pi-crew 0.9.50 → 0.9.52

package/docs/pi-crew-investigation-report.md

@@ -1,411 +0,0 @@
-# pi-crew v0.2.20 — Investigation Results and Analysis
-
-**Date:** 2026-05-19  
-**Environment:** linux/x64, Node v22.22.0, Pi CLI v0.75.3  
-**Models:** zai/glm-5.1 (planner, executor, test-engineer), minimax/MiniMax-M2.7-highspeed (explorer, analyst, reviewer, verifier, writer, critic)  
-**pi-crew version:** 0.2.20
-
----
-
-## 1. pi-crew architecture overview
-
-### 1.1 Source code structure
-
-```
-pi-crew/src/
-├── adapters/          — Adapters for external systems
-├── agents/            — Agent discovery & config (10 agents)
-├── config/            — Configuration, defaults, drift detection
-├── extension/         — Pi extension registration
-├── hooks/             — Lifecycle hooks (before_run_start, before_task_start, task_result, etc.)
-├── observability/     — Metrics, correlation, exporters (OTLP, Prometheus)
-├── prompt/            — Prompt runtime & pipeline
-├── runtime/           — Core runtime (~30+ files)
-│   ├── async-runner.ts      — Background process spawning with jiti loader
-│   ├── background-runner.ts — Background entry point, team execution
-│   ├── child-pi.ts          — Child Pi process lifecycle, stdout capture, timeout
-│   ├── child-pi-pool.ts     — Warm pool skeleton (disabled, size=0)
-│   ├── live-session-runtime.ts — Live-session (reuses the parent Pi)
-│   ├── team-runner.ts       — Main team run orchestrator
-│   ├── worker-heartbeat.ts  — Heartbeat state tracking
-│   ├── worker-startup.ts    — Startup failure classification
-│   ├── pi-spawn.ts          — Pi binary resolution & spawn command
-│   ├── pi-args.ts           — Build args for child Pi workers
-│   ├── runtime-resolver.ts  — Resolve live-session vs child-process
-│   ├── crash-recovery.ts    — Crash recovery logic
-│   ├── deadletter.ts        — Dead letter queue
-│   └── ...
-├── schema/            — Config & team-tool schema validation
-├── skills/            — Built-in skills
-├── state/             — State store, manifests, event logs
-├── subagents/         — Subagent index, spawn, manager
-├── teams/             — Team discovery (6 teams)
-├── types/             — Shared TypeScript types
-├── ui/                — TUI: widgets, overlays, dashboard, powerbar
-├── utils/             — Utilities (sleep, shell resolve, redaction, env-filter)
-├── workflows/         — Workflow discovery (6 workflows)
-└── worktree/          — Git worktree isolation
-```
-
-### 1.2 Resource inventory
-
-| Resource | Count | Details |
-|---|---|---|
-| **Teams** | 6 | default, fast-fix, implementation, parallel-research, research, review |
-| **Workflows** | 6 | default, fast-fix, implementation, parallel-research, research, review |
-| **Agents** | 10 | explorer, planner, analyst, critic, executor, reviewer, security-reviewer, test-engineer, verifier, writer |
-| **Skills** | 27 | async-worker-recovery, child-pi-spawning, orchestration, systematic-debugging, verification-before-done, ... |
-| **Hooks** | 5+ | before_run_start, before_task_start, before_retry, task_result, ... |
-
-### 1.3 Runtime modes
-
-pi-crew supports 2 runtime modes:
-
-| Mode | Description | Pros | Cons |
-|---|---|---|---|
-| **live-session** | Reuses the current Pi session | Fast, shares the provider connection | Cannot run async/background |
-| **child-process** | Spawns a new Pi process | Can run background/async | Needs its own provider connection |
-
-**Runtime resolution flow:**
-```
-team action='run' + async=true
-  → runtime-resolver.ts: resolveCrewRuntime()
-  → live-session available? NO (background cannot use live-session)
-  → Fallback: child-process
-  → spawn new Pi via jiti loader
-```
-
----
-
-## 2. Comprehensive test results
-
-### 2.1 Summary table
-
-| Category | Tests | Pass | Fail | Partial |
-|---|---|---|---|---|
-| Resource Discovery (list, get, recommend) | 5 | ✅ 5 | 0 | 0 |
-| Subagent Lifecycle (Agent, crew_agent) | 4 | 0 | ❌ 4 | 0 |
-| Team Run Lifecycle (run, cancel, retry) | 4 | ✅ 1 | ❌ 2 | ⚠️ 1 |
-| Planning (plan) | 1 | ✅ 1 | 0 | 0 |
-| State Management (status, events, artifacts, summary, prune) | 6 | ✅ 6 | 0 | 0 |
-| Diagnostics (doctor, validate, help) | 3 | ✅ 3 | 0 | 0 |
-| Portability (export, import) | 2 | ✅ 2 | 0 | 0 |
-| Configuration (settings, autonomy) | 2 | ✅ 2 | 0 | 0 |
-| **Total** | **27** | **20** | **6** | **1** |
-
-### 2.2 Per-test details
-
-#### ✅ Resource Discovery — 5/5 PASS
-
-| Test | Input | Output | Result |
-|---|---|---|---|
-| `team list` | List all resources | 6 teams, 6 workflows, 10 agents | ✅ |
-| `team get` team | Get team=default | 4 roles (explorer→planner→executor→verifier) | ✅ |
-| `team get` workflow | Get workflow for implementation | Implementation workflow steps | ✅ |
-| `team get` agent | Get agent=explorer | Full profile: model, description, instructions | ✅ |
-| `team recommend` | goal="test all features" | Recommended implementation team, high confidence | ✅ |
-
-#### ✅ Diagnostics — 3/3 PASS
-
-| Test | Input | Output | Result |
-|---|---|---|---|
-| `team doctor` | Full diagnostics | 17/17 checks OK (runtime, filesystem, discovery, validation, drift, schema, async, worktrees) | ✅ |
-| `team validate` | Validate all resources | 10 agents, 6 teams, 6 workflows, 0 issues | ✅ |
-| `team help` | Show help | Full command reference (core, inspection, maintenance, portability, diagnostics) | ✅ |
-
-Detailed doctor checks:
-- Runtime: cwd, platform, node, pi, git, config, model — all OK
-- Filesystem: user state, project state, artifacts — all OK
-- Discovery: 10 agents, 6 teams, 6 workflows, 10 model hints — all OK
-- Drift: no config drift detected
-- Schema: strict-provider schema compatible
-- Async: fs.watch with polling fallback, completion notifications enabled
-- Worktrees: leader repository OK, dirty worktrees preserved policy
-
-#### ✅ State Management — 6/6 PASS
-
-| Test | Input | Output | Result |
-|---|---|---|---|
-| `team status` | Check run state | Detailed: task graph, events, artifacts, policy decisions | ✅ |
-| `team events` | Get event log | 20+ events from run.created → task.failed with timestamps | ✅ |
-| `team artifacts` | List artifacts | 14 artifacts (prompts, results, metadata, logs, shared) | ✅ |
-| `team summary` | Run overview | Status, goal, tasks, usage summary | ✅ |
-| `team prune` | keep=2, confirm=true | 9 runs pruned, 2 kept, audit trail in prune.jsonl | ✅ |
-| `team worktrees` | Without runId | Correctly required runId parameter | ✅ |
-
-#### ✅ Portability — 2/2 PASS
-
-| Test | Input | Output | Result |
-|---|---|---|---|
-| `team export` | Export completed run | run-export.json + run-export.md created | ✅ |
-| `team import` | Import exported bundle | Bundle imported to .crew/imports/ with README.md | ✅ |
-
-#### ✅ Configuration — 2/2 PASS
-
-| Test | Input | Output | Result |
-|---|---|---|---|
-| `team settings` | Show effective settings | Complete: agent overrides, UI config, autonomous mode | ✅ |
-| `team autonomy` | Show autonomy profile | Profile=suggested, enabled=true, inject policy=true | ✅ |
-
-#### ✅ Planning — 1/1 PASS
-
-| Test | Input | Output | Result |
-|---|---|---|---|
-| `team plan` | goal="Add health-check endpoint" | 4-step plan: explore → plan → execute → verify | ✅ |
-
-#### ❌ Subagent Lifecycle — 0/4 FAIL
-
-| Test | Agent ID | Type | Duration | Output | Result |
-|---|---|---|---|---|---|
-| Agent(explorer) | agent_mpc423rq_1 | explorer | 305s | Empty | ❌ |
-| Agent(planner) | agent_mpc423rv_2 | planner | 305s | Empty | ❌ |
-| Agent(analyst) | agent_mpc423rw_3 | analyst | 305s | Empty | ❌ |
-| crew_agent(explorer) | agent_mpc423rw_4 | explorer | 305s | Empty | ❌ |
-
-All of them: spawn successfully (PID exists) → zero output → 305s heartbeat timeout → failed.
-
-#### ❌ Team Run Lifecycle — 1 PASS, 2 FAIL, 1 PARTIAL
-
-| Test | Team | Runtime | Result | Details |
-|---|---|---|---|---|
-| implementation async | implementation | child-process | ❌ FAIL | 01_assess heartbeat dead after 300s |
-| `team retry` | — | — | ✅ PASS | Task re-queued successfully |
-| fast-fix foreground | fast-fix | live-session | ⚠️ PARTIAL | 01_explore completed, run cancelled before execute |
-| `team cancel` | — | — | ✅ PASS | Run successfully cancelled |
-
----
-
-## 3. Critical issue: `pi --print` hangs
-
-### 3.1 Description
-
-**All 6 background worker failures share the same root cause:** `pi --print` (non-interactive mode) hangs indefinitely.
-
-### 3.2 Reproduce
-
-```bash
-$ timeout 10 pi --print "say hi"
-[context-mode] WARNING: skipping MCP bridge — CONTEXT_MODE_BRIDGE_DEPTH=1 indicates recursion
-# ... hangs indefinitely ...
-EXIT_CODE: 124  (timeout)
-```
-
-Result: **100% reproducible**. The Pi CLI starts (prints the context-mode warning) but blocks on the provider/model call.
-
-### 3.3 Chain of failure
-
-```
-pi-crew background run
-  → runtime-resolver.ts: fallback to child-process
-  → async-runner.ts: resolve jiti-register.mjs
-  → spawn("pi", [...args], { cwd, env })
-  → Pi CLI starts, prints "[pi-crew] background loader=jiti"
-  → Pi tries to connect to the model provider
-  → BLOCKS INDEFINITELY — no stdout, no stderr, no error
-  → 300,000ms (5 min) heartbeat timeout
-  → worker.response_timeout: "No output for 300000ms"
-  → task.failed → run.failed
-```
-
-### 3.4 Why does live-session still work?
-
-| Aspect | Live-session | Child-process |
-|---|---|---|
-| Provider connection | **Reuses** the parent Pi's connection | Creates a new connection |
-| Auth context | Shared with parent | Must set up itself |
-| Startup time | Fast (no new process) | Slow (spawn + init) |
-| Background capable | ❌ No | ✅ Yes (if the provider works) |
-
-### 3.5 Possible causes
-
-| # | Cause | Likelihood | How to verify |
-|---|---|---|---|
-| 1 | **API key not inherited** by the child process env | High | Check whether `sanitizeEnvSecrets()` filters too aggressively |
-| 2 | **Provider endpoint unreachable** from the child process | Medium | `curl` to the provider API from the child env |
-| 3 | **Provider rate limiting** (parent + child concurrent) | Medium | Check provider response headers |
-| 4 | **jiti loader stall** — TS compilation hangs | Low | jiti import succeeded (log confirmed) |
-
-### 3.6 Related key files
-
-```
-pi-crew/src/runtime/
-├── async-runner.ts       — resolveTypeScriptLoader(), spawn args with --import jiti-register.mjs
-├── child-pi.ts           — runChildPi(), response timeout, stdout capture
-│                           buildChildPiSpawnOptions() → { cwd, env: sanitizeEnvSecrets(env) }
-├── background-runner.ts  — Background entry point
-├── pi-spawn.ts           — getPiSpawnCommand() → { command: "pi", args }
-├── pi-args.ts            — buildPiWorkerArgs() → args array
-└── worker-heartbeat.ts   — Heartbeat stale check (5 min default)
-
-pi-crew/src/config/defaults.ts
-└── DEFAULT_CHILD_PI.responseTimeoutMs = 5 * 60_000  (300s)
-
-pi-crew/src/utils/env-filter.ts
-└── sanitizeEnvSecrets()  — Filter secret env vars (possibly too aggressive?)
-```
-
-### 3.7 Recommended fix
-
-1. **Immediate:** Run `pi --print "test"` in a terminal to confirm the provider connection issue
-2. **Check `sanitizeEnvSecrets()`:** Verify that API keys (GOOGLE_API_KEY, MINIMAX_API_KEY, ZAI_API_KEY, etc.) are not filtered out
-3. **Add error logging:** Capture stderr from the child Pi process into background.log
-4. **Add a connection timeout:** The Pi CLI should time out after ~30s if the provider does not respond, instead of blocking indefinitely
-5. **Test workaround:** Set `PI_TEAMS_MOCK_CHILD_PI=success` to bypass the provider call and verify pi-crew logic in isolation
-
----
-
-## 4. Secondary issue: Stale heartbeat notifications after prune
-
-### 4.1 Description
-
-After running `team prune`, the background watcher still emits "Task heartbeat dead" notifications for removed runs.
-
-### 4.2 Pattern
-
-```
-team prune --keep=0 --confirm=true   → 9 runs removed
-→ Notification: "agent_mpc423rq_1 heartbeat dead" (pruned run)
-→ Notification: "agent_mpc423rv_2 heartbeat dead" (pruned run)
-→ Notification: "agent_mpc423rw_3 heartbeat dead" (pruned run)
-→ Notification: "agent_mpc423rw_4 heartbeat dead" (pruned run)
-→ ... (6+ stale notifications total)
-```
-
-### 4.3 Cause
-
-The background watcher maintains a queue of worker health checks. When runs are pruned, the watcher does not deregister immediately — notifications already in the queue are still emitted.
-
-### 4.4 Severity: LOW (cosmetic)
-
-### 4.5 Recommendation
-
-- The background watcher should check run existence before emitting heartbeat alerts
-- Or: the watcher should deregister workers when runs are pruned
-
----
-
-## 5. Secondary issue: Live-session run cancelled mid-execution
-
-### 5.1 Description
-
-A fast-fix team ran in a live-session; task `01_explore` completed successfully but the run was cancelled before `02_execute` started.
-
-### 5.2 Events
-
-```
-04:12:20 live-session.prompt_start 01_explore
-04:12:51 live-session.prompt_done 01_explore
-04:12:51 live_agent.terminated 01_explore (status=cancelled)
-04:12:51 task.completed 01_explore
-04:12:51 run.cancelled: "This operation was aborted"
-```
-
-### 5.3 Possible causes
-
-- Session concurrency limit (only 1 active live-session)
-- User-initiated cancellation
-- Conflict with concurrent test operations
-
-### 5.4 Severity: MEDIUM
-
----
-
-## 6. Features working stably
-
-A list of features that have been tested and work correctly:
-
-### Resource Discovery
-- ✅ `team list` — Lists teams, workflows, agents, recent runs
-- ✅ `team get` — Details of team/workflow/agent
-- ✅ `team recommend` — Suggests a suitable team based on a goal
-- ✅ `team validate` — Validates all resources
-
-### Diagnostics
-- ✅ `team doctor` — 17 checks (runtime, filesystem, discovery, drift, schema, async, worktrees)
-- ✅ `team help` — Full command reference
-
-### State Management
-- ✅ `team status` — Run state with task graph, events, policy decisions
-- ✅ `team events` — Detailed chronological event log
-- ✅ `team artifacts` — Lists artifact files (prompts, results, metadata, logs)
-- ✅ `team summary` — Concise run overview
-- ✅ `team prune` — Cleanup runs with audit trail (prune.jsonl)
-- ✅ `team cancel` — Cancel running/queued runs
-
-### Portability
-- ✅ `team export` — Export a run to JSON + Markdown
-- ✅ `team import` — Import a run bundle, create a README.md summary
-
-### Configuration
-- ✅ `team settings` — Show effective settings (agent overrides, UI, autonomous)
-- ✅ `team autonomy` — Show/set the autonomous mode profile
-
-### Planning
-- ✅ `team plan` — Create an execution plan with structured steps
-
-### Retry
-- ✅ `team retry` — Re-queue failed tasks
-
----
-
-## 7. Current configuration
-
-### Autonomous Mode
-```
-Profile: suggested
-Enabled: true
-Inject policy: true
-Prefer async for long tasks: false
-Allow worktree suggestion: true
-```
-
-### Agent Model Overrides
-| Agent | Model | Thinking |
-|---|---|---|
-| explorer | minimax/MiniMax-M2.7-highspeed | off |
-| writer | minimax/MiniMax-M2.7-highspeed | off |
-| planner | zai/glm-5.1 | medium |
-| analyst | minimax/MiniMax-M2.7-highspeed | off |
-| critic | minimax/MiniMax-M2.7 | low |
-| executor | zai/glm-5.1 | medium |
-| reviewer | minimax/MiniMax-M2.7 | off |
-| security-reviewer | minimax/MiniMax-M2.7 | medium |
-| test-engineer | zai/glm-5.1 | low |
-| verifier | minimax/MiniMax-M2.7 | off |
-
-### Timeouts
-```
-DEFAULT_CHILD_PI.responseTimeoutMs = 300,000 (5 min)
-DEFAULT_LIVE_SESSION.responseTimeoutMs = 600,000 (10 min)
-```
-
----
-
-## 8. Related files
-
-| File | Description |
-|---|---|
-| `/home/bom/source/my_pi/pi-crew-test-results.md` | Detailed test report |
-| `/home/bom/.pi/agent/pi-crew.json` | pi-crew config |
-| `/home/bom/.pi/agent/agents/explorer.md` | Explorer agent config |
-| `/home/bom/.pi/agent/agents/security-reviewer.md` | Security reviewer config |
-| `/home/bom/.pi/agent/agents/test-engineer.md` | Test engineer config |
-| `/home/bom/.pi/agent/agents/verifier.md` | Verifier config |
-| `/home/bom/source/my_pi/.crew/audit/prune.jsonl` | Prune audit trail (381 entries) |
-
----
-
-## 9. Next Steps
-
-### High priority
-1. **Fix `pi --print` hangs:** Investigate the provider connection in the child process
-2. **Check `sanitizeEnvSecrets()`:** Verify it does not filter out necessary API keys
-3. **Add stderr logging:** background.log should capture stderr from the child Pi
-
-### Medium priority
-4. **Test a foreground team to completion:** Verify the full workflow lifecycle (explore→plan→execute→verify)
-5. **Stale notification fix:** Background watcher deregister on prune
-
-### Low priority
-6. **Configurable heartbeat timeout:** Replace the hardcoded 300s with a config value
-7. **Warm pool implementation:** Currently disabled (size=0), needs Pi-side support

package/docs/product/README.md

@@ -1,31 +0,0 @@
-# Product Docs
-
-Product documentation for pi-crew. Each file describes a product domain —
-what it does, how it behaves, and what contracts it maintains.
-
-## Update Rule
-
-When behavior changes:
-1. Update the affected product doc
-2. Update or create the story packet
-3. Update `docs/TEST_MATRIX.md`
-4. Record a decision if it affects architecture, scope, risk, or settled rules
-
-## Domain Index
-
-| File | Domain | Description |
-|------|--------|-------------|
-| `team-run.md` | Core | Team run lifecycle: start, execute, complete |
-| `team-tool.md` | API | Team tool actions: run, status, list, plan |
-| `child-process.md` | Runtime | Child Pi process spawning and management |
-| `live-session.md` | Runtime | In-process agent execution |
-| `async-runner.md` | Runtime | Background/async run execution |
-| `state.md` | State | Durable state: manifests, tasks, events |
-| `worktree.md` | Isolation | Git worktree isolation for parallel work |
-| `group-join.md` | Coordination | Agent result grouping and delivery |
-| `model-fallback.md` | Runtime | Model selection and fallback chain |
-| `conflict-detect.md` | Utils | Merge conflict detection in file edits |
-| `crash-recovery.md` | Reliability | Crash recovery and stale reconciliation |
-| `effectiveness.md` | Quality | Effectiveness guard for worker activity |
-| `platform.md` | Platform | Cross-platform considerations (Windows) |
-| `runtime-safety.md` | Safety | Runtime safety: depth guard, resource limits |

package/docs/product/platform.md

@@ -1,27 +0,0 @@
-# Platform (Cross-Platform)
-
-## Behavior
-
-pi-crew runs on Windows, macOS, and Linux. Primary development is on Windows.
-
-### Windows Considerations
-
-- **EBUSY/EPERM**: Files locked by antivirus, shell, or indexer
-  - `rmSyncRetry()` with exponential backoff (50ms, 100ms, 200ms, 400ms)
-  - `existsSync` check before cleanup in finally blocks
-- **Path separators**: Use `path.join()` everywhere, never hardcoded `/`
-- **Shell**: `resolve-shell.ts` handles `cmd.exe` vs `bash` detection
-- **Case sensitivity**: Windows is case-insensitive for file paths
-
-### Unix Considerations
-
-- `unref()` on timers to prevent blocking process exit
-- POSIX shell compatibility in any shell scripts
-- Signal handling (SIGTERM, SIGINT) for graceful shutdown
-
-### CI Matrix
-
-All changes validated on:
-- `ubuntu-latest` / Node 22
-- `windows-latest` / Node 22
-- `macos-latest` / Node 22

package/docs/product/runtime-safety.md

@@ -1,37 +0,0 @@
-# Runtime Safety
-
-## Behavior
-
-pi-crew enforces multiple safety layers to prevent resource leaks, crashes,
-and runaway execution.
-
-### Depth Guard
-
-- Tracks `PI_CREW_SESSION_DEPTH` environment variable
-- Depth >= 2 forces `child-process` mode instead of `live-session`
-- Prevents stack overflow from nested team runs
-
-### Resource Limits
-
-- Memory cap on live-session agents
-- Prompt timeout for agent responses
-- Tool count restoration after session error
-
-### Process Cleanup
-
-- `cleanupTempDir()` with `existsSync` guard against double cleanup
-- `safeDisposeLiveSession()` for clean resource teardown
-- `removeLiveAgentHandle()` for registry cleanup
-
-### Error Handling
-
-- `try/catch` around all I/O operations in UI code
-- `rmSyncRetry()` with exponential backoff for Windows EBUSY
-- `rejectRunPromise` (not `resolveRunPromise`) in error paths
-
-### State Integrity
-
-- `withRunLockSync` for all state mutations
-- Atomic write helpers (`writeJsonAtomic`, `appendJsonAtomic`)
-- `markActiveTasksAndAgentsFailed()` for crash recovery
-- Event log append-only for audit trail

package/docs/product/team-run.md

@@ -1,39 +0,0 @@
-# Team Run
-
-## Behavior
-
-A team run executes a workflow with multiple agents, tracking progress through
-durable state on disk.
-
-### Lifecycle
-
-1. User invokes `team action='run'` with a goal
-2. Team runner creates a manifest, resolves team/workflow
-3. Task graph is built from workflow steps
-4. Tasks execute (parallel or sequential per workflow)
-5. Results are collected, artifacts written
-6. Run completes with final status
-
-### Statuses
-
-| Status | Meaning |
-|--------|---------|
-| pending | Manifest created, not yet executing |
-| running | Tasks executing |
-| completed | All tasks finished successfully |
-| failed | One or more tasks failed |
-| cancelled | User cancelled the run |
-| partial | Some tasks completed, others still pending |
-
-### Concurrency
-
-- Tasks without dependencies run in parallel (up to concurrency limit)
-- Tasks with `dependsOn` wait for predecessors
-- Workflow phases enforce ordering
-
-### Artifacts
-
-- `results/{taskId}.txt` — task output
-- `logs/{taskId}.log` — full transcript
-- `metadata/` — task metadata files
-- `shared/` — inter-agent shared context

package/docs/product/team-tool.md

@@ -1,37 +0,0 @@
-# Team Tool API
-
-## Behavior
-
-The `team` tool is the primary interface for users to interact with pi-crew.
-
-### Actions
-
-| Action | Description |
-|--------|-------------|
-| `run` | Start a team run |
-| `plan` | Create a plan without executing |
-| `status` | Check run/task status |
-| `list` | List teams, agents, workflows |
-| `get` | Get resource details |
-| `cancel` | Cancel a running task/run |
-| `resume` | Resume a paused run |
-| `respond` | Respond to a waiting task |
-| `recommend` | Get team/workflow recommendations |
-| `create/update/delete` | Manage resources |
-| `doctor` | Diagnose configuration issues |
-
-### Parameters
-
-- `action` (required): The action to perform
-- `team`: Team name for run operations
-- `goal`: High-level objective
-- `runId`: Run ID for status/cancel/resume
-- `taskId`: Task ID for respond operations
-- `confirm: true`: Required for destructive actions
-
-### Safety Rules
-
-- Delete operations require `confirm: true`
-- Referenced resources blocked unless `force: true`
-- Cancel requires explicit run ID
-- Respond requires task ID + message