AI WorkflowĐộ khó: intermediate

Autopilot

Full autonomous execution from idea to working code. Autopilot runs a structured plan-execute-review-test-fix loop until your task is complete.

Dùng được với:Claude

Khi nào dùng

Use when you have a clear feature or task and want the AI to handle the full implementation cycle autonomously — planning, coding, testing, and cleanup.

Ví dụ sử dụng

In Claude Code: invoke the autopilot skill, describe your task, and let it run phases 0–5 until done.

Select Model

<skill name="Autopilot">
<purpose>
# Autopilot Skill

## Overview
Autopilot is a full autonomous execution engine for Claude Code (OMC framework). It converts a natural-language task description into working, tested code through a structured multi-phase pipeline.

## Phases
1. **Phase 0 — Expansion**: Analyst + Architect expand your idea into a structured spec
2. **Phase 1 — Planning**: Planner creates a concrete implementation plan
3. **Phase 2 — Execution**: Ralph loops + Ultrawork parallelise implementation
4. **Phase 3 — QA Cycling**: Tests written and iterated until they pass
5. **Phase 4 — Validation**: Multi-perspective review (correctness, security, UX)
6. **Phase 5 — Cleanup**: Formatting, dead code removal, final commit

## How to Invoke
```
Skill: autopilot
Task: &lt;your task description&gt;
```

## Best Practices
- Provide file paths or acceptance criteria for faster planning
- Pair with **deep-interview** first for vague tasks — autopilot works best with clear specs
- Use **ralplan** before autopilot for architecturally complex tasks
- Monitor the QA phase — if tests keep failing, use **debug** to investigate

## When NOT to Use
- Quick single-file fixes — use ralph instead
- Pure research tasks — use external-context or autoresearch
- When you need step-by-step control — use ralph for a supervised loop
</purpose>
</skill>

Tải .md

Khi nào dùng

Ví dụ sử dụng

Skill liên quan