Autopilot - Unified Autonomous Workflow

Execute a complete workflow from initial request to validated code. This skill replaces and extends the previous ultrawork skill.

Invocation

/autopilot "Add user authentication with JWT"
/autopilot --fast "Fix all TypeScript errors"      # Fast mode (ulw alias)
/autopilot --swarm 5 "Implement all API endpoints" # Parallel execution
/autopilot --ui "Build dashboard page"             # With UI verification

Aliases

• ulw - Alias for /autopilot --fast
• ultrawork - Alias for /autopilot --fast

Modes

Mode	Phases	Use Case
Full (default)	0-4	Complex features, new systems
Fast (--fast)	1-2 only	Bug fixes, simple changes
Swarm (--swarm N)	Parallel execution	Multiple independent tasks
UI (--ui)	+ UI verification	Frontend development

5 Phases

Phase	Name	Agent	Gate Criteria
0	Expansion	Metis (GPT)	spec.md created
1	Planning	Prometheus → Momus	plan approved
2	Execution	Atlas → Junior	all tasks done
3	QA	Junior	build + lint + tests (+ ui) pass
4	Validation	Oracle (Codex)	security + code review

Options

Flag	Description
--fast	Skip Metis/Momus (equivalent to ulw)
--swarm N	Use N parallel agents in execution
--ui	Enable Playwright + Gemini UI verification in QA
--no-qa	Skip QA phase
--no-validation	Skip Oracle validation

Workmode

When autopilot starts, workmode is automatically enabled:

• Sisyphus cannot directly modify code
• All code changes must go through Atlas → Junior
• This ensures consistent orchestration

To stop: /autopilot off or mcp__chronos__workmode_disable()

Workflow

Initialization

1. Start autopilot with workmode:
   mcp__chronos__autopilot_start(name, request, options)
   → Automatically enables workmode

2. Start Ralph Loop for continuity:
   mcp__chronos__ralph_start(completion_promise, max_iterations)

Phase 0: Expansion (Metis)

Skip if: --fast

1. Delegate to Metis agent (GPT-5.2 xhigh reasoning)
2. Metis creates: .sisyphus/specs/{name}.md
3. Set output: mcp__chronos__autopilot_set_output(0, spec_path)
4. Advance: mcp__chronos__autopilot_advance()

Phase 1: Planning (Prometheus + Momus)

1. Delegate to Prometheus agent with spec
2. Prometheus creates plan: .sisyphus/plans/{name}.md
3. Momus reviews (Codex-5.2 xhigh) unless --fast
4. Set output and advance

Phase 2: Execution (Atlas or Swarm)

If --swarm N:
  1. Parse plan into tasks
  2. Initialize swarm: mcp__swarm__swarm_init(tasks)
  3. Launch N junior agents in parallel
  4. Monitor: mcp__swarm__swarm_stats()

Else:
  1. Delegate to Atlas
  2. Atlas creates todos and delegates to Junior
  3. Monitor via TaskList

Update progress: mcp__chronos__autopilot_update_progress(2, {done, total})

Phase 3: QA

Standard checks:
1. Build: npm run build / tsc
2. Lint: npm run lint
3. Tests: npm test

If --ui enabled:
4. Playwright screenshot capture
5. Gemini analyzes screenshot
6. Compare against expectations

Update: mcp__chronos__autopilot_update_progress(3, {build, lint, tests, ui})

All must pass to advance.

Phase 4: Validation (Oracle)

Skip if: --no-validation

1. Delegate to Oracle (Codex primary)
2. Oracle checks:
   - Security vulnerabilities
   - Code quality
   - Architecture compliance

Update: mcp__chronos__autopilot_update_progress(4, {
  blocking_issues: 0,
  warnings: [],
  approved: true
})

Completion

1. All phases pass
2. Disable workmode: mcp__chronos__workmode_disable()
3. Clear autopilot: mcp__chronos__autopilot_clear()
4. Stop Ralph Loop: mcp__chronos__ralph_stop(reason="completed")
5. Report summary to user

State Tracking

State file: .sisyphus/autopilot.json

{
  "id": "autopilot-20260131-auth",
  "current_phase": 2,
  "status": "running",
  "options": {
    "fast": false,
    "ui": true,
    "use_swarm": true,
    "swarm_agents": 3,
    "skip_qa": false,
    "skip_validation": false
  },
  "phases": {
    "0": { "status": "completed", "output": ".sisyphus/specs/auth.md" },
    "1": { "status": "completed", "output": ".sisyphus/plans/auth.md" },
    "2": { "status": "in_progress", "progress": { "done": 3, "total": 7 } },
    "3": { "status": "pending" },
    "4": { "status": "pending" }
  }
}

Model Routing

External models are used to reduce Claude API costs:

Agent	Primary Model	Fallback
Metis	GPT-5.2 (xhigh)	Claude Sonnet
Momus	Codex-5.2 (xhigh)	Claude Sonnet
Oracle	Codex	Claude Sonnet
Multimodal-looker	Gemini	Claude Sonnet
Librarian	GLM-4.7	Claude Haiku

Junior agent tiers based on task complexity:

Complexity	Criteria	Agent
Low	1 file, <20 lines	junior-low (Sonnet)
Medium	2-5 files, 20-100 lines	junior (Sonnet)
High	6+ files, 100+ lines	junior-high (Opus)

Examples

Full Run (Default)

/autopilot "Add user profile page with avatar upload"

Phase 0: Metis → spec
Phase 1: Prometheus → plan, Momus → review
Phase 2: Atlas → Junior (parallel tasks)
Phase 3: Build/Lint/Tests pass
Phase 4: Oracle approves

Fast Mode (ulw alias)

ulw "Fix login button styling"

Phase 1: Prometheus → quick plan
Phase 2: Junior executes
→ Complete (QA/Validation skipped by default in fast mode)

Parallel Swarm

/autopilot --swarm 5 "Implement all REST endpoints"

Phase 2: 5 junior agents work in parallel
→ Each claims tasks atomically from swarm pool
→ Monitor: mcp__swarm__swarm_stats()

UI Verification

/autopilot --ui "Build dashboard page"

Phase 3 (QA):
1. Playwright captures screenshot
2. Gemini analyzes UI
3. Compares against expectations
4. Reports issues

Error Handling

If phase fails:

1. mcp__chronos__autopilot_fail(error="Description")
2. Status becomes "failed"
3. Workmode remains active
4. Report to user with details
5. User can fix and retry with /autopilot --resume

Related Commands

• /autopilot on "task" - Start autopilot (same as /autopilot "task")
• /autopilot off - Stop and clear autopilot, disable workmode
• /autopilot status - Show current autopilot state
• /ecomode on - Enable resource-efficient mode