| name | omg-autopilot |
| description | Full autonomous execution from idea to working code. Activate when user says: autopilot, auto-pilot, autonomous, build me, create me, make me, full auto, handle it all, or "I want a/an..."
|
| argument-hint | <product idea or task description> |
OMG Autopilot
OMG Autopilot takes a brief product idea and autonomously handles the full lifecycle: requirements analysis, technical design, planning, parallel implementation, QA cycling, and multi-perspective validation.
When to Use
- User wants end-to-end autonomous execution from an idea to working code
- Task requires multiple phases: planning, coding, testing, and validation
- User wants hands-off execution
When NOT to Use
- User wants to explore options → use
/plan
- Single focused code change → use
/ralph or delegate to @executor
- Quick fix or small bug → delegate directly to @executor
Interactive Hook Protocol
Use vscode_askQuestions at critical decision gates during autopilot execution (when available).
If vscode_askQuestions is NOT available (e.g., Copilot CLI), present numbered options in markdown.
Autopilot is designed to be mostly autonomous, but hooks fire at ambiguity points and failure recovery.
When to Fire Hooks
| Trigger Point | Question Type |
|---|
| Phase 0: Vague input | Redirect to deep-interview or proceed with assumptions |
| Phase 0: Spec review | Confirm extracted spec before planning |
| Phase 3: Repeated QA failure | Ask user for direction on persistent errors |
| Phase 4: Validation rejection | Surface reviewer concerns for user decision |
Execution Pipeline
Phase 0 - Expansion
Turn the user's idea into a detailed spec.
- If ralplan consensus plan exists (
.omg/plans/ralplan-*.md): Skip Phase 0 AND Phase 1 → jump to Phase 2
- If deep-interview spec exists (
.omg/specs/deep-interview-*.md): Use pre-validated spec, skip to Phase 1
- If input is vague: HOOK via
vscode_askQuestions:
header: "autopilot-vague-input"
question: "Your request is broad. How should I proceed?"
options: [
{ label: "Run /deep-interview first", description: "Thorough requirements gathering", recommended: true },
{ label: "Proceed with my best interpretation", description: "I'll extract requirements and show you for approval" },
{ label: "Let me clarify — here's more detail..." }
]
allowFreeformInput: true
- Otherwise: @analyst extracts requirements, @architect creates technical specification
- HOOK: Spec confirmation before proceeding to planning:
header: "autopilot-spec-review"
question: "Here's what I understood: [1-line summary]. Proceed?"
options: [
{ label: "Looks good — proceed", recommended: true },
{ label: "Mostly right, but adjust..." },
{ label: "Wrong direction — let me re-explain" }
]
allowFreeformInput: true
- Output:
.omg/autopilot/spec.md
- Track phase:
omg_write_state(phase="expansion_done")
Phase 1 - Planning
Create an implementation plan from the spec.
- @architect creates plan (direct mode)
- @critic validates plan
- Output:
.omg/plans/autopilot-impl.md
- Track:
omg_write_state(phase="planning_done")
Phase 2 - Execution
Implement the plan using parallel execution.
- Route tasks by complexity to @executor
- Run independent tasks in parallel
- Track:
omg_write_state(phase="execution_done")
Phase 3 - QA
Cycle until all tests pass (max 5 cycles).
- Build, lint, test, fix failures
- Stop early if the same error repeats 3 times (fundamental issue)
- HOOK: On 2nd repeat of same failure via
vscode_askQuestions:
header: "autopilot-qa-stuck"
question: "Same error persists after [N] cycles: [error summary]. How to proceed?"
options: [
{ label: "Try a different approach", recommended: true },
{ label: "Skip this check — it's acceptable" },
{ label: "I know the fix — let me explain..." },
{ label: "Abort autopilot" }
]
allowFreeformInput: true
- Track:
omg_write_state(phase="qa_done")
Phase 4 - Validation
Multi-perspective review in parallel.
- @architect: Functional completeness
- @security-reviewer: Vulnerability check
- @code-reviewer: Quality review
- All must approve; on rejection:
HOOK via
vscode_askQuestions:
header: "autopilot-validation-rejection"
question: "[Reviewer] flagged: [issue summary]. Action?"
options: [
{ label: "Fix the issues", recommended: true },
{ label: "Accept risk — proceed anyway" },
{ label: "I need to review this myself first" }
]
allowFreeformInput: true
- Fix and re-validate on rejection
- Track:
omg_write_state(phase="validation_done")
Phase 5 - Cleanup
Delete all state files on successful completion.
- Run
/cancel for clean exit
Stop Conditions
- Same QA error persists across 3 cycles → report fundamental issue
- Validation keeps failing after 3 rounds → report issues
- User says "stop", "cancel", or "abort"
Checklist
