| name | debug-autofix-skills |
| description | Investigate agent-side failures in the autofix-skills plugin. Covers wrong verdicts, review loop issues, state machine bugs, and prompt problems. Use when the agent ran but produced wrong results. |
| allowed-tools | Bash Read Grep Glob |
Debug Autofix-Skills
Investigate failures in the agent-side layer: skill prompts, review orchestration, verdict generation, and state management.
Key entry points
| Area | Entry point |
|---|
| Orchestration | skills/autofix-resolve/SKILL.md, skills/autofix-cve-resolve/SKILL.md |
| Triage | skills/autofix-triage/SKILL.md |
| State machine | scripts/state.py (duplicated in each skill's scripts/ dir) |
| Review merging | scripts/merge_findings.py |
| Verdict writing | scripts/write_json.py + schemas/*.json |
| Prompts | prompts/*.md within each skill directory |
| Hook recovery | hooks/hooks.json, tmp/dispatch-recovery.sh |
Protocol
- Read this repo's
AGENTS.md for architecture context.
- Check references/symptoms.md for known patterns matching the failure.
- Investigate the relevant skill's SKILL.md, prompts, scripts, and schemas.
- Write RCA using assets/rca-template.md.
