=400,
+        )
+    base_url = str(body.get("base_url", "")).strip()
+    api_key = str(body.get("api_key", "")).strip()
+    ctx_raw = body.get("context_window", 32768)
+    context_window = max(0, int(ctx_raw)) if isinstance(ctx_raw, (int, float)) else 0
+    caps = body.get("capabilities", {})
+    capabilities = json.dumps(caps) if isinstance(caps, dict) else "{}"
+    enabled = bool(body.get("enabled", True))
+
+    storage.create_model_definition(
+        definition_id=definition_id,
+        alias=alias,
+        model=model_name,
+        provider=provider,
+        base_url=base_url,
+        api_key=api_key,
+        context_window=context_window,
+        capabilities=capabilities,
+        enabled=enabled,
+        created_by=audit_uid,
+    )
+
+    record_audit(
+        storage,
+        audit_uid,
+        "model_definition.create",
+        "model_definition",
+        definition_id,
+        {"alias": alias},
+        ip,
+    )
+
+    created = storage.get_model_definition(definition_id)
+    if created is None:
+        return JSONResponse(
+            {"error": f"Model alias '{alias}' already exists (concurrent insert)"},
+            status_code=409,
+        )
+    return JSONResponse(_mask_model_secrets(created))
+
+
+async def admin_get_model_definition(request: Request) -> JSONResponse:
+    """GET /v1/api/admin/model-definitions/{definition_id}."""
+    from turnstone.core.auth import require_permission
+    from turnstone.core.web_helpers import require_storage_or_503
+
+    storage, err = require_storage_or_503(request)
+    if err:
+        return err
+    err = require_permission(request, "admin.models")
+    if err:
+        return err
+
+    definition_id = request.path_params["definition_id"]
+    model_def = storage.get_model_definition(definition_id)
+    if model_def is None:
+        return JSONResponse({"error": "Model definition not found"}, status_code=404)
+
+    return JSONResponse(_mask_model_secrets(model_def))
+
+
+async def admin_update_model_definition(request: Request) -> JSONResponse:
+    """PUT /v1/api/admin/model-definitions/{definition_id}."""
+    from turnstone.core.audit import record_audit
+    from turnstone.core.auth import require_permission
+    from turnstone.core.web_helpers import read_json_or_400, require_storage_or_503
+
+    storage, err = require_storage_or_503(request)
+    if err:
+        return err
+    err = require_permission(request, "admin.models")
+    if err:
+        return err
+
+    definition_id = request.path_params["definition_id"]
+    existing = storage.get_model_definition(definition_id)
+    if existing is None:
+        return JSONResponse({"error": "Model definition not found"}, status_code=404)
+
+    body = await read_json_or_400(request)
+    if isinstance(body, JSONResponse):
+        return body
+
+    updates: dict[str, Any] = {}
+    if "alias" in body:
+        alias = str(body["alias"]).strip()[:64]
+        if not alias:
+            return JSONResponse({"error": "alias cannot be empty"}, status_code=400)
+        if not _MODEL_ALIAS_RE.match(alias):
+            return JSONResponse(
+                {"error": "alias must match [a-zA-Z0-9._-]+"},
+                status_code=400,
+            )
+        if alias != existing["alias"] and storage.get_model_definition_by_alias(alias):
+            return JSONResponse(
+                {"error": f"Model alias '{alias}' already exists"},
+                status_code=409,
+            )
+        updates["alias"] = alias
+    if "model" in body:
+        model_val = str(body["model"]).strip()[:128]
+        if not model_val:
+            return JSONResponse({"error": "model cannot be empty"}, status_code=400)
+        updates["model"] = model_val
+    if "provider" in body:
+        prov = str(body["provider"]).strip()
+        if prov not in _MODEL_PROVIDERS:
+            return JSONResponse(
+                {"error": f"Unknown provider: {prov!r}"},
+                status_code=400,
+            )
+        updates["provider"] = prov
+    if "base_url" in body:
+        updates["base_url"] = str(body["base_url"]).strip()
+    if "api_key" in body:
+        api_key = str(body["api_key"]).strip()
+        # Sentinel "***" or empty string means "keep existing"
+        if api_key and api_key != "***":
+            updates["api_key"] = api_key
+    if "context_window" in body:
+        ctx_raw = body["context_window"]
+        updates["context_window"] = max(0, int(ctx_raw)) if isinstance(ctx_raw, (int, float)) else 0
+    if "capabilities" in body:
+        caps = body["capabilities"]
+        updates["capabilities"] = json.dumps(caps) if isinstance(caps, dict) else "{}"
+    if "enabled" in body:
+        updates["enabled"] = bool(body["enabled"])
+
+    if updates:
+        storage.update_model_definition(definition_id, **updates)
+
+    audit_uid, ip = _audit_context(request)
+    audit_detail = dict(updates)
+    if "api_key" in audit_detail:
+        audit_detail["api_key"] = "(updated)"
+    record_audit(
+        storage,
+        audit_uid,
+        "model_definition.update",
+        "model_definition",
+        definition_id,
+        audit_detail,
+        ip,
+    )
+
+    model_def = storage.get_model_definition(definition_id)
+    return JSONResponse(_mask_model_secrets(model_def or {}))
+
+
+async def admin_delete_model_definition(request: Request) -> JSONResponse:
+    """DELETE /v1/api/admin/model-definitions/{definition_id}."""
+    from turnstone.core.audit import record_audit
+    from turnstone.core.auth import require_permission
+    from turnstone.core.web_helpers import require_storage_or_503
+
+    storage, err = require_storage_or_503(request)
+    if err:
+        return err
+    err = require_permission(request, "admin.models")
+    if err:
+        return err
+
+    definition_id = request.path_params["definition_id"]
+    existing = storage.get_model_definition(definition_id)
+    if existing is None:
+        return JSONResponse({"error": "Model definition not found"}, status_code=404)
+
+    storage.delete_model_definition(definition_id)
+
+    audit_uid, ip = _audit_context(request)
+    record_audit(
+        storage,
+        audit_uid,
+        "model_definition.delete",
+        "model_definition",
+        definition_id,
+        {"alias": existing.get("alias", "")},
+        ip,
+    )
+
+    return JSONResponse({"status": "ok", "definition_id": definition_id})
+
+
+async def admin_model_reload(request: Request) -> JSONResponse:
+    """POST /v1/api/admin/model-definitions/reload — tell nodes to re-read DB."""
+    from turnstone.core.auth import require_permission
+    from turnstone.core.web_helpers import require_storage_or_503
+
+    storage, err = require_storage_or_503(request)
+    if err:
+        return err
+    err = require_permission(request, "admin.models")
+    if err:
+        return err
+
+    results = await _notify_nodes_model_reload(request)
+    return JSONResponse({"status": "ok", "results": results})
+
+
+async def admin_detect_model(request: Request) -> JSONResponse:
+    """POST /v1/api/admin/model-definitions/detect — stateless endpoint probe."""
+    import asyncio
+
+    from turnstone.core.auth import require_permission
+    from turnstone.core.model_registry import probe_model_endpoint
+    from turnstone.core.web_helpers import read_json_or_400, require_storage_or_503
+
+    storage, err = require_storage_or_503(request)
+    if err:
+        return err
+    err = require_permission(request, "admin.models")
+    if err:
+        return err
+
+    body = await read_json_or_400(request)
+    if isinstance(body, JSONResponse):
+        return body
+
+    provider = str(body.get("provider", "openai")).strip()
+    base_url = str(body.get("base_url", "")).strip()
+    api_key = str(body.get("api_key", "")).strip()
+    model = str(body.get("model", "")).strip()
+    definition_id = str(body.get("definition_id", "")).strip()
+
+    if provider not in _MODEL_PROVIDERS:
+        return JSONResponse({"error": f"Unknown provider: {provider!r}"}, status_code=400)
+
+    # Resolve api_key from DB when the UI sends the masked sentinel
+    if (not api_key or api_key == "***") and definition_id:
+        row = storage.get_model_definition(definition_id)
+        if row:
+            api_key = row.get("api_key", "")
+            if not base_url:
+                base_url = row.get("base_url", "")
+
+    # For commercial endpoints an api_key is required
+    if not api_key and (
+        not base_url or "api.openai.com" in base_url or "api.anthropic.com" in base_url
+    ):
+        return JSONResponse({"error": "api_key is required"}, status_code=400)
+
+    loop = asyncio.get_running_loop()
+    result = await loop.run_in_executor(
+        None, probe_model_endpoint, provider, base_url, api_key, model
+    )
+    return JSONResponse(result)
+
+
+async def admin_model_capabilities(request: Request) -> JSONResponse:
+    """GET /v1/api/admin/model-capabilities — static capability lookup."""
+    from turnstone.core.auth import require_permission
+    from turnstone.core.providers import lookup_model_capabilities
+
+    err = require_permission(request, "admin.models")
+    if err:
+        return err
+
+    provider = request.query_params.get("provider", "").strip()
+    model = request.query_params.get("model", "").strip()
+
+    if provider not in _MODEL_PROVIDERS:
+        return JSONResponse({"error": f"Unknown provider: {provider!r}"}, status_code=400)
+    if not model:
+        return JSONResponse({"error": "model is required"}, status_code=400)
+
+    caps = lookup_model_capabilities(provider, model)
+    return JSONResponse(
+        {
+            "model": model,
+            "provider": provider,
+            "known": caps is not None,
+            "capabilities": caps or {},
+        }
+    )
+
+
+async def admin_known_models(request: Request) -> JSONResponse:
+    """GET /v1/api/admin/model-capabilities/known — list known model name prefixes."""
+    from turnstone.core.auth import require_permission
+    from turnstone.core.providers import list_known_models
+
+    err = require_permission(request, "admin.models")
+    if err:
+        return err
+
+    provider = request.query_params.get("provider", "").strip()
+    if provider not in _MODEL_PROVIDERS:
+        return JSONResponse({"error": f"Unknown provider: {provider!r}"}, status_code=400)
+
+    return JSONResponse({"provider": provider, "models": list_known_models(provider)})
+
+
 # ---------------------------------------------------------------------------
 # TLS endpoints
 # ---------------------------------------------------------------------------
@@ -4854,6 +5365,7 @@ def create_app(
                     Route("/api/cluster/node/{node_id}", cluster_node_detail),
                     Route("/api/cluster/snapshot", cluster_snapshot),
                     Route("/api/cluster/events", cluster_events_sse),
+                    Route("/api/models", list_available_models),
                     Route("/api/skills", list_skills_summary),
                     Route("/api/auth/login", auth_login, methods=["POST"]),
                     Route("/api/auth/logout", auth_logout, methods=["POST"]),
@@ -5046,6 +5558,42 @@ def create_app(
                         admin_delete_mcp_server,
                         methods=["DELETE"],
                     ),
+                    # System: Model Definitions
+                    Route("/api/admin/model-definitions", admin_list_model_definitions),
+                    Route(
+                        "/api/admin/model-definitions",
+                        admin_create_model_definition,
+                        methods=["POST"],
+                    ),
+                    Route(
+                        "/api/admin/model-definitions/reload",
+                        admin_model_reload,
+                        methods=["POST"],
+                    ),
+                    Route(
+                        "/api/admin/model-definitions/detect",
+                        admin_detect_model,
+                        methods=["POST"],
+                    ),
+                    Route(
+                        "/api/admin/model-definitions/{definition_id}",
+                        admin_get_model_definition,
+                    ),
+                    Route(
+                        "/api/admin/model-definitions/{definition_id}",
+                        admin_update_model_definition,
+                        methods=["PUT"],
+                    ),
+                    Route(
+                        "/api/admin/model-definitions/{definition_id}",
+                        admin_delete_model_definition,
+                        methods=["DELETE"],
+                    ),
+                    Route("/api/admin/model-capabilities", admin_model_capabilities),
+                    Route(
+                        "/api/admin/model-capabilities/known",
+                        admin_known_models,
+                    ),
                     # Governance: Usage & Audit
                     Route("/api/admin/usage", admin_usage),
                     Route("/api/admin/audit", admin_audit),
@@ -5321,8 +5869,31 @@ def main() -> None:
                 log.info("TLS enabled")
         except ImportError:
             log.warning("TLS enabled but lacme not installed — pip install turnstone[tls]")
+            tls_mgr = None
         except Exception:
             log.warning("TLS initialization failed", exc_info=True)
+            tls_mgr = None
+
+        # Sync TLS state to ConfigStore so server nodes see the correct value.
+        # Three cases:
+        # 1. TLS succeeded → write true
+        # 2. TLS not configured (DB false/unset) → write false (definitive)
+        # 3. TLS configured (DB true) but init failed → don't overwrite
+        #    (transient failure shouldn't permanently disable TLS)
+        try:
+            db_enabled = _cs.get("tls.enabled")
+            if tls_mgr is not None:
+                if not db_enabled:
+                    _cs.set("tls.enabled", True, changed_by="console-startup")
+            elif db_enabled:
+                log.warning(
+                    "tls.enabled is true in ConfigStore but TLS init failed — "
+                    "server nodes will attempt TLS and fall back to plain HTTP"
+                )
+            else:
+                _cs.set("tls.enabled", False, changed_by="console-startup")
+        except Exception:
+            log.debug("Failed to sync TLS state to ConfigStore", exc_info=True)
 
     app = create_app(
         collector=collector,

turnstone/console/static/admin.js

@@ -66,6 +66,7 @@ function showAdmin() {
     settings: "admin.settings",
     tls: "admin.settings",
     mcp: "admin.mcp",
+    models: "admin.models",
   };
   if (perms) {
     var permSet = perms.split(",");
@@ -193,6 +194,7 @@ function switchAdminTab(tab) {
     "usage",
     "audit",
     "memories",
+    "models",
     "settings",
     "tls",
     "mcp",
@@ -216,6 +218,7 @@ function switchAdminTab(tab) {
     loadGovAudit();
   }
   if (tab === "memories") loadAdminMemories();
+  if (tab === "models") loadAdminModels();
   if (tab === "settings") loadSettings();
   if (tab === "tls") loadTlsCerts();
   if (tab === "mcp") loadAdminMcp();
@@ -1844,6 +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();
       }
     };
   }
@@ -1932,6 +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]);
@@ -4025,3 +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/index.html

@@ -109,6 +109,7 @@
         </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>
@@ -394,6 +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">
@@ -523,7 +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>
@@ -1189,6 +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

@@ -7,6 +7,8 @@
    Header overrides — wider padding for console layout
    ========================================================================== */
 #header { padding: 10px 20px; gap: 16px; }
+#status-bar { font-size: 11px; color: var(--fg-dim); margin-left: auto; }
+#status-bar.disconnected { color: var(--red); }
 .header-dim {
   color: var(--fg-dim);
   font-weight: 400;
@@ -1189,7 +1191,8 @@
 .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; }
@@ -1369,6 +1372,7 @@ 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;
@@ -1394,7 +1398,8 @@ 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);
@@ -2116,7 +2121,7 @@ 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 -------------------------------------------------- */
@@ -2183,7 +2188,8 @@ 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%}
@@ -2287,6 +2293,31 @@ 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
    ========================================================================== */
@@ -2310,5 +2341,5 @@ 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

@@ -80,7 +80,12 @@ class TLSManager:
 
             setup_metrics(self._event_dispatcher)
         except ImportError:
-            pass  # prometheus_client not installed
+            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."""
@@ -322,7 +327,7 @@ class TLSManager:
         _require_lacme()
         from lacme.mtls import server_ssl_context
 
-        return server_ssl_context(  # type: ignore[no-any-return]
+        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(),
@@ -339,7 +344,7 @@ class TLSManager:
         _require_lacme()
         from lacme.mtls import client_ssl_context
 
-        return client_ssl_context(  # type: ignore[no-any-return]
+        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(),

turnstone/core/auth.py

@@ -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/"
 

turnstone/core/healthcheck.py

@@ -5,11 +5,13 @@ from __future__ import annotations
 import enum
 import threading
 import time
-from typing import TYPE_CHECKING
+from typing import TYPE_CHECKING, Any
 
 from turnstone.core.log import get_logger
 
 if TYPE_CHECKING:
+    from collections.abc import Callable
+
     from openai import OpenAI
 
 log = get_logger(__name__)
@@ -37,6 +39,10 @@ class BackendHealthMonitor:
         probe_timeout: float = 5.0,
         failure_threshold: int = 5,
         cooldown: float = 60.0,
+        *,
+        provider: str = "openai",
+        initial_model: str = "",
+        on_model_changed: Callable[[str, int | None], None] | None = None,
     ) -> None:
         self._client = client
         self._probe_interval = probe_interval
@@ -44,6 +50,11 @@ class BackendHealthMonitor:
         self._failure_threshold = failure_threshold
         self._cooldown = cooldown
 
+        # Model change detection
+        self._provider = provider
+        self._last_detected_model = initial_model
+        self._on_model_changed = on_model_changed
+
         self._lock = threading.Lock()
         self._state = CircuitState.CLOSED
         self._consecutive_failures = 0
@@ -197,11 +208,39 @@ class BackendHealthMonitor:
     def _probe_once(self) -> bool:
         """Single probe: call ``client.models.list()``.  Returns True on success."""
         try:
-            self._client.with_options(timeout=self._probe_timeout).models.list()
+            resp = self._client.with_options(timeout=self._probe_timeout).models.list()
+            self._check_model_change(resp)
             return True
         except Exception:
             return False
 
+    def _check_model_change(self, resp: Any) -> None:
+        """Compare detected model against last known and fire callback if changed."""
+        if not self._on_model_changed or not resp.data:
+            return
+        try:
+            from turnstone.core.model_registry import (
+                _extract_context_window,
+                _select_best_model,
+            )
+
+            all_ids = [m.id for m in resp.data]
+            selected = _select_best_model(all_ids, self._provider)
+            if selected == self._last_detected_model:
+                return
+            model_obj = next((m for m in resp.data if m.id == selected), None)
+            ctx = _extract_context_window(model_obj, self._provider) if model_obj else None
+            log.info(
+                "Backend model changed: %s -> %s (ctx=%s)",
+                self._last_detected_model,
+                selected,
+                ctx,
+            )
+            self._last_detected_model = selected
+            self._on_model_changed(selected, ctx)
+        except Exception:
+            log.debug("Model change check failed", exc_info=True)
+
     # ------------------------------------------------------------------
     # Metrics
     # ------------------------------------------------------------------

turnstone/core/judge.py

@@ -189,6 +189,20 @@ _CRITICAL_RULES: list[_HeuristicRule] = [
             "This is a two-step variant of pipe-to-shell."
         ),
     ),
+    _HeuristicRule(
+        name="proc-environ-exfil",
+        risk_level="critical",
+        confidence=0.95,
+        recommendation="deny",
+        tool_pattern="bash",
+        arg_patterns=[r"/proc/\d+/environ", r"/proc/self/environ"],
+        intent_template="Process environment exfiltration: {arg_snippet}",
+        reasoning_template=(
+            "Reading /proc/*/environ exposes all environment variables of the "
+            "target process, which may include database credentials, API keys, "
+            "and JWT secrets. This is a credential exfiltration vector."
+        ),
+    ),
 ]
 
 # -- High (confidence 0.80, review) ----------------------------------------
@@ -606,7 +620,10 @@ def _get_arg_text(func_name: str, func_args: dict[str, object]) -> str:
     if func_name == "bash":
         return str(func_args.get("command", ""))
     if func_name in ("write_file", "edit_file"):
-        return str(func_args.get("path", ""))
+        path = str(func_args.get("path", ""))
+        expanded = os.path.expanduser(path) if path else ""
+        resolved = os.path.realpath(expanded) if expanded else ""
+        return f"{path} {resolved}" if resolved != os.path.abspath(expanded) else path
     try:
         return json.dumps(func_args, ensure_ascii=False, separators=(",", ":"))
     except (TypeError, ValueError):
@@ -853,6 +870,10 @@ 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 +933,12 @@ 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 +951,10 @@ 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 +977,7 @@ 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 +991,44 @@ 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()
@@ -1010,16 +1048,20 @@ class IntentJudge:
         # Prepare context
         judge_messages = self._prepare_context(item, messages)
 
-        # Prepare tools (only if read_only_tools enabled)
+        # Prepare tools (only if read_only_tools enabled).
+        # Pass raw OpenAI-format schemas — create_completion handles conversion.
         tools: list[dict[str, Any]] | None = None
         if self._config.read_only_tools:
-            tools = self._provider.convert_tools(_JUDGE_TOOL_SCHEMAS)
+            tools = _JUDGE_TOOL_SCHEMAS
 
         # Multi-turn judge loop
         timeout_budget = self._config.timeout
         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 +1085,7 @@ 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 +1095,24 @@ 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
@@ -1181,10 +1230,20 @@ class IntentJudge:
             f"{json.dumps(func_args, indent=2, ensure_ascii=False)}\n```"
         )
 
-        # FIFO truncation of conversation history (keep most recent)
+        # Trim to messages from the last user message onward — the judge
+        # only needs the immediate request context, not the full history.
+        # This keeps latency bounded as conversations grow.
+        last_user_idx = None
+        for i in range(len(messages) - 1, -1, -1):
+            if messages[i].get("role") == "user":
+                last_user_idx = i
+                break
+        recent = messages[last_user_idx:] if last_user_idx is not None else messages
+
+        # Apply FIFO budget cap on the trimmed context
         truncated: list[dict[str, Any]] = []
         total_chars = 0
-        for msg in reversed(messages):
+        for msg in reversed(recent):
             content = msg.get("content", "") or ""
             if isinstance(content, list):
                 content = " ".join(p.get("text", "") for p in content if isinstance(p, dict))

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 ----------------------------------------------------
 
 
@@ -252,10 +267,10 @@ def update_workstream_title(ws_id: str, title: str) -> None:
 # -- Conversation search -------------------------------------------------------
 
 
-def search_history(query: str, limit: int = 20) -> list[Any]:
+def search_history(query: str, limit: int = 20, offset: int = 0) -> list[Any]:
     """Search conversation history."""
     try:
-        return get_storage().search_history(query, limit)
+        return get_storage().search_history(query, limit, offset)
     except Exception:
         log.warning("Failed to search history", exc_info=True)
         return []
@@ -316,6 +331,18 @@ def save_structured_memory(
         return "", None
 
 
+def get_structured_memory_by_name(
+    name: str, scope: str = "global", scope_id: str = ""
+) -> dict[str, str] | None:
+    """Retrieve a single structured memory by name+scope. Returns full content."""
+    name = normalize_key(name)
+    try:
+        return get_storage().get_structured_memory_by_name(name, scope, scope_id)
+    except Exception:
+        log.warning("Failed to get structured memory name=%s", name, exc_info=True)
+        return None
+
+
 def delete_structured_memory(name: str, scope: str = "global", scope_id: str = "") -> bool:
     """Delete a structured memory by name+scope. Returns True if existed."""
     name = normalize_key(name)

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")
@@ -274,6 +366,28 @@ def _select_best_model(model_ids: list[str], provider: str) -> str:
     return model_ids[0]
 
 
+def _extract_context_window(model_obj: Any, provider: str) -> int | None:
+    """Extract context window from a model object returned by ``/v1/models``.
+
+    Handles Anthropic (static capability table), vLLM (``max_model_len``),
+    and llama.cpp (``meta.n_ctx_train``).  Returns ``None`` when not available.
+    """
+    if provider == "anthropic":
+        from turnstone.core.providers._anthropic import AnthropicProvider
+
+        return AnthropicProvider().get_capabilities(model_obj.id).context_window
+    model_data = model_obj.model_dump()
+    max_len = model_data.get("max_model_len")
+    if isinstance(max_len, int) and max_len > 0:
+        return max_len
+    meta = model_data.get("meta")
+    if isinstance(meta, dict):
+        n_ctx = meta.get("n_ctx_train")
+        if isinstance(n_ctx, int) and n_ctx > 0:
+            return n_ctx
+    return None
+
+
 def detect_model(
     client: Any,
     log_fn: Any = print,
@@ -319,18 +433,7 @@ def detect_model(
             log_fn(f"Available models: {', '.join(all_ids)}")
             log_fn(f"Using: {m.id} (override with --model)")
 
-        ctx: int | None = None
-        if provider == "anthropic":
-            from turnstone.core.providers._anthropic import AnthropicProvider
-
-            ctx = AnthropicProvider().get_capabilities(m.id).context_window
-        else:
-            # OpenAI-compatible: extract from backend metadata (llama.cpp, vLLM)
-            meta = m.model_dump().get("meta")
-            if isinstance(meta, dict):
-                n_ctx = meta.get("n_ctx_train")
-                if isinstance(n_ctx, int) and n_ctx > 0:
-                    ctx = n_ctx
+        ctx = _extract_context_window(m, provider)
         return m.id, ctx
     except SystemExit:
         raise
@@ -342,3 +445,114 @@ 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 = ""
+    dumped: dict[str, Any] = {}
+    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.
+    if model_obj is not None:
+        ctx = _extract_context_window(model_obj, "openai")
+        if ctx is not None:
+            result["context_window"] = 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

@@ -51,11 +51,19 @@ _RE_PRIVATE_KEY_BLOCK = re.compile(
     r"-----END\s+(?:RSA\s+|EC\s+|OPENSSH\s+|PGP\s+)?PRIVATE\s+KEY-----",
 )
 _RE_CONNECTION_STRING = re.compile(
-    r"(?:postgresql|mysql|mongodb|redis|amqp)://[^:@\s]+:[^@\s]+@",
+    r"(?:postgresql\+?(?:psycopg)?|mysql|mongodb|redis|amqp|sqlite)://[^:@\s]+:[^@\s]+@",
 )
 _RE_ENV_SECRET_LINE = re.compile(r"[A-Z][A-Z_0-9]+=\S+")
 _RE_ENV_SECRET_KEY = re.compile(
-    r"(?:^|_)(?:SECRET|TOKEN|PASSWORD|CREDENTIAL)(?:_|$)|(?:^|_)KEY(?:_|$)",
+    r"(?:^|_)(?:SECRET|TOKEN|PASSWORD|CREDENTIAL|DSN)(?:_|$)"
+    r"|(?:^|_)KEY(?:_|$)"
+    r"|^(?:DATABASE_URL|TURNSTONE_DB_URL|DB_URL)$",
+    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,
 )
 
@@ -228,6 +236,15 @@ 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 +264,15 @@ 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."""
@@ -80,7 +84,7 @@ _ANTHROPIC_DEFAULT = ModelCapabilities(
 
 _ANTHROPIC_CAPABILITIES: dict[str, ModelCapabilities] = {
     "claude-opus-4-6": ModelCapabilities(
-        context_window=200000,
+        context_window=1000000,
         max_output_tokens=128000,
         token_param="max_tokens",
         thinking_mode="adaptive",
@@ -91,7 +95,7 @@ _ANTHROPIC_CAPABILITIES: dict[str, ModelCapabilities] = {
         supports_vision=True,
     ),
     "claude-sonnet-4-6": ModelCapabilities(
-        context_window=200000,
+        context_window=1000000,
         max_output_tokens=64000,
         token_param="max_tokens",
         thinking_mode="adaptive",
@@ -127,24 +131,6 @@ _ANTHROPIC_CAPABILITIES: dict[str, ModelCapabilities] = {
         supports_web_search=True,
         supports_vision=True,
     ),
-    "claude-opus-4": ModelCapabilities(
-        context_window=200000,
-        max_output_tokens=32000,
-        token_param="max_tokens",
-        thinking_mode="manual",
-        supports_web_search=True,
-        supports_tool_search=True,
-        supports_vision=True,
-    ),
-    "claude-sonnet-4": ModelCapabilities(
-        context_window=200000,
-        max_output_tokens=64000,
-        token_param="max_tokens",
-        thinking_mode="manual",
-        supports_web_search=True,
-        supports_tool_search=True,
-        supports_vision=True,
-    ),
 }
 
 
@@ -286,6 +272,7 @@ 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 +286,50 @@ 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
 
@@ -328,27 +354,105 @@ 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
 
             if role == "tool":
-                # Anthropic: tool results are content blocks in a user message
+                # Anthropic: tool results are content blocks in a user message.
+                # Collect valid tool_use IDs from the preceding assistant message
+                # so we can drop orphaned tool_results that have no matching
+                # tool_use (e.g. from compaction boundary, old cancel stripping).
+                prev_tool_use_ids: set[str] = set()
+                if converted and converted[-1].get("role") == "assistant":
+                    prev_content = converted[-1].get("content", [])
+                    if isinstance(prev_content, list):
+                        for block in prev_content:
+                            if isinstance(block, dict) and block.get("type") == "tool_use":
+                                bid = block.get("id", "")
+                                if bid:
+                                    prev_tool_use_ids.add(bid)
+
                 tool_results: list[dict[str, Any]] = []
                 while i < len(messages) and messages[i]["role"] == "tool":
                     tool_msg = messages[i]
+                    tc_id = tool_msg.get("tool_call_id", "")
+                    # Drop orphaned tool_results with no matching tool_use.
+                    # When prev_tool_use_ids is empty (no preceding assistant
+                    # tool_use), let all results through — avoids false drops
+                    # from unexpected message ordering.
+                    if prev_tool_use_ids and tc_id not in prev_tool_use_ids:
+                        log.debug(
+                            "Dropping orphaned tool_result (no matching tool_use): %s",
+                            tc_id,
+                        )
+                        i += 1
+                        continue
                     content = tool_msg.get("content", "")
                     # 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": tc_id,
+                        "content": content,
+                    }
+                    if tool_msg.get("is_error"):
+                        result_block["is_error"] = True
+                    tool_results.append(result_block)
                     i += 1
-                converted.append({"role": "user", "content": tool_results})
+                # Append any deferred synthetic results after real ones
+                if pending_orphan_results:
+                    tool_results.extend(pending_orphan_results)
+                    pending_orphan_results = []
+                if tool_results:
+                    converted.append({"role": "user", "content": tool_results})
                 continue
 
             if role == "user":
@@ -459,6 +563,7 @@ 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 +581,25 @@ 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 +663,12 @@ 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 +775,24 @@ 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(

turnstone/core/safety.py

@@ -11,12 +11,30 @@ BLOCKED_PATTERNS = [
     "reboot",
     "halt",
     "poweroff",
-    "dd if=",
+    "of=/dev/sd",
+    "of=/dev/nvme",
+    "of=/dev/vd",
+    "of=/dev/xvd",
+    "of=/dev/hd",
+    "of=/dev/dm-",
+    "of=/dev/md",
+    "of=/dev/loop",
+    "of=/dev/disk/",
     ":(){ :|:& };:",  # fork bomb
-    "> /dev/sda",
+    "> /dev/sd",
+    "> /dev/nvme",
+    "> /dev/vd",
+    "> /dev/xvd",
+    "> /dev/hd",
+    "> /dev/dm-",
+    "> /dev/md",
+    "> /dev/disk/",
     "mv / ",
     "chmod -R 777 /",
     "chown -R ",
+    # Credential exfiltration via procfs
+    "/proc/1/environ",
+    "/proc/self/environ",
 ]
 
 

turnstone/core/sandbox.py

@@ -19,6 +19,15 @@ _MATH_BLOCKED_BUILTINS = {
     "globals",
     "locals",
     "vars",
+    # Reflection primitives — bypass AST dunder checks via runtime strings
+    "getattr",
+    "setattr",
+    "delattr",
+    # Type system — can reconstruct arbitrary classes
+    "type",
+    # Import — the replaced _safe_import is in the namespace, but block the
+    # name so direct __import__ calls are caught by the AST validator
+    "__import__",
 }
 
 _MATH_BLOCKED_MODULES = {
@@ -85,6 +94,9 @@ class _ASTValidator(ast.NodeVisitor):
             and node.attr not in {"__name__", "__doc__", "__class__"}
         ):
             self.errors.append(f"Access to '{node.attr}' is not allowed")
+        # Block operator.attrgetter/itemgetter which act as getattr bypasses
+        if node.attr in ("attrgetter", "itemgetter"):
+            self.errors.append(f"Access to '{node.attr}' is not allowed")
         self.generic_visit(node)
 
 
@@ -109,6 +121,7 @@ def validate_math_code(code: str) -> list[str]:
 
 def _math_exec_in_process(code: str, result_queue: multiprocessing.Queue[tuple[str, str]]) -> None:
     """Execute code in a subprocess, put (status, output) in queue."""
+    import contextlib
     import signal as _signal
     import sys as _sys
     from io import StringIO
@@ -124,7 +137,14 @@ def _math_exec_in_process(code: str, result_queue: multiprocessing.Queue[tuple[s
         def _safe_import(name: str, *args: Any, **kwargs: Any) -> Any:
             if name.split(".")[0] in _MATH_BLOCKED_MODULES:
                 raise ImportError(f"Import of '{name}' is blocked")
-            return original_import(name, *args, **kwargs)
+            mod = original_import(name, *args, **kwargs)
+            # Strip __builtins__ from every imported module so
+            # module.__builtins__['__import__'] can't bypass _safe_import
+            # (covers operator.attrgetter('__builtins__') and similar).
+            if hasattr(mod, "__builtins__"):
+                with contextlib.suppress(AttributeError, TypeError):
+                    mod.__builtins__ = {}  # type: ignore[attr-defined]
+            return mod
 
         original_import = (
             __builtins__["__import__"]
@@ -242,7 +262,14 @@ def _math_exec_in_process(code: str, result_queue: multiprocessing.Queue[tuple[s
         except ImportError:
             pass
 
-        exec(code, ns)
+        # Strip __builtins__ from all pre-imported modules so
+        # module.__builtins__['__import__'] can't bypass _safe_import.
+        for v in list(ns.values()):
+            if hasattr(v, "__builtins__"):
+                with contextlib.suppress(AttributeError, TypeError):
+                    v.__builtins__ = {}
+
+        exec(code, ns)  # noqa: S102
 
         _sys.stdout = _sys.__stdout__
         printed = captured.getvalue()

turnstone/core/storage/_postgresql.py

@@ -17,6 +17,7 @@ from turnstone.core.storage._schema import (
     intent_verdicts,
     mcp_servers,
     metadata,
+    model_definitions,
     oidc_identities,
     oidc_pending_states,
     orgs,
@@ -44,6 +45,9 @@ from turnstone.core.storage._schema import (
 from turnstone.core.storage._utils import (
     MCP_SERVER_MUTABLE as _MCP_SERVER_MUTABLE,
 )
+from turnstone.core.storage._utils import (
+    MODEL_DEFINITION_MUTABLE as _MODEL_DEF_MUTABLE,
+)
 from turnstone.core.storage._utils import (
     ORG_MUTABLE as _ORG_MUTABLE,
 )
@@ -68,6 +72,7 @@ from turnstone.core.storage._utils import (
 from turnstone.core.storage._utils import (
     row_to_dict as _row_to_dict,
 )
+from turnstone.core.storage._utils import sanitize_text
 from turnstone.core.storage._utils import (
     scan_skill_content as _scan_skill_content,
 )
@@ -103,12 +108,13 @@ class PostgreSQLBackend:
         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,
     ) -> None:
         now = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%S")
+        content = sanitize_text(content)
+        provider_data = sanitize_text(provider_data)
         with self._engine.connect() as conn:
             conn.execute(
                 sa.insert(conversations),
@@ -118,7 +124,6 @@ class PostgreSQLBackend:
                     "role": role,
                     "content": content,
                     "tool_name": tool_name,
-                    "tool_args": tool_args,
                     "tool_call_id": tool_call_id,
                     "provider_data": provider_data,
                     "tool_calls": tool_calls,
@@ -136,7 +141,6 @@ class PostgreSQLBackend:
                     conversations.c.role,
                     conversations.c.content,
                     conversations.c.tool_name,
-                    conversations.c.tool_args,
                     conversations.c.tool_call_id,
                     conversations.c.provider_data,
                     conversations.c.tool_calls,
@@ -146,6 +150,29 @@ class PostgreSQLBackend:
             ).fetchall()
         return _reconstruct_messages(list(rows), ws_id)
 
+    def delete_messages_after(self, ws_id: str, keep_count: int) -> int:
+        with self._engine.connect() as conn:
+            cutoff_row = conn.execute(
+                sa.select(conversations.c.id)
+                .where(conversations.c.ws_id == ws_id)
+                .order_by(conversations.c.id)
+                .limit(1)
+                .offset(keep_count)
+            ).fetchone()
+            if cutoff_row is None:
+                return 0
+            cutoff_id = cutoff_row[0]
+            result = conn.execute(
+                sa.delete(conversations).where(
+                    sa.and_(
+                        conversations.c.ws_id == ws_id,
+                        conversations.c.id >= cutoff_id,
+                    )
+                )
+            )
+            conn.commit()
+            return result.rowcount
+
     # -- Workstream management -------------------------------------------------
 
     def list_workstreams_with_history(self, limit: int = 20) -> list[Any]:
@@ -382,10 +409,11 @@ class PostgreSQLBackend:
 
     # -- Conversation search ---------------------------------------------------
 
-    def search_history(self, query: str, limit: int = 20) -> list[Any]:
+    def search_history(self, query: str, limit: int = 20, offset: int = 0) -> list[Any]:
         if not query or not query.strip():
             return []
-        capped = min(limit, 100)
+        capped = min(int(limit), 100)
+        capped_offset = max(0, int(offset))
         with self._engine.connect() as conn:
             # Use PostgreSQL full-text search if search_vector column exists
             try:
@@ -398,9 +426,9 @@ class PostgreSQLBackend:
                             "   @@ plainto_tsquery('english', :query) "
                             "ORDER BY ts_rank(to_tsvector('english', COALESCE(c.content, '')), "
                             "   plainto_tsquery('english', :query)) DESC "
-                            "LIMIT :limit"
+                            "LIMIT :limit OFFSET :offset"
                         ),
-                        {"query": query, "limit": capped},
+                        {"query": query, "limit": capped, "offset": capped_offset},
                     ).fetchall()
                 )
             except Exception:
@@ -410,9 +438,9 @@ class PostgreSQLBackend:
                         sa.text(
                             "SELECT timestamp, ws_id, role, content, tool_name "
                             "FROM conversations WHERE content ILIKE :pattern "
-                            "ORDER BY timestamp DESC LIMIT :limit"
+                            "ORDER BY timestamp DESC LIMIT :limit OFFSET :offset"
                         ),
-                        {"pattern": f"%{query}%", "limit": capped},
+                        {"pattern": f"%{query}%", "limit": capped, "offset": capped_offset},
                     ).fetchall()
                 )
 
@@ -2644,6 +2672,102 @@ class PostgreSQLBackend:
             conn.commit()
             return result.rowcount > 0
 
+    # -- Model definitions -----------------------------------------------------
+
+    def create_model_definition(
+        self,
+        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,
+        created_by: str = "",
+    ) -> None:
+        from sqlalchemy.dialects import postgresql
+
+        now = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%S")
+        with self._engine.connect() as conn:
+            conn.execute(
+                postgresql.insert(model_definitions)
+                .values(
+                    definition_id=definition_id,
+                    alias=alias,
+                    model=model,
+                    provider=provider,
+                    base_url=base_url,
+                    api_key=api_key,
+                    context_window=context_window,
+                    capabilities=capabilities,
+                    enabled=1 if enabled else 0,
+                    created_by=created_by,
+                    created=now,
+                    updated=now,
+                )
+                .on_conflict_do_nothing()
+            )
+            conn.commit()
+
+    def get_model_definition(self, definition_id: str) -> dict[str, Any] | None:
+
+        with self._engine.connect() as conn:
+            row = conn.execute(
+                sa.select(model_definitions).where(
+                    model_definitions.c.definition_id == definition_id
+                )
+            ).fetchone()
+            if row is None:
+                return None
+            return _row_to_dict(row, "enabled")
+
+    def get_model_definition_by_alias(self, alias: str) -> dict[str, Any] | None:
+
+        with self._engine.connect() as conn:
+            row = conn.execute(
+                sa.select(model_definitions).where(model_definitions.c.alias == alias)
+            ).fetchone()
+            if row is None:
+                return None
+            return _row_to_dict(row, "enabled")
+
+    def list_model_definitions(self, enabled_only: bool = False) -> list[dict[str, Any]]:
+
+        with self._engine.connect() as conn:
+            q = sa.select(model_definitions).order_by(model_definitions.c.alias)
+            if enabled_only:
+                q = q.where(model_definitions.c.enabled == 1)
+            rows = conn.execute(q).fetchall()
+            return [_row_to_dict(r, "enabled") for r in rows]
+
+    def update_model_definition(self, definition_id: str, **fields: Any) -> bool:
+
+        fields = {k: v for k, v in fields.items() if k in _MODEL_DEF_MUTABLE}
+        fields["updated"] = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%S")
+        if "enabled" in fields:
+            fields["enabled"] = 1 if fields["enabled"] else 0
+        with self._engine.connect() as conn:
+            result = conn.execute(
+                sa.update(model_definitions)
+                .where(model_definitions.c.definition_id == definition_id)
+                .values(**fields)
+            )
+            conn.commit()
+            return result.rowcount > 0
+
+    def delete_model_definition(self, definition_id: str) -> bool:
+
+        with self._engine.connect() as conn:
+            result = conn.execute(
+                sa.delete(model_definitions).where(
+                    model_definitions.c.definition_id == definition_id
+                )
+            )
+            conn.commit()
+            return result.rowcount > 0
+
     # -- OIDC identity ---------------------------------------------------------
 
     def create_oidc_identity(self, issuer: str, subject: str, user_id: str, email: str) -> None:

turnstone/core/storage/_protocol.py

@@ -21,7 +21,6 @@ class StorageBackend(Protocol):
         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,
@@ -33,6 +32,15 @@ class StorageBackend(Protocol):
         """Load messages for a workstream and reconstruct OpenAI message format."""
         ...
 
+    def delete_messages_after(self, ws_id: str, keep_count: int) -> int:
+        """Delete conversation rows beyond the first *keep_count* rows for a workstream.
+
+        Rows are ordered by auto-increment ``id``.  If the workstream has
+        N rows total and ``keep_count`` < N, the last N - keep_count rows
+        are deleted.  Returns the number of rows deleted.
+        """
+        ...
+
     # -- Workstream management -------------------------------------------------
 
     def list_workstreams_with_history(self, limit: int = 20) -> list[Any]:
@@ -181,7 +189,7 @@ class StorageBackend(Protocol):
 
     # -- Conversation search ---------------------------------------------------
 
-    def search_history(self, query: str, limit: int = 20) -> list[Any]:
+    def search_history(self, query: str, limit: int = 20, offset: int = 0) -> list[Any]:
         """Search conversation history. Returns (timestamp, ws_id, role, content, tool_name)."""
         ...
 
@@ -952,6 +960,44 @@ class StorageBackend(Protocol):
         """Delete an MCP server definition. Returns True if existed."""
         ...
 
+    # -- Model definitions -----------------------------------------------------
+
+    def create_model_definition(
+        self,
+        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,
+        created_by: str = "",
+    ) -> None:
+        """Create a model definition. No-op if definition_id already exists."""
+        ...
+
+    def get_model_definition(self, definition_id: str) -> dict[str, Any] | None:
+        """Return model definition dict or None."""
+        ...
+
+    def get_model_definition_by_alias(self, alias: str) -> dict[str, Any] | None:
+        """Return model definition dict by alias or None."""
+        ...
+
+    def list_model_definitions(self, enabled_only: bool = False) -> list[dict[str, Any]]:
+        """Return model definitions ordered by alias."""
+        ...
+
+    def update_model_definition(self, definition_id: str, **fields: Any) -> bool:
+        """Update specified fields on a model definition. Returns True if found."""
+        ...
+
+    def delete_model_definition(self, definition_id: str) -> bool:
+        """Delete a model definition. Returns True if existed."""
+        ...
+
     # -- TLS / ACME (lacme Store) ----------------------------------------------
 
     def save_tls_account_key(self, key_id: str, key_pem: str) -> None:

turnstone/core/storage/_schema.py

@@ -35,7 +35,6 @@ conversations = sa.Table(
     sa.Column("role", sa.Text, nullable=False),
     sa.Column("content", sa.Text),
     sa.Column("tool_name", sa.Text),
-    sa.Column("tool_args", sa.Text),
     sa.Column("tool_call_id", sa.Text),
     sa.Column("provider_data", sa.Text),
     sa.Column("tool_calls", sa.Text),
@@ -520,6 +519,29 @@ sa.Index(
     postgresql_where=mcp_servers.c.registry_name.isnot(None),
 )
 
+# ---------------------------------------------------------------------------
+# Model definitions — database-backed model configuration
+# ---------------------------------------------------------------------------
+
+model_definitions = sa.Table(
+    "model_definitions",
+    metadata,
+    sa.Column("definition_id", sa.Text, primary_key=True),
+    sa.Column("alias", sa.Text, nullable=False, unique=True),
+    sa.Column("model", sa.Text, nullable=False),
+    sa.Column("provider", sa.Text, nullable=False, server_default="openai"),
+    sa.Column("base_url", sa.Text, nullable=False, server_default=""),
+    sa.Column("api_key", sa.Text, nullable=False, server_default=""),
+    sa.Column("context_window", sa.Integer, nullable=False, server_default="32768"),
+    sa.Column("capabilities", sa.Text, nullable=False, server_default="{}"),
+    sa.Column("enabled", sa.Integer, nullable=False, server_default="1"),
+    sa.Column("created_by", sa.Text, nullable=False, server_default=""),
+    sa.Column("created", sa.Text, nullable=False),
+    sa.Column("updated", sa.Text, nullable=False),
+)
+
+sa.Index("idx_model_definitions_enabled", model_definitions.c.enabled)
+
 # ---------------------------------------------------------------------------
 # OIDC identity tables
 # ---------------------------------------------------------------------------

turnstone/core/storage/_sqlite.py

@@ -17,6 +17,7 @@ from turnstone.core.storage._schema import (
     intent_verdicts,
     mcp_servers,
     metadata,
+    model_definitions,
     oidc_identities,
     oidc_pending_states,
     orgs,
@@ -44,6 +45,9 @@ from turnstone.core.storage._schema import (
 from turnstone.core.storage._utils import (
     MCP_SERVER_MUTABLE as _MCP_SERVER_MUTABLE,
 )
+from turnstone.core.storage._utils import (
+    MODEL_DEFINITION_MUTABLE as _MODEL_DEF_MUTABLE,
+)
 from turnstone.core.storage._utils import (
     ORG_MUTABLE as _ORG_MUTABLE,
 )
@@ -68,6 +72,7 @@ from turnstone.core.storage._utils import (
 from turnstone.core.storage._utils import (
     row_to_dict as _row_to_dict,
 )
+from turnstone.core.storage._utils import sanitize_text
 from turnstone.core.storage._utils import (
     scan_skill_content as _scan_skill_content,
 )
@@ -154,12 +159,13 @@ class SQLiteBackend:
         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,
     ) -> None:
         now = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%S")
+        content = sanitize_text(content)
+        provider_data = sanitize_text(provider_data)
         with self._engine.connect() as conn:
             result = conn.execute(
                 sa.insert(conversations),
@@ -169,7 +175,6 @@ class SQLiteBackend:
                     "role": role,
                     "content": content,
                     "tool_name": tool_name,
-                    "tool_args": tool_args,
                     "tool_call_id": tool_call_id,
                     "provider_data": provider_data,
                     "tool_calls": tool_calls,
@@ -200,7 +205,6 @@ class SQLiteBackend:
                     conversations.c.role,
                     conversations.c.content,
                     conversations.c.tool_name,
-                    conversations.c.tool_args,
                     conversations.c.tool_call_id,
                     conversations.c.provider_data,
                     conversations.c.tool_calls,
@@ -211,6 +215,43 @@ class SQLiteBackend:
 
         return _reconstruct_messages(list(rows), ws_id)
 
+    def delete_messages_after(self, ws_id: str, keep_count: int) -> int:
+        with self._engine.connect() as conn:
+            # Find the id of the first row to delete (the row at offset keep_count)
+            cutoff_row = conn.execute(
+                sa.select(conversations.c.id)
+                .where(conversations.c.ws_id == ws_id)
+                .order_by(conversations.c.id)
+                .limit(1)
+                .offset(keep_count)
+            ).fetchone()
+            if cutoff_row is None:
+                return 0  # nothing to delete
+            cutoff_id = cutoff_row[0]
+            # Remove FTS5 entries first (external content table doesn't auto-sync)
+            if self._fts5_available:
+                try:
+                    conn.execute(
+                        sa.text(
+                            "DELETE FROM conversations_fts WHERE rowid IN "
+                            "(SELECT id FROM conversations "
+                            " WHERE ws_id = :ws_id AND id >= :cutoff_id)"
+                        ),
+                        {"ws_id": ws_id, "cutoff_id": cutoff_id},
+                    )
+                except Exception:
+                    self._fts5_available = False
+            result = conn.execute(
+                sa.delete(conversations).where(
+                    sa.and_(
+                        conversations.c.ws_id == ws_id,
+                        conversations.c.id >= cutoff_id,
+                    )
+                )
+            )
+            conn.commit()
+            return result.rowcount
+
     # -- Workstream management -------------------------------------------------
 
     def list_workstreams_with_history(self, limit: int = 20) -> list[Any]:
@@ -456,10 +497,11 @@ class SQLiteBackend:
 
     # -- Conversation search ---------------------------------------------------
 
-    def search_history(self, query: str, limit: int = 20) -> list[Any]:
+    def search_history(self, query: str, limit: int = 20, offset: int = 0) -> list[Any]:
         if not query or not query.strip():
             return []
-        capped = min(limit, 100)
+        capped = min(int(limit), 100)
+        capped_offset = max(0, int(offset))
         with self._engine.connect() as conn:
             if self._fts5_available:
                 return list(
@@ -469,9 +511,9 @@ class SQLiteBackend:
                             "FROM conversations_fts f "
                             "JOIN conversations c ON c.id = f.rowid "
                             "WHERE conversations_fts MATCH :query "
-                            "ORDER BY f.rank ASC LIMIT :limit"
+                            "ORDER BY f.rank ASC LIMIT :limit OFFSET :offset"
                         ),
-                        {"query": _fts5_query(query), "limit": capped},
+                        {"query": _fts5_query(query), "limit": capped, "offset": capped_offset},
                     ).fetchall()
                 )
             return list(
@@ -479,9 +521,13 @@ class SQLiteBackend:
                     sa.text(
                         "SELECT timestamp, ws_id, role, content, tool_name "
                         "FROM conversations WHERE content LIKE :pattern ESCAPE '\\' "
-                        "ORDER BY timestamp DESC LIMIT :limit"
+                        "ORDER BY timestamp DESC LIMIT :limit OFFSET :offset"
                     ),
-                    {"pattern": f"%{_escape_like(query)}%", "limit": capped},
+                    {
+                        "pattern": f"%{_escape_like(query)}%",
+                        "limit": capped,
+                        "offset": capped_offset,
+                    },
                 ).fetchall()
             )
 
@@ -2693,6 +2739,100 @@ class SQLiteBackend:
             conn.commit()
             return result.rowcount > 0
 
+    # -- Model definitions -----------------------------------------------------
+
+    def create_model_definition(
+        self,
+        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,
+        created_by: str = "",
+    ) -> None:
+
+        now = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%S")
+        with self._engine.connect() as conn:
+            conn.execute(
+                sa.insert(model_definitions).prefix_with("OR IGNORE"),
+                {
+                    "definition_id": definition_id,
+                    "alias": alias,
+                    "model": model,
+                    "provider": provider,
+                    "base_url": base_url,
+                    "api_key": api_key,
+                    "context_window": context_window,
+                    "capabilities": capabilities,
+                    "enabled": 1 if enabled else 0,
+                    "created_by": created_by,
+                    "created": now,
+                    "updated": now,
+                },
+            )
+            conn.commit()
+
+    def get_model_definition(self, definition_id: str) -> dict[str, Any] | None:
+
+        with self._engine.connect() as conn:
+            row = conn.execute(
+                sa.select(model_definitions).where(
+                    model_definitions.c.definition_id == definition_id
+                )
+            ).fetchone()
+            if row is None:
+                return None
+            return _row_to_dict(row, "enabled")
+
+    def get_model_definition_by_alias(self, alias: str) -> dict[str, Any] | None:
+
+        with self._engine.connect() as conn:
+            row = conn.execute(
+                sa.select(model_definitions).where(model_definitions.c.alias == alias)
+            ).fetchone()
+            if row is None:
+                return None
+            return _row_to_dict(row, "enabled")
+
+    def list_model_definitions(self, enabled_only: bool = False) -> list[dict[str, Any]]:
+
+        with self._engine.connect() as conn:
+            q = sa.select(model_definitions).order_by(model_definitions.c.alias)
+            if enabled_only:
+                q = q.where(model_definitions.c.enabled == 1)
+            rows = conn.execute(q).fetchall()
+            return [_row_to_dict(r, "enabled") for r in rows]
+
+    def update_model_definition(self, definition_id: str, **fields: Any) -> bool:
+
+        fields = {k: v for k, v in fields.items() if k in _MODEL_DEF_MUTABLE}
+        fields["updated"] = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%S")
+        if "enabled" in fields:
+            fields["enabled"] = 1 if fields["enabled"] else 0
+        with self._engine.connect() as conn:
+            result = conn.execute(
+                sa.update(model_definitions)
+                .where(model_definitions.c.definition_id == definition_id)
+                .values(**fields)
+            )
+            conn.commit()
+            return result.rowcount > 0
+
+    def delete_model_definition(self, definition_id: str) -> bool:
+
+        with self._engine.connect() as conn:
+            result = conn.execute(
+                sa.delete(model_definitions).where(
+                    model_definitions.c.definition_id == definition_id
+                )
+            )
+            conn.commit()
+            return result.rowcount > 0
+
     # -- OIDC identity ---------------------------------------------------------
 
     def create_oidc_identity(self, issuer: str, subject: str, user_id: str, email: str) -> None:

turnstone/core/storage/_utils.py

@@ -10,6 +10,22 @@ from turnstone.core.log import get_logger
 
 log = get_logger(__name__)
 
+# ---------------------------------------------------------------------------
+# Text sanitization
+# ---------------------------------------------------------------------------
+
+
+def sanitize_text(value: str | None) -> str | None:
+    """Strip NUL bytes that PostgreSQL text fields cannot store.
+
+    SQLite tolerates NUL in TEXT but they cause downstream issues (API
+    payloads, web UI rendering), so both backends use this.
+    """
+    if value and "\x00" in value:
+        return value.replace("\x00", "")
+    return value
+
+
 # ---------------------------------------------------------------------------
 # Row helper
 # ---------------------------------------------------------------------------
@@ -80,6 +96,18 @@ MCP_SERVER_MUTABLE = frozenset(
         "registry_meta",
     }
 )
+MODEL_DEFINITION_MUTABLE = frozenset(
+    {
+        "alias",
+        "model",
+        "provider",
+        "base_url",
+        "api_key",
+        "context_window",
+        "capabilities",
+        "enabled",
+    }
+)
 VERDICT_MUTABLE = frozenset(
     {
         "user_decision",
@@ -136,8 +164,8 @@ def scan_skill_content(content: str, allowed_tools: str) -> tuple[str, str, str]
 def reconstruct_messages(rows: list[Any], ws_id: str) -> list[dict[str, Any]]:
     """Reconstruct OpenAI message format from stored conversation rows.
 
-    Each *row* is a 7-element tuple of ``(role, content, tool_name,
-    tool_args, tool_call_id, provider_data, tool_calls_json)`` ordered
+    Each *row* is a 6-element tuple of ``(role, content, tool_name,
+    tool_call_id, provider_data, tool_calls_json)`` ordered
     chronologically by row ID.
 
     Post-migration 013 the only roles are ``user``, ``assistant``, and
@@ -146,13 +174,13 @@ def reconstruct_messages(rows: list[Any], ws_id: str) -> list[dict[str, Any]]:
     """
     messages: list[dict[str, Any]] = []
     for row in rows:
-        role, content, _tool_name, _tool_args, tc_id, provider_data, tool_calls_json = row
+        role, content, _tool_name, tc_id, provider_data, tool_calls_json = row
 
         if role == "user":
             messages.append({"role": "user", "content": content or ""})
 
         elif role == "assistant":
-            msg: dict[str, Any] = {"role": "assistant", "content": content}
+            msg: dict[str, Any] = {"role": "assistant", "content": content or ""}
             if provider_data:
                 with contextlib.suppress(json.JSONDecodeError, TypeError):
                     msg["_provider_content"] = json.loads(provider_data)
@@ -188,4 +216,45 @@ def reconstruct_messages(rows: list[Any], ws_id: str) -> list[dict[str, Any]]:
             break
         del messages[asst_idx:]
 
+    # Repair: synthesize tool results for mid-conversation orphaned tool calls.
+    # This happens when a cancel interrupts tool execution — the assistant
+    # message with tool_calls is saved to DB but GenerationCancelled prevents
+    # tool results from being created.  Both Anthropic (strict) and OpenAI
+    # (lenient today, may tighten) benefit from well-formed histories.
+    i = 0
+    while i < len(messages):
+        msg = messages[i]
+        if msg.get("role") == "assistant" and msg.get("tool_calls"):
+            expected_ids = [tc.get("id", "") for tc in msg["tool_calls"] if tc.get("id")]
+            # Collect tool result IDs that follow
+            j = i + 1
+            result_ids: set[str] = set()
+            while j < len(messages) and messages[j].get("role") == "tool":
+                tc_id = messages[j].get("tool_call_id", "")
+                if tc_id:
+                    result_ids.add(tc_id)
+                j += 1
+            # Synthesize results for any missing IDs
+            orphaned = [uid for uid in expected_ids if uid not in result_ids]
+            if orphaned:
+                synthetic = [
+                    {
+                        "role": "tool",
+                        "tool_call_id": uid,
+                        "content": "Tool execution was cancelled.",
+                        "is_error": True,
+                    }
+                    for uid in orphaned
+                ]
+                # Insert after the last existing tool result (or after assistant)
+                messages[j:j] = synthetic
+            if orphaned:
+                i = j + len(orphaned)  # skip past spliced synthetics
+            elif j > i + 1:
+                i = j  # skip past existing tool block
+            else:
+                i += 1  # no tools followed; just advance
+        else:
+            i += 1
+
     return messages

turnstone/core/storage/migrations/versions/027_drop_tool_args.py

@@ -0,0 +1,28 @@
+"""Drop vestigial tool_args column from conversations table.
+
+The tool_args column has not been written since migration 013 moved
+tool call data into the tool_calls JSON column on assistant rows.
+All existing rows have NULL in this column.
+
+Revision ID: 027
+Revises: 026
+Create Date: 2026-03-28
+"""
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = "027"
+down_revision = "026"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    with op.batch_alter_table("conversations") as batch_op:
+        batch_op.drop_column("tool_args")
+
+
+def downgrade() -> None:
+    with op.batch_alter_table("conversations") as batch_op:
+        batch_op.add_column(sa.Column("tool_args", sa.Text))

turnstone/core/storage/migrations/versions/028_model_definitions.py

@@ -0,0 +1,54 @@
+"""Create model_definitions table and grant admin.models permission.
+
+Revision ID: 028
+Revises: 027
+Create Date: 2026-03-29
+"""
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = "028"
+down_revision = "027"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.create_table(
+        "model_definitions",
+        sa.Column("definition_id", sa.Text, primary_key=True),
+        sa.Column("alias", sa.Text, nullable=False, unique=True),
+        sa.Column("model", sa.Text, nullable=False),
+        sa.Column("provider", sa.Text, nullable=False, server_default="openai"),
+        sa.Column("base_url", sa.Text, nullable=False, server_default=""),
+        sa.Column("api_key", sa.Text, nullable=False, server_default=""),
+        sa.Column("context_window", sa.Integer, nullable=False, server_default="32768"),
+        sa.Column("capabilities", sa.Text, nullable=False, server_default="{}"),
+        sa.Column("enabled", sa.Integer, nullable=False, server_default="1"),
+        sa.Column("created_by", sa.Text, nullable=False, server_default=""),
+        sa.Column("created", sa.Text, nullable=False),
+        sa.Column("updated", sa.Text, nullable=False),
+    )
+    op.create_index("idx_model_definitions_enabled", "model_definitions", ["enabled"])
+
+    # Grant admin.models permission to the built-in admin role
+    conn = op.get_bind()
+    conn.execute(
+        sa.text(
+            "UPDATE roles SET permissions = permissions || ',admin.models' "
+            "WHERE role_id = 'builtin-admin' "
+            "AND permissions NOT LIKE '%admin.models%'"
+        )
+    )
+
+
+def downgrade() -> None:
+    conn = op.get_bind()
+    conn.execute(
+        sa.text(
+            "UPDATE roles SET permissions = REPLACE(permissions, ',admin.models', '') "
+            "WHERE role_id = 'builtin-admin'"
+        )
+    )
+    op.drop_table("model_definitions")

turnstone/core/storage/migrations/versions/029_conversation_modify_permission.py

@@ -0,0 +1,44 @@
+"""Grant conversation.modify permission to admin and operator roles.
+
+Revision ID: 029
+Revises: 028
+Create Date: 2026-03-29
+"""
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = "029"
+down_revision = "028"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+    # Grant to admin role
+    conn.execute(
+        sa.text(
+            "UPDATE roles SET permissions = permissions || ',conversation.modify' "
+            "WHERE role_id = 'builtin-admin' "
+            "AND permissions NOT LIKE '%conversation.modify%'"
+        )
+    )
+    # Grant to operator role
+    conn.execute(
+        sa.text(
+            "UPDATE roles SET permissions = permissions || ',conversation.modify' "
+            "WHERE role_id = 'builtin-operator' "
+            "AND permissions NOT LIKE '%conversation.modify%'"
+        )
+    )
+
+
+def downgrade() -> None:
+    conn = op.get_bind()
+    conn.execute(
+        sa.text(
+            "UPDATE roles SET permissions = REPLACE(permissions, ',conversation.modify', '') "
+            "WHERE role_id IN ('builtin-admin', 'builtin-operator')"
+        )
+    )

turnstone/core/tls.py

@@ -79,7 +79,12 @@ class TLSClient:
 
             setup_metrics(self._event_dispatcher)
         except ImportError:
-            pass
+            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
 
     async def init(self) -> None:
         """Fetch CA root cert and request a service certificate.
@@ -220,7 +225,7 @@ class TLSClient:
         _require_lacme()
         from lacme.mtls import server_ssl_context
 
-        return server_ssl_context(  # type: ignore[no-any-return]
+        return server_ssl_context(  # type: ignore[no-any-return,unused-ignore]
             cert_pem=self._bundle.fullchain_pem,
             key_pem=self._bundle.key_pem,
             ca_cert_pem=self._ca_pem,
@@ -233,7 +238,7 @@ class TLSClient:
         _require_lacme()
         from lacme.mtls import client_ssl_context
 
-        return client_ssl_context(  # type: ignore[no-any-return]
+        return client_ssl_context(  # type: ignore[no-any-return,unused-ignore]
             cert_pem=self._bundle.cert_pem,
             key_pem=self._bundle.key_pem,
             ca_cert_pem=self._ca_pem,

turnstone/core/web.py

@@ -21,16 +21,32 @@ def strip_html(html: str) -> str:
 
 
 def check_ssrf(url: str) -> str | None:
-    """Return error string if URL resolves to a private/link-local address, else None."""
+    """Return error string if URL resolves to a private/link-local address, else None.
+
+    Checks both IPv4 and IPv6 addresses via getaddrinfo to prevent bypasses
+    using IPv6 loopback (``::1``), link-local (``fe80::``), or unique-local
+    (``fd00::``/``fc00::``) addresses.
+    """
     try:
         parsed = urlparse(url)
         hostname = parsed.hostname
         if not hostname:
             return "Invalid URL: no hostname"
-        addr = socket.gethostbyname(hostname)
-        ip = ipaddress.ip_address(addr)
-        if ip.is_private or ip.is_loopback or ip.is_link_local:
-            return f"Blocked: URL resolves to private/internal address ({addr})"
-    except (socket.gaierror, ValueError):
-        pass  # DNS failure or invalid IP — let the actual fetch handle it
+        # Resolve all address families (IPv4 + IPv6)
+        results = socket.getaddrinfo(hostname, parsed.port or 80, proto=socket.IPPROTO_TCP)
+        for _family, _type, _proto, _canonname, sockaddr in results:
+            addr = str(sockaddr[0])
+            # Strip IPv6 zone/scope identifier (e.g. "fe80::1%lo0")
+            addr_clean = addr.split("%", 1)[0] if "%" in addr else addr
+            try:
+                ip = ipaddress.ip_address(addr_clean)
+            except ValueError:
+                return f"Blocked: unable to parse resolved address ({addr})"
+            # Normalize IPv4-mapped IPv6 (e.g. ::ffff:127.0.0.1)
+            if isinstance(ip, ipaddress.IPv6Address) and ip.ipv4_mapped is not None:
+                ip = ip.ipv4_mapped
+            if ip.is_private or ip.is_loopback or ip.is_link_local:
+                return f"Blocked: URL resolves to private/internal address ({addr})"
+    except (socket.gaierror, OSError):
+        pass  # DNS failure — let the actual fetch handle it
     return None

turnstone/eval.py

@@ -105,7 +105,14 @@ class NullUI:
     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,
+        *,
+        is_error: bool = False,
+    ) -> None:
         pass
 
     def on_tool_output_chunk(self, call_id: str, chunk: str) -> None:

turnstone/mq/bridge.py

@@ -665,6 +665,8 @@ class Bridge:
                     effort=data.get("effort", ""),
                     cache_creation_tokens=data.get("cache_creation_tokens", 0),
                     cache_read_tokens=data.get("cache_read_tokens", 0),
+                    tool_calls_this_turn=data.get("tool_calls_this_turn", 0),
+                    turn_count=data.get("turn_count", 0),
                 ),
             )
         elif etype == "error":

turnstone/mq/protocol.py

@@ -252,6 +252,8 @@ class StatusEvent(OutboundEvent):
     effort: str = ""
     cache_creation_tokens: int = 0
     cache_read_tokens: int = 0
+    tool_calls_this_turn: int = 0
+    turn_count: int = 0
 
 
 @dataclass

turnstone/sdk/console.py

@@ -50,6 +50,7 @@ from turnstone.api.schemas import (
     AuthLoginResponse,
     AuthSetupResponse,
     AuthStatusResponse,
+    DeleteSettingResponse,
     ListScheduleRunsResponse,
     ListSchedulesResponse,
     ScheduleInfo,
@@ -642,13 +643,16 @@ class AsyncTurnstoneConsole(_BaseClient):
             "PUT", f"/v1/api/admin/settings/{key}", json_body=body, response_model=SettingInfo
         )
 
-    async def delete_setting(self, key: str, *, node_id: str = "") -> StatusResponse:
+    async def delete_setting(self, key: str, *, node_id: str = "") -> DeleteSettingResponse:
         """Reset a setting to its default value."""
         params: dict[str, Any] = {}
         if node_id:
             params["node_id"] = node_id
         return await self._request(
-            "DELETE", f"/v1/api/admin/settings/{key}", params=params, response_model=StatusResponse
+            "DELETE",
+            f"/v1/api/admin/settings/{key}",
+            params=params,
+            response_model=DeleteSettingResponse,
         )
 
     # -- MCP servers -------------------------------------------------------
@@ -1205,7 +1209,7 @@ class TurnstoneConsole:
     def update_setting(self, key: str, value: Any, *, node_id: str = "") -> SettingInfo:
         return self._runner.run(self._async.update_setting(key, value, node_id=node_id))
 
-    def delete_setting(self, key: str, *, node_id: str = "") -> StatusResponse:
+    def delete_setting(self, key: str, *, node_id: str = "") -> DeleteSettingResponse:
         return self._runner.run(self._async.delete_setting(key, node_id=node_id))
 
     # -- MCP servers -------------------------------------------------------