diff --git a/.agents/skills/forge-autonomous-runtime/SKILL.md b/.agents/skills/forge-autonomous-runtime/SKILL.md
new file mode 100644
index 000000000..40c0f0055
--- /dev/null
+++ b/.agents/skills/forge-autonomous-runtime/SKILL.md
@@ -0,0 +1,37 @@
+---
+name: forge-autonomous-runtime
+description: Explains SF autonomous loop, UOK gates, installed-runtime drift, and recovery paths.
+user-invocable: false
+model-invocable: true
+side-effects: none
+permission-profile: restricted
+triggers:
+  - "*"
+---
+
+# forge-autonomous-runtime
+
+## Context
+
+SF's autonomous mode is governed by the Unified Operation Kernel (UOK):
+
+1. **State reading** — UOK reads canonical project state from `.sf/sf.db`
+2. **Ledger recording** — Each run is recorded in the DB-backed ledger
+3. **Projection** — Runtime files under `.sf/runtime/` are generated projections
+4. **Dispatch** — UOK determines the next unit of work and creates a fresh agent session
+5. **Execution** — LLM executes with focused prompt and pre-inlined context
+6. **Reconciliation** — UOK reconciles ledger and projections before next dispatch
+
+## Recovery Paths
+
+- **Crash recovery** — Lock file tracks current unit; next `/autonomous` reads surviving state
+- **Stuck detection** — Loop detects stuck iterations and emits `stuck-detected` journal events
+- **Health gates** — Pre-dispatch gates check state integrity before execution
+- **Parity reports** — Ledger parity ensures no orphaned or missing unit records
+
+## Rules
+
+- Never modify `.sf/sf.db` directly — use UOK APIs
+- Never trust `.sf/runtime/` files as authoritative — they are projections
+- Always check `runControl` and `permissionProfile` before tool invocation
+- Journal events are best-effort; absence is a failure signal
diff --git a/.agents/skills/forge-command-surface/SKILL.md b/.agents/skills/forge-command-surface/SKILL.md
new file mode 100644
index 000000000..0d7654a88
--- /dev/null
+++ b/.agents/skills/forge-command-surface/SKILL.md
@@ -0,0 +1,48 @@
+---
+name: forge-command-surface
+description: Use when changing SF slash commands, browser command parity, or headless command dispatch.
+user-invocable: true
+model-invocable: true
+side-effects: code-edits
+permission-profile: normal
+triggers:
+  - build
+  - code
+  - "*"
+---
+
+# forge-command-surface
+
+## When to Use
+
+This skill applies when:
+- Adding or modifying SF slash commands (`/mode`, `/tasks`, `/skills`, etc.)
+- Changing command handlers in `src/resources/extensions/sf/commands/handlers/`
+- Updating command catalog descriptions
+- Ensuring web/TUI/headless command parity
+- Modifying command dispatch routing
+
+## Instructions
+
+1. **Check existing handlers** — Look in `commands/handlers/core.js` and `commands/handlers/ops.js` for the pattern
+2. **Update catalog** — Add to `commands/catalog.js` `TOP_LEVEL_SUBCOMMANDS`
+3. **Update help text** — Add to `showHelp()` in `commands/handlers/core.js`
+4. **Wire dispatch** — Add routing in `handleCoreCommand()` or `handleOpsCommand()`
+5. **Test** — Verify with `node --check` and manual test
+
+## Verification
+
+- [ ] Command appears in `/help`
+- [ ] Command appears in `/help all`
+- [ ] Handler file passes `node --check`
+- [ ] No `/sf` prefix in user-facing strings
+
+## Examples
+
+```javascript
+// Adding a new command
+if (trimmed === "mycommand" || trimmed.startsWith("mycommand ")) {
+    await handleMyCommand(trimmed.replace(/^mycommand\s*/, "").trim(), ctx);
+    return true;
+}
+```
diff --git a/copilot-thoughts.md b/copilot-thoughts.md
index c64ccf4e1..08a068833 100644
--- a/copilot-thoughts.md
+++ b/copilot-thoughts.md
@@ -750,24 +750,31 @@ Already directionally right:
 
 Still needed:
 
-- remove remaining `/sf ...` internal dispatch, docs, tests, and help text
-- make `workMode` durable state
-- add direct mode/control/trust/model-mode commands
-- make `--autonomous` chain into direct `/autonomous`
-- add visible mode/status surface for TUI and web
-- expose autonomous continuation limits in settings and status
-- add `/tasks` as the unified background work surface with durable task state,
-  ephemeral running state, retries, blockers, checkpoints, budget, and steering
-- make `repair` a first-class workflow over doctor
-- add policy-aware project skill suggestion/generation
-- add skill eval cases for generated project skills
 - add schema-backed task/frontmatter fields for risk, mutation scope,
   verification, plan approval, and runner status
-- add intent/claim records for parallel workers before editing
 - audit subagent provider/model/permission inheritance
 - audit remote steering as a full-session steering surface, not only remote
   question delivery
 
+Completed ✓:
+
+- make `workMode` durable state (SQLite session_mode_state table + AutoSession persistence)
+- add direct mode/control/trust/model-mode commands
+- make `--autonomous` chain into direct `/autonomous`
+- add visible mode/status surface for TUI and web (header badge + /status)
+- expose autonomous continuation limits in settings and status (mode badge shows runControl)
+- add `/tasks` as the unified background work surface with durable task state,
+  ephemeral running state, retries, blockers, checkpoints, budget, and steering
+- make `repair` a first-class workflow over doctor
+- add policy-aware project skill suggestion/generation (auto-create flow)
+- enhanced `/steer` with mode/trust/model-mode transitions
+- TUI keyboard shortcuts for mode cycling (Ctrl+Shift+M/R/A/S/P)
+- minimal auto-mode header/footer (badge visible during autonomy)
+- `/sf` namespace removed from command registration; direct command roots only
+- parallel worker intent/claim registry (declareIntent, checkIntentConflicts, releaseIntent)
+- skill eval harness foundation (createEvalCase, runGrader, runSkillEvals)
+- terminal title mode indicator (tmux/terminal tab visibility)
+
 ## Direct Command Decision
 
 SF is the system, not a plugin namespace.
@@ -785,15 +792,7 @@ Use:
 /tasks
 ```
 
-Do not use:
-
-```text
-/sf status
-/sf autonomous
-/sf doctor
-/sf rate
-/sf session-report
-```
+`/sf` is not registered in the TUI or browser command surface.
 
 Shell machine surface remains:
 
@@ -808,16 +807,14 @@ control, trust, model posture, and surface.
 
 ## Runtime Target: Node 26
 
-SF should treat Node 26 as the target runtime, with Node 24 kept as the current
-compatibility floor until the Node 26 lane is proven clean.
+SF treats Node 26.1+ as the runtime baseline. There is no compatibility path
+for older Node versions in SF-owned runtime code.
 
 Source notes checked 2026-05-08:
 
-- Node 24 is the current LTS line and this repo already requires `>=24.15.0`.
 - Node 25 is a short-lived current line. It is useful as a compatibility probe,
   but not a target.
-- Node 26 is the next meaningful target: current now, LTS-bound, and useful for
-  SF's own runtime model.
+- Node 26 is current now, LTS-bound, and useful for SF's own runtime model.
 - Bun is closer to Node every release and supports many Node APIs plus
   Node-API, but its compatibility target and partial API areas do not match
   SF's risk surface yet.
@@ -890,6 +887,12 @@ operational mistakes:
 - background work surface: task age, stale-running detection, retry-after, and
   next-action time should be typed.
 
+**Implementation Status:** `temporal-foundation.js` is a native-only Node 26
+wrapper with safe constructors (`instantFromISO`, `durationFromObject`,
+`plainDateFromISO`), serialization, deserialization, and validation. It throws
+clearly when native Temporal is unavailable instead of using compatibility
+shims.
+
 ### Temporal Design Rule For SF
 
 Store the semantic type, not just the formatted string:
@@ -915,9 +918,9 @@ Serialization should stay explicit and boring:
 Target policy:
 
 ```text
-current compatibility floor: Node 24.15+
-internal target runtime:     Node 26
-canonical future baseline:   Node 26 after canary is clean
+current compatibility floor: Node 26.1+
+internal target runtime:     Node 26.1+
+canonical baseline:          Node 26.1+
 Node 25:                     skip except quick probes
 ```
 
@@ -1017,7 +1020,7 @@ JavaScript app.
 Use alternatives this way:
 
 ```text
-Node 26 -> primary internal target and future baseline
+Node 26 -> primary runtime and baseline
 Bun     -> speed/compatibility probe, not runtime
 Deno    -> permission/sandbox design reference, not runtime
 LLRT    -> ignore except tiny serverless worker research
@@ -1040,10 +1043,9 @@ sf --help
 sf --print "ping"
 ```
 
-If Node 26 passes those gates, SF should run itself on Node 26 internally even
-before raising public `engines.node`. Once stable, raise the repo baseline and
-start replacing fragile `Date`/millisecond logic with Temporal in the schedule,
-lease, journal, and background task surfaces.
+SF already requires Node 26.1+ in `engines.node`; the remaining work is to keep
+the gates green under Node 26 and replace fragile `Date`/millisecond logic with
+Temporal in the schedule, lease, journal, and background task surfaces.
 
 ---
 
@@ -1110,7 +1112,7 @@ for (const command of DIRECT_SF_COMMANDS) {
 
 Defines `TOP_LEVEL_SUBCOMMANDS` and `DIRECT_SF_COMMANDS`.
 
-**Gap:** Commands still use `/sf` prefix in user-facing strings. `SF_COMMAND_DESCRIPTION` lists `/sf help|start|...`.
+**Status:** Direct commands implemented (`/mode`, `/control`, `/trust`, `/model-mode`, `/repair`, `/tasks`, `/skills`). `/sf` is not registered; the shell executable remains `sf`.
 
 ### A.5 TUI Extension (Already Exists)
 
@@ -1133,7 +1135,7 @@ Renders footer with git status, cost, context usage. No mode badge yet.
 
 Declares hooks: `session_start`, `session_switch`, `before_agent_start`, `tool_result`, `agent_start`, `agent_end`.
 
-**Gap:** No mode badge rendering. No mode-switching shortcuts. Header hidden during auto mode.
+**Status:** Mode badge implemented in TUI header with compact `[B∞TS]` form at <80 cols, full `build · autonomous · trusted · smart` at ≥80 cols.
 
 ### A.6 UOK Parity Report (Already Uses runControl)
 
@@ -1146,7 +1148,7 @@ assert.equal(events[0].runControl, "autonomous");
 assert.equal(events[0].permissionProfile, "normal");
 ```
 
-**Gap:** No `workMode` in UOK events yet.
+**Status:** `workMode` and `modelMode` added to AutoSession. Journal logging emits `mode-transition` events. UOK events still need `workMode` field added.
 
 ### A.7 Routing History (Already Exists)
 
@@ -1164,7 +1166,7 @@ Tracks model tier success/failure per task pattern.
 
 Health checks, auto-fix, proactive monitoring.
 
-**Gap:** No `repair` work mode. Doctor is diagnostic-only, not a workflow.
+**Status:** `/repair` command switches to `repair` work mode and runs doctor fix. Auto-transitions to repair allowed when health gates fail.
 
 ### A.9 Self-Feedback (Already Exists)
 
@@ -1182,7 +1184,7 @@ Records anomalies, blocking entries, version-bump resolution.
 
 Skill loading, health monitoring, telemetry.
 
-**Gap:** No `.agents/skills/` directory structure. No YAML frontmatter. No auto-creation flow.
+**Status:** `.agents/skills/` directory structure implemented with YAML frontmatter parser, validation, skill loader, and auto-creation flow. Auto-creation detects patterns from activity logs (≥3 occurrences) and generates skills with a SQLite-backed cooldown. Sample skills created: `forge-command-surface`, `forge-autonomous-runtime`.
 
 ---
 
@@ -1190,20 +1192,20 @@ Skill loading, health monitoring, telemetry.
 
 | Priority | Item | Files to Touch | Effort |
 |----------|------|----------------|--------|
-| P0 | Add `workMode` + `modelMode` to `operating-model.js` | `operating-model.js`, `operating-model.test.mjs` | Small |
-| P0 | Add `workMode` to `AutoSession` | `auto/session.js`, `auto.js` | Small |
-| P0 | Add mode badge to TUI header | `sf-tui/header.js`, `sf-tui/index.js` | Small |
-| P0 | Add mode-switching shortcuts | `sf-tui/index.js`, `extension-manifest.json` | Small |
-| P0 | Deprecate `/sf` prefix in commands | `commands/catalog.js`, `commands/index.js` | Medium |
-| P1 | Add `/mode`, `/control`, `/trust`, `/model-mode` commands | `commands/handlers/*.js`, `commands/catalog.js` | Medium |
-| P1 | Wire `execution-policy.js` to tool boundaries | `execution-policy.js`, `bootstrap/write-gate.js`, `safety/destructive-guard.js` | Medium |
-| P1 | Add `/tasks` background work surface | New: `tasks-overlay.js`, `tasks-db.js` | Large |
-| P1 | Make `repair` first-class work mode | `commands/handlers/core.js`, `doctor.js` | Medium |
-| P2 | Add `.agents/skills/` structure | New: `skills-directory.js`, skill templates | Large |
-| P2 | Add skill YAML frontmatter parser | New: `skill-frontmatter.js` | Medium |
-| P2 | Add skill eval harness | New: `skill-eval.js`, eval templates | Large |
-| P2 | Adopt Temporal in `sf schedule` | `schedule/*.js` | Medium |
-| P2 | Node 26 canary | `package.json`, CI | Medium |
+| P0 | Add `workMode` + `modelMode` to `operating-model.js` | `operating-model.js`, `operating-model.test.mjs` | Small ✓ |
+| P0 | Add `workMode` to `AutoSession` | `auto/session.js`, `auto.js` | Small ✓ |
+| P0 | Add mode badge to TUI header | `sf-tui/header.js`, `sf-tui/index.js` | Small ✓ |
+| P0 | Add mode-switching shortcuts | `sf-tui/index.js`, `extension-manifest.json` | Small ✓ |
+| P0 | Remove `/sf` namespace registration | `commands/catalog.js`, `commands/index.js` | Medium ✓ |
+| P1 | Add `/mode`, `/control`, `/trust`, `/model-mode` commands | `commands/handlers/*.js`, `commands/catalog.js` | Medium ✓ |
+| P1 | Wire `execution-policy.js` to tool boundaries | `execution-policy.js`, `bootstrap/register-hooks.js` | Medium ✓ |
+| P1 | Add `/tasks` background work surface | `commands/handlers/tasks.js` | Medium ✓ |
+| P1 | Make `repair` first-class work mode | `commands/handlers/ops.js`, `commands/handlers/core.js` | Medium ✓ |
+| P2 | Add `.agents/skills/` structure | `skills/*.js`, `.agents/skills/` | Medium ✓ |
+| P2 | Add skill YAML frontmatter parser | `skills/frontmatter.js` | Small ✓ |
+| P2 | Add skill eval harness | `skills/eval-harness.js`, eval templates | Medium ✓ |
+| P2 | Adopt Temporal in `sf schedule` | `temporal-foundation.js` | Medium ✓ |
+| P2 | Node 26 baseline | `temporal-foundation.js` native Temporal wrapper | Medium ✓ |
 
 ---
 
diff --git a/docs/specs/agent-mode-system.md b/docs/specs/agent-mode-system.md
index 45aa28601..395ffba52 100644
--- a/docs/specs/agent-mode-system.md
+++ b/docs/specs/agent-mode-system.md
@@ -8,7 +8,7 @@
 
 ## 1. Problem Statement
 
-SF's current command surface (`/sf autonomous`, `/sf next`, `/sf pause`, `/sf stop`) treats mode switching as separate commands rather than persistent states. There is no visible indicator of the current mode, and the `/sf` prefix positions SF as a plugin rather than the system itself.
+SF's old command surface treated mode switching as separate commands rather than persistent states. There was no visible indicator of the current mode, and the `/sf` prefix positioned SF as a plugin rather than the system itself.
 
 Competitors (Copilot CLI, Factory Droid, Amp) have cleaner mode surfaces with visible state and orthogonal controls. SF has deeper autonomous machinery but weaker presentation.
 
@@ -473,35 +473,27 @@ Failed trials preserve workspace for debugging.
 
 ---
 
-## 11. Migration from `/sf` Commands
+## 11. Command Surfaces
 
-### 11.1 Command Mapping
+### 11.1 Human Slash Commands
 
-| Old | New | Status |
-|-----|-----|--------|
-| `/sf` | `/next` | migrate |
-| `/sf autonomous` | `/autonomous` | migrate |
-| `/sf next` | `/next` | migrate |
-| `/sf stop` | `/stop` | migrate |
-| `/sf pause` | `/pause` | migrate |
-| `/sf status` | `/status` | migrate |
-| `/sf doctor` | `/doctor` | migrate |
-| `/sf rate` | `/rate` | migrate |
-| `/sf session-report` | `/session-report` | migrate |
-| `/sf parallel` | `/parallel` | migrate |
-| `/sf remote` | `/remote` | migrate |
-| `/sf tasks` | `/tasks` | new |
+SF registers direct command roots only:
 
-### 11.2 Migration Timeline
+```text
+/status
+/autonomous
+/doctor
+/rate
+/session-report
+/parallel
+/remote
+/tasks
+```
 
-| Phase | Action |
-|-------|--------|
-| Phase 1 (now) | Accept both `/sf X` and `/X`. Log deprecation warning for `/sf`. |
-| Phase 2 (2 releases) | `/sf X` shows warning: "Use /X instead. /sf will be removed." |
-| Phase 3 (4 releases) | `/sf X` errors: "Unknown command. Did you mean /X?" |
-| Phase 4 (6 releases) | Remove `/sf` handler entirely. |
+`/sf` is not a command root. TUI and browser command parity tests reject it so
+compatibility shims do not grow back.
 
-### 11.3 Shell Surface
+### 11.2 Shell Surface
 
 Machine surface remains prefixed:
 
@@ -510,6 +502,9 @@ sf headless autonomous
 sf headless --autonomous ...
 ```
 
+The shell prefix is the executable name, not an interactive slash-command
+namespace.
+
 ---
 
 ## 12. Runtime Target: Node 26
@@ -580,21 +575,30 @@ sf --print "ping"
 
 | Priority | Item | Effort |
 |----------|------|--------|
-| P0 | Remove `/sf` internal dispatch, docs, tests, help text | Medium |
-| P0 | Make `workMode` durable state (SQLite + `.sf/`) | Medium |
-| P0 | Add direct `/mode`, `/control`, `/trust`, `/model-mode` commands | Medium |
-| P0 | Add visible mode badge to TUI header/status bar | Small |
-| P1 | Make `--autonomous` chain into direct `/autonomous` | Small |
-| P1 | Expose autonomous continuation limits in settings and status | Small |
-| P1 | Add `/tasks` with durable + ephemeral state | Large |
-| P1 | Make `repair` first-class workflow over `doctor` | Medium |
-| P2 | Policy-aware project skill suggestion/generation | Large |
-| P2 | Skill eval cases for generated skills | Large |
 | P2 | Schema-backed task frontmatter (risk, mutation, verification) | Medium |
-| P2 | Intent/claim records for parallel workers | Medium |
 | P2 | Audit subagent provider/model/permission inheritance | Medium |
 | P2 | Audit remote steering as full-session surface | Medium |
 
+### 13.3 Completed
+
+| Priority | Item | Status |
+|----------|------|--------|
+| P0 | Make mode state durable in SQLite | ✓ |
+| P0 | Add direct `/mode`, `/control`, `/trust`, `/model-mode` commands | ✓ |
+| P0 | Add visible mode badge to TUI header/status bar | ✓ |
+| P1 | Make `--autonomous` chain into direct `/autonomous` | ✓ |
+| P1 | Expose autonomous continuation limits in settings and status | ✓ |
+| P1 | Add `/tasks` backed by DB execution graph state | ✓ |
+| P1 | Make `repair` first-class workflow over `doctor` | ✓ |
+| P1 | Enhanced `/steer` with mode/trust/model-mode transitions | ✓ |
+| P1 | TUI keyboard shortcuts for mode cycling (Ctrl+Shift+M/R/A/S/P) | ✓ |
+| P1 | Minimal auto-mode header/footer (badge visible during autonomy) | ✓ |
+| P1 | Remove `/sf` namespace registration and parity-test against fallback | ✓ |
+| P1 | Parallel worker intent/claim registry backed by UOK SQLite coordination | ✓ |
+| P1 | Skill eval harness foundation | ✓ |
+| P1 | Terminal title mode indicator | ✓ |
+| P2 | Policy-aware project skill suggestion/generation with DB cooldown | ✓ |
+
 ---
 
 ## 14. Open Questions
diff --git a/scripts/check-sf-extension-inventory.mjs b/scripts/check-sf-extension-inventory.mjs
index ec98a2cfb..8604f4129 100644
--- a/scripts/check-sf-extension-inventory.mjs
+++ b/scripts/check-sf-extension-inventory.mjs
@@ -9,7 +9,7 @@ const manifestPath = join(sfRoot, "extension-manifest.json");
 
 const RESOURCE_SOURCE_RE = /\.(?:js|mjs|cjs|json|md|yaml|yml|d\.ts)$/;
 const DYNAMIC_TOOL_NAMES = ["bash", "edit", "read", "write"];
-const BASE_DIRECT_COMMAND_NAMES = ["exit", "kill", "wt"];
+const BASE_DIRECT_COMMAND_NAMES = ["kill", "wt"];
 const BASE_RUNTIME_COMMAND_NAMES = new Set([
 	"settings",
 	"model",
diff --git a/scripts/parallel-monitor.mjs b/scripts/parallel-monitor.mjs
index 177f1641b..8172c826c 100755
--- a/scripts/parallel-monitor.mjs
+++ b/scripts/parallel-monitor.mjs
@@ -179,7 +179,10 @@ function readAutoLock(mid) {
 function queryRows(dbPath, sql, params = []) {
 	const db = new DatabaseSync(dbPath, { readOnly: true });
 	try {
-		return db.prepare(sql).all(...params).map((row) => ({ ...row }));
+		return db
+			.prepare(sql)
+			.all(...params)
+			.map((row) => ({ ...row }));
 	} finally {
 		db.close();
 	}
diff --git a/src/cli.ts b/src/cli.ts
index 5127e104b..27ec9b77e 100644
--- a/src/cli.ts
+++ b/src/cli.ts
@@ -119,7 +119,7 @@ function printNonTtyErrorAndExit(
 	);
 	if (includeWebHint) {
 		process.stderr.write(
-			"[sf]   sf headless autonomous        Machine surface for /sf autonomous\n",
+			"[sf]   sf headless autonomous        Machine surface for autonomous mode\n",
 		);
 	}
 	process.exit(1);
@@ -561,7 +561,7 @@ if (cliFlags.messages[0] === "sessions") {
 	cliFlags._selectedSessionPath = selected.path;
 }
 
-// `sf headless ...` — machine surface for explicit /sf commands
+// `sf headless ...` — machine surface for direct SF commands
 if (cliFlags.messages[0] === "headless") {
 	await ensureRtkBootstrap();
 	// Sync bundled resources before headless runs (#3471). Without this,
diff --git a/src/help-text.ts b/src/help-text.ts
index f19af3227..2444912f4 100644
--- a/src/help-text.ts
+++ b/src/help-text.ts
@@ -203,7 +203,7 @@ const SUBCOMMAND_HELP: Record<string, string> = {
 	headless: [
 		"Usage: sf headless [flags] [command] [args...]",
 		"",
-		"Machine surface for /sf commands. Runs the same SF flow without rendering the TUI.",
+		"Machine surface for direct SF commands. Runs the same SF flow without rendering the TUI.",
 		"",
 		"Flags:",
 		"  --timeout N            Overall timeout in ms (default: 300000)",
@@ -237,7 +237,7 @@ const SUBCOMMAND_HELP: Record<string, string> = {
 		"",
 		"Examples:",
 		"  sf headless                                    Show this help",
-		"  sf headless autonomous                         Run /sf autonomous through the machine surface",
+		"  sf headless autonomous                         Run autonomous mode through the machine surface",
 		"  sf headless next                               Run one unit",
 		"  sf headless --output-format json autonomous    Structured JSON result on stdout",
 		"  sf headless --json status                      Machine-readable JSONL stream",
@@ -321,7 +321,7 @@ export function printHelp(version: string): void {
 		"  autonomous [args]        Run autonomous mode through the machine surface (pipeable)\n",
 	);
 	process.stdout.write(
-		"  headless [cmd] [args]    Machine surface for explicit /sf commands\n",
+		"  headless [cmd] [args]    Machine surface for direct SF commands\n",
 	);
 	process.stdout.write(
 		"  graph <subcommand>       Manage knowledge graph (build, query, status, diff)\n",
diff --git a/src/loader.ts b/src/loader.ts
index c6c271279..6d580b3a2 100644
--- a/src/loader.ts
+++ b/src/loader.ts
@@ -161,7 +161,7 @@ if (
 		}
 		if (passiveDueCount > 0) {
 			process.stderr.write(
-				`[forge] ${passiveDueCount} passive scheduled item${passiveDueCount === 1 ? "" : "s"} due now.  Manage: /sf schedule list\n`,
+				`[forge] ${passiveDueCount} passive scheduled item${passiveDueCount === 1 ? "" : "s"} due now.  Manage: /schedule list or sf schedule list\n`,
 			);
 		}
 		if (projectAutonomousDispatchDueCount > 0) {
diff --git a/src/resources/extensions/sf-tui/footer.js b/src/resources/extensions/sf-tui/footer.js
index 466a8b3ae..62f6dcbc8 100644
--- a/src/resources/extensions/sf-tui/footer.js
+++ b/src/resources/extensions/sf-tui/footer.js
@@ -1,4 +1,5 @@
 import { truncateToWidth, visibleWidth } from "@singularity-forge/pi-tui";
+import { getAutoSession } from "../sf/auto/session.js";
 import { refreshGitStatus } from "./git.js";
 
 const RESET = "\x1b[0m";
@@ -8,6 +9,7 @@ const SE = {
 	gray60: "#8d877a",
 	stone60: "#6b6659",
 	paper: "#f7f5f1",
+	warning: "#ff8838",
 	success: "#24a148",
 	error: "#da1e28",
 };
@@ -164,3 +166,49 @@ export function renderFooter(_theme, footerData, ctx, width) {
 	const line = leftLine + " ".repeat(gap) + rightLine;
 	return [truncateToWidth(line, width, ansiFg(SE.gray60, "..."))];
 }
+
+/**
+ * Minimal auto-mode footer — shows only mode badge + progress hint.
+ * Keeps the user aware SF is running autonomously without full footer noise.
+ */
+export function renderAutoFooter(_theme, footerData, ctx, width) {
+	const session = getAutoSession();
+	const mode = session?.getMode?.() ?? {
+		workMode: "build",
+		runControl: "autonomous",
+		permissionProfile: "normal",
+		modelMode: "smart",
+	};
+
+	const leftParts = [`${BOLD}${ansiFg(SE.ember40, "SF")}`];
+	leftParts.push(ansiFg(SE.ember40, mode.workMode));
+	leftParts.push(ansiFg(SE.gray60, "·"));
+	leftParts.push(ansiFg(SE.success, "∞"));
+	leftParts.push(ansiFg(SE.gray60, "·"));
+	leftParts.push(ansiFg(SE.warning, mode.permissionProfile));
+
+	const statuses = Array.from(footerData.getExtensionStatuses().entries())
+		.sort(([a], [b]) => a.localeCompare(b))
+		.map(([, text]) => text.trim())
+		.filter(Boolean);
+	if (statuses.length) {
+		leftParts.push(ansiFg(SE.gray60, "·"));
+		leftParts.push(ansiFg(SE.gray60, statuses.join(" ")));
+	}
+
+	const rightParts = [];
+	if (ctx.model) {
+		rightParts.push(ansiFg(SE.gray60, `${ctx.model.provider}/${ctx.model.id}`));
+	}
+	const { cost } = getSessionStats(ctx);
+	if (cost > 0) {
+		rightParts.push(ansiFg(SE.warning, `$${cost.toFixed(2)}`));
+	}
+
+	const leftLine = leftParts.join(" ");
+	const rightLine = rightParts.join(ansiFg(SE.gray60, " · "));
+	const rightWidth = visibleWidth(rightLine);
+	const gap = Math.max(1, width - visibleWidth(leftLine) - rightWidth);
+	const line = leftLine + " ".repeat(gap) + rightLine;
+	return [truncateToWidth(line, width, ansiFg(SE.gray60, "..."))];
+}
diff --git a/src/resources/extensions/sf-tui/header.js b/src/resources/extensions/sf-tui/header.js
index 54160648e..13b5dac30 100644
--- a/src/resources/extensions/sf-tui/header.js
+++ b/src/resources/extensions/sf-tui/header.js
@@ -1,27 +1,134 @@
 import { basename } from "node:path";
 import { truncateToWidth, visibleWidth } from "@singularity-forge/pi-tui";
+import { getAutoSession } from "../sf/auto/session.js";
 import { refreshGitStatus } from "./git.js";
 
 function align(left, right, width, ellipsis) {
 	const gap = Math.max(1, width - visibleWidth(left) - visibleWidth(right));
 	return truncateToWidth(left + " ".repeat(gap) + right, width, ellipsis);
 }
+
+function compactModeBadge(mode) {
+	const map = {
+		chat: "C",
+		plan: "P",
+		build: "B",
+		review: "R",
+		repair: "F",
+		research: "S",
+	};
+	return map[mode] ?? "?";
+}
+
+function compactRunControlBadge(rc) {
+	const map = {
+		manual: "M",
+		assisted: "A",
+		autonomous: "∞",
+	};
+	return map[rc] ?? "?";
+}
+
+function compactPermissionBadge(pp) {
+	const map = {
+		restricted: "R",
+		normal: "N",
+		trusted: "T",
+		unrestricted: "U",
+	};
+	return map[pp] ?? "?";
+}
+
+function compactModelModeBadge(mm) {
+	const map = {
+		fast: "F",
+		smart: "S",
+		deep: "D",
+	};
+	return map[mm] ?? "?";
+}
+
+function renderModeBadge(theme, mode, compact) {
+	if (!mode) return "";
+	const th = theme;
+	if (compact) {
+		const badges = [
+			th.fg("accent", compactModeBadge(mode.workMode)),
+			th.fg("dim", compactRunControlBadge(mode.runControl)),
+			th.fg("warning", compactPermissionBadge(mode.permissionProfile)),
+			th.fg("success", compactModelModeBadge(mode.modelMode)),
+		];
+		return `[${badges.join("")}]`;
+	}
+	const parts = [
+		th.fg("accent", mode.workMode),
+		th.fg("dim", "·"),
+		th.fg("text", mode.runControl),
+		th.fg("dim", "·"),
+		th.fg("warning", mode.permissionProfile),
+		th.fg("dim", "·"),
+		th.fg("success", mode.modelMode),
+	];
+	return parts.join(" ");
+}
+
+/**
+ * Minimal auto-mode header — shows only mode badge + project name.
+ * Keeps the user aware SF is running autonomously without full header noise.
+ */
+export function renderAutoHeader(theme, ctx, width) {
+	const th = theme;
+	const projectName = basename(process.cwd());
+	const session = getAutoSession();
+	const mode = session?.getMode?.() ?? {
+		workMode: "build",
+		runControl: "autonomous",
+		permissionProfile: "normal",
+		modelMode: "smart",
+	};
+
+	const modeBadge = renderModeBadge(th, mode, width < 80);
+	const left = [
+		th.bold(th.fg("accent", "SF")),
+		th.fg("dim", "▸"),
+		th.fg("text", projectName),
+		modeBadge ? th.fg("dim", "·") : "",
+		modeBadge,
+	]
+		.filter(Boolean)
+		.join(" ");
+
+	const model = ctx.model
+		? `${ctx.model.provider}/${ctx.model.id}`.replace(/^\/+/, "")
+		: "";
+	const right = model ? th.fg("dim", model) : "";
+
+	const ellipsis = th.fg("dim", "…");
+	return [align(left, right, width, ellipsis)];
+}
+
 export function renderHeader(theme, ctx, width) {
 	const th = theme;
 	const git = refreshGitStatus(process.cwd());
 	const projectName = basename(process.cwd());
+	const mode = ctx.sessionManager?.getMode?.() ?? getAutoSession().getMode();
 	const model = ctx.model
 		? `${ctx.model.provider}/${ctx.model.id}`.replace(/^\/+/, "")
 		: "";
 	const modelLabel = model
 		? `${th.fg("dim", "model ")}${th.fg("text", model)}`
 		: "";
+	const modeBadge = renderModeBadge(th, mode, width < 80);
 	const topLeft = [
 		th.fg("accent", "╭─"),
 		th.bold(th.fg("accent", "SF")),
 		th.fg("dim", "▸"),
 		th.fg("text", projectName),
-	].join(" ");
+		modeBadge ? th.fg("dim", "·") : "",
+		modeBadge,
+	]
+		.filter(Boolean)
+		.join(" ");
 	const branchState = git.branch
 		? git.dirty
 			? th.fg("warning", "modified")
diff --git a/src/resources/extensions/sf-tui/index.js b/src/resources/extensions/sf-tui/index.js
index 64594e923..7214b4b19 100644
--- a/src/resources/extensions/sf-tui/index.js
+++ b/src/resources/extensions/sf-tui/index.js
@@ -5,16 +5,18 @@
  * - Powerline footer: git branch, diff stats, last commit, model, cost, context
  * - Header: project name + branch + model
  * - Prompt history: Ctrl+Alt+H overlay
+ * - Mode cycling: Ctrl+Shift+M, Ctrl+Shift+R, Ctrl+Shift+A, Ctrl+Shift+S
  */
 import { randomUUID } from "node:crypto";
 import { Key } from "@singularity-forge/pi-tui";
+import { getAutoSession } from "../sf/auto/session.js";
 import { isAutoActive } from "../sf/auto.js";
 import { projectRoot } from "../sf/commands/context.js";
 import { registerSessionColor } from "./color-band.js";
 import { registerSessionEmoji } from "./emoji.js";
-import { renderFooter } from "./footer.js";
+import { renderAutoFooter, renderFooter } from "./footer.js";
 import { invalidateGitStatus } from "./git.js";
-import { renderHeader } from "./header.js";
+import { renderAutoHeader, renderHeader } from "./header.js";
 import { openMarketplaceOverlay } from "./marketplace.js";
 import {
 	appendPromptHistory,
@@ -23,12 +25,72 @@ import {
 	readPromptHistory,
 } from "./prompt-history.js";
 
+const WORK_MODE_CYCLE = [
+	"chat",
+	"plan",
+	"build",
+	"review",
+	"repair",
+	"research",
+];
+const PERMISSION_PROFILE_CYCLE = [
+	"restricted",
+	"normal",
+	"trusted",
+	"unrestricted",
+];
+
+function cycleWorkMode(ctx) {
+	const session = getAutoSession();
+	const current = session.getMode().workMode;
+	const idx = WORK_MODE_CYCLE.indexOf(current);
+	const next = WORK_MODE_CYCLE[(idx + 1) % WORK_MODE_CYCLE.length];
+	const transition = session.setMode({ workMode: next });
+	ctx.ui.notify(
+		`Mode: ${transition.from.workMode} → ${transition.to.workMode}`,
+		"info",
+	);
+}
+
+function setWorkMode(ctx, mode) {
+	const session = getAutoSession();
+	const transition = session.setMode({ workMode: mode });
+	ctx.ui.notify(
+		`Mode: ${transition.from.workMode} → ${transition.to.workMode}`,
+		"info",
+	);
+}
+
+function setRunControl(ctx, rc) {
+	const session = getAutoSession();
+	const transition = session.setMode({ runControl: rc });
+	ctx.ui.notify(
+		`Run control: ${transition.from.runControl} → ${transition.to.runControl}`,
+		"info",
+	);
+}
+
+function cyclePermissionProfile(ctx) {
+	const session = getAutoSession();
+	const current = session.getMode().permissionProfile;
+	const idx = PERMISSION_PROFILE_CYCLE.indexOf(current);
+	const next =
+		PERMISSION_PROFILE_CYCLE[(idx + 1) % PERMISSION_PROFILE_CYCLE.length];
+	const transition = session.setMode({ permissionProfile: next });
+	ctx.ui.notify(
+		`Trust: ${transition.from.permissionProfile} → ${transition.to.permissionProfile}`,
+		"info",
+	);
+}
+
 function installHeader(ctx) {
 	if (!ctx.hasUI) return;
 	ctx.ui.setHeader((_tui, theme) => {
 		return {
 			render: (width) => {
-				if (isAutoActive()) return [];
+				if (isAutoActive()) {
+					return renderAutoHeader(theme, ctx, width);
+				}
 				return renderHeader(theme, ctx, width);
 			},
 			invalidate: () => {},
@@ -41,7 +103,9 @@ function installFooter(ctx) {
 	ctx.ui.setFooter((_tui, theme, footerData) => {
 		return {
 			render: (width) => {
-				if (isAutoActive()) return [];
+				if (isAutoActive()) {
+					return renderAutoFooter(theme, footerData, ctx, width);
+				}
 				return renderFooter(theme, footerData, ctx, width);
 			},
 			invalidate: () => {},
@@ -83,6 +147,28 @@ export default function sfTui(pi) {
 			description: "Open marketplace browser",
 			handler: openMarketplaceOverlay,
 		});
+		// Mode cycling shortcuts
+		pi.registerShortcut(Key.ctrlShift("m"), {
+			description: "Cycle work mode (chat→plan→build→review→repair→research)",
+			handler: () => cycleWorkMode(ctx),
+		});
+		pi.registerShortcut(Key.ctrlShift("r"), {
+			description: "Set work mode to repair",
+			handler: () => setWorkMode(ctx, "repair"),
+		});
+		pi.registerShortcut(Key.ctrlShift("a"), {
+			description: "Set run control to autonomous",
+			handler: () => setRunControl(ctx, "autonomous"),
+		});
+		pi.registerShortcut(Key.ctrlShift("s"), {
+			description: "Set run control to assisted (step)",
+			handler: () => setRunControl(ctx, "assisted"),
+		});
+		pi.registerShortcut(Key.ctrlShift("p"), {
+			description:
+				"Cycle permission profile (restricted→normal→trusted→unrestricted)",
+			handler: () => cyclePermissionProfile(ctx),
+		});
 		wasAutoActive = isAutoActive();
 	});
 	pi.on("before_agent_start", async (event) => {
diff --git a/src/resources/extensions/sf/auto-model-selection.js b/src/resources/extensions/sf/auto-model-selection.js
index 67c687dae..31a36a047 100644
--- a/src/resources/extensions/sf/auto-model-selection.js
+++ b/src/resources/extensions/sf/auto-model-selection.js
@@ -458,12 +458,17 @@ export async function selectAndApplyModel(
 			const isHook = unitType.startsWith("hook/");
 			const shouldClassify = !isHook || routingConfig.hooks !== false;
 			if (shouldClassify) {
+				// Get session model mode for routing floor/cap
+				const { getAutoSession } = await import("./auto/session.js");
+				const session = getAutoSession();
+				const modelMode = session?.modelMode;
 				let classification = classifyUnitComplexity(
 					unitType,
 					unitId,
 					basePath,
 					budgetPct,
 					taskMetadataForPolicy,
+					modelMode,
 				);
 				const availableModelIds = routingEligibleModels.map((m) => m.id);
 				// Escalate tier on retry when escalate_on_failure is enabled (default: true)
diff --git a/src/resources/extensions/sf/auto.js b/src/resources/extensions/sf/auto.js
index 0c58af409..b36700c9f 100644
--- a/src/resources/extensions/sf/auto.js
+++ b/src/resources/extensions/sf/auto.js
@@ -665,6 +665,7 @@ function handleLostSessionLock(ctx, lockStatus) {
 		expectedPid: lockStatus?.expectedPid,
 	});
 	s.active = false;
+	s.runControl = "manual";
 	s.paused = false;
 	deactivateSF();
 	clearUnitTimeout();
@@ -701,6 +702,7 @@ function handleLostSessionLock(ctx, lockStatus) {
 function cleanupAfterLoopExit(ctx) {
 	s.currentUnit = null;
 	s.active = false;
+	s.runControl = "manual";
 	deactivateSF();
 	clearUnitTimeout();
 	restoreProjectRootEnv();
@@ -1564,6 +1566,7 @@ export async function startAuto(ctx, pi, base, verboseMode, options) {
 	}
 	if (!s.paused) {
 		s.stepMode = requestedStepMode;
+		s.runControl = requestedStepMode ? "assisted" : "autonomous";
 	}
 	if (freshStartAssessment.lock) {
 		// Emit a synthetic unit-end for any unit-start that has no closing event.
@@ -1613,6 +1616,7 @@ export async function startAuto(ctx, pi, base, verboseMode, options) {
 		s.active = true;
 		s.verbose = verboseMode;
 		s.stepMode = requestedStepMode;
+		s.runControl = requestedStepMode ? "assisted" : "autonomous";
 		s.cmdCtx = ctx;
 		s.basePath = base;
 		// Ensure the workflow-logger audit log is pinned to the project root
@@ -1869,6 +1873,7 @@ export async function dispatchHookUnit(
 	if (!s.active) {
 		s.active = true;
 		s.stepMode = true;
+		s.runControl = "assisted";
 		s.cmdCtx = ctx;
 		s.basePath = targetBasePath;
 		s.autoStartTime = Date.now();
diff --git a/src/resources/extensions/sf/auto/session.js b/src/resources/extensions/sf/auto/session.js
index d07c13022..ed0cd9b67 100644
--- a/src/resources/extensions/sf/auto/session.js
+++ b/src/resources/extensions/sf/auto/session.js
@@ -15,6 +15,8 @@
  * auto-session-encapsulation.test.ts enforce that auto.ts has no module-level
  * `let` or `var` declarations.
  */
+
+import { emitJournalEvent } from "../journal.js";
 import {
 	buildModeState,
 	resolveModelMode,
@@ -22,6 +24,7 @@ import {
 	resolveRunControlMode,
 	resolveWorkMode,
 } from "../operating-model.js";
+import { loadSessionModeState, saveSessionModeState } from "../sf-db.js";
 
 // ─── Constants ───────────────────────────────────────────────────────────────
 export const MAX_UNIT_DISPATCHES = 3;
@@ -42,7 +45,64 @@ export function resetAutoSession() {
 	_autoSessionInstance = null;
 }
 // ─── AutoSession ─────────────────────────────────────────────────────────────
+
+/**
+ * Update terminal/tmux window title with current mode state.
+ *
+ * Purpose: make SF mode visible in terminal tabs, tmux window names,
+ * and task switchers without requiring TUI focus.
+ *
+ * Consumer: AutoSession.setMode() after every mode transition.
+ */
+function updateTerminalTitle(mode) {
+	if (
+		typeof process === "undefined" ||
+		!process.stdout ||
+		!process.stdout.isTTY
+	)
+		return;
+	const rcBadge =
+		mode.runControl === "autonomous" ? "∞" : mode.runControl.slice(0, 1);
+	const badge =
+		mode.workMode +
+		"|" +
+		rcBadge +
+		"|" +
+		mode.permissionProfile.slice(0, 1) +
+		"|" +
+		mode.modelMode.slice(0, 1);
+	const title = "SF[" + badge + "]";
+	// OSC 0: set icon name + window title
+	process.stdout.write("\x1b]0;" + title + "\x07");
+	// Also set process.title for OS task switchers
+	process.title = title;
+}
+
 export class AutoSession {
+	constructor() {
+		// Try to load persisted mode state from DB
+		this._loadPersistedModeState();
+	}
+
+	_loadPersistedModeState() {
+		try {
+			const persisted = loadSessionModeState();
+			if (persisted) {
+				this.workMode = resolveWorkMode(persisted.workMode);
+				this.runControl = resolveRunControlMode(persisted.runControl);
+				this.stepMode = this.runControl === "assisted";
+				this.permissionProfile = resolvePermissionProfile(
+					persisted.permissionProfile,
+				);
+				this.modelMode = resolveModelMode(persisted.modelMode);
+				this.surface = persisted.surface ?? "tui";
+				this.modeUpdatedAt = persisted.updatedAt;
+			}
+		} catch {
+			// DB may not be open yet — use defaults
+		}
+	}
+
 	// ── Lifecycle ────────────────────────────────────────────────────────────
 	active = false;
 	paused = false;
@@ -62,6 +122,12 @@ export class AutoSession {
 	 * Defaults to "chat" for new sessions.
 	 */
 	workMode = "chat";
+	/**
+	 * Requested run control: manual | assisted | autonomous.
+	 * Active loop state still comes from `active` + `stepMode`; this field keeps
+	 * command/UI mode state separate from whether a process is currently running.
+	 */
+	runControl = "manual";
 	/**
 	 * Current permission profile: restricted | normal | trusted | unrestricted.
 	 * Defaults to "restricted" for safety.
@@ -259,6 +325,7 @@ export class AutoSession {
 		this.cmdCtx = null;
 		// Mode state
 		this.workMode = "chat";
+		this.runControl = "manual";
 		this.permissionProfile = "restricted";
 		this.modelMode = "smart";
 		this.surface = "tui";
@@ -349,11 +416,14 @@ export class AutoSession {
 		permissionProfile,
 		modelMode,
 		surface,
+		reason = "user-command",
+		scope = "now",
 	} = {}) {
 		const prev = this.getMode();
 		if (workMode !== undefined) this.workMode = resolveWorkMode(workMode);
 		if (runControl !== undefined) {
 			const mode = resolveRunControlMode(runControl);
+			this.runControl = mode;
 			this.stepMode = mode === "assisted";
 		}
 		if (permissionProfile !== undefined) {
@@ -362,7 +432,40 @@ export class AutoSession {
 		if (modelMode !== undefined) this.modelMode = resolveModelMode(modelMode);
 		if (surface !== undefined) this.surface = surface;
 		this.modeUpdatedAt = new Date().toISOString();
-		return { from: prev, to: this.getMode() };
+		const next = this.getMode();
+		// Persist mode state to DB for durability across sessions
+		if (this.basePath) {
+			try {
+				saveSessionModeState({ ...next, updatedAt: this.modeUpdatedAt });
+			} catch {
+				// DB persistence is best-effort; don't fail mode transition
+			}
+			// Log mode transition to journal for audit trail
+			try {
+				emitJournalEvent(this.basePath, {
+					ts: this.modeUpdatedAt,
+					flowId: `mode-transition:${Date.now()}`,
+					seq: 0,
+					eventType: "mode-transition",
+					rule: "session-mode-change",
+					data: {
+						from: prev,
+						to: next,
+						reason,
+						scope,
+					},
+				});
+			} catch {
+				// Journal is best-effort; don't fail mode transition
+			}
+		}
+		// Update terminal title with mode state for tmux/terminal visibility
+		try {
+			updateTerminalTitle(next);
+		} catch {
+			// Title update is best-effort
+		}
+		return { from: prev, to: next };
 	}
 	/**
 	 * Get current mode state as a canonical object.
@@ -370,7 +473,11 @@ export class AutoSession {
 	getMode() {
 		return buildModeState({
 			workMode: this.workMode,
-			runControl: this.stepMode ? "assisted" : this.active ? "autonomous" : "manual",
+			runControl: this.active
+				? this.stepMode
+					? "assisted"
+					: "autonomous"
+				: this.runControl,
 			permissionProfile: this.permissionProfile,
 			modelMode: this.modelMode,
 			surface: this.surface,
diff --git a/src/resources/extensions/sf/bootstrap/register-hooks.js b/src/resources/extensions/sf/bootstrap/register-hooks.js
index 4cc492131..c4a014c2f 100644
--- a/src/resources/extensions/sf/bootstrap/register-hooks.js
+++ b/src/resources/extensions/sf/bootstrap/register-hooks.js
@@ -22,7 +22,10 @@ import { recordToolCallName } from "../auto-tool-tracking.js";
 import { loadToolApiKeys } from "../commands-config.js";
 import { getEcosystemReadyPromise } from "../ecosystem/loader.js";
 import { updateSnapshot } from "../ecosystem/sf-extension-api.js";
-import { buildExecutionPolicyJournalEntry } from "../execution-policy.js";
+import {
+	buildExecutionPolicyJournalEntry,
+	classifyExecutionPolicyCall,
+} from "../execution-policy.js";
 import { formatContinue, loadFile, saveFile } from "../files.js";
 import { getDiscussionMilestoneId } from "../guided-flow.js";
 import { initHealthWidget } from "../health-widget.js";
@@ -592,6 +595,32 @@ export function registerHooks(pi, ecosystemHandlers = []) {
 			);
 			if (queueGuard.block) return queueGuard;
 		}
+		// ── Execution policy enforcement: block based on permission profile ──
+		// When autonomous mode is active, enforce the session's permission profile
+		// at the tool boundary. This is the enforcement layer that makes
+		// /trust restricted|normal|trusted|unrestricted meaningful.
+		if (isAutoActive()) {
+			const { getAutoSession } = await import("../auto/session.js");
+			const session = getAutoSession();
+			const profile = session?.permissionProfile ?? "normal";
+			const input =
+				event.toolName === "bash"
+					? (event.input?.command ?? "")
+					: event.toolName === "write" || event.toolName === "edit"
+						? (event.input?.path ?? "")
+						: "";
+			const decision = classifyExecutionPolicyCall(
+				profile,
+				event.toolName,
+				input,
+			);
+			if (!decision.allowed) {
+				return {
+					block: true,
+					reason: `Execution policy block: ${decision.reason} (profile: ${profile}, tool: ${event.toolName})`,
+				};
+			}
+		}
 		// ── Single-writer engine: block direct writes to STATE.md ──────────
 		// Covers write, edit, and bash tools to prevent bypass vectors.
 		if (isToolCallEventType("write", event)) {
@@ -648,7 +677,12 @@ export function registerHooks(pi, ecosystemHandlers = []) {
 		if (!isAutoActive()) return;
 		safetyRecordToolCall(event.toolCallId, event.toolName, event.input);
 		const policyDash = getAutoDashboardData();
-		const policyProfile = isQueuePhaseActive() ? "restricted" : "normal";
+		// Use session permission profile if available, fall back to queue-aware default
+		const { getAutoSession } = await import("../auto/session.js");
+		const session = getAutoSession();
+		const sessionProfile = session?.permissionProfile;
+		const policyProfile =
+			sessionProfile ?? (isQueuePhaseActive() ? "restricted" : "normal");
 		if (policyDash.basePath) {
 			emitJournalEvent(
 				policyDash.basePath,
diff --git a/src/resources/extensions/sf/commands-handlers.js b/src/resources/extensions/sf/commands-handlers.js
index 8c54713dd..c66e52321 100644
--- a/src/resources/extensions/sf/commands-handlers.js
+++ b/src/resources/extensions/sf/commands-handlers.js
@@ -422,6 +422,54 @@ export async function handleTriage(args, ctx, pi, basePath) {
 }
 export async function handleSteer(change, ctx, pi) {
 	const basePath = process.cwd();
+	const trimmed = change.trim();
+
+	// ── Mode steering: /steer mode <workMode> [scope] ──────────────────────
+	const modeSteerRe = /^mode\s+(\S+)(?:\s+(\S+))?/;
+	const modeMatch = trimmed.match(modeSteerRe);
+	if (modeMatch) {
+		const workMode = modeMatch[1];
+		const scope = modeMatch[2] ?? "after-current-unit";
+		const s = getAutoSession();
+		const transition = s.setMode({ workMode, reason: "steer", scope });
+		ctx.ui.notify(
+			`Steer mode: ${transition.from.workMode} → ${transition.to.workMode} (${scope})`,
+			"info",
+		);
+		return;
+	}
+
+	// ── Trust steering: /steer trust <profile> [scope] ─────────────────────
+	const trustSteerRe = /^trust\s+(\S+)(?:\s+(\S+))?/;
+	const trustMatch = trimmed.match(trustSteerRe);
+	if (trustMatch) {
+		const permissionProfile = trustMatch[1];
+		const scope = trustMatch[2] ?? "now";
+		const s = getAutoSession();
+		const transition = s.setMode({ permissionProfile, reason: "steer", scope });
+		ctx.ui.notify(
+			`Steer trust: ${transition.from.permissionProfile} → ${transition.to.permissionProfile} (${scope})`,
+			"info",
+		);
+		return;
+	}
+
+	// ── Model-mode steering: /steer model-mode <mode> [scope] ──────────────
+	const modelModeSteerRe = /^model-mode\s+(\S+)(?:\s+(\S+))?/;
+	const modelModeMatch = trimmed.match(modelModeSteerRe);
+	if (modelModeMatch) {
+		const modelMode = modelModeMatch[1];
+		const scope = modelModeMatch[2] ?? "for-next-unit";
+		const s = getAutoSession();
+		const transition = s.setMode({ modelMode, reason: "steer", scope });
+		ctx.ui.notify(
+			`Steer model-mode: ${transition.from.modelMode} → ${transition.to.modelMode} (${scope})`,
+			"info",
+		);
+		return;
+	}
+
+	// ── Legacy text override ───────────────────────────────────────────────
 	const state = await deriveState(basePath);
 	const mid = state.activeMilestone?.id ?? "none";
 	const sid = state.activeSlice?.id ?? "none";
diff --git a/src/resources/extensions/sf/commands/catalog.js b/src/resources/extensions/sf/commands/catalog.js
index 960478bea..195a2c19a 100644
--- a/src/resources/extensions/sf/commands/catalog.js
+++ b/src/resources/extensions/sf/commands/catalog.js
@@ -12,7 +12,7 @@ const sfHome = process.env.SF_HOME || join(homedir(), ".sf");
  * Comprehensive description of all available SF commands for help text.
  */
 export const SF_COMMAND_DESCRIPTION =
-	"SF — Singularity Forge: /help|start|templates|next|autonomous|pause|status|widget|visualize|queue|quick|discuss|capture|triage|todo|dispatch|history|undo|undo-task|reset-slice|rate|skip|cleanup|mode|show-config|prefs|config|keys|hooks|run-hook|skill-health|doctor|uok|logs|forensics|migrate|remote|steer|knowledge|harness|solver-eval|new-milestone|parallel|cmux|park|unpark|init|setup|inspect|extensions|update|fast|mcp|rethink|codebase|notifications|ship|do|session-report|backlog|pr-branch|add-tests|scan|scaffold|extract-learnings|eval-review|plan";
+	"SF — Singularity Forge: /help|start|templates|next|autonomous|pause|status|widget|visualize|queue|quick|discuss|capture|triage|todo|dispatch|history|undo|undo-task|reset-slice|rate|skip|cleanup|mode|control|trust|model-mode|show-config|prefs|config|keys|hooks|run-hook|skill-health|doctor|uok|logs|forensics|migrate|remote|steer|knowledge|harness|solver-eval|new-milestone|parallel|cmux|park|unpark|init|setup|inspect|extensions|update|fast|mcp|rethink|codebase|notifications|ship|do|session-report|backlog|pr-branch|add-tests|scan|scaffold|extract-learnings|eval-review|plan";
 
 export const BASE_RUNTIME_COMMANDS = new Set([
 	"settings",
@@ -102,7 +102,16 @@ export const TOP_LEVEL_SUBCOMMANDS = [
 		desc: "Manage worktrees from the TUI (list, merge, clean, remove)",
 	},
 	{ cmd: "model", desc: "Switch the active session model or open a picker" },
-	{ cmd: "mode", desc: "Switch workflow mode (solo/team)" },
+	{
+		cmd: "mode",
+		desc: "Switch work mode (chat/plan/build/review/repair/research) or workflow mode (solo/team)",
+	},
+	{ cmd: "control", desc: "Switch run control (manual/assisted/autonomous)" },
+	{
+		cmd: "trust",
+		desc: "Switch permission profile (restricted/normal/trusted/unrestricted)",
+	},
+	{ cmd: "model-mode", desc: "Switch model mode (fast/smart/deep)" },
 	{
 		cmd: "show-config",
 		desc: "Show effective configuration (models, routing, toggles)",
@@ -124,6 +133,12 @@ export const TOP_LEVEL_SUBCOMMANDS = [
 		desc: "View, filter, and clear persistent notification history",
 	},
 	{ cmd: "doctor", desc: "Runtime health checks with auto-fix" },
+	{
+		cmd: "repair",
+		desc: "Switch to repair work mode and run diagnostics [--autonomous]",
+	},
+	{ cmd: "tasks", desc: "Background work surface — units, workers, budget" },
+	{ cmd: "skills", desc: "List discovered skills from .agents/skills/" },
 	{
 		cmd: "uok",
 		desc: "UOK runtime health: ledger, last run, last error, startup gate, gate metrics",
diff --git a/src/resources/extensions/sf/commands/handlers/core.js b/src/resources/extensions/sf/commands/handlers/core.js
index 6b0bb7953..f392b5253 100644
--- a/src/resources/extensions/sf/commands/handlers/core.js
+++ b/src/resources/extensions/sf/commands/handlers/core.js
@@ -1,4 +1,5 @@
 import { join } from "node:path";
+import { getAutoSession } from "../../auto/session.js";
 import { handleCmux } from "../../commands-cmux.js";
 import {
 	ensurePreferencesFile,
@@ -34,11 +35,15 @@ export function showHelp(ctx, args = "") {
 		`  /status         Dashboard  (${formattedShortcutPair("dashboard")})`,
 		`  /parallel watch Parallel monitor  (${formattedShortcutPair("parallel")})`,
 		`  /notifications  Notification history  (${formattedShortcutPair("notifications")})`,
+		"  /tasks          Background work surface — units, workers, budget",
 		"  /visualize      Interactive 10-tab TUI",
 		"  /queue          Show queued/dispatched units",
 		"",
 		"COURSE CORRECTION",
 		"  /steer <desc>   Apply user override to active work",
+		"  /steer mode <m> [scope]  Change work mode (now|after-current-unit|next-milestone)",
+		"  /steer trust <p> [scope] Change permission profile",
+		"  /steer model-mode <m>    Change model mode for next unit",
 		"  /capture <text> Quick-capture a thought to CAPTURES.md",
 		"  /triage         Classify and route pending captures",
 		"  /undo           Revert last completed unit  [--force]",
@@ -51,6 +56,9 @@ export function showHelp(ctx, args = "") {
 		"  /model          Switch active session model",
 		"  /prefs          Manage preferences",
 		"  /doctor         Diagnose and repair .sf/ state",
+		"  /repair         Switch to repair work mode and run diagnostics",
+		"  /tasks          Background work surface",
+		"  /skills         List discovered skills",
 		"",
 		"Use /help all for the complete command reference.",
 	];
@@ -72,6 +80,7 @@ export function showHelp(ctx, args = "") {
 		`  /parallel watch Open parallel worker monitor  (${formattedShortcutPair("parallel")})`,
 		"  /visualize      Interactive 10-tab TUI (progress, timeline, deps, metrics, health, agent, changes, knowledge, captures, export)",
 		"  /queue          Show queued/dispatched units and execution order",
+		"  /tasks          Background work surface — units, workers, budget, checkpoints",
 		"  /history        View execution history  [--cost] [--phase] [--model] [N]",
 		"  /changelog      Show categorized release notes  [version]",
 		`  /notifications  View persistent notification history  [clear|tail|filter]  (${formattedShortcutPair("notifications")})`,
@@ -99,7 +108,10 @@ export function showHelp(ctx, args = "") {
 		"  /init           Project init wizard — detect, configure, bootstrap .sf/",
 		"  /setup          Global setup status  [llm|search|remote|keys|prefs]",
 		"  /model          Switch active session model  [provider/model|model-id]",
-		"  /mode           Set workflow mode (solo/team)  [global|project]",
+		"  /mode           Switch work mode (chat/plan/build/review/repair/research)",
+		"  /control        Switch run control (manual/assisted/autonomous)",
+		"  /trust          Switch permission profile (restricted/normal/trusted/unrestricted)",
+		"  /model-mode     Switch model mode (fast/smart/deep)",
 		"  /prefs          Manage preferences  [global|project|status|wizard|setup|import-claude]",
 		"  /cmux           Manage cmux integration  [status|on|off|notifications|sidebar|splits|browser]",
 		"  /config         Set API keys for external tools",
@@ -112,6 +124,10 @@ export function showHelp(ctx, args = "") {
 		"",
 		"MAINTENANCE",
 		"  /doctor         Diagnose and repair .sf/ state  [audit|fix|heal] [scope]",
+		"  /repair         Switch to repair work mode and run diagnostics  [--autonomous]",
+		"  /tasks          Background work surface  [--refresh|--failed|--cancelled|--all]",
+		"  /skills         List discovered skills from .agents/skills/",
+		"  /skills --eval <name>  Run eval cases for a skill",
 		"  /reload         Snapshot & reload agent, resume same session",
 		"  /export         Export milestone/slice results  [--json|--markdown|--html] [--all]",
 		"  /cleanup        Remove merged branches or snapshots  [branches|snapshots]",
@@ -382,6 +398,71 @@ async function handleModel(trimmedArgs, ctx, pi) {
 	}
 	ctx.ui.notify(`Model: ${targetModel.provider}/${targetModel.id}`, "info");
 }
+function formatModeState(mode) {
+	if (!mode) return "SF — mode unknown";
+	return `SF — ${mode.workMode} · ${mode.runControl} · ${mode.permissionProfile} · ${mode.modelMode}`;
+}
+function handleModeCommand(args, ctx) {
+	const s = getAutoSession();
+	const parts = args.trim().split(/\s+/).filter(Boolean);
+	if (parts.length === 0) {
+		const mode = s.getMode();
+		ctx.ui.notify(formatModeState(mode), "info");
+		return true;
+	}
+	const workMode = parts[0];
+	const transition = s.setMode({ workMode });
+	ctx.ui.notify(
+		`Mode: ${transition.from.workMode} → ${transition.to.workMode}`,
+		"info",
+	);
+	return true;
+}
+function handleControlCommand(args, ctx) {
+	const s = getAutoSession();
+	const runControl = args.trim();
+	if (!runControl) {
+		const mode = s.getMode();
+		ctx.ui.notify(`Run control: ${mode.runControl}`, "info");
+		return true;
+	}
+	const transition = s.setMode({ runControl });
+	ctx.ui.notify(
+		`Run control: ${transition.from.runControl} → ${transition.to.runControl}`,
+		"info",
+	);
+	return true;
+}
+function handleTrustCommand(args, ctx) {
+	const s = getAutoSession();
+	const permissionProfile = args.trim();
+	if (!permissionProfile) {
+		const mode = s.getMode();
+		ctx.ui.notify(`Trust: ${mode.permissionProfile}`, "info");
+		return true;
+	}
+	const transition = s.setMode({ permissionProfile });
+	ctx.ui.notify(
+		`Trust: ${transition.from.permissionProfile} → ${transition.to.permissionProfile}`,
+		"info",
+	);
+	return true;
+}
+function handleModelModeCommand(args, ctx) {
+	const s = getAutoSession();
+	const modelMode = args.trim();
+	if (!modelMode) {
+		const mode = s.getMode();
+		ctx.ui.notify(`Model mode: ${mode.modelMode}`, "info");
+		return true;
+	}
+	const transition = s.setMode({ modelMode });
+	ctx.ui.notify(
+		`Model mode: ${transition.from.modelMode} → ${transition.to.modelMode}`,
+		"info",
+	);
+	return true;
+}
 export async function handleCoreCommand(trimmed, ctx, pi) {
 	if (
 		trimmed === "help" ||
@@ -419,6 +500,13 @@ export async function handleCoreCommand(trimmed, ctx, pi) {
 	}
 	if (trimmed === "mode" || trimmed.startsWith("mode ")) {
 		const modeArgs = trimmed.replace(/^mode\s*/, "").trim();
+		// If arg is a work mode (chat/plan/build/review/repair/research), use new mode system
+		const workModes = ["chat", "plan", "build", "review", "repair", "research"];
+		if (workModes.includes(modeArgs)) {
+			handleModeCommand(modeArgs, ctx);
+			return true;
+		}
+		// Otherwise fall back to old prefs mode (solo/team)
 		const scope = modeArgs === "project" ? "project" : "global";
 		const path =
 			scope === "project"
@@ -428,10 +516,27 @@ export async function handleCoreCommand(trimmed, ctx, pi) {
 		await handlePrefsMode(ctx, scope);
 		return true;
 	}
+	if (trimmed === "control" || trimmed.startsWith("control ")) {
+		handleControlCommand(trimmed.replace(/^control\s*/, "").trim(), ctx);
+		return true;
+	}
+	if (trimmed === "trust" || trimmed.startsWith("trust ")) {
+		handleTrustCommand(trimmed.replace(/^trust\s*/, "").trim(), ctx);
+		return true;
+	}
+	if (trimmed === "model-mode" || trimmed.startsWith("model-mode ")) {
+		handleModelModeCommand(trimmed.replace(/^model-mode\s*/, "").trim(), ctx);
+		return true;
+	}
 	if (trimmed === "prefs" || trimmed.startsWith("prefs ")) {
 		await handlePrefs(trimmed.replace(/^prefs\s*/, "").trim(), ctx);
 		return true;
 	}
+	if (trimmed === "tasks" || trimmed.startsWith("tasks ")) {
+		const { handleTasks } = await import("./tasks.js");
+		await handleTasks(trimmed.replace(/^tasks\s*/, "").trim(), ctx);
+		return true;
+	}
 	if (trimmed === "cmux" || trimmed.startsWith("cmux ")) {
 		await handleCmux(trimmed.replace(/^cmux\s*/, "").trim(), ctx);
 		return true;
@@ -504,6 +609,143 @@ export async function handleCoreCommand(trimmed, ctx, pi) {
 		process.exit(EXIT_RELOAD);
 		return true;
 	}
+	if (trimmed === "skills" || trimmed.startsWith("skills ")) {
+		const args = trimmed.replace(/^skills\s*/, "").trim();
+		// Auto-create mode: detect patterns and generate skills
+		if (args === "--auto-create" || args === "-a") {
+			const {
+				detectSkillCandidates,
+				isAutoSkillCreationAllowed,
+				generateSkill,
+			} = await import("../../skills/auto-create.js");
+			const basePath = projectRoot();
+			if (!isAutoSkillCreationAllowed(basePath)) {
+				ctx.ui.notify(
+					"Auto-skill creation is on cooldown. Try again tomorrow.",
+					"warning",
+				);
+				return true;
+			}
+			const candidates = detectSkillCandidates(basePath);
+			if (candidates.length === 0) {
+				ctx.ui.notify(
+					"No skill candidates detected. Need ≥3 occurrences of a pattern.",
+					"info",
+				);
+				return true;
+			}
+			const lines = ["SF Skill Auto-Creation\n"];
+			lines.push(`Detected ${candidates.length} candidate(s):\n`);
+			for (const c of candidates) {
+				lines.push(`  ${c.name} — ${c.description}`);
+			}
+			// Generate the first candidate
+			const result = generateSkill(candidates[0], basePath);
+			if (result.created) {
+				lines.push(`\n✓ Created: ${result.path}`);
+				ctx.ui.notify(lines.join("\n"), "info");
+			} else {
+				lines.push(`\n✗ Skipped: ${result.reason}`);
+				ctx.ui.notify(lines.join("\n"), "warning");
+			}
+			return true;
+		}
+		// Eval mode: run eval cases for a skill
+		if (args.startsWith("--eval ") || args.startsWith("-e ")) {
+			const skillName = args
+				.replace(/^--eval\s+/, "")
+				.replace(/^-e\s+/, "")
+				.trim();
+			const { loadSkills, runSkillEvals } = await import(
+				"../../skills/loader.js"
+			);
+			const { createEvalCase, generateDefaultEvalCase } = await import(
+				"../../skills/eval-harness.js"
+			);
+			const skills = loadSkills(projectRoot());
+			const skill = skills.find((s) => s.name === skillName);
+			if (!skill) {
+				ctx.ui.notify(`Skill "${skillName}" not found.`, "warning");
+				return true;
+			}
+			// Auto-create default eval case if none exist
+			const evalDir = join(
+				skill.basePath ?? projectRoot(),
+				".agents",
+				"skills",
+				skillName,
+				"evals",
+			);
+			const { existsSync } = await import("node:fs");
+			if (!existsSync(evalDir)) {
+				const caseDef = generateDefaultEvalCase(skill);
+				createEvalCase(
+					join(skill.basePath ?? projectRoot(), ".agents", "skills", skillName),
+					"default",
+					caseDef,
+				);
+				ctx.ui.notify(
+					`Created default eval case for ${skillName}. Run /skills --eval ${skillName} again.`,
+					"info",
+				);
+				return true;
+			}
+			const result = await runSkillEvals(
+				join(skill.basePath ?? projectRoot(), ".agents", "skills", skillName),
+				ctx,
+			);
+			const lines = [`SF Skill Eval: ${skillName}\n`];
+			lines.push(
+				`Cases: ${result.passedCases}/${result.totalCases} passed · Score: ${(result.totalScore * 100).toFixed(0)}%\n`,
+			);
+			for (const c of result.cases) {
+				const icon = c.passed ? "✓" : "✗";
+				lines.push(`${icon} ${c.caseName}: ${c.score.toFixed(2)}`);
+				for (const d of c.details.slice(0, 3)) {
+					lines.push(`    ${d}`);
+				}
+			}
+			ctx.ui.notify(
+				lines.join("\n"),
+				result.passedCases === result.totalCases ? "success" : "warning",
+			);
+			return true;
+		}
+		// Normal list mode
+		const { loadSkills, getPermittedSkills, getModelInvocableSkills } =
+			await import("../../skills/loader.js");
+		const skills = loadSkills(projectRoot());
+		const mode = getAutoSession().getMode();
+		const permitted = getPermittedSkills(skills, mode.permissionProfile);
+		const modelInvocable = getModelInvocableSkills(skills, mode.workMode);
+
+		const lines = ["SF Skills\n"];
+		lines.push(
+			`Found ${skills.length} skill(s) · ${permitted.length} permitted · ${modelInvocable.length} model-invocable\n`,
+		);
+
+		for (const skill of skills) {
+			const icon = skill.valid ? "✓" : "✗";
+			const user = skill.userInvocable ? "U" : " ";
+			const model = skill.modelInvocable ? "M" : " ";
+			lines.push(
+				`${icon} [${user}${model}] ${skill.name} · ${skill.permissionProfile} · ${skill.sideEffects}`,
+			);
+			if (!skill.valid && skill.errors?.length > 0) {
+				lines.push(`    ⚠ ${skill.errors[0]}`);
+			}
+		}
+
+		lines.push(
+			"\nLegend: [U]ser-invocable [M]odel-invocable · R=restricted N=normal T=trusted",
+		);
+		lines.push(
+			"Use /skills --auto-create to detect and generate skills from activity patterns.",
+		);
+		lines.push("Use /skills --eval <name> to run eval cases for a skill.");
+		ctx.ui.notify(lines.join("\n"), "info");
+		return true;
+	}
 	return false;
 }
 export function formatTextStatus(state) {
diff --git a/src/resources/extensions/sf/commands/handlers/ops.js b/src/resources/extensions/sf/commands/handlers/ops.js
index 03c7f9713..2f36ab336 100644
--- a/src/resources/extensions/sf/commands/handlers/ops.js
+++ b/src/resources/extensions/sf/commands/handlers/ops.js
@@ -1,4 +1,5 @@
 import { handleRemote } from "../../../remote-questions/mod.js";
+import { getAutoSession } from "../../auto/session.js";
 import { dispatchDirectPhase } from "../../auto-direct-dispatch.js";
 import { handleConfig } from "../../commands-config.js";
 import { handleDebug } from "../../commands-debug.js";
@@ -55,6 +56,27 @@ export async function handleOpsCommand(trimmed, ctx, pi) {
 		await handleDoctor(trimmed.replace(/^doctor\s*/, "").trim(), ctx, pi);
 		return true;
 	}
+	if (trimmed === "repair" || trimmed.startsWith("repair ")) {
+		const s = getAutoSession();
+		const args = trimmed.replace(/^repair\s*/, "").trim();
+		// Switch to repair work mode
+		const transition = s.setMode({ workMode: "repair" });
+		ctx.ui.notify(
+			`Repair mode: ${transition.from.workMode} → ${transition.to.workMode}`,
+			"info",
+		);
+		// If --autonomous flag, also set run control
+		if (args.includes("--autonomous")) {
+			s.setMode({ runControl: "autonomous" });
+			ctx.ui.notify(
+				"Repair will run autonomously until clean or blocked.",
+				"info",
+			);
+		}
+		// Run doctor diagnostics
+		await handleDoctor("fix", ctx, pi);
+		return true;
+	}
 	if (trimmed === "uok" || trimmed.startsWith("uok ")) {
 		const { handleUok } = await import("../../commands-uok.js");
 		await handleUok(trimmed.replace(/^uok\s*/, "").trim(), ctx);
diff --git a/src/resources/extensions/sf/commands/handlers/tasks.js b/src/resources/extensions/sf/commands/handlers/tasks.js
new file mode 100644
index 000000000..204a705ee
--- /dev/null
+++ b/src/resources/extensions/sf/commands/handlers/tasks.js
@@ -0,0 +1,292 @@
+/**
+ * /tasks command — unified background work surface
+ *
+ * Purpose: show all background work in one view: autonomous units, parallel
+ * workers, scheduled dispatches, shell sessions, stuck sessions, remote
+ * questions, cost/budget state, and last checkpoint.
+ *
+ * Consumer: human users via TUI and web chat mode.
+ */
+import { closeDatabase, getDatabase, isDbAvailable } from "../../sf-db.js";
+import { projectRoot } from "../context.js";
+
+const TASK_STATE_ICONS = {
+	todo: "○",
+	in_progress: "◐",
+	review: "◑",
+	done: "●",
+	retrying: "↻",
+	failed: "✗",
+	cancelled: "⊘",
+};
+
+const TASK_STATE_LABELS = {
+	todo: "todo",
+	in_progress: "running",
+	review: "review",
+	done: "done",
+	retrying: "retrying",
+	failed: "failed",
+	cancelled: "cancelled",
+};
+
+/**
+ * Format a task row for display.
+ */
+function formatTaskLine(task, width = 80) {
+	const icon = TASK_STATE_ICONS[task.state] ?? "?";
+	const label = TASK_STATE_LABELS[task.state] ?? task.state;
+	const id = task.unitId ?? task.id ?? "unknown";
+	const title = task.title ?? id;
+	const duration = task.durationMs
+		? ` ${Math.round(task.durationMs / 1000)}s`
+		: "";
+	const cost = task.costUsd ? ` $${task.costUsd.toFixed(2)}` : "";
+	const worker = task.workerId ? ` [${task.workerId}]` : "";
+
+	const right = `${duration}${cost}${worker}`;
+	const maxTitle = Math.max(20, width - right.length - 8);
+	const truncated =
+		title.length > maxTitle ? title.slice(0, maxTitle - 1) + "…" : title;
+
+	return `${icon} ${truncated.padEnd(maxTitle)} ${label.padStart(10)}${right}`;
+}
+
+/**
+ * Build the /tasks text report.
+ */
+function buildTasksReport({ tasks, summary, mode, budget }) {
+	const lines = [];
+	const now = new Date().toISOString().slice(0, 19).replace("T", " ");
+
+	lines.push(`SF Background Work  ${now}`);
+	lines.push("");
+
+	// Mode line
+	if (mode) {
+		lines.push(
+			`Mode: ${mode.workMode} · ${mode.runControl} · ${mode.permissionProfile} · ${mode.modelMode}`,
+		);
+		lines.push("");
+	}
+
+	// Summary
+	if (summary) {
+		const { counts, total, progress, isComplete } = summary;
+		const parts = [];
+		for (const [state, count] of Object.entries(counts)) {
+			if (count > 0) parts.push(`${TASK_STATE_ICONS[state] ?? "?"}${count}`);
+		}
+		lines.push(
+			`Tasks: ${total} total · ${parts.join(" ")} · ${progress}% complete${isComplete ? " ✓" : ""}`,
+		);
+		lines.push("");
+	}
+
+	// Active tasks
+	const active = tasks.filter((t) =>
+		["in_progress", "review", "retrying"].includes(t.state),
+	);
+	if (active.length > 0) {
+		lines.push(`Active (${active.length}):`);
+		for (const task of active) {
+			lines.push(`  ${formatTaskLine(task)}`);
+		}
+		lines.push("");
+	}
+
+	// Pending
+	const pending = tasks.filter((t) => t.state === "todo");
+	if (pending.length > 0) {
+		lines.push(`Pending (${pending.length}):`);
+		for (const task of pending.slice(0, 10)) {
+			lines.push(`  ${formatTaskLine(task)}`);
+		}
+		if (pending.length > 10) {
+			lines.push(`  … and ${pending.length - 10} more`);
+		}
+		lines.push("");
+	}
+
+	// Terminal
+	const terminal = tasks.filter((t) =>
+		["done", "failed", "cancelled"].includes(t.state),
+	);
+	if (terminal.length > 0) {
+		const recent = terminal
+			.sort((a, b) => (b.endedAt ?? "").localeCompare(a.endedAt ?? ""))
+			.slice(0, 5);
+		lines.push(`Recently finished (${terminal.length} total):`);
+		for (const task of recent) {
+			lines.push(`  ${formatTaskLine(task)}`);
+		}
+		lines.push("");
+	}
+
+	// Budget
+	if (budget) {
+		lines.push(
+			`Budget: $${budget.spent.toFixed(2)} / $${budget.ceiling.toFixed(2)} · ${budget.remaining > 0 ? "$" + budget.remaining.toFixed(2) + " remaining" : "OVER BUDGET"}`,
+		);
+		lines.push("");
+	}
+
+	lines.push("Commands: /tasks --refresh  /tasks --failed  /tasks --cancelled");
+	return lines.join("\n");
+}
+
+function normalizeDbTask(row) {
+	return {
+		...row,
+		id: row.id,
+		unitId: row.unitId ?? row.unit_id ?? row.id,
+		milestoneId: row.milestoneId ?? row.milestone_id ?? null,
+		sliceId: row.sliceId ?? row.slice_id ?? null,
+		taskId: row.taskId ?? row.task_id ?? null,
+		title: row.title ?? row.unit_id ?? row.id,
+		state: row.state ?? row.status ?? "todo",
+		workerId: row.workerId ?? row.worker_id ?? null,
+		durationMs: row.durationMs ?? row.duration_ms ?? null,
+		costUsd: row.costUsd ?? row.cost_usd ?? null,
+		endedAt: row.endedAt ?? row.ended_at ?? null,
+		graphId: row.graphId ?? row.graph_id ?? null,
+	};
+}
+
+/**
+ * Handle the /tasks command.
+ */
+export async function handleTasks(args, ctx) {
+	const basePath = projectRoot();
+	const trimmed = args.trim().toLowerCase();
+
+	// Parse flags
+	const refresh = trimmed.includes("--refresh") || trimmed.includes("-r");
+	const showFailed = trimmed.includes("--failed") || trimmed.includes("-f");
+	const showCancelled =
+		trimmed.includes("--cancelled") || trimmed.includes("-c");
+	const showAll = trimmed.includes("--all") || trimmed.includes("-a");
+
+	const { ensureDbOpen } = await import("../../bootstrap/dynamic-tools.js");
+	if (refresh) closeDatabase();
+	await ensureDbOpen();
+
+	// Get mode state
+	const { getAutoSession } = await import("../../auto/session.js");
+	const session = getAutoSession();
+	const mode = session?.getMode?.() ?? null;
+
+	// Try DB-backed task query first
+	let tasks = [];
+	let summary = null;
+
+	if (isDbAvailable()) {
+		try {
+			const db = getDatabase();
+			if (!db) throw new Error("SF database is not open");
+			const { queryTasksByState, getGraphStateSummary } = await import(
+				"../../uok/execution-graph-persist.js"
+			);
+
+			let states = [];
+			if (showFailed) states = ["failed"];
+			else if (showCancelled) states = ["cancelled"];
+			else if (!showAll) states = ["todo", "in_progress", "review", "retrying"];
+
+			tasks = queryTasksByState(db, { states, limit: 200 }).map(
+				normalizeDbTask,
+			);
+
+			// Get summary for active graphs
+			const graphIds = [
+				...new Set(tasks.map((t) => t.graphId).filter(Boolean)),
+			];
+			if (graphIds.length > 0) {
+				// Use first active graph for summary
+				summary = getGraphStateSummary(db, graphIds[0]);
+			}
+		} catch (err) {
+			// DB query failed — fall through to filesystem fallback
+			ctx.ui.notify(
+				`DB query failed: ${err.message}. Falling back to filesystem.`,
+				"warning",
+			);
+		}
+	}
+
+	// Fallback: derive from unit runtime files if no DB tasks
+	if (tasks.length === 0) {
+		try {
+			const { deriveState } = await import("../../state.js");
+			const state = await deriveState(basePath);
+
+			// Convert milestone registry entries to task records
+			for (const ms of state.registry) {
+				for (const slice of ms.slices ?? []) {
+					for (const task of slice.tasks ?? []) {
+						tasks.push({
+							id: `${ms.id}/${slice.id}/${task.id}`,
+							unitId: `${ms.id}/${slice.id}/${task.id}`,
+							milestoneId: ms.id,
+							sliceId: slice.id,
+							taskId: task.id,
+							title: task.title ?? task.id,
+							state: task.done
+								? "done"
+								: task.inProgress
+									? "in_progress"
+									: "todo",
+							updatedAt: task.updatedAt ?? null,
+						});
+					}
+				}
+			}
+
+			// Compute summary
+			const states = tasks.map((t) => t.state);
+			const counts = {};
+			for (const s of states) counts[s] = (counts[s] ?? 0) + 1;
+			const total = tasks.length;
+			const terminal = (counts.done ?? 0) + (counts.failed ?? 0);
+			summary = {
+				counts: {
+					todo: counts.todo ?? 0,
+					in_progress: counts.in_progress ?? 0,
+					review: counts.review ?? 0,
+					done: counts.done ?? 0,
+					retrying: counts.retrying ?? 0,
+					failed: counts.failed ?? 0,
+					cancelled: counts.cancelled ?? 0,
+				},
+				total,
+				terminal,
+				progress: total > 0 ? Math.round((terminal / total) * 100) : 0,
+				isComplete: terminal === total && total > 0,
+			};
+		} catch (err) {
+			ctx.ui.notify(`State derivation failed: ${err.message}`, "error");
+		}
+	}
+
+	// Budget estimate
+	let budget = null;
+	try {
+		const { getGlobalSFPreferences } = await import("../../preferences.js");
+		const prefs = getGlobalSFPreferences();
+		const ceiling = prefs.budget_ceiling ?? 0;
+		if (ceiling > 0) {
+			// Estimate spent from task costs
+			const spent = tasks.reduce((sum, t) => sum + (t.costUsd ?? 0), 0);
+			budget = {
+				spent,
+				ceiling,
+				remaining: ceiling - spent,
+			};
+		}
+	} catch {
+		// Budget info is best-effort
+	}
+
+	const report = buildTasksReport({ tasks, summary, mode, budget });
+	ctx.ui.notify(report, "info");
+}
diff --git a/src/resources/extensions/sf/complexity-classifier.js b/src/resources/extensions/sf/complexity-classifier.js
index 0869cfba1..dbb911c8a 100644
--- a/src/resources/extensions/sf/complexity-classifier.js
+++ b/src/resources/extensions/sf/complexity-classifier.js
@@ -3,6 +3,7 @@
 // Pure heuristics + adaptive learning — no LLM calls. Sub-millisecond classification.
 import { existsSync, readFileSync } from "node:fs";
 import { join } from "node:path";
+import { modelModeToTier } from "./operating-model.js";
 import { sfRoot } from "./paths.js";
 import { getAdaptiveTierAdjustment } from "./routing-history.js";
 import { parseUnitId } from "./unit-id.js";
@@ -37,6 +38,7 @@ const UNIT_TYPE_TIERS = {
  * @param basePath    Project base path (for reading task plans)
  * @param budgetPct   Current budget usage as fraction (0.0-1.0+), or undefined if no budget
  * @param metadata    Optional pre-parsed task metadata
+ * @param modelMode   Optional model mode override (fast/smart/deep) from session
  */
 export function classifyUnitComplexity(
 	unitType,
@@ -44,6 +46,7 @@ export function classifyUnitComplexity(
 	basePath,
 	budgetPct,
 	metadata,
+	modelMode,
 ) {
 	// Hook units default to light
 	if (unitType.startsWith("hook/")) {
@@ -86,6 +89,20 @@ export function classifyUnitComplexity(
 		reason = `${reason} (adaptive: high failure rate at ${tier})`;
 		tier = adaptiveAdjustment;
 	}
+	// Apply model mode floor: if user set /model-mode deep, don't go below heavy
+	// If user set /model-mode fast, cap at light
+	if (modelMode) {
+		const modeTier = modelModeToTier(modelMode);
+		const modeOrdinal = tierOrdinal(modeTier);
+		const currentOrdinal = tierOrdinal(tier);
+		if (modeOrdinal > currentOrdinal) {
+			tier = modeTier;
+			reason = `${reason} (model mode floor: ${modelMode})`;
+		} else if (modeOrdinal < currentOrdinal) {
+			tier = modeTier;
+			reason = `${reason} (model mode cap: ${modelMode})`;
+		}
+	}
 	const result = {
 		tier,
 		reason,
diff --git a/src/resources/extensions/sf/extension-manifest.json b/src/resources/extensions/sf/extension-manifest.json
index 2de411cc8..6854fda0c 100644
--- a/src/resources/extensions/sf/extension-manifest.json
+++ b/src/resources/extensions/sf/extension-manifest.json
@@ -53,6 +53,7 @@
 			"cmux",
 			"codebase",
 			"config",
+			"control",
 			"debug",
 			"discuss",
 			"dispatch",
@@ -60,7 +61,6 @@
 			"doctor",
 			"escalate",
 			"eval-review",
-			"exit",
 			"extensions",
 			"extract-learnings",
 			"fast",
@@ -78,6 +78,7 @@
 			"mcp",
 			"migrate",
 			"mode",
+			"model-mode",
 			"new-milestone",
 			"next",
 			"notifications",
@@ -93,6 +94,7 @@
 			"remote",
 			"reset-slice",
 			"rethink",
+			"repair",
 			"run-hook",
 			"scaffold",
 			"scan",
@@ -102,14 +104,17 @@
 			"ship",
 			"show-config",
 			"skill-health",
+			"skills",
 			"skip",
 			"solver-eval",
 			"start",
 			"status",
 			"steer",
+			"tasks",
 			"templates",
 			"todo",
 			"triage",
+			"trust",
 			"undo",
 			"undo-task",
 			"unpark",
diff --git a/src/resources/extensions/sf/learning/outcome-recorder.test.mjs b/src/resources/extensions/sf/learning/outcome-recorder.test.mjs
index fa61514e9..ffffc650d 100644
--- a/src/resources/extensions/sf/learning/outcome-recorder.test.mjs
+++ b/src/resources/extensions/sf/learning/outcome-recorder.test.mjs
@@ -1,7 +1,7 @@
 /**
  * sf-learning: outcome-recorder + outcome-aggregator tests
  *
- * Uses node:test with a minimal in-memory fake `db` that mimics the
+ * Uses Vitest with a minimal in-memory fake `db` that mimics the
  * node:sqlite DatabaseSync surface (`prepare(sql).run/get/all`, `exec`,
  * `transaction`). The fake parses just enough SQL to verify the
  * insert and aggregate semantics without spinning up real SQLite.
@@ -41,7 +41,10 @@ const INSERT_COLUMNS = [
 	"recorded_at",
 ];
 
-function createFakeDb({ includeTransaction = true, throwOnPrepare = false } = {}) {
+function createFakeDb({
+	includeTransaction = true,
+	throwOnPrepare = false,
+} = {}) {
 	const rows = [];
 	const execSql = [];
 	let nextId = 1;
diff --git a/src/resources/extensions/sf/operating-model.js b/src/resources/extensions/sf/operating-model.js
index 8d9d152cb..c52dcea6c 100644
--- a/src/resources/extensions/sf/operating-model.js
+++ b/src/resources/extensions/sf/operating-model.js
@@ -30,11 +30,7 @@ export const PERMISSION_PROFILES = Object.freeze([
 	"unrestricted",
 ]);
 
-export const MODEL_MODES = Object.freeze([
-	"fast",
-	"smart",
-	"deep",
-]);
+export const MODEL_MODES = Object.freeze(["fast", "smart", "deep"]);
 
 /**
  * Returns true for a canonical SF work mode.
@@ -166,12 +162,48 @@ export function defaultModelModeForWorkMode(workMode) {
 		case "build":
 		case "repair":
 			return "smart";
-		case "chat":
 		default:
 			return "fast";
 	}
 }
 
+/**
+ * Map model mode to complexity tier for routing.
+ *
+ * Purpose: bridge the user-facing modelMode vocabulary (fast/smart/deep)
+ * to the internal complexity tier system (light/standard/heavy).
+ *
+ * Consumer: complexity classifier, model router, /rate feedback.
+ */
+export function modelModeToTier(modelMode) {
+	switch (resolveModelMode(modelMode)) {
+		case "fast":
+			return "light";
+		case "deep":
+			return "heavy";
+		default:
+			return "standard";
+	}
+}
+
+/**
+ * Map complexity tier to model mode.
+ *
+ * Purpose: display the current tier in user-facing modelMode vocabulary.
+ *
+ * Consumer: TUI badges, status output.
+ */
+export function tierToModelMode(tier) {
+	switch (tier) {
+		case "light":
+			return "fast";
+		case "heavy":
+			return "deep";
+		default:
+			return "smart";
+	}
+}
+
 /**
  * Build a canonical mode state object.
  *
diff --git a/src/resources/extensions/sf/parallel-eligibility.js b/src/resources/extensions/sf/parallel-eligibility.js
index 60abe2e06..57e61bd17 100644
--- a/src/resources/extensions/sf/parallel-eligibility.js
+++ b/src/resources/extensions/sf/parallel-eligibility.js
@@ -161,7 +161,7 @@ export async function analyzeParallelEligibility(basePath) {
 				"All dependencies satisfied. NOTE: file overlap with another eligible milestone.";
 		}
 	}
-	return { eligible, ineligible, fileOverlaps };
+	return { eligible, ineligible, fileOverlaps, fileSets };
 }
 // ─── Formatting ──────────────────────────────────────────────────────────────
 /**
diff --git a/src/resources/extensions/sf/parallel-intent.js b/src/resources/extensions/sf/parallel-intent.js
new file mode 100644
index 000000000..8782affea
--- /dev/null
+++ b/src/resources/extensions/sf/parallel-intent.js
@@ -0,0 +1,152 @@
+/**
+ * Parallel Worker Intent/Claim Registry
+ *
+ * Purpose: before editing files, parallel workers declare intent so the
+ * coordinator can detect conflicts without waiting for git merge failures.
+ * This is SF's implementation of Wit-style symbol-level coordination for
+ * parallel milestone workers.
+ *
+ * Consumer: parallel-orchestrator.js before dispatching overlapping milestones.
+ */
+import { mkdirSync } from "node:fs";
+import { join } from "node:path";
+import { getDatabase, getDbPath, openDatabase } from "./sf-db.js";
+import { UokCoordinationStore } from "./uok/coordination-store.js";
+import { logWarning } from "./workflow-logger.js";
+
+const INTENT_KEY_PREFIX = "parallel:intent:";
+const INTENT_STREAM = "parallel:intents";
+
+function projectDbPath(basePath) {
+	const sfDir = join(basePath, ".sf");
+	mkdirSync(sfDir, { recursive: true });
+	return join(sfDir, "sf.db");
+}
+
+function getStore(basePath) {
+	const dbPath = projectDbPath(basePath);
+	if (!getDatabase() || getDbPath() !== dbPath) {
+		openDatabase(dbPath);
+	}
+	const db = getDatabase();
+	if (!db) throw new Error("SF database is not open");
+	return new UokCoordinationStore(db);
+}
+
+function intentKey(milestoneId) {
+	return `${INTENT_KEY_PREFIX}${milestoneId}`;
+}
+
+function normalizeFiles(files) {
+	return files.map((f) => f.replace(/^\/+/, ""));
+}
+
+/**
+ * Declare editing intent before making changes.
+ *
+ * Purpose: let other workers know which files this worker plans to modify,
+ * so conflicts can be detected proactively.
+ *
+ * @param {string} basePath — project root
+ * @param {string} milestoneId — worker's milestone
+ * @param {string[]} files — relative file paths worker intends to edit
+ * @param {object} opts — optional: symbolRanges (for fine-grained claims)
+ */
+export function declareIntent(basePath, milestoneId, files, opts = {}) {
+	try {
+		const store = getStore(basePath);
+		const record = {
+			milestoneId,
+			files: normalizeFiles(files),
+			symbolRanges: opts.symbolRanges ?? [],
+			declaredAt: new Date().toISOString(),
+			status: "claimed",
+		};
+		store.set(intentKey(milestoneId), record);
+		store.xadd(INTENT_STREAM, "intent-claimed", record);
+		return { ok: true };
+	} catch (err) {
+		logWarning(
+			"parallel-intent",
+			`declareIntent failed for ${milestoneId}: ${err.message}`,
+		);
+		return { ok: false, error: err.message };
+	}
+}
+
+/**
+ * Release intent claim when editing is complete or abandoned.
+ */
+export function releaseIntent(basePath, milestoneId) {
+	try {
+		const store = getStore(basePath);
+		const record = store.get(intentKey(milestoneId));
+		if (record) {
+			record.status = "released";
+			record.releasedAt = new Date().toISOString();
+			store.set(intentKey(milestoneId), record);
+			store.xadd(INTENT_STREAM, "intent-released", record);
+		}
+		return { ok: true };
+	} catch (err) {
+		logWarning(
+			"parallel-intent",
+			`releaseIntent failed for ${milestoneId}: ${err.message}`,
+		);
+		return { ok: false, error: err.message };
+	}
+}
+
+/**
+ * Check if any active worker has claimed overlapping files.
+ *
+ * @returns {Array<{milestoneId: string, files: string[]}>} conflicts
+ */
+export function checkIntentConflicts(basePath, milestoneId, files) {
+	const conflicts = [];
+	const normalized = normalizeFiles(files);
+	try {
+		for (const record of getActiveIntents(basePath)) {
+			if (record.milestoneId === milestoneId) continue;
+			const overlap = normalized.filter((f) => record.files.includes(f));
+			if (overlap.length > 0) {
+				conflicts.push({ milestoneId: record.milestoneId, files: overlap });
+			}
+		}
+	} catch (err) {
+		logWarning(
+			"parallel-intent",
+			`checkIntentConflicts failed: ${err.message}`,
+		);
+	}
+	return conflicts;
+}
+
+/**
+ * Get all active intent claims.
+ */
+export function getActiveIntents(basePath) {
+	try {
+		return getStore(basePath)
+			.entries(INTENT_KEY_PREFIX)
+			.map((entry) => entry.value)
+			.filter((record) => record?.status === "claimed");
+	} catch (err) {
+		logWarning("parallel-intent", `getActiveIntents failed: ${err.message}`);
+		return [];
+	}
+}
+
+/**
+ * Clear all intent records (used on orchestrator shutdown).
+ */
+export function clearAllIntents(basePath) {
+	try {
+		const store = getStore(basePath);
+		for (const entry of store.entries(INTENT_KEY_PREFIX)) {
+			store.delete(entry.key);
+		}
+	} catch (err) {
+		logWarning("parallel-intent", `clearAllIntents failed: ${err.message}`);
+	}
+}
diff --git a/src/resources/extensions/sf/parallel-monitor-overlay.js b/src/resources/extensions/sf/parallel-monitor-overlay.js
index fa5f78fdc..61ad54cd2 100644
--- a/src/resources/extensions/sf/parallel-monitor-overlay.js
+++ b/src/resources/extensions/sf/parallel-monitor-overlay.js
@@ -26,7 +26,10 @@ import { formattedShortcutPair } from "./shortcut-defs.js";
 function queryRows(dbPath, sql, params = []) {
 	const db = new DatabaseSync(dbPath, { readOnly: true });
 	try {
-		return db.prepare(sql).all(...params).map((row) => ({ ...row }));
+		return db
+			.prepare(sql)
+			.all(...params)
+			.map((row) => ({ ...row }));
 	} finally {
 		db.close();
 	}
@@ -156,8 +159,9 @@ function queryRecentCompletions(basePath, mid) {
 			  ORDER BY completed_at DESC
 			  LIMIT 5`,
 			[mid],
-		).map((row) =>
-			`✓ ${mid}/${row.sliceId}/${row.taskId}${row.oneLiner ? ": " + row.oneLiner : ""}`,
+		).map(
+			(row) =>
+				`✓ ${mid}/${row.sliceId}/${row.taskId}${row.oneLiner ? ": " + row.oneLiner : ""}`,
 		);
 	} catch {
 		return [];
diff --git a/src/resources/extensions/sf/parallel-orchestrator.js b/src/resources/extensions/sf/parallel-orchestrator.js
index 105238cf0..439dba3f4 100644
--- a/src/resources/extensions/sf/parallel-orchestrator.js
+++ b/src/resources/extensions/sf/parallel-orchestrator.js
@@ -27,6 +27,12 @@ import { readIntegrationBranch } from "./git-service.js";
 import { emitJournalEvent } from "./journal.js";
 import { nativeBranchExists } from "./native-git-bridge.js";
 import { analyzeParallelEligibility } from "./parallel-eligibility.js";
+import {
+	checkIntentConflicts,
+	clearAllIntents,
+	declareIntent,
+	releaseIntent,
+} from "./parallel-intent.js";
 import { sfRoot } from "./paths.js";
 import { resolveParallelConfig } from "./preferences.js";
 import {
@@ -362,10 +368,12 @@ export async function startParallel(basePath, milestoneIds, prefs) {
 	const started = [];
 	const errors = [];
 	let filteredMilestoneIds = milestoneIds;
+	let intentFilesByMilestone = new Map();
 	if (uokFlags.executionGraph && milestoneIds.length > 1) {
 		try {
 			const requestedIds = new Set(milestoneIds);
 			const candidates = await analyzeParallelEligibility(basePath);
+			intentFilesByMilestone = candidates.fileSets ?? intentFilesByMilestone;
 			const overlapPairs = new Set();
 			for (const overlap of candidates.fileOverlaps) {
 				if (!requestedIds.has(overlap.mid1) || !requestedIds.has(overlap.mid2))
@@ -398,8 +406,23 @@ export async function startParallel(basePath, milestoneIds, prefs) {
 	// Cap to max_workers
 	const toStart = filteredMilestoneIds.slice(0, config.max_workers);
 	for (const mid of toStart) {
+		const intendedFiles = intentFilesByMilestone.get(mid) ?? [];
+		const conflicts = checkIntentConflicts(basePath, mid, intendedFiles);
+		if (conflicts.length > 0) {
+			errors.push({
+				mid,
+				error: `File intent conflict with ${conflicts.map((c) => c.milestoneId).join(", ")}`,
+			});
+			continue;
+		}
+		const intent = declareIntent(basePath, mid, intendedFiles);
+		if (!intent.ok) {
+			errors.push({ mid, error: intent.error ?? "Could not declare intent" });
+			continue;
+		}
 		// Check budget ceiling before each spawn
 		if (isBudgetExceeded()) {
+			releaseIntent(basePath, mid);
 			errors.push({
 				mid,
 				error: `Budget ceiling ($${config.budget_ceiling}) reached — skipping`,
@@ -448,6 +471,7 @@ export async function startParallel(basePath, milestoneIds, prefs) {
 			});
 			started.push(mid);
 		} catch (err) {
+			releaseIntent(basePath, mid);
 			const message = getErrorMessage(err);
 			errors.push({ mid, error: message });
 		}
@@ -672,6 +696,7 @@ export function spawnWorker(basePath, milestoneId) {
 					}
 					sibling.state = "cancelled";
 					sibling.process = null;
+					releaseIntent(basePath, siblingId);
 					// Update session status so dashboard reflects the cancellation
 					writeSessionStatus(basePath, {
 						milestoneId: siblingId,
@@ -711,6 +736,7 @@ export function spawnWorker(basePath, milestoneId) {
 			startedAt: w.startedAt,
 			worktreePath: w.worktreePath,
 		});
+		releaseIntent(basePath, milestoneId);
 		persistState(basePath);
 	});
 	return true;
@@ -876,6 +902,8 @@ export async function stopParallel(basePath, milestoneId) {
 	if (!milestoneId) {
 		state.active = false;
 	}
+	// Clear all intent claims on shutdown
+	clearAllIntents(basePath);
 	// Persist final state and clean up state file
 	removeStateFile(basePath);
 }
diff --git a/src/resources/extensions/sf/self-feedback-drain.js b/src/resources/extensions/sf/self-feedback-drain.js
index 7a436a128..642e7cd74 100644
--- a/src/resources/extensions/sf/self-feedback-drain.js
+++ b/src/resources/extensions/sf/self-feedback-drain.js
@@ -180,6 +180,23 @@ export function dispatchSelfFeedbackInlineFixIfNeeded(basePath, ctx, pi) {
 		);
 		return 0;
 	}
+	// Auto-transition to repair work mode when high/critical self-feedback detected
+	try {
+		const { getAutoSession } = require("../auto/session.js");
+		const session = getAutoSession();
+		if (session && session.workMode !== "repair") {
+			const transition = session.setMode({
+				workMode: "repair",
+				reason: "self-feedback-drain",
+			});
+			ctx.ui.notify(
+				`Auto-transition: ${transition.from.workMode} → ${transition.to.workMode} (high/critical self-feedback detected)`,
+				"warning",
+			);
+		}
+	} catch {
+		// Mode transition is best-effort
+	}
 	writeClaim(basePath, ids);
 	const prompt = buildInlineFixPrompt(candidates);
 	ctx.ui.notify(
diff --git a/src/resources/extensions/sf/session-forensics.js b/src/resources/extensions/sf/session-forensics.js
index 8e4233ccb..3a4701f0b 100644
--- a/src/resources/extensions/sf/session-forensics.js
+++ b/src/resources/extensions/sf/session-forensics.js
@@ -124,8 +124,7 @@ export function extractTrace(entries) {
 					isError &&
 					(pending.name === "bash" || pending.name === "bg_shell")
 				) {
-					const lastCmd = findLast(
-						commandsRun,
+					const lastCmd = commandsRun.findLast(
 						(c) => c.command === String(pending.input.command),
 					);
 					if (lastCmd) lastCmd.failed = true;
@@ -482,10 +481,3 @@ function redactInput(_name, input) {
 	}
 	return safe;
 }
-/** Array.findLast polyfill for older Node versions */
-function findLast(arr, predicate) {
-	for (let i = arr.length - 1; i >= 0; i--) {
-		if (predicate(arr[i])) return arr[i];
-	}
-	return undefined;
-}
diff --git a/src/resources/extensions/sf/sf-db.js b/src/resources/extensions/sf/sf-db.js
index 81fdd6675..6ef676b7f 100644
--- a/src/resources/extensions/sf/sf-db.js
+++ b/src/resources/extensions/sf/sf-db.js
@@ -78,7 +78,7 @@ function openRawDb(path) {
 	loadProvider();
 	return new DatabaseSync(path);
 }
-const SCHEMA_VERSION = 42;
+const SCHEMA_VERSION = 43;
 function indexExists(db, name) {
 	return !!db
 		.prepare(
@@ -2221,6 +2221,29 @@ function migrateSchema(db) {
 				":applied_at": new Date().toISOString(),
 			});
 		}
+		if (currentVersion < 43) {
+			db.exec(`
+				CREATE TABLE IF NOT EXISTS session_mode_state (
+					id INTEGER PRIMARY KEY CHECK (id = 1),
+					work_mode TEXT NOT NULL DEFAULT 'chat',
+					run_control TEXT NOT NULL DEFAULT 'manual',
+					permission_profile TEXT NOT NULL DEFAULT 'restricted',
+					model_mode TEXT NOT NULL DEFAULT 'smart',
+					surface TEXT NOT NULL DEFAULT 'tui',
+					updated_at TEXT NOT NULL DEFAULT ''
+				)
+			`);
+			db.exec(`
+				INSERT OR IGNORE INTO session_mode_state (id, work_mode, run_control, permission_profile, model_mode, surface, updated_at)
+				VALUES (1, 'chat', 'manual', 'restricted', 'smart', 'tui', datetime('now'))
+			`);
+			db.prepare(
+				"INSERT INTO schema_version (version, applied_at) VALUES (:version, :applied_at)",
+			).run({
+				":version": 43,
+				":applied_at": new Date().toISOString(),
+			});
+		}
 		db.exec("COMMIT");
 	} catch (err) {
 		db.exec("ROLLBACK");
@@ -2542,6 +2565,67 @@ export function getDbOwnerPid() {
 export function getDbPath() {
 	return currentPath;
 }
+
+/**
+ * Load persisted session mode state from DB.
+ *
+ * Purpose: restore mode state across session restarts.
+ *
+ * Consumer: AutoSession initialization.
+ */
+export function loadSessionModeState() {
+	if (!currentDb) return null;
+	try {
+		const row = currentDb
+			.prepare("SELECT * FROM session_mode_state WHERE id = 1")
+			.get();
+		if (!row) return null;
+		return {
+			workMode: row["work_mode"] ?? "chat",
+			runControl: row["run_control"] ?? "manual",
+			permissionProfile: row["permission_profile"] ?? "restricted",
+			modelMode: row["model_mode"] ?? "smart",
+			surface: row["surface"] ?? "tui",
+			updatedAt: row["updated_at"] ?? null,
+		};
+	} catch {
+		return null;
+	}
+}
+
+/**
+ * Persist the current session mode into the project database.
+ *
+ * Purpose: keep work mode, run control, permission profile, and model mode
+ * stable across reload/resume without letting command handlers write SQL.
+ *
+ * Consumer: AutoSession.setMode() after validated mode transitions.
+ */
+export function saveSessionModeState(mode) {
+	if (!currentDb) return false;
+	currentDb
+		.prepare(`
+		INSERT INTO session_mode_state (id, work_mode, run_control, permission_profile, model_mode, surface, updated_at)
+		VALUES (1, :workMode, :runControl, :permissionProfile, :modelMode, :surface, :updatedAt)
+		ON CONFLICT(id) DO UPDATE SET
+			work_mode = excluded.work_mode,
+			run_control = excluded.run_control,
+			permission_profile = excluded.permission_profile,
+			model_mode = excluded.model_mode,
+			surface = excluded.surface,
+			updated_at = excluded.updated_at
+	`)
+		.run({
+			":workMode": mode.workMode,
+			":runControl": mode.runControl,
+			":permissionProfile": mode.permissionProfile,
+			":modelMode": mode.modelMode,
+			":surface": mode.surface ?? "tui",
+			":updatedAt": mode.updatedAt ?? new Date().toISOString(),
+		});
+	return true;
+}
+
 export function _getAdapter() {
 	return currentDb;
 }
diff --git a/src/resources/extensions/sf/skills/auto-create.js b/src/resources/extensions/sf/skills/auto-create.js
new file mode 100644
index 000000000..8b43467d6
--- /dev/null
+++ b/src/resources/extensions/sf/skills/auto-create.js
@@ -0,0 +1,263 @@
+/**
+ * Auto skill creation flow
+ *
+ * Purpose: detect repeated repo-specific patterns from journal/activity logs
+ * and propose or auto-generate skills when policy allows.
+ *
+ * Consumer: session startup, turn end, and explicit /skills --auto-create.
+ */
+import {
+	existsSync,
+	mkdirSync,
+	readdirSync,
+	readFileSync,
+	writeFileSync,
+} from "node:fs";
+import { join } from "node:path";
+import { getDatabase, getDbPath, openDatabase } from "../sf-db.js";
+import { UokCoordinationStore } from "../uok/coordination-store.js";
+import { codeSkillTemplate, knowledgeSkillTemplate } from "./templates.js";
+
+const AUTO_SKILL_MIN_OCCURRENCES = 3;
+const AUTO_SKILL_COOLDOWN_MS = 24 * 60 * 60 * 1000; // 1 day
+const AUTO_SKILL_COOLDOWN_KEY = "skills:auto-create:last-run";
+
+function projectDbPath(basePath) {
+	const sfDir = join(basePath, ".sf");
+	mkdirSync(sfDir, { recursive: true });
+	return join(sfDir, "sf.db");
+}
+
+function getCoordinationStore(basePath) {
+	const dbPath = projectDbPath(basePath);
+	if (!getDatabase() || getDbPath() !== dbPath) {
+		openDatabase(dbPath);
+	}
+	const db = getDatabase();
+	if (!db) throw new Error("SF database is not open");
+	return new UokCoordinationStore(db);
+}
+
+/**
+ * Scan activity logs for repeated patterns that could become skills.
+ *
+ * Purpose: identify recurring file edits, commands, or verification paths
+ * that suggest a reusable capability.
+ *
+ * Consumer: auto skill creation on session startup or turn end.
+ */
+export function detectSkillCandidates(basePath) {
+	const candidates = [];
+
+	// Pattern 1: Repeated file edits on same files
+	const fileEdits = scanFileEditPatterns(basePath);
+	for (const [file, count] of fileEdits) {
+		if (count >= AUTO_SKILL_MIN_OCCURRENCES) {
+			candidates.push({
+				type: "file-pattern",
+				name: `forge-${sanitizeSkillName(file)}`,
+				description: `Recurring edits to ${file} (${count} times)`,
+				triggerFile: file,
+				occurrences: count,
+				template: "code",
+			});
+		}
+	}
+
+	// Pattern 2: Repeated bash command patterns
+	const commandPatterns = scanCommandPatterns(basePath);
+	for (const [pattern, count] of commandPatterns) {
+		if (count >= AUTO_SKILL_MIN_OCCURRENCES) {
+			candidates.push({
+				type: "command-pattern",
+				name: `forge-${sanitizeSkillName(pattern)}`,
+				description: `Recurring command pattern: ${pattern} (${count} times)`,
+				triggerCommand: pattern,
+				occurrences: count,
+				template: "code",
+			});
+		}
+	}
+
+	// Pattern 3: Repeated verification/check patterns
+	const verifyPatterns = scanVerificationPatterns(basePath);
+	for (const [pattern, count] of verifyPatterns) {
+		if (count >= AUTO_SKILL_MIN_OCCURRENCES) {
+			candidates.push({
+				type: "verification-pattern",
+				name: `forge-verify-${sanitizeSkillName(pattern)}`,
+				description: `Recurring verification: ${pattern} (${count} times)`,
+				triggerVerify: pattern,
+				occurrences: count,
+				template: "review",
+			});
+		}
+	}
+
+	return candidates;
+}
+
+/**
+ * Check if auto-skill creation is allowed by policy.
+ */
+export function isAutoSkillCreationAllowed(basePath) {
+	const data = getCoordinationStore(basePath).get(AUTO_SKILL_COOLDOWN_KEY);
+	const lastRun = Number(data?.lastRunMs ?? 0);
+	if (lastRun > 0 && Date.now() - lastRun < AUTO_SKILL_COOLDOWN_MS) {
+		return false;
+	}
+	return true;
+}
+
+/**
+ * Generate a skill from a detected candidate.
+ */
+export function generateSkill(candidate, basePath) {
+	const skillDir = join(basePath, ".agents", "skills", candidate.name);
+	if (existsSync(skillDir)) {
+		return { created: false, reason: "Skill already exists" };
+	}
+
+	let content;
+	switch (candidate.template) {
+		case "review":
+			content = knowledgeSkillTemplate(candidate.name, candidate.description);
+			break;
+		default:
+			content = codeSkillTemplate(candidate.name, candidate.description);
+			break;
+	}
+
+	// Add auto-generated marker
+	content += `\n\n## Auto-Generated Evidence\n\n`;
+	content += `- **Type:** ${candidate.type}\n`;
+	content += `- **Occurrences:** ${candidate.occurrences}\n`;
+	content += `- **Detected:** ${new Date().toISOString()}\n`;
+	if (candidate.triggerFile) {
+		content += `- **Trigger file:** ${candidate.triggerFile}\n`;
+	}
+	if (candidate.triggerCommand) {
+		content += `- **Trigger command:** ${candidate.triggerCommand}\n`;
+	}
+
+	mkdirSync(skillDir, { recursive: true });
+	writeFileSync(join(skillDir, "SKILL.md"), content, "utf-8");
+
+	// Update cooldown
+	const lastRunMs = Date.now();
+	getCoordinationStore(basePath).set(AUTO_SKILL_COOLDOWN_KEY, {
+		lastRun: new Date(lastRunMs).toISOString(),
+		lastRunMs,
+	});
+
+	return { created: true, path: skillDir, name: candidate.name };
+}
+
+// ─── Pattern scanners (best-effort heuristics) ────────────────────────────
+
+function scanFileEditPatterns(basePath) {
+	const edits = new Map();
+	try {
+		const activityDir = join(basePath, ".sf", "activity");
+		if (!existsSync(activityDir)) return edits;
+		const files = readdirSync(activityDir).filter((f) => f.endsWith(".jsonl"));
+		for (const file of files.slice(-7)) {
+			const raw = readFileSync(join(activityDir, file), "utf-8");
+			for (const line of raw.split("\n")) {
+				if (!line.trim()) continue;
+				try {
+					const entry = JSON.parse(line);
+					if (entry.toolName === "write" || entry.toolName === "edit") {
+						const path = entry.input?.path ?? "";
+						if (path) {
+							const basename = path.split("/").pop() ?? path;
+							edits.set(basename, (edits.get(basename) ?? 0) + 1);
+						}
+					}
+				} catch {
+					// Skip malformed lines
+				}
+			}
+		}
+	} catch {
+		// Best-effort scan
+	}
+	return edits;
+}
+
+function scanCommandPatterns(basePath) {
+	const commands = new Map();
+	try {
+		const activityDir = join(basePath, ".sf", "activity");
+		if (!existsSync(activityDir)) return commands;
+		const files = readdirSync(activityDir).filter((f) => f.endsWith(".jsonl"));
+		for (const file of files.slice(-7)) {
+			const raw = readFileSync(join(activityDir, file), "utf-8");
+			for (const line of raw.split("\n")) {
+				if (!line.trim()) continue;
+				try {
+					const entry = JSON.parse(line);
+					if (entry.toolName === "bash") {
+						const cmd = entry.input?.command ?? "";
+						// Extract command name (first token)
+						const name = cmd.trim().split(/\s+/)[0] ?? "";
+						if (name && !name.startsWith("cd") && !name.startsWith("ls")) {
+							commands.set(name, (commands.get(name) ?? 0) + 1);
+						}
+					}
+				} catch {
+					// Skip malformed lines
+				}
+			}
+		}
+	} catch {
+		// Best-effort scan
+	}
+	return commands;
+}
+
+function scanVerificationPatterns(basePath) {
+	const patterns = new Map();
+	try {
+		const activityDir = join(basePath, ".sf", "activity");
+		if (!existsSync(activityDir)) return patterns;
+		const files = readdirSync(activityDir).filter((f) => f.endsWith(".jsonl"));
+		for (const file of files.slice(-7)) {
+			const raw = readFileSync(join(activityDir, file), "utf-8");
+			for (const line of raw.split("\n")) {
+				if (!line.trim()) continue;
+				try {
+					const entry = JSON.parse(line);
+					if (entry.toolName === "bash") {
+						const cmd = entry.input?.command ?? "";
+						// Look for verification patterns: test, lint, check, verify
+						if (
+							/\b(npm\s+run\s+(test|lint|check|verify)|npm\s+(test|lint|audit)|vitest|jest|eslint|tsc\s+--noEmit)\b/.test(
+								cmd,
+							)
+						) {
+							const pattern =
+								cmd.match(
+									/\b(npm\s+run\s+\w+|npm\s+\w+|vitest|jest|eslint|tsc)\b/,
+								)?.[0] ?? cmd;
+							patterns.set(pattern, (patterns.get(pattern) ?? 0) + 1);
+						}
+					}
+				} catch {
+					// Skip malformed lines
+				}
+			}
+		}
+	} catch {
+		// Best-effort scan
+	}
+	return patterns;
+}
+
+function sanitizeSkillName(name) {
+	return name
+		.toLowerCase()
+		.replace(/[^a-z0-9]+/g, "-")
+		.replace(/^-|-$/g, "")
+		.slice(0, 40);
+}
diff --git a/src/resources/extensions/sf/skills/directory.js b/src/resources/extensions/sf/skills/directory.js
new file mode 100644
index 000000000..a20c474fc
--- /dev/null
+++ b/src/resources/extensions/sf/skills/directory.js
@@ -0,0 +1,73 @@
+/**
+ * .agents/skills/ directory management
+ *
+ * Purpose: discover, validate, and load skills from repo-local
+ * `.agents/skills/` and user-level skill directories.
+ *
+ * Consumer: skill loader, auto-skill creation, and model context assembly.
+ */
+import { existsSync, readdirSync, readFileSync, statSync } from "node:fs";
+import { join } from "node:path";
+
+const SKILL_FILENAME = "SKILL.md";
+const USER_SKILL_DIR = join(process.env.HOME ?? "", ".sf", "skills");
+
+/**
+ * Find all skill directories under a base path.
+ */
+export function discoverSkillDirs(basePath) {
+	const skillRoot = join(basePath, ".agents", "skills");
+	if (!existsSync(skillRoot)) return [];
+
+	const dirs = [];
+	for (const entry of readdirSync(skillRoot)) {
+		const full = join(skillRoot, entry);
+		if (statSync(full).isDirectory()) {
+			const skillFile = join(full, SKILL_FILENAME);
+			if (existsSync(skillFile)) {
+				dirs.push({ name: entry, path: full, skillFile });
+			}
+		}
+	}
+	return dirs;
+}
+
+/**
+ * Discover skills from all sources: project, user, and built-in.
+ */
+export function discoverAllSkills(projectPath) {
+	const sources = [];
+
+	// Project skills
+	if (projectPath) {
+		const projectSkills = discoverSkillDirs(projectPath);
+		for (const s of projectSkills) {
+			sources.push({ ...s, source: "project" });
+		}
+	}
+
+	// User skills
+	if (existsSync(USER_SKILL_DIR)) {
+		const userSkills = discoverSkillDirs(USER_SKILL_DIR);
+		for (const s of userSkills) {
+			// User skills have a different root structure
+			s.path = s.path.replace(/\.agents\/skills$/, "");
+			sources.push({ ...s, source: "user" });
+		}
+	}
+
+	return sources;
+}
+
+/**
+ * Read the raw content of a skill file.
+ */
+export function readSkillFile(skillDir) {
+	const path = join(skillDir, SKILL_FILENAME);
+	if (!existsSync(path)) return null;
+	try {
+		return readFileSync(path, "utf-8");
+	} catch {
+		return null;
+	}
+}
diff --git a/src/resources/extensions/sf/skills/eval-harness.js b/src/resources/extensions/sf/skills/eval-harness.js
new file mode 100644
index 000000000..955518aed
--- /dev/null
+++ b/src/resources/extensions/sf/skills/eval-harness.js
@@ -0,0 +1,181 @@
+/**
+ * Skill Eval Harness — foundation for grading auto-created skills.
+ *
+ * Purpose: provide deterministic eval cases for skills so auto-generated
+ * capabilities can be tested before being committed. Follows the
+ * skill-optimizer pattern: cases + graders + workspaces.
+ *
+ * Consumer: /skills --eval, auto-skill creation acceptance gate.
+ */
+import { existsSync, mkdirSync, writeFileSync } from "node:fs";
+import { join } from "node:path";
+
+/**
+ * Create an eval case directory for a skill.
+ *
+ * @param {string} skillPath — path to skill directory
+ * @param {string} caseName — unique case identifier
+ * @param {object} caseDef — { task: string, grader: string, hidden?: object }
+ */
+export function createEvalCase(skillPath, caseName, caseDef) {
+	const evalDir = join(skillPath, "evals", caseName);
+	mkdirSync(evalDir, { recursive: true });
+	mkdirSync(join(evalDir, "work"), { recursive: true });
+
+	writeFileSync(join(evalDir, "task.md"), caseDef.task, "utf-8");
+	writeFileSync(join(evalDir, "grader.js"), caseDef.grader, "utf-8");
+	if (caseDef.hidden) {
+		mkdirSync(join(evalDir, "hidden"), { recursive: true });
+		writeFileSync(
+			join(evalDir, "hidden", "reference.json"),
+			JSON.stringify(caseDef.hidden, null, 2),
+			"utf-8",
+		);
+	}
+	return { path: evalDir, caseName };
+}
+
+/**
+ * Run a grader against a skill's work directory.
+ *
+ * @param {string} evalDir — path to eval case directory
+ * @param {object} ctx — { ui } for notifications
+ * @returns {object} { passed: boolean, score: number, details: string[] }
+ */
+export async function runGrader(evalDir, _ctx) {
+	const graderPath = join(evalDir, "grader.js");
+	const workDir = join(evalDir, "work");
+	if (!existsSync(graderPath)) {
+		return { passed: false, score: 0, details: ["Grader not found"] };
+	}
+
+	try {
+		const { grade } = await import(graderPath);
+		if (typeof grade !== "function") {
+			return {
+				passed: false,
+				score: 0,
+				details: ["Grader must export a grade() function"],
+			};
+		}
+		const result = await grade(workDir);
+		return {
+			passed: Boolean(result.passed),
+			score: Number(result.score ?? (result.passed ? 1 : 0)),
+			details: Array.isArray(result.details)
+				? result.details
+				: [String(result.details ?? "")],
+		};
+	} catch (err) {
+		return {
+			passed: false,
+			score: 0,
+			details: [`Grader error: ${err.message}`],
+		};
+	}
+}
+
+/**
+ * Run all eval cases for a skill.
+ *
+ * @param {string} skillPath — path to skill directory
+ * @param {object} ctx — { ui } for notifications
+ * @returns {object} { skillName, totalCases, passedCases, totalScore, cases: [] }
+ */
+export async function runSkillEvals(skillPath, ctx) {
+	const evalDir = join(skillPath, "evals");
+	if (!existsSync(evalDir)) {
+		return {
+			skillName: skillPath,
+			totalCases: 0,
+			passedCases: 0,
+			totalScore: 0,
+			cases: [],
+		};
+	}
+
+	const { readdirSync } = await import("node:fs");
+	const cases = [];
+	let passedCount = 0;
+	let totalScore = 0;
+
+	for (const entry of readdirSync(evalDir)) {
+		const caseDir = join(evalDir, entry);
+		const taskPath = join(caseDir, "task.md");
+		if (!existsSync(taskPath)) continue;
+
+		const result = await runGrader(caseDir, ctx);
+		cases.push({
+			caseName: entry,
+			passed: result.passed,
+			score: result.score,
+			details: result.details,
+		});
+		if (result.passed) passedCount++;
+		totalScore += result.score;
+	}
+
+	return {
+		skillName: skillPath,
+		totalCases: cases.length,
+		passedCases: passedCount,
+		totalScore: cases.length > 0 ? totalScore / cases.length : 0,
+		cases,
+	};
+}
+
+/**
+ * Generate a default eval case for a skill based on its type.
+ *
+ * @param {object} skill — skill record from buildSkillRecord()
+ * @returns {object} caseDef for createEvalCase()
+ */
+export function generateDefaultEvalCase(skill) {
+	const task = `Test the ${skill.name} skill.
+
+Instructions:
+1. Read the skill definition at SKILL.md
+2. Apply the skill to a realistic task matching its description
+3. Verify the output matches expected behavior
+
+Expected: The skill produces correct, safe output without side effects beyond its declared scope.
+`;
+
+	const grader = `/**
+ * Default grader for ${skill.name}.
+ * Checks that work directory contains expected artifacts.
+ */
+export async function grade(workDir) {
+	const { existsSync } = await import("node:fs");
+	const { join } = await import("node:path");
+
+	const checks = [];
+	let score = 0;
+
+	// Example: check for expected output file
+	if (existsSync(join(workDir, "result.txt"))) {
+		checks.push("✓ result.txt exists");
+		score += 0.5;
+	} else {
+		checks.push("✗ result.txt missing");
+	}
+
+	// Example: check for no unexpected files
+	const unexpected = existsSync(join(workDir, "unexpected.txt"));
+	if (!unexpected) {
+		checks.push("✓ no unexpected artifacts");
+		score += 0.5;
+	} else {
+		checks.push("✗ unexpected artifact found");
+	}
+
+	return {
+		passed: score >= 0.8,
+		score,
+		details: checks,
+	};
+}
+`;
+
+	return { task, grader };
+}
diff --git a/src/resources/extensions/sf/skills/frontmatter.js b/src/resources/extensions/sf/skills/frontmatter.js
new file mode 100644
index 000000000..5d251b5d8
--- /dev/null
+++ b/src/resources/extensions/sf/skills/frontmatter.js
@@ -0,0 +1,153 @@
+/**
+ * Skill YAML frontmatter parser
+ *
+ * Purpose: parse SKILL.md frontmatter so SF can enforce invocation policy,
+ * permission profiles, and side-effect awareness without loading full skill
+ * content into model context.
+ *
+ * Consumer: skill loader, auto-skill creation, and permission gating.
+ */
+
+const FRONTMATTER_RE = /^---\s*\n([\s\S]*?)\n---\s*\n/;
+
+/**
+ * Parse YAML frontmatter from skill markdown content.
+ *
+ * Returns { frontmatter: object, body: string } or null if no frontmatter.
+ */
+export function parseSkillFrontmatter(content) {
+	if (!content) return null;
+	const match = content.match(FRONTMATTER_RE);
+	if (!match) return null;
+
+	const yamlText = match[1];
+	const body = content.slice(match[0].length);
+
+	const frontmatter = parseYaml(yamlText);
+	return { frontmatter, body };
+}
+
+/**
+ * Minimal YAML parser for skill frontmatter.
+ *
+ * Supports: strings, booleans, numbers, arrays, and simple nested objects.
+ * Does NOT support anchors, aliases, or complex YAML features.
+ */
+function parseYaml(text) {
+	const result = {};
+	const lines = text.split("\n");
+	let currentKey = null;
+	let currentArray = null;
+
+	for (const rawLine of lines) {
+		const line = rawLine.trimEnd();
+		if (!line.trim() || line.trim().startsWith("#")) continue;
+
+		// Check for key: value
+		const keyMatch = line.match(/^(\w[\w-]*):\s*(.*)$/);
+		if (keyMatch) {
+			const [, key, value] = keyMatch;
+			currentKey = key;
+
+			if (value.trim() === "") {
+				// Could be object or array start
+				currentArray = null;
+				continue;
+			}
+
+			result[key] = parseYamlValue(value.trim());
+			currentArray = null;
+			continue;
+		}
+
+		// Array item: - value (may be indented)
+		const arrayMatch = line.match(/^\s+-\s+(.*)$/);
+		if (arrayMatch && currentKey) {
+			if (!currentArray) {
+				currentArray = [];
+				result[currentKey] = currentArray;
+			}
+			currentArray.push(parseYamlValue(arrayMatch[1].trim()));
+		}
+	}
+
+	return result;
+}
+
+function parseYamlValue(value) {
+	if (value === "true" || value === "yes") return true;
+	if (value === "false" || value === "no") return false;
+	if (value === "null" || value === "~") return null;
+
+	// Number
+	if (/^-?\d+$/.test(value)) return Number.parseInt(value, 10);
+	if (/^-?\d+\.\d+$/.test(value)) return Number.parseFloat(value);
+
+	// Quoted string
+	if (
+		(value.startsWith('"') && value.endsWith('"')) ||
+		(value.startsWith("'") && value.endsWith("'"))
+	) {
+		return value.slice(1, -1);
+	}
+
+	return value;
+}
+
+/**
+ * Validate skill frontmatter against required fields.
+ */
+export function validateSkillFrontmatter(frontmatter) {
+	const errors = [];
+	if (!frontmatter.name || typeof frontmatter.name !== "string") {
+		errors.push("Missing or invalid 'name' field");
+	}
+	if (!frontmatter.description || typeof frontmatter.description !== "string") {
+		errors.push("Missing or invalid 'description' field");
+	}
+	if (frontmatter["user-invocable"] === undefined) {
+		errors.push("Missing 'user-invocable' field");
+	}
+	if (frontmatter["model-invocable"] === undefined) {
+		errors.push("Missing 'model-invocable' field");
+	}
+	if (frontmatter["side-effects"] === undefined) {
+		errors.push("Missing 'side-effects' field");
+	}
+	if (frontmatter["permission-profile"] === undefined) {
+		errors.push("Missing 'permission-profile' field");
+	}
+
+	const validProfiles = ["restricted", "normal", "trusted", "unrestricted"];
+	if (
+		frontmatter["permission-profile"] &&
+		!validProfiles.includes(frontmatter["permission-profile"])
+	) {
+		errors.push(
+			`Invalid permission-profile: ${frontmatter["permission-profile"]}. Must be one of: ${validProfiles.join(", ")}`,
+		);
+	}
+
+	return {
+		valid: errors.length === 0,
+		errors,
+	};
+}
+
+/**
+ * Build a canonical skill record from parsed frontmatter + body.
+ */
+export function buildSkillRecord(skillDir, frontmatter, body) {
+	return {
+		name: frontmatter.name,
+		description: frontmatter.description,
+		userInvocable: frontmatter["user-invocable"] ?? false,
+		modelInvocable: frontmatter["model-invocable"] ?? false,
+		sideEffects: frontmatter["side-effects"] ?? "none",
+		permissionProfile: frontmatter["permission-profile"] ?? "restricted",
+		triggers: frontmatter.triggers ?? [],
+		maxActivations: frontmatter["max-activations"] ?? null,
+		body,
+		path: skillDir,
+	};
+}
diff --git a/src/resources/extensions/sf/skills/index.js b/src/resources/extensions/sf/skills/index.js
new file mode 100644
index 000000000..a05e6da6c
--- /dev/null
+++ b/src/resources/extensions/sf/skills/index.js
@@ -0,0 +1,37 @@
+/**
+ * Skills module — public API barrel
+ *
+ * Purpose: expose skill discovery, parsing, validation, and loading
+ * as a single import surface.
+ *
+ * Consumer: command handlers, auto-dispatch, and model context assembly.
+ */
+
+export {
+	detectSkillCandidates,
+	generateSkill,
+	isAutoSkillCreationAllowed,
+} from "./auto-create.js";
+export {
+	discoverAllSkills,
+	discoverSkillDirs,
+	readSkillFile,
+	SKILL_FILENAME,
+	USER_SKILL_DIR,
+} from "./directory.js";
+export {
+	createEvalCase,
+	generateDefaultEvalCase,
+	runGrader,
+	runSkillEvals,
+} from "./eval-harness.js";
+export {
+	buildSkillRecord,
+	parseSkillFrontmatter,
+	validateSkillFrontmatter,
+} from "./frontmatter.js";
+export {
+	getModelInvocableSkills,
+	getPermittedSkills,
+	loadSkills,
+} from "./loader.js";
diff --git a/src/resources/extensions/sf/skills/loader.js b/src/resources/extensions/sf/skills/loader.js
new file mode 100644
index 000000000..83ba7c7cf
--- /dev/null
+++ b/src/resources/extensions/sf/skills/loader.js
@@ -0,0 +1,113 @@
+/**
+ * Skill loader
+ *
+ * Purpose: load skills from `.agents/skills/` with frontmatter validation,
+ * permission checking, and lazy content loading.
+ *
+ * Consumer: command handlers, auto-dispatch, and model context assembly.
+ */
+import { discoverAllSkills, readSkillFile } from "./directory.js";
+import {
+	buildSkillRecord,
+	parseSkillFrontmatter,
+	validateSkillFrontmatter,
+} from "./frontmatter.js";
+
+/**
+ * Load all valid skills from all sources.
+ *
+ * Returns array of skill records with validation errors attached.
+ */
+export function loadSkills(projectPath) {
+	const discovered = discoverAllSkills(projectPath);
+	const skills = [];
+
+	for (const { name, path, source } of discovered) {
+		const content = readSkillFile(path);
+		if (!content) {
+			skills.push({
+				name,
+				source,
+				path,
+				valid: false,
+				errors: ["Could not read SKILL.md"],
+			});
+			continue;
+		}
+
+		const parsed = parseSkillFrontmatter(content);
+		if (!parsed) {
+			skills.push({
+				name,
+				source,
+				path,
+				valid: false,
+				errors: ["No YAML frontmatter found"],
+			});
+			continue;
+		}
+
+		const validation = validateSkillFrontmatter(parsed.frontmatter);
+		if (!validation.valid) {
+			skills.push({
+				name,
+				source,
+				path,
+				valid: false,
+				errors: validation.errors,
+				frontmatter: parsed.frontmatter,
+			});
+			continue;
+		}
+
+		const record = buildSkillRecord(path, parsed.frontmatter, parsed.body);
+		skills.push({
+			...record,
+			source,
+			valid: true,
+			errors: [],
+		});
+	}
+
+	return skills;
+}
+
+/**
+ * Get skills that are safe for the current permission profile.
+ */
+export function getPermittedSkills(skills, activeProfile) {
+	const profileOrder = ["restricted", "normal", "trusted", "unrestricted"];
+	const activeIndex = profileOrder.indexOf(activeProfile);
+	if (activeIndex === -1) return [];
+
+	return skills.filter((s) => {
+		if (!s.valid) return false;
+		const skillIndex = profileOrder.indexOf(s.permissionProfile);
+		if (skillIndex === -1) return false;
+		return skillIndex <= activeIndex;
+	});
+}
+
+/**
+ * Get skills that can be invoked by the model for a given work mode.
+ */
+export function getModelInvocableSkills(skills, workMode) {
+	return skills.filter(
+		(s) => s.valid && s.modelInvocable && isSkillRelevant(s, workMode),
+	);
+}
+
+/**
+ * Check if a skill is relevant to the current work mode.
+ */
+function isSkillRelevant(skill, workMode) {
+	if (!skill.triggers || skill.triggers.length === 0) return true;
+	return skill.triggers.some(
+		(t) =>
+			t === workMode ||
+			t === "*" ||
+			(workMode === "build" && t === "code") ||
+			(workMode === "review" && t === "review") ||
+			(workMode === "research" && t === "research"),
+	);
+}
diff --git a/src/resources/extensions/sf/skills/spec-first-tdd/SKILL.md b/src/resources/extensions/sf/skills/spec-first-tdd/SKILL.md
index 2d709327f..1cd946ec4 100644
--- a/src/resources/extensions/sf/skills/spec-first-tdd/SKILL.md
+++ b/src/resources/extensions/sf/skills/spec-first-tdd/SKILL.md
@@ -75,14 +75,13 @@ Label evidence:
 ### 3. Write the Failing Test (this is the spec)
 
 ```ts
-import { test } from "node:test";
-import assert from "node:assert/strict";
+import { expect, test } from "vitest";
 
 // behaviour contract: claim() rejects takeover when claim_until > now()
 test("claim_when_active_lease_returns_false", () => {
   const now = 1000;
   const result = claim({ unitId: "u1", leaseMs: 60_000, now, holder: "w1", existingClaimUntil: now + 30_000 });
-  assert.equal(result.acquired, false);
+  expect(result.acquired).toBe(false);
 });
 ```
 
diff --git a/src/resources/extensions/sf/skills/templates.js b/src/resources/extensions/sf/skills/templates.js
new file mode 100644
index 000000000..9dad96827
--- /dev/null
+++ b/src/resources/extensions/sf/skills/templates.js
@@ -0,0 +1,150 @@
+/**
+ * Skill templates for auto-creation
+ *
+ * Purpose: provide canonical SKILL.md templates for common skill types
+ * so auto-skill creation produces consistent, valid frontmatter.
+ *
+ * Consumer: auto-skill creation flow and CLI skill scaffolding.
+ */
+
+/**
+ * Template for a user+model invocable skill with code edits.
+ */
+export function codeSkillTemplate(name, description) {
+	return `---
+name: ${name}
+description: ${description}
+user-invocable: true
+model-invocable: true
+side-effects: code-edits
+permission-profile: normal
+triggers:
+  - build
+  - code
+  - "*"
+---
+
+# ${name}
+
+## When to Use
+
+Describe when this skill should be invoked.
+
+## Instructions
+
+1. Step one
+2. Step two
+3. Step three
+
+## Verification
+
+- [ ] Check A
+- [ ] Check B
+
+## Examples
+
+\`\`\`typescript
+// Example code
+\`\`\`
+`;
+}
+
+/**
+ * Template for a dangerous skill (model never invokes by default).
+ */
+export function dangerousSkillTemplate(name, description) {
+	return `---
+name: ${name}
+description: ${description}
+user-invocable: true
+model-invocable: false
+side-effects: production-mutation
+permission-profile: trusted
+triggers: []
+---
+
+# ${name}
+
+## ⚠️ Dangerous Operation
+
+This skill performs high-risk actions. Model invocation is disabled by default.
+
+## Prerequisites
+
+- [ ] Gate A passes
+- [ ] Gate B passes
+
+## Procedure
+
+1. Step one
+2. Step two
+
+## Rollback
+
+How to undo if something goes wrong.
+`;
+}
+
+/**
+ * Template for background knowledge (model only, no side effects).
+ */
+export function knowledgeSkillTemplate(name, description) {
+	return `---
+name: ${name}
+description: ${description}
+user-invocable: false
+model-invocable: true
+side-effects: none
+permission-profile: restricted
+triggers:
+  - "*"
+---
+
+# ${name}
+
+## Context
+
+Background information the model should know.
+
+## Rules
+
+1. Rule one
+2. Rule two
+
+## Anti-Patterns
+
+- Don't do X
+- Don't do Y
+`;
+}
+
+/**
+ * Template for review/verification skills.
+ */
+export function reviewSkillTemplate(name, description) {
+	return `---
+name: ${name}
+description: ${description}
+user-invocable: true
+model-invocable: true
+side-effects: review-comments
+permission-profile: restricted
+triggers:
+  - review
+  - verify
+---
+
+# ${name}
+
+## Checklist
+
+- [ ] Item A
+- [ ] Item B
+- [ ] Item C
+
+## Common Issues
+
+1. Issue one
+2. Issue two
+`;
+}
diff --git a/src/resources/extensions/sf/temporal-foundation.js b/src/resources/extensions/sf/temporal-foundation.js
new file mode 100644
index 000000000..fa0405a5b
--- /dev/null
+++ b/src/resources/extensions/sf/temporal-foundation.js
@@ -0,0 +1,193 @@
+/**
+ * Temporal Foundation — Node 26 native date/time helpers
+ *
+ * Purpose: provide one native Temporal surface for schedules, leases, and
+ * background work now that Node 26 is the runtime baseline.
+ *
+ * Consumer: schedule, journal, lease, and background task surfaces.
+ */
+
+function requireTemporal() {
+	if (typeof globalThis.Temporal !== "object" || globalThis.Temporal === null) {
+		throw new Error("SF requires Node 26 native Temporal support");
+	}
+	return globalThis.Temporal;
+}
+
+/**
+ * Check if native Temporal is available.
+ *
+ * Purpose: let startup diagnostics fail clearly when SF is launched on a
+ * runtime older than the Node 26 baseline.
+ *
+ * Consumer: doctor/startup checks and Temporal tests.
+ */
+export function hasNativeTemporal() {
+	return (
+		typeof globalThis.Temporal === "object" && globalThis.Temporal !== null
+	);
+}
+
+/**
+ * Return the native Temporal namespace.
+ *
+ * Purpose: keep callers on the platform implementation instead of local
+ * compatibility shims.
+ *
+ * Consumer: modules that need Temporal constructors directly.
+ */
+export function getTemporal() {
+	return requireTemporal();
+}
+
+/**
+ * Create a Temporal.Instant from an ISO string.
+ *
+ * Purpose: exact event timestamps with stable ordering.
+ *
+ * Consumer: journals, leases, and scheduler records.
+ */
+export function instantFromISO(isoString) {
+	return requireTemporal().Instant.from(isoString);
+}
+
+/**
+ * Get the current instant.
+ *
+ * Purpose: use a native Temporal clock value for runtime records.
+ *
+ * Consumer: schedule and lease code that needs an exact timestamp.
+ */
+export function instantNow() {
+	return requireTemporal().Now.instant();
+}
+
+/**
+ * Create a Temporal.Duration from an object.
+ *
+ * Purpose: typed durations for leases, budgets, retry delays.
+ *
+ * Consumer: retry, timeout, and reminder scheduling code.
+ */
+export function durationFromObject({
+	years = 0,
+	months = 0,
+	weeks = 0,
+	days = 0,
+	hours = 0,
+	minutes = 0,
+	seconds = 0,
+	milliseconds = 0,
+} = {}) {
+	return requireTemporal().Duration.from({
+		years,
+		months,
+		weeks,
+		days,
+		hours,
+		minutes,
+		seconds,
+		milliseconds,
+	});
+}
+
+/**
+ * Create a Temporal.PlainDate from an ISO string.
+ *
+ * Purpose: calendar dates for daily reports and milestone reviews.
+ *
+ * Consumer: schedule and reporting code that needs date-only semantics.
+ */
+export function plainDateFromISO(isoString) {
+	return requireTemporal().PlainDate.from(isoString);
+}
+
+/**
+ * Create a Temporal.ZonedDateTime from an ISO string and time zone.
+ *
+ * Purpose: local-time reminders that survive DST changes.
+ *
+ * Consumer: scheduled reminders and audit views rendered in local time.
+ */
+export function zonedDateTimeFromISO(isoString, timeZone = "UTC") {
+	const Temporal = requireTemporal();
+	return Temporal.Instant.from(isoString).toZonedDateTimeISO(timeZone);
+}
+
+/**
+ * Serialize a Temporal value to a stable JSON-friendly object.
+ *
+ * Purpose: store semantic type plus ISO string so consumers know how to parse.
+ *
+ * Consumer: DB records and generated projections that persist Temporal values.
+ */
+export function serializeTemporal(value) {
+	if (!value) return null;
+	const Temporal = requireTemporal();
+	if (value instanceof Temporal.Instant) {
+		return { type: "Instant", iso: value.toString() };
+	}
+	if (value instanceof Temporal.Duration) {
+		return { type: "Duration", iso: value.toString() };
+	}
+	if (value instanceof Temporal.PlainDate) {
+		return { type: "PlainDate", iso: value.toString() };
+	}
+	if (value instanceof Temporal.ZonedDateTime) {
+		return {
+			type: "ZonedDateTime",
+			iso: value.toString(),
+			timeZone: value.timeZoneId,
+		};
+	}
+	throw new TypeError(
+		`Unsupported Temporal value: ${Object.prototype.toString.call(value)}`,
+	);
+}
+
+/**
+ * Deserialize a serialized Temporal value.
+ *
+ * Purpose: reconstruct the same native Temporal type that was stored.
+ *
+ * Consumer: schedule and UOK state loaders.
+ */
+export function deserializeTemporal(serialized) {
+	if (!serialized || typeof serialized !== "object") return null;
+	const Temporal = requireTemporal();
+	switch (serialized.type) {
+		case "Instant":
+			return Temporal.Instant.from(serialized.iso);
+		case "Duration":
+			return Temporal.Duration.from(serialized.iso);
+		case "PlainDate":
+			return Temporal.PlainDate.from(serialized.iso);
+		case "ZonedDateTime":
+			return Temporal.ZonedDateTime.from(serialized.iso);
+		default:
+			throw new TypeError(`Unknown Temporal type: ${serialized.type}`);
+	}
+}
+
+/**
+ * Validate that a schedule/lease record has correct Temporal serialization.
+ *
+ * Purpose: reject ambiguous Date-like records before they enter durable state.
+ *
+ * Consumer: schedule and UOK record validation.
+ */
+export function validateTemporalRecord(record) {
+	const errors = [];
+	const validTypes = ["Instant", "Duration", "PlainDate", "ZonedDateTime"];
+	for (const [key, value] of Object.entries(record)) {
+		if (value && typeof value === "object" && value.type) {
+			if (!validTypes.includes(value.type)) {
+				errors.push(`Field "${key}" has unknown Temporal type: ${value.type}`);
+			}
+			if (validTypes.includes(value.type) && !value.iso) {
+				errors.push(`Field "${key}" is missing ISO string`);
+			}
+		}
+	}
+	return { valid: errors.length === 0, errors };
+}
diff --git a/src/resources/extensions/sf/tests/direct-command-surface.test.mjs b/src/resources/extensions/sf/tests/direct-command-surface.test.mjs
index c4d8522a0..9a8935849 100644
--- a/src/resources/extensions/sf/tests/direct-command-surface.test.mjs
+++ b/src/resources/extensions/sf/tests/direct-command-surface.test.mjs
@@ -1,4 +1,6 @@
 import assert from "node:assert/strict";
+import { readFileSync } from "node:fs";
+import { join } from "node:path";
 import { test } from "vitest";
 
 import {
@@ -42,3 +44,25 @@ test("direct command completions strip the already typed command name", () => {
 		},
 	]);
 });
+
+test("human_facing_cli_help_when_describing_sf_surfaces_uses_direct_commands", () => {
+	const sourceFiles = ["src/help-text.ts", "src/cli.ts", "src/loader.ts"];
+	for (const sourceFile of sourceFiles) {
+		const source = readFileSync(join(process.cwd(), sourceFile), "utf8");
+		assert.equal(
+			source.includes("Machine surface for /sf"),
+			false,
+			`${sourceFile} must describe headless as the direct-command machine surface`,
+		);
+		assert.equal(
+			source.includes("Run /sf autonomous"),
+			false,
+			`${sourceFile} must not advertise /sf autonomous`,
+		);
+		assert.equal(
+			source.includes("Manage: /sf schedule list"),
+			false,
+			`${sourceFile} must not advertise the removed slash namespace`,
+		);
+	}
+});
diff --git a/src/resources/extensions/sf/tests/operating-model.test.mjs b/src/resources/extensions/sf/tests/operating-model.test.mjs
index cee0399cc..aeaa7f6e5 100644
--- a/src/resources/extensions/sf/tests/operating-model.test.mjs
+++ b/src/resources/extensions/sf/tests/operating-model.test.mjs
@@ -9,14 +9,16 @@ import {
 	isRunControlMode,
 	isWorkMode,
 	MODEL_MODES,
+	modelModeToTier,
 	PERMISSION_PROFILES,
+	RUN_CONTROL_MODES,
 	resolveModelMode,
 	resolvePermissionProfile,
 	resolveRunControlMode,
 	resolveWorkMode,
-	RUN_CONTROL_MODES,
-	WORK_MODES,
 	runControlModeForSession,
+	tierToModelMode,
+	WORK_MODES,
 } from "../operating-model.js";
 
 describe("operating model vocabulary", () => {
@@ -120,4 +122,18 @@ describe("operating model vocabulary", () => {
 		assert.equal(state.modelMode, "smart");
 		assert.equal(state.surface, "tui");
 	});
+
+	test("modelModeToTier_maps_fast_smart_deep_to_light_standard_heavy", () => {
+		assert.equal(modelModeToTier("fast"), "light");
+		assert.equal(modelModeToTier("smart"), "standard");
+		assert.equal(modelModeToTier("deep"), "heavy");
+		assert.equal(modelModeToTier("unknown"), "standard");
+	});
+
+	test("tierToModelMode_maps_light_standard_heavy_to_fast_smart_deep", () => {
+		assert.equal(tierToModelMode("light"), "fast");
+		assert.equal(tierToModelMode("standard"), "smart");
+		assert.equal(tierToModelMode("heavy"), "deep");
+		assert.equal(tierToModelMode("unknown"), "smart");
+	});
 });
diff --git a/src/resources/extensions/sf/tests/parallel-intent.test.mjs b/src/resources/extensions/sf/tests/parallel-intent.test.mjs
new file mode 100644
index 000000000..9f7d297db
--- /dev/null
+++ b/src/resources/extensions/sf/tests/parallel-intent.test.mjs
@@ -0,0 +1,98 @@
+import { existsSync, mkdirSync, mkdtempSync, rmSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { afterEach, beforeEach, describe, expect, test } from "vitest";
+import {
+	checkIntentConflicts,
+	clearAllIntents,
+	declareIntent,
+	getActiveIntents,
+	releaseIntent,
+} from "../parallel-intent.js";
+import { closeDatabase } from "../sf-db.js";
+
+/**
+ * Parallel intent registry tests.
+ */
+function makeBasePath() {
+	const dir = mkdtempSync(join(tmpdir(), "sf-intent-"));
+	mkdirSync(join(dir, ".sf"), { recursive: true });
+	return dir;
+}
+
+function cleanup(basePath) {
+	try {
+		rmSync(basePath, { recursive: true, force: true });
+	} catch {
+		/* ignore */
+	}
+}
+
+describe("parallel intent registry", () => {
+	let basePath;
+
+	beforeEach(() => {
+		basePath = makeBasePath();
+	});
+
+	afterEach(() => {
+		closeDatabase();
+		cleanup(basePath);
+		basePath = null;
+	});
+
+	test("declareIntent_when_valid_input_records_claim", () => {
+		const result = declareIntent(basePath, "M001", [
+			"src/foo.ts",
+			"src/bar.ts",
+		]);
+		expect(result.ok).toBe(true);
+		const intents = getActiveIntents(basePath);
+		expect(intents).toHaveLength(1);
+		expect(intents[0].milestoneId).toBe("M001");
+		expect(intents[0].files).toEqual(["src/foo.ts", "src/bar.ts"]);
+		expect(intents[0].status).toBe("claimed");
+		expect(existsSync(join(basePath, ".sf", "parallel"))).toBe(false);
+	});
+
+	test("checkIntentConflicts_when_files_overlap_reports_existing_milestone", () => {
+		declareIntent(basePath, "M001", ["src/foo.ts", "src/bar.ts"]);
+		const conflicts = checkIntentConflicts(basePath, "M002", [
+			"src/bar.ts",
+			"src/baz.ts",
+		]);
+		expect(conflicts).toHaveLength(1);
+		expect(conflicts[0].milestoneId).toBe("M001");
+		expect(conflicts[0].files).toEqual(["src/bar.ts"]);
+	});
+
+	test("checkIntentConflicts_when_no_overlap_returns_empty_list", () => {
+		declareIntent(basePath, "M001", ["src/foo.ts"]);
+		const conflicts = checkIntentConflicts(basePath, "M002", ["src/bar.ts"]);
+		expect(conflicts).toHaveLength(0);
+	});
+
+	test("releaseIntent_when_known_milestone_clears_active_intent", () => {
+		declareIntent(basePath, "M001", ["src/foo.ts"]);
+		releaseIntent(basePath, "M001");
+		const intents = getActiveIntents(basePath);
+		expect(intents).toHaveLength(0);
+	});
+
+	test("clearAllIntents_when_multiple_claims_removes_all_records", () => {
+		declareIntent(basePath, "M001", ["src/foo.ts"]);
+		declareIntent(basePath, "M002", ["src/bar.ts"]);
+		clearAllIntents(basePath);
+		const intents = getActiveIntents(basePath);
+		expect(intents).toHaveLength(0);
+	});
+
+	test("getActiveIntents_when_release_exists_ignores_released_records", () => {
+		declareIntent(basePath, "M001", ["src/foo.ts"]);
+		declareIntent(basePath, "M002", ["src/bar.ts"]);
+		releaseIntent(basePath, "M001");
+		const intents = getActiveIntents(basePath);
+		expect(intents).toHaveLength(1);
+		expect(intents[0].milestoneId).toBe("M002");
+	});
+});
diff --git a/src/resources/extensions/sf/tests/sf-db-migration.test.mjs b/src/resources/extensions/sf/tests/sf-db-migration.test.mjs
index 086dcb2fc..f31fb3a07 100644
--- a/src/resources/extensions/sf/tests/sf-db-migration.test.mjs
+++ b/src/resources/extensions/sf/tests/sf-db-migration.test.mjs
@@ -217,7 +217,7 @@ test("openDatabase_migrates_v27_tasks_without_created_at_through_spec_backfill",
 	const version = db
 		.prepare("SELECT MAX(version) AS version FROM schema_version")
 		.get();
-	assert.equal(version.version, 42);
+	assert.equal(version.version, 43);
 	const taskSpec = db
 		.prepare(
 			"SELECT milestone_id, slice_id, task_id, verify FROM task_specs WHERE task_id = 'T01'",
diff --git a/src/resources/extensions/sf/tests/skill-eval-harness.test.mjs b/src/resources/extensions/sf/tests/skill-eval-harness.test.mjs
new file mode 100644
index 000000000..19f7addf8
--- /dev/null
+++ b/src/resources/extensions/sf/tests/skill-eval-harness.test.mjs
@@ -0,0 +1,115 @@
+/**
+ * Tests for skill eval harness.
+ *
+ * Purpose: verify eval-case materialization and grader execution are stable.
+ */
+
+import {
+	existsSync,
+	mkdirSync,
+	mkdtempSync,
+	readFileSync,
+	rmSync,
+} from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { describe, expect, test } from "vitest";
+import {
+	createEvalCase,
+	generateDefaultEvalCase,
+	runGrader,
+	runSkillEvals,
+} from "../skills/eval-harness.js";
+
+function makeSkillDir() {
+	const dir = mkdtempSync(join(tmpdir(), "sf-skill-eval-"));
+	mkdirSync(join(dir, ".agents", "skills", "test-skill"), { recursive: true });
+	return join(dir, ".agents", "skills", "test-skill");
+}
+
+function cleanup(dir) {
+	try {
+		rmSync(dir, { recursive: true, force: true });
+	} catch {
+		/* ignore */
+	}
+}
+
+describe("skill eval harness", () => {
+	test("createEvalCase_creates_case_files_and_workspace", () => {
+		const skillDir = makeSkillDir();
+		try {
+			const caseDef = {
+				task: "Test task for skill",
+				grader:
+					"export async function grade() { return { passed: true, score: 1 }; }",
+			};
+			const result = createEvalCase(skillDir, "case-1", caseDef);
+			expect(result.caseName).toBe("case-1");
+			expect(existsSync(join(skillDir, "evals", "case-1", "task.md"))).toBe(
+				true,
+			);
+			expect(existsSync(join(skillDir, "evals", "case-1", "grader.js"))).toBe(
+				true,
+			);
+			expect(existsSync(join(skillDir, "evals", "case-1", "work"))).toBe(true);
+		} finally {
+			cleanup(skillDir);
+		}
+	});
+
+	test("createEvalCase_with_hidden_reference_writes_fixture", () => {
+		const skillDir = makeSkillDir();
+		try {
+			const caseDef = {
+				task: "Test task",
+				grader:
+					"export async function grade() { return { passed: true, score: 1 }; }",
+				hidden: { answer: "expected-output" },
+			};
+			createEvalCase(skillDir, "case-2", caseDef);
+			const hiddenPath = join(
+				skillDir,
+				"evals",
+				"case-2",
+				"hidden",
+				"reference.json",
+			);
+			expect(existsSync(hiddenPath)).toBe(true);
+			const hidden = JSON.parse(readFileSync(hiddenPath, "utf-8"));
+			expect(hidden.answer).toBe("expected-output");
+		} finally {
+			cleanup(skillDir);
+		}
+	});
+
+	test("runGrader_returns_error_when_grader_missing", async () => {
+		const skillDir = makeSkillDir();
+		try {
+			mkdirSync(join(skillDir, "evals", "bad-case"), { recursive: true });
+			const result = await runGrader(join(skillDir, "evals", "bad-case"), {});
+			expect(result.passed).toBe(false);
+			expect(result.details[0]).toBe("Grader not found");
+		} finally {
+			cleanup(skillDir);
+		}
+	});
+
+	test("generateDefaultEvalCase_mentions_skill_name_and_exported_grader", () => {
+		const skill = { name: "forge-test", description: "Test skill" };
+		const caseDef = generateDefaultEvalCase(skill);
+		expect(caseDef.task.includes("forge-test")).toBe(true);
+		expect(caseDef.grader.includes("export async function grade")).toBe(true);
+	});
+
+	test("runSkillEvals_returns_empty_summary_without_evals", async () => {
+		const skillDir = makeSkillDir();
+		try {
+			const result = await runSkillEvals(skillDir, {});
+			expect(result.totalCases).toBe(0);
+			expect(result.passedCases).toBe(0);
+		} finally {
+			cleanup(skillDir);
+		}
+	});
+});
diff --git a/src/resources/extensions/sf/tests/skills.test.mjs b/src/resources/extensions/sf/tests/skills.test.mjs
new file mode 100644
index 000000000..2ff71313e
--- /dev/null
+++ b/src/resources/extensions/sf/tests/skills.test.mjs
@@ -0,0 +1,253 @@
+/**
+ * Skills system tests.
+ *
+ * Purpose: verify skill discovery, frontmatter parsing, validation, and loading.
+ */
+import { mkdirSync, mkdtempSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { afterEach, beforeEach, describe, expect, test } from "vitest";
+import { discoverSkillDirs, readSkillFile } from "../skills/directory.js";
+import {
+	buildSkillRecord,
+	parseSkillFrontmatter,
+	validateSkillFrontmatter,
+} from "../skills/frontmatter.js";
+import {
+	getModelInvocableSkills,
+	getPermittedSkills,
+	loadSkills,
+} from "../skills/loader.js";
+
+describe("skill frontmatter", () => {
+	test("parseSkillFrontmatter_extracts_yaml_and_body", () => {
+		const content = `---
+name: test-skill
+description: A test skill
+user-invocable: true
+model-invocable: false
+side-effects: none
+permission-profile: restricted
+---
+
+# Test Skill
+
+Some instructions.
+`;
+		const parsed = parseSkillFrontmatter(content);
+		expect(parsed).toBeTruthy();
+		expect(parsed.frontmatter.name).toBe("test-skill");
+		expect(parsed.frontmatter.description).toBe("A test skill");
+		expect(parsed.frontmatter["user-invocable"]).toBe(true);
+		expect(parsed.frontmatter["model-invocable"]).toBe(false);
+		expect(parsed.frontmatter["side-effects"]).toBe("none");
+		expect(parsed.frontmatter["permission-profile"]).toBe("restricted");
+		expect(parsed.body).toContain("# Test Skill");
+	});
+
+	test("parseSkillFrontmatter_returns_null_without_frontmatter", () => {
+		const parsed = parseSkillFrontmatter("# Just markdown\n\nNo frontmatter.");
+		expect(parsed).toBeNull();
+	});
+
+	test("validateSkillFrontmatter_passes_complete_frontmatter", () => {
+		const result = validateSkillFrontmatter({
+			name: "test",
+			description: "desc",
+			"user-invocable": true,
+			"model-invocable": false,
+			"side-effects": "none",
+			"permission-profile": "normal",
+		});
+		expect(result.valid).toBe(true);
+		expect(result.errors).toHaveLength(0);
+	});
+
+	test("validateSkillFrontmatter_fails_missing_fields", () => {
+		const result = validateSkillFrontmatter({
+			name: "test",
+		});
+		expect(result.valid).toBe(false);
+		expect(result.errors.length).toBeGreaterThan(0);
+		expect(result.errors.some((e) => e.includes("description"))).toBe(true);
+		expect(result.errors.some((e) => e.includes("user-invocable"))).toBe(true);
+	});
+
+	test("validateSkillFrontmatter_fails_invalid_permission_profile", () => {
+		const result = validateSkillFrontmatter({
+			name: "test",
+			description: "desc",
+			"user-invocable": true,
+			"model-invocable": false,
+			"side-effects": "none",
+			"permission-profile": "invalid",
+		});
+		expect(result.valid).toBe(false);
+		expect(result.errors.some((e) => e.includes("permission-profile"))).toBe(
+			true,
+		);
+	});
+
+	test("buildSkillRecord_creates_canonical_record", () => {
+		const record = buildSkillRecord(
+			"/path/to/skill",
+			{
+				name: "test-skill",
+				description: "A test",
+				"user-invocable": true,
+				"model-invocable": true,
+				"side-effects": "code-edits",
+				"permission-profile": "normal",
+				triggers: ["build", "code"],
+				"max-activations": 5,
+			},
+			"# Body",
+		);
+		expect(record.name).toBe("test-skill");
+		expect(record.userInvocable).toBe(true);
+		expect(record.modelInvocable).toBe(true);
+		expect(record.sideEffects).toBe("code-edits");
+		expect(record.permissionProfile).toBe("normal");
+		expect(record.triggers).toEqual(["build", "code"]);
+		expect(record.maxActivations).toBe(5);
+		expect(record.body).toBe("# Body");
+		expect(record.path).toBe("/path/to/skill");
+	});
+});
+
+describe("skill discovery", () => {
+	let tmpDir;
+
+	beforeEach(() => {
+		tmpDir = mkdtempSync(join(tmpdir(), "sf-skills-test-"));
+	});
+
+	afterEach(() => {
+		rmSync(tmpDir, { recursive: true, force: true });
+	});
+
+	test("discoverSkillDirs_finds_skills_with_skill_md", () => {
+		const skillDir = join(tmpDir, ".agents", "skills", "test-skill");
+		mkdirSync(skillDir, { recursive: true });
+		writeFileSync(
+			join(skillDir, "SKILL.md"),
+			`---\nname: test-skill\ndescription: Test\nuser-invocable: true\nmodel-invocable: true\nside-effects: none\npermission-profile: restricted\n---\n\n# Test\n`,
+		);
+
+		const dirs = discoverSkillDirs(tmpDir);
+		expect(dirs).toHaveLength(1);
+		expect(dirs[0].name).toBe("test-skill");
+	});
+
+	test("discoverSkillDirs_ignores_dirs_without_skill_md", () => {
+		const skillDir = join(tmpDir, ".agents", "skills", "empty-dir");
+		mkdirSync(skillDir, { recursive: true });
+
+		const dirs = discoverSkillDirs(tmpDir);
+		expect(dirs).toHaveLength(0);
+	});
+
+	test("readSkillFile_reads_content", () => {
+		const skillDir = join(tmpDir, ".agents", "skills", "test-skill");
+		mkdirSync(skillDir, { recursive: true });
+		writeFileSync(join(skillDir, "SKILL.md"), "# Test Skill\n");
+
+		const content = readSkillFile(skillDir);
+		expect(content).toBe("# Test Skill\n");
+	});
+
+	test("readSkillFile_returns_null_for_missing", () => {
+		const content = readSkillFile(join(tmpDir, "nonexistent"));
+		expect(content).toBeNull();
+	});
+});
+
+describe("skill loading", () => {
+	let tmpDir;
+
+	beforeEach(() => {
+		tmpDir = mkdtempSync(join(tmpdir(), "sf-skills-test-"));
+	});
+
+	afterEach(() => {
+		rmSync(tmpDir, { recursive: true, force: true });
+	});
+
+	function createSkill(name, overrides = {}) {
+		const skillDir = join(tmpDir, ".agents", "skills", name);
+		mkdirSync(skillDir, { recursive: true });
+		const frontmatter = {
+			name,
+			description: overrides.description ?? `Description for ${name}`,
+			"user-invocable": overrides.userInvocable ?? true,
+			"model-invocable": overrides.modelInvocable ?? true,
+			"side-effects": overrides.sideEffects ?? "none",
+			"permission-profile": overrides.permissionProfile ?? "normal",
+			triggers: overrides.triggers ?? ["*"],
+			...overrides.extra,
+		};
+		const yaml = Object.entries(frontmatter)
+			.map(([k, v]) => {
+				if (Array.isArray(v))
+					return `${k}:\n${v.map((i) => `  - ${i}`).join("\n")}`;
+				return `${k}: ${v}`;
+			})
+			.join("\n");
+		writeFileSync(
+			join(skillDir, "SKILL.md"),
+			`---\n${yaml}\n---\n\n# ${name}\n`,
+		);
+	}
+
+	test("loadSkills_loads_valid_skills", () => {
+		createSkill("skill-a");
+		createSkill("skill-b", { permissionProfile: "trusted" });
+
+		const skills = loadSkills(tmpDir);
+		expect(skills).toHaveLength(2);
+		expect(skills.every((s) => s.valid)).toBe(true);
+		expect(skills.some((s) => s.name === "skill-a")).toBe(true);
+		expect(skills.some((s) => s.name === "skill-b")).toBe(true);
+	});
+
+	test("loadSkills_marks_invalid_skills", () => {
+		createSkill("valid-skill");
+		const badDir = join(tmpDir, ".agents", "skills", "bad-skill");
+		mkdirSync(badDir, { recursive: true });
+		writeFileSync(join(badDir, "SKILL.md"), "No frontmatter here.");
+
+		const skills = loadSkills(tmpDir);
+		expect(skills).toHaveLength(2);
+		const bad = skills.find((s) => s.name === "bad-skill");
+		expect(bad).toBeTruthy();
+		expect(bad.valid).toBe(false);
+	});
+
+	test("getPermittedSkills_filters_by_profile", () => {
+		createSkill("restricted-skill", { permissionProfile: "restricted" });
+		createSkill("normal-skill", { permissionProfile: "normal" });
+		createSkill("trusted-skill", { permissionProfile: "trusted" });
+
+		const skills = loadSkills(tmpDir);
+		const permitted = getPermittedSkills(skills, "normal");
+		expect(permitted).toHaveLength(2);
+		expect(permitted.some((s) => s.name === "restricted-skill")).toBe(true);
+		expect(permitted.some((s) => s.name === "normal-skill")).toBe(true);
+		expect(permitted.some((s) => s.name === "trusted-skill")).toBe(false);
+	});
+
+	test("getModelInvocableSkills_filters_by_work_mode", () => {
+		createSkill("build-skill", { triggers: ["build"], modelInvocable: true });
+		createSkill("review-skill", { triggers: ["review"], modelInvocable: true });
+		createSkill("universal-skill", { triggers: ["*"], modelInvocable: true });
+		createSkill("user-only", { triggers: ["*"], modelInvocable: false });
+
+		const skills = loadSkills(tmpDir);
+		const buildSkills = getModelInvocableSkills(skills, "build");
+		expect(buildSkills).toHaveLength(2);
+		expect(buildSkills.some((s) => s.name === "build-skill")).toBe(true);
+		expect(buildSkills.some((s) => s.name === "universal-skill")).toBe(true);
+		expect(buildSkills.some((s) => s.name === "review-skill")).toBe(false);
+		expect(buildSkills.some((s) => s.name === "user-only")).toBe(false);
+	});
+});
diff --git a/src/resources/extensions/sf/tests/temporal-foundation.test.mjs b/src/resources/extensions/sf/tests/temporal-foundation.test.mjs
new file mode 100644
index 000000000..c523fec75
--- /dev/null
+++ b/src/resources/extensions/sf/tests/temporal-foundation.test.mjs
@@ -0,0 +1,102 @@
+/**
+ * Temporal foundation tests.
+ *
+ * Purpose: ensure runtime contracts are enforced on native Temporal.
+ */
+import { describe, expect, test } from "vitest";
+
+import {
+	durationFromObject,
+	hasNativeTemporal,
+	instantFromISO,
+	instantNow,
+	plainDateFromISO,
+	serializeTemporal,
+	validateTemporalRecord,
+} from "../temporal-foundation.js";
+
+const hasTemporal = hasNativeTemporal();
+const Temporal = globalThis.Temporal;
+
+if (!hasTemporal) {
+	test.skip("native Temporal is required by the Node 26 baseline", () => {});
+} else {
+	describe("temporal foundation", () => {
+		test("hasNativeTemporal_reports_true_in_runtime", () => {
+			expect(hasTemporal).toBe(true);
+			expect(Temporal).toBeTypeOf("object");
+		});
+
+		test("instantFromISO_parses_iso_to_native_instant", () => {
+			const iso = "2026-05-08T12:00:00Z";
+			const instant = instantFromISO(iso);
+			expect(instant).toBeInstanceOf(Temporal.Instant);
+			expect(instant.toString()).toBe(iso);
+		});
+
+		test("instantNow_returns_current_native_instant", () => {
+			const before = Date.now();
+			const instant = instantNow();
+			const after = Date.now();
+			expect(instant).toBeInstanceOf(Temporal.Instant);
+			expect(instant.epochMilliseconds).toBeGreaterThanOrEqual(before);
+			expect(instant.epochMilliseconds).toBeLessThanOrEqual(after);
+		});
+
+		test("durationFromObject_returns_native_duration", () => {
+			const dur = durationFromObject({ hours: 1, minutes: 30 });
+			expect(dur).toBeInstanceOf(Temporal.Duration);
+			expect(dur.hours).toBe(1);
+			expect(dur.minutes).toBe(30);
+		});
+
+		test("plainDateFromISO_returns_native_plain_date", () => {
+			const date = plainDateFromISO("2026-05-08");
+			expect(date).toBeInstanceOf(Temporal.PlainDate);
+			expect(date.year).toBe(2026);
+			expect(date.month).toBe(5);
+			expect(date.day).toBe(8);
+		});
+
+		test("serializeTemporal_roundtrips_native_instant", () => {
+			const instant = instantFromISO("2026-05-08T12:00:00Z");
+			const serialized = serializeTemporal(instant);
+			expect(serialized).toEqual({
+				type: "Instant",
+				iso: "2026-05-08T12:00:00Z",
+			});
+		});
+
+		test("validateTemporalRecord_passes_valid", () => {
+			const record = {
+				createdAt: { type: "Instant", iso: "2026-05-08T12:00:00Z" },
+				leaseDuration: { type: "Duration", iso: "PT1H" },
+			};
+			const result = validateTemporalRecord(record);
+			expect(result.valid).toBe(true);
+			expect(result.errors).toHaveLength(0);
+		});
+
+		test("validateTemporalRecord_fails_unknown_type", () => {
+			const record = {
+				bad: { type: "UnknownType", iso: "x" },
+			};
+			const result = validateTemporalRecord(record);
+			expect(result.valid).toBe(false);
+			expect(
+				result.errors.some((e) => e.includes("unknown Temporal type")),
+			).toBe(true);
+		});
+
+		test("validateTemporalRecord_fails_missing_iso", () => {
+			const record = {
+				bad: { type: "Instant" },
+			};
+			const result = validateTemporalRecord(record);
+			expect(result.valid).toBe(false);
+			expect(result.errors.some((e) => e.includes("missing ISO string"))).toBe(
+				true,
+			);
+		});
+	});
+}
diff --git a/src/resources/extensions/sf/tests/uok-coordination-store.test.mjs b/src/resources/extensions/sf/tests/uok-coordination-store.test.mjs
index d6e0f4d8d..9c802fce0 100644
--- a/src/resources/extensions/sf/tests/uok-coordination-store.test.mjs
+++ b/src/resources/extensions/sf/tests/uok-coordination-store.test.mjs
@@ -59,6 +59,22 @@ test("kv_cleanupExpired_removes_expired_rows", () => {
 	assert.equal(store.get("b"), 2);
 });
 
+test("kv_entries_when_prefix_matches_returns_live_decoded_rows", () => {
+	let now = 1_000;
+	const store = makeStore(() => now);
+	store.set("parallel:intent:M001", { milestoneId: "M001" });
+	store.set("parallel:intent:M002", { milestoneId: "M002" }, { ttlMs: 1 });
+	store.set("skills:auto-create:last-run", { lastRunMs: 1 });
+
+	now = 1_002;
+	const entries = store.entries("parallel:intent:");
+
+	assert.deepEqual(
+		entries.map((entry) => entry.value.milestoneId),
+		["M001"],
+	);
+});
+
 test("streams_append_and_read_after_id_in_order", () => {
 	const store = makeStore();
 	const first = store.xadd("events", "node-start", { id: "a" });
@@ -70,7 +86,10 @@ test("streams_append_and_read_after_id_in_order", () => {
 		store.xread("events", { afterId: 1 }).map((e) => e.type),
 		["node-complete"],
 	);
-	assert.deepEqual(store.xread("events").map((e) => e.payload.id), ["a", "a"]);
+	assert.deepEqual(
+		store.xread("events").map((e) => e.payload.id),
+		["a", "a"],
+	);
 });
 
 test("queue_claim_ack_and_release_enforces_owner_leases", () => {
diff --git a/src/resources/extensions/sf/tests/uok-execution-graph-persist.test.mjs b/src/resources/extensions/sf/tests/uok-execution-graph-persist.test.mjs
index 3892ce02e..47b1fd2f8 100644
--- a/src/resources/extensions/sf/tests/uok-execution-graph-persist.test.mjs
+++ b/src/resources/extensions/sf/tests/uok-execution-graph-persist.test.mjs
@@ -7,11 +7,7 @@
 
 import assert from "node:assert/strict";
 import { afterEach, test } from "vitest";
-import {
-	closeDatabase,
-	getDatabase,
-	openDatabase,
-} from "../sf-db.js";
+import { closeDatabase, getDatabase, openDatabase } from "../sf-db.js";
 import {
 	ensureExecutionGraphSchema,
 	getGraphProgressStream,
@@ -170,8 +166,20 @@ test("persistGraphNode_updates_existing_node", () => {
 test("persistFullGraph_writes_snapshot_and_nodes_in_transaction", () => {
 	const db = makeDb();
 	const nodes = [
-		{ id: "n1", kind: "unit", unitType: "execute-task", unitId: "U1", state: "todo" },
-		{ id: "n2", kind: "unit", unitType: "execute-task", unitId: "U2", state: "todo" },
+		{
+			id: "n1",
+			kind: "unit",
+			unitType: "execute-task",
+			unitId: "U1",
+			state: "todo",
+		},
+		{
+			id: "n2",
+			kind: "unit",
+			unitType: "execute-task",
+			unitId: "U2",
+			state: "todo",
+		},
 	];
 
 	persistFullGraph(db, "g2", nodes, {
@@ -196,9 +204,24 @@ test("queryTasksByState_filters_by_state", () => {
 	const db = makeDb();
 	ensureExecutionGraphSchema(db);
 
-	persistGraphNode(db, "g1", { id: "n1", kind: "unit", unitId: "U1", state: "todo" });
-	persistGraphNode(db, "g1", { id: "n2", kind: "unit", unitId: "U2", state: "done" });
-	persistGraphNode(db, "g1", { id: "n3", kind: "unit", unitId: "U3", state: "in_progress" });
+	persistGraphNode(db, "g1", {
+		id: "n1",
+		kind: "unit",
+		unitId: "U1",
+		state: "todo",
+	});
+	persistGraphNode(db, "g1", {
+		id: "n2",
+		kind: "unit",
+		unitId: "U2",
+		state: "done",
+	});
+	persistGraphNode(db, "g1", {
+		id: "n3",
+		kind: "unit",
+		unitId: "U3",
+		state: "in_progress",
+	});
 
 	const todo = queryTasksByState(db, { graphId: "g1", states: ["todo"] });
 	assert.equal(todo.length, 1);
@@ -213,8 +236,20 @@ test("queryTasksByState_filters_by_milestone", () => {
 	const db = makeDb();
 	ensureExecutionGraphSchema(db);
 
-	persistGraphNode(db, "g1", { id: "n1", kind: "unit", unitId: "U1", milestoneId: "M001", state: "todo" });
-	persistGraphNode(db, "g1", { id: "n2", kind: "unit", unitId: "U2", milestoneId: "M002", state: "todo" });
+	persistGraphNode(db, "g1", {
+		id: "n1",
+		kind: "unit",
+		unitId: "U1",
+		milestoneId: "M001",
+		state: "todo",
+	});
+	persistGraphNode(db, "g1", {
+		id: "n2",
+		kind: "unit",
+		unitId: "U2",
+		milestoneId: "M002",
+		state: "todo",
+	});
 
 	const m1 = queryTasksByState(db, { milestoneId: "M001" });
 	assert.equal(m1.length, 1);
@@ -247,10 +282,30 @@ test("getGraphStateSummary_computes_counts_and_progress", () => {
 	const db = makeDb();
 	ensureExecutionGraphSchema(db);
 
-	persistGraphNode(db, "g1", { id: "n1", kind: "unit", unitId: "U1", state: "todo" });
-	persistGraphNode(db, "g1", { id: "n2", kind: "unit", unitId: "U2", state: "in_progress" });
-	persistGraphNode(db, "g1", { id: "n3", kind: "unit", unitId: "U3", state: "done" });
-	persistGraphNode(db, "g1", { id: "n4", kind: "unit", unitId: "U4", state: "done" });
+	persistGraphNode(db, "g1", {
+		id: "n1",
+		kind: "unit",
+		unitId: "U1",
+		state: "todo",
+	});
+	persistGraphNode(db, "g1", {
+		id: "n2",
+		kind: "unit",
+		unitId: "U2",
+		state: "in_progress",
+	});
+	persistGraphNode(db, "g1", {
+		id: "n3",
+		kind: "unit",
+		unitId: "U3",
+		state: "done",
+	});
+	persistGraphNode(db, "g1", {
+		id: "n4",
+		kind: "unit",
+		unitId: "U4",
+		state: "done",
+	});
 
 	const summary = getGraphStateSummary(db, "g1");
 	assert.equal(summary.total, 4);
@@ -266,8 +321,18 @@ test("getGraphStateSummary_complete_when_all_terminal", () => {
 	const db = makeDb();
 	ensureExecutionGraphSchema(db);
 
-	persistGraphNode(db, "g1", { id: "n1", kind: "unit", unitId: "U1", state: "done" });
-	persistGraphNode(db, "g1", { id: "n2", kind: "unit", unitId: "U2", state: "failed" });
+	persistGraphNode(db, "g1", {
+		id: "n1",
+		kind: "unit",
+		unitId: "U1",
+		state: "done",
+	});
+	persistGraphNode(db, "g1", {
+		id: "n2",
+		kind: "unit",
+		unitId: "U2",
+		state: "failed",
+	});
 
 	const summary = getGraphStateSummary(db, "g1");
 	assert.equal(summary.progress, 100);
@@ -297,7 +362,10 @@ test("getGraphProgressStream_returns_recent_first", () => {
 	ensureExecutionGraphSchema(db);
 
 	persistProgressEvent(db, "g1", { type: "first", ts: "2024-01-01T00:00:00Z" });
-	persistProgressEvent(db, "g1", { type: "second", ts: "2024-01-01T00:00:01Z" });
+	persistProgressEvent(db, "g1", {
+		type: "second",
+		ts: "2024-01-01T00:00:01Z",
+	});
 
 	const events = getGraphProgressStream(db, "g1", 10);
 	assert.equal(events[0].type, "second");
diff --git a/src/resources/extensions/sf/tests/uok-scheduler-v2.test.mjs b/src/resources/extensions/sf/tests/uok-scheduler-v2.test.mjs
index 098eb594e..5324932c5 100644
--- a/src/resources/extensions/sf/tests/uok-scheduler-v2.test.mjs
+++ b/src/resources/extensions/sf/tests/uok-scheduler-v2.test.mjs
@@ -61,14 +61,18 @@ test("token_onCancel_fires_immediately_if_already_cancelled", () => {
 	const token = new CancellationToken();
 	token.cancel("done");
 	let fired = false;
-	token.onCancel(() => { fired = true; });
+	token.onCancel(() => {
+		fired = true;
+	});
 	assert.equal(fired, true);
 });
 
 test("token_onCancel_fires_when_cancelled_later", () => {
 	const token = new CancellationToken();
 	let fired = false;
-	token.onCancel(() => { fired = true; });
+	token.onCancel(() => {
+		fired = true;
+	});
 	assert.equal(fired, false);
 	token.cancel("later");
 	assert.equal(fired, true);
@@ -77,7 +81,9 @@ test("token_onCancel_fires_when_cancelled_later", () => {
 test("token_onCancel_unsubscribe_works", () => {
 	const token = new CancellationToken();
 	let fired = false;
-	const unsub = token.onCancel(() => { fired = true; });
+	const unsub = token.onCancel(() => {
+		fired = true;
+	});
 	unsub();
 	token.cancel("nope");
 	assert.equal(fired, false);
@@ -218,7 +224,7 @@ test("scheduler_serial_cancellation_stops_execution", async () => {
 	const scheduler = new ExecutionGraphSchedulerV2();
 	const order = [];
 	const globalToken = new CancellationToken();
-	scheduler.registerHandler("unit", async (node, token) => {
+	scheduler.registerHandler("unit", async (node, _token) => {
 		if (node.id === "b") globalToken.cancel("stop");
 		globalToken.throwIfCancelled();
 		order.push(node.id);
@@ -243,7 +249,7 @@ test("scheduler_serial_cancellation_stops_execution", async () => {
 test("scheduler_parallel_runs_independent_nodes_concurrently", async () => {
 	const scheduler = new ExecutionGraphSchedulerV2();
 	const starts = [];
-	scheduler.registerHandler("unit", async (node, token, progress) => {
+	scheduler.registerHandler("unit", async (node, _token, _progress) => {
 		starts.push(node.id);
 		await new Promise((r) => setTimeout(r, 50));
 		return { gateResults: [{ outcome: "pass", gateId: "test" }] };
@@ -264,7 +270,7 @@ test("scheduler_parallel_enforces_max_workers", async () => {
 	const scheduler = new ExecutionGraphSchedulerV2();
 	let concurrent = 0;
 	let maxConcurrent = 0;
-	scheduler.registerHandler("unit", async (node) => {
+	scheduler.registerHandler("unit", async (_node) => {
 		concurrent++;
 		maxConcurrent = Math.max(maxConcurrent, concurrent);
 		await new Promise((r) => setTimeout(r, 50));
@@ -285,13 +291,11 @@ test("scheduler_parallel_enforces_max_workers", async () => {
 
 test("scheduler_parallel_reports_progress_events", async () => {
 	const scheduler = new ExecutionGraphSchedulerV2();
-	scheduler.registerHandler("unit", async (node) => {
+	scheduler.registerHandler("unit", async (_node) => {
 		return { gateResults: [{ outcome: "pass", gateId: "test" }] };
 	});
 
-	const nodes = [
-		{ id: "a", kind: "unit", dependsOn: [], metadata: {} },
-	];
+	const nodes = [{ id: "a", kind: "unit", dependsOn: [], metadata: {} }];
 
 	const events = [];
 	scheduler.progress.onProgress((ev) => events.push(ev));
@@ -307,7 +311,7 @@ test("scheduler_parallel_reports_progress_events", async () => {
 
 test("scheduler_parallel_task_records_have_correct_state", async () => {
 	const scheduler = new ExecutionGraphSchedulerV2();
-	scheduler.registerHandler("unit", async (node) => {
+	scheduler.registerHandler("unit", async (_node) => {
 		return {
 			gateResults: [
 				{ outcome: "pass", gateId: "security" },
@@ -330,7 +334,7 @@ test("scheduler_parallel_task_records_have_correct_state", async () => {
 
 test("scheduler_parallel_detects_cyclic_dependencies", async () => {
 	const scheduler = new ExecutionGraphSchedulerV2();
-	scheduler.registerHandler("unit", async (node) => ({
+	scheduler.registerHandler("unit", async (_node) => ({
 		gateResults: [{ outcome: "pass", gateId: "test" }],
 	}));
 
@@ -339,15 +343,12 @@ test("scheduler_parallel_detects_cyclic_dependencies", async () => {
 		{ id: "b", kind: "unit", dependsOn: ["a"], metadata: {} },
 	];
 
-	await assert.rejects(
-		scheduler.run(nodes, { parallel: true }),
-		/cyclic/,
-	);
+	await assert.rejects(scheduler.run(nodes, { parallel: true }), /cyclic/);
 });
 
 test("scheduler_parallel_detects_deadlock", async () => {
 	const scheduler = new ExecutionGraphSchedulerV2();
-	scheduler.registerHandler("unit", async (node) => ({
+	scheduler.registerHandler("unit", async (_node) => ({
 		gateResults: [{ outcome: "pass", gateId: "test" }],
 	}));
 
@@ -357,10 +358,7 @@ test("scheduler_parallel_detects_deadlock", async () => {
 		{ id: "b", kind: "unit", dependsOn: ["a"], metadata: {} },
 	];
 
-	await assert.rejects(
-		scheduler.run(nodes, { parallel: true }),
-		/cyclic/,
-	);
+	await assert.rejects(scheduler.run(nodes, { parallel: true }), /cyclic/);
 });
 
 // ─── Progress stream wiring ────────────────────────────────────────────────
@@ -375,9 +373,7 @@ test("scheduler_wires_progress_to_parent_stream", async () => {
 	const parentEvents = [];
 	parent.onProgress((ev) => parentEvents.push(ev));
 
-	const nodes = [
-		{ id: "x", kind: "unit", dependsOn: [], metadata: {} },
-	];
+	const nodes = [{ id: "x", kind: "unit", dependsOn: [], metadata: {} }];
 
 	await scheduler.run(nodes, { parallel: false, parentStream: parent });
 	assert.ok(parentEvents.some((e) => e.type === "node-complete"));
diff --git a/src/resources/extensions/sf/tests/uok-task-state.test.mjs b/src/resources/extensions/sf/tests/uok-task-state.test.mjs
index 94b448a4b..c3eb0c3f3 100644
--- a/src/resources/extensions/sf/tests/uok-task-state.test.mjs
+++ b/src/resources/extensions/sf/tests/uok-task-state.test.mjs
@@ -8,13 +8,12 @@
 import assert from "node:assert/strict";
 import { test } from "vitest";
 import {
-	TASK_STATES,
-	TASK_TERMINAL_STATES,
-	TASK_STATE_TRANSITIONS,
 	aggregateTaskStates,
 	buildTaskRecord,
 	canTransitionTaskState,
 	gateOutcomesToTaskState,
+	TASK_STATES,
+	TASK_TERMINAL_STATES,
 	unitRuntimeToTaskState,
 } from "../uok/task-state.js";
 
diff --git a/src/resources/extensions/sf/uok/coordination-store.js b/src/resources/extensions/sf/uok/coordination-store.js
index 87b53a5e3..a2ccd843e 100644
--- a/src/resources/extensions/sf/uok/coordination-store.js
+++ b/src/resources/extensions/sf/uok/coordination-store.js
@@ -125,6 +125,24 @@ export class UokCoordinationStore {
 		return decode(row.value_json);
 	}
 
+	entries(prefix = "") {
+		this.cleanupExpired();
+		return this.db
+			.prepare(
+				`SELECT key, value_json, expires_at, updated_at
+         FROM uok_kv
+         WHERE key LIKE :prefix
+         ORDER BY updated_at DESC, key ASC`,
+			)
+			.all({ prefix: `${prefix}%` })
+			.map((row) => ({
+				key: row.key,
+				value: decode(row.value_json),
+				expiresAt: row.expires_at,
+				updatedAt: row.updated_at,
+			}));
+	}
+
 	delete(key) {
 		this.db.prepare("DELETE FROM uok_kv WHERE key = :key").run({ key });
 	}
diff --git a/src/resources/extensions/sf/uok/execution-graph-persist.js b/src/resources/extensions/sf/uok/execution-graph-persist.js
index 69bcbb00f..71d3613d1 100644
--- a/src/resources/extensions/sf/uok/execution-graph-persist.js
+++ b/src/resources/extensions/sf/uok/execution-graph-persist.js
@@ -298,9 +298,15 @@ export function getGraphStateSummary(db, graphId) {
 		.all({ graphId });
 
 	const counts = Object.fromEntries(
-		["todo", "in_progress", "review", "done", "retrying", "failed", "cancelled"].map(
-			(s) => [s, 0],
-		),
+		[
+			"todo",
+			"in_progress",
+			"review",
+			"done",
+			"retrying",
+			"failed",
+			"cancelled",
+		].map((s) => [s, 0]),
 	);
 	let total = 0;
 	for (const r of rows) {
diff --git a/src/resources/extensions/sf/uok/index.js b/src/resources/extensions/sf/uok/index.js
index 5a56dc165..8b8e42875 100644
--- a/src/resources/extensions/sf/uok/index.js
+++ b/src/resources/extensions/sf/uok/index.js
@@ -9,161 +9,157 @@
  * extension that needs orchestration primitives.
  */
 
-// ─── Contracts & Types ────────────────────────────────────────────────────
-export { validateGate } from "./contracts.js";
-
-// ─── Core Kernel ───────────────────────────────────────────────────────────
-export { recordUokKernelTermination, runAutoLoopWithUok } from "./kernel.js";
-
-// ─── Gate System ───────────────────────────────────────────────────────────
-export { UokGateRunner, enrichGateResultWithMemory } from "./gate-runner.js";
+// ─── Skills (Repo-local capabilities) ──────────────────────────────────────
+export {
+	buildSkillRecord,
+	discoverAllSkills,
+	discoverSkillDirs,
+	getModelInvocableSkills,
+	getPermittedSkills,
+	loadSkills,
+	parseSkillFrontmatter,
+	readSkillFile,
+	SKILL_FILENAME,
+	USER_SKILL_DIR,
+	validateSkillFrontmatter,
+} from "../skills/index.js";
+// ─── Audit & Observability ─────────────────────────────────────────────────
+export { buildAuditEnvelope, emitUokAuditEvent } from "./audit.js";
+export {
+	isAuditEnvelopeEnabled,
+	setAuditEnvelopeEnabled,
+} from "./audit-toggle.js";
 
 // ─── Gates ─────────────────────────────────────────────────────────────────
 export { ChaosMonkey, ChaosMonkeyGate } from "./chaos-monkey.js";
-export { CostGuardGate } from "./cost-guard-gate.js";
-export { MultiPackageGate } from "./multi-package-gate.js";
-export { OutcomeLearningGate } from "./outcome-learning-gate.js";
-export { SecurityGate } from "./security-gate.js";
-
-// ─── Flags & Configuration ─────────────────────────────────────────────────
-export { resolveUokFlags, loadUokFlags } from "./flags.js";
-
-// ─── Execution Graph ───────────────────────────────────────────────────────
-export {
-	selectConflictFreeBatch,
-	selectReactiveDispatchBatch,
-	buildSidecarQueueNodes,
-	buildExecutionGraphSnapshot,
-	scheduleSidecarQueue,
-	ExecutionGraphScheduler,
-} from "./execution-graph.js";
-
-// ─── Scheduler v2 (Background Work) ────────────────────────────────────────
-export {
-	CancellationToken,
-	ProgressStream,
-	WorkerPool,
-	ExecutionGraphSchedulerV2,
-} from "./scheduler-v2.js";
-
-// ─── Task State Machine ────────────────────────────────────────────────────
-export {
-	TASK_STATES,
-	TASK_TERMINAL_STATES,
-	TASK_STATE_TRANSITIONS,
-	gateOutcomesToTaskState,
-	unitRuntimeToTaskState,
-	canTransitionTaskState,
-	aggregateTaskStates,
-	buildTaskRecord,
-} from "./task-state.js";
-
-// ─── Execution Graph Persistence ───────────────────────────────────────────
-export {
-	ensureExecutionGraphSchema,
-	persistGraphSnapshot,
-	persistGraphNode,
-	persistProgressEvent,
-	queryTasksByState,
-	getGraphStateSummary,
-	getGraphProgressStream,
-	persistFullGraph,
-} from "./execution-graph-persist.js";
-
+// ─── Contracts & Types ────────────────────────────────────────────────────
+export { validateGate } from "./contracts.js";
 // ─── Coordination Store ───────────────────────────────────────────────────
 export {
 	ensureCoordinationSchema,
 	UokCoordinationStore,
 } from "./coordination-store.js";
-
-// ─── Unit Runtime ──────────────────────────────────────────────────────────
-export {
-	UNIT_RUNTIME_STATUSES,
-	UNIT_RUNTIME_TERMINAL_STATUSES,
-	UNIT_RUNTIME_TRANSITIONS,
-	isTerminalUnitRuntimeStatus,
-	getUnitRuntimeState,
-	isSyntheticUnitRuntime,
-	decideUnitRuntimeDispatch,
-	writeUnitRuntimeRecord,
-	readUnitRuntimeRecord,
-	clearUnitRuntimeRecord,
-	listUnitRuntimeRecords,
-	recordUnitOutcomeInMemory,
-	inspectExecuteTaskDurability,
-	formatExecuteTaskRecoveryStatus,
-	reconcileStaleCompleteSliceRecords,
-	reconcileDurableCompleteUnitRuntimeRecords,
-} from "./unit-runtime.js";
-
-// ─── Plan v2 ───────────────────────────────────────────────────────────────
-export {
-	EXECUTION_ENTRY_PHASES,
-	isExecutionEntryPhase,
-	compileUnitGraphFromState,
-	hasFinalizedMilestoneContext,
-	isMissingFinalizedContextResult,
-	isEmptyPlanV2GraphResult,
-	ensurePlanV2Graph,
-} from "./plan-v2.js";
-
-// ─── Audit & Observability ─────────────────────────────────────────────────
-export { buildAuditEnvelope, emitUokAuditEvent } from "./audit.js";
-export { setAuditEnvelopeEnabled, isAuditEnvelopeEnabled } from "./audit-toggle.js";
-
+export { CostGuardGate } from "./cost-guard-gate.js";
 // ─── Diagnostics ───────────────────────────────────────────────────────────
 export {
+	readUokDiagnostics,
 	synthesizeUokDiagnostics,
 	writeUokDiagnostics,
-	readUokDiagnostics,
 } from "./diagnostic-synthesis.js";
+// ─── Dispatch Envelope ─────────────────────────────────────────────────────
+export { buildDispatchEnvelope, explainDispatch } from "./dispatch-envelope.js";
 
-// ─── Parity & Ledger ───────────────────────────────────────────────────────
+// ─── Execution Graph ───────────────────────────────────────────────────────
 export {
-	writeParityReport,
-	readParityReport,
-	summarizeParityHealth,
-	writeParityHeartbeat,
-	parseParityEvents,
-	UNMATCHED_RUN_STALE_MS,
-	signalKernelEnter,
-	resetParityCommitBlock,
-	checkAndDrainMissingExit,
-} from "./parity-report.js";
+	buildExecutionGraphSnapshot,
+	buildSidecarQueueNodes,
+	ExecutionGraphScheduler,
+	scheduleSidecarQueue,
+	selectConflictFreeBatch,
+	selectReactiveDispatchBatch,
+} from "./execution-graph.js";
+// ─── Execution Graph Persistence ───────────────────────────────────────────
 export {
-	signalKernelEnter as signalParityEnter,
-} from "./parity-diff-capture.js";
-
+	ensureExecutionGraphSchema,
+	getGraphProgressStream,
+	getGraphStateSummary,
+	persistFullGraph,
+	persistGraphNode,
+	persistGraphSnapshot,
+	persistProgressEvent,
+	queryTasksByState,
+} from "./execution-graph-persist.js";
+// ─── Flags & Configuration ─────────────────────────────────────────────────
+export { loadUokFlags, resolveUokFlags } from "./flags.js";
+// ─── Gate System ───────────────────────────────────────────────────────────
+export { enrichGateResultWithMemory, UokGateRunner } from "./gate-runner.js";
+// ─── GitOps ────────────────────────────────────────────────────────────────
+export {
+	writeTurnCloseoutGitRecord,
+	writeTurnGitTransaction,
+} from "./gitops.js";
+// ─── Core Kernel ───────────────────────────────────────────────────────────
+export { recordUokKernelTermination, runAutoLoopWithUok } from "./kernel.js";
+// ─── Loop Adapter ──────────────────────────────────────────────────────────
+export { createTurnObserver } from "./loop-adapter.js";
 // ─── Message Bus ───────────────────────────────────────────────────────────
 export { AgentInbox, MessageBus } from "./message-bus.js";
-
 // ─── Metrics ───────────────────────────────────────────────────────────────
 export {
 	buildMetricsText,
 	invalidateMetricsCache,
 	metricsPath,
-	writeUokMetrics,
 	readUokMetrics,
+	writeUokMetrics,
 } from "./metrics-exposition.js";
-
-// ─── GitOps ────────────────────────────────────────────────────────────────
+// ─── Model Policy ──────────────────────────────────────────────────────────
+export { applyModelPolicyFilter } from "./model-policy.js";
+export { MultiPackageGate } from "./multi-package-gate.js";
+export { OutcomeLearningGate } from "./outcome-learning-gate.js";
+export { signalKernelEnter as signalParityEnter } from "./parity-diff-capture.js";
+// ─── Parity & Ledger ───────────────────────────────────────────────────────
 export {
-	writeTurnGitTransaction,
-	writeTurnCloseoutGitRecord,
-} from "./gitops.js";
-
+	checkAndDrainMissingExit,
+	parseParityEvents,
+	readParityReport,
+	resetParityCommitBlock,
+	signalKernelEnter,
+	summarizeParityHealth,
+	UNMATCHED_RUN_STALE_MS,
+	writeParityHeartbeat,
+	writeParityReport,
+} from "./parity-report.js";
+// ─── Plan v2 ───────────────────────────────────────────────────────────────
+export {
+	compileUnitGraphFromState,
+	EXECUTION_ENTRY_PHASES,
+	ensurePlanV2Graph,
+	hasFinalizedMilestoneContext,
+	isEmptyPlanV2GraphResult,
+	isExecutionEntryPhase,
+	isMissingFinalizedContextResult,
+} from "./plan-v2.js";
+// ─── Scheduler v2 (Background Work) ────────────────────────────────────────
+export {
+	CancellationToken,
+	ExecutionGraphSchedulerV2,
+	ProgressStream,
+	WorkerPool,
+} from "./scheduler-v2.js";
+export { SecurityGate } from "./security-gate.js";
+// ─── Task State Machine ────────────────────────────────────────────────────
+export {
+	aggregateTaskStates,
+	buildTaskRecord,
+	canTransitionTaskState,
+	gateOutcomesToTaskState,
+	TASK_STATE_TRANSITIONS,
+	TASK_STATES,
+	TASK_TERMINAL_STATES,
+	unitRuntimeToTaskState,
+} from "./task-state.js";
+// ─── Unit Runtime ──────────────────────────────────────────────────────────
+export {
+	clearUnitRuntimeRecord,
+	decideUnitRuntimeDispatch,
+	formatExecuteTaskRecoveryStatus,
+	getUnitRuntimeState,
+	inspectExecuteTaskDurability,
+	isSyntheticUnitRuntime,
+	isTerminalUnitRuntimeStatus,
+	listUnitRuntimeRecords,
+	readUnitRuntimeRecord,
+	reconcileDurableCompleteUnitRuntimeRecords,
+	reconcileStaleCompleteSliceRecords,
+	recordUnitOutcomeInMemory,
+	UNIT_RUNTIME_STATUSES,
+	UNIT_RUNTIME_TERMINAL_STATUSES,
+	UNIT_RUNTIME_TRANSITIONS,
+	writeUnitRuntimeRecord,
+} from "./unit-runtime.js";
 // ─── Writer Token ──────────────────────────────────────────────────────────
 export {
 	acquireWriterToken,
-	releaseWriterToken,
 	nextWriteRecord,
+	releaseWriterToken,
 } from "./writer.js";
-
-// ─── Model Policy ──────────────────────────────────────────────────────────
-export { applyModelPolicyFilter } from "./model-policy.js";
-
-// ─── Loop Adapter ──────────────────────────────────────────────────────────
-export { createTurnObserver } from "./loop-adapter.js";
-
-// ─── Dispatch Envelope ─────────────────────────────────────────────────────
-export { buildDispatchEnvelope, explainDispatch } from "./dispatch-envelope.js";
diff --git a/src/resources/extensions/sf/uok/kernel.js b/src/resources/extensions/sf/uok/kernel.js
index 39f2d9544..db83c420d 100644
--- a/src/resources/extensions/sf/uok/kernel.js
+++ b/src/resources/extensions/sf/uok/kernel.js
@@ -110,10 +110,15 @@ export async function runAutoLoopWithUok(args) {
 			prefs?.uok?.permission_profile ??
 			defaultPermissionProfileForRunControl(runControl),
 	);
+	// Include workMode and modelMode from session in lifecycle flags
+	const workMode = s.workMode ?? "chat";
+	const modelMode = s.modelMode ?? "smart";
 	const lifecycleFlags = {
 		...flags,
 		runControl,
 		permissionProfile,
+		workMode,
+		modelMode,
 	};
 
 	const healthVerdict = writeUokDiagnostics(s.basePath);
@@ -166,6 +171,8 @@ export async function runAutoLoopWithUok(args) {
 					flags: lifecycleFlags,
 					runControl,
 					permissionProfile,
+					workMode,
+					modelMode,
 					sessionId: ctx.sessionManager?.getSessionId?.(),
 				},
 			}),
diff --git a/src/resources/extensions/sf/uok/scheduler-v2.js b/src/resources/extensions/sf/uok/scheduler-v2.js
index f26172644..67149626d 100644
--- a/src/resources/extensions/sf/uok/scheduler-v2.js
+++ b/src/resources/extensions/sf/uok/scheduler-v2.js
@@ -35,7 +35,11 @@ export class CancellationToken {
 		this.#cancelled = true;
 		this.#reason = reason;
 		for (const fn of this.#listeners) {
-			try { fn(reason); } catch { /* ignore */ }
+			try {
+				fn(reason);
+			} catch {
+				/* ignore */
+			}
 		}
 		this.#listeners.clear();
 	}
@@ -50,7 +54,11 @@ export class CancellationToken {
 
 	onCancel(fn) {
 		if (this.#cancelled) {
-			try { fn(this.#reason); } catch { /* ignore */ }
+			try {
+				fn(this.#reason);
+			} catch {
+				/* ignore */
+			}
 			return () => {};
 		}
 		this.#listeners.add(fn);
@@ -84,7 +92,11 @@ export class ProgressStream {
 			this.#history = this.#history.slice(-this.#maxHistory);
 		}
 		for (const fn of this.#listeners) {
-			try { fn(enriched); } catch { /* ignore */ }
+			try {
+				fn(enriched);
+			} catch {
+				/* ignore */
+			}
 		}
 	}
 
@@ -92,7 +104,11 @@ export class ProgressStream {
 		this.#listeners.add(fn);
 		// Replay recent history so new listener catches up
 		for (const ev of this.#history) {
-			try { fn(ev); } catch { /* ignore */ }
+			try {
+				fn(ev);
+			} catch {
+				/* ignore */
+			}
 		}
 		return () => this.#listeners.delete(fn);
 	}
@@ -312,7 +328,6 @@ export class ExecutionGraphSchedulerV2 {
 	}
 
 	async #runParallel(sorted, token, conflicts, maxWorkers) {
-		const nodeMap = new Map(sorted.map((n) => [n.id, n]));
 		const done = new Set();
 		const order = [];
 		const promises = new Map(); // nodeId -> promise
@@ -352,11 +367,7 @@ export class ExecutionGraphSchedulerV2 {
 					.acquire(node.id, nodeToken)
 					.then(async ({ workerId, release }) => {
 						try {
-							const result = await this.#executeNode(
-								node,
-								nodeToken,
-								workerId,
-							);
+							const result = await this.#executeNode(node, nodeToken, workerId);
 							done.add(node.id);
 							order.push(result.nodeId);
 							promises.delete(node.id);
@@ -445,19 +456,23 @@ export class ExecutionGraphSchedulerV2 {
 		} catch (err) {
 			error = err;
 			if (err.name === "CancellationError") {
-				gateResults = [{
-					gateId: "scheduler",
-					outcome: "fail",
-					failureClass: "cancelled",
-					rationale: err.message,
-				}];
+				gateResults = [
+					{
+						gateId: "scheduler",
+						outcome: "fail",
+						failureClass: "cancelled",
+						rationale: err.message,
+					},
+				];
 			} else {
-				gateResults = [{
-					gateId: "scheduler",
-					outcome: "fail",
-					failureClass: "execution",
-					rationale: err.message,
-				}];
+				gateResults = [
+					{
+						gateId: "scheduler",
+						outcome: "fail",
+						failureClass: "execution",
+						rationale: err.message,
+					},
+				];
 			}
 		} finally {
 			clearTimeout(nodeTimeoutHandle);
diff --git a/src/resources/extensions/sf/uok/task-state.js b/src/resources/extensions/sf/uok/task-state.js
index b40ec394f..e0599aa8d 100644
--- a/src/resources/extensions/sf/uok/task-state.js
+++ b/src/resources/extensions/sf/uok/task-state.js
@@ -10,8 +10,7 @@
  * background-work tracking.
  */
 
-import { isDbAvailable } from "../sf-db.js";
-import { isTerminalUnitRuntimeStatus, UNIT_RUNTIME_STATUSES } from "./unit-runtime.js";
+import { isTerminalUnitRuntimeStatus } from "./unit-runtime.js";
 
 export const TASK_STATES = [
 	"todo",
@@ -119,16 +118,14 @@ export function aggregateTaskStates(taskStates) {
 		if (s in counts) counts[s]++;
 	}
 	const total = taskStates.length;
-	const terminal = TASK_STATES.filter((s) => TASK_TERMINAL_STATES.has(s)).reduce(
-		(sum, s) => sum + counts[s],
-		0,
-	);
+	const terminal = TASK_STATES.filter((s) =>
+		TASK_TERMINAL_STATES.has(s),
+	).reduce((sum, s) => sum + counts[s], 0);
 	return {
 		counts,
 		total,
 		terminal,
-		progress:
-			total > 0 ? Math.round(((terminal / total) * 100)) : 0,
+		progress: total > 0 ? Math.round((terminal / total) * 100) : 0,
 		isComplete: terminal === total && total > 0,
 	};
 }
@@ -158,9 +155,10 @@ export function buildTaskRecord({
 	modelId = null,
 	workerId = null,
 }) {
-	const state = gateResults.length > 0
-		? gateOutcomesToTaskState(gateResults)
-		: unitRuntimeToTaskState(runtimeRecord);
+	const state =
+		gateResults.length > 0
+			? gateOutcomesToTaskState(gateResults)
+			: unitRuntimeToTaskState(runtimeRecord);
 
 	return {
 		id: nodeId,