How nick-vels/skills works
Autopilot is an AI-agent skill that turns an explicitly supplied project idea or brief into a built project. Its workflow moves from setup and clarification through specification and task planning, then implementation, review/repair, and final acceptance. The skill documents modes, depth controls, resume/memory, subagents, and a live dashboard. ap.py supports run state, tickets, requirement tracking, and dashboard serving; measure-run.py analyzes Claude Code session logs. The phase and prompt files are present in the tree but were not sampled, so their internal call relationships are not asserted.
Main entry points: SKILL.md Autopilot command · 5-subagents.md Task execution · ap.py Run state and tickets · measure-run.py Run measurement
Components 18
Project workflow
- Preflight and modes
skills/autopilot/phases - Briefing and questions
skills/autopilot/phases/2-briefing.md - Specification
skills/autopilot/phases/3-spec.md - Task planning
skills/autopilot/phases/4-plan.md - Task execution
skills/autopilot/phases/5-subagents.md - Review and repair
skills/autopilot/phases/6-review.md - Final acceptance
skills/autopilot/phases/8-final.md - Generated project
Skill guidance
- Autopilot command
skills/autopilot/SKILL.md - Project memory
skills/autopilot/phases/9-memory.md - Agent prompts
skills/autopilot/prompts - Instrument guidance
skills/autopilot/phases/0-instruments.md
Run state and visibility
- Run state and tickets
skills/autopilot/tools/ap.py - Live dashboard
skills/autopilot/phases/dashboard-template.html - Run artifacts
Run analysis
- Claude session logs
- Run measurement
tools/measure-run.py
Other components
- Project requester
Connections 17
- Project requester → Autopilot command: starts explicitly
README.md - Autopilot command → Preflight and modes: begins with
README.md - Preflight and modes → Briefing and questions: configures
- Briefing and questions → Specification: informs
- Specification → Task planning: breaks into
- Task planning → Task execution: assigns tasks
- Task execution → Generated project: builds
- Review and repair → Task execution: requests fixes
- Task execution → Final acceptance: submits result
- Autopilot command → Project memory: maintains
- Autopilot command → Agent prompts: uses prompts
- Autopilot command → Instrument guidance: documents
- Run state and tickets → Run artifacts: writes state
skills/autopilot/tools/ap.py - Run state and tickets → Live dashboard: serves
skills/autopilot/tools/ap.py - Run measurement → Claude session logs: reads logs
tools/measure-run.py - Run measurement → Run measurement: summarizes usage
tools/measure-run.py - Autopilot command → Run state and tickets: records events