Analyze And Fix All Skills
Coordinate the existing single-skill analysis and repair skills across a full local skills inventory.
UTILITY SKILL. INVOKES: analyze-customization-evaluations, fix-customization-evaluation-diagnostics, Waza CLI, quick_validate.py, local skill files.
Read batch repair workflow for command details, result classification, patch rules, and the final report format.
Dependencies
Read analyze-customization-evaluations for read-only Waza/skill-health analysis and fix-customization-evaluation-diagnostics for actionable repair policy. Do not create a separate repair policy here.
Target Discovery
Use scripts/list_skill_targets.py to build the worklist:
python3 <this-skill>/scripts/list_skill_targets.py --json
Defaults include ${CODEX_HOME:-$HOME/.codex}/skills and $HOME/.agents/skills, skip hidden/system directories, and skip plugin cache paths unless explicitly requested. If this coordinator appears in the worklist, process it last.
Batch Workflow
- Discover targets and report the count before editing.
- Locate Waza, run
waza models when available, and choose an available model.
- Run analyzer baselines per target root and
waza coverage when Waza is available.
- For each skill, run
quick_validate.py, run waza check when available, inspect relevant result JSON, and classify before editing.
- Record model/session/auth failures as runner blockers; patch only readiness, trigger, behavior, metadata, reference, or token-budget failures.
- Re-run validation after each patch, using at most two repair passes per skill unless the user asks for deeper iteration.
- Continue through the worklist when one skill is blocked.
Patch Rules
- Preserve each skill's purpose and useful workflow details.
- Keep
SKILL.md compact; move substantial detail into linked references/*.md.
- Keep reference files one hop from
SKILL.md.
- Update
agents/openai.yaml when UI metadata no longer matches.
- Do not add README, changelog, installation, or process-history files.
- Do not commit changes unless asked.
1---2name: analyze-and-fix-all-skills3description: Batch analyze and repair Codex skill evaluation health across installed skill folders. Use when the user says "analyze and fix all skills", "fix every skill", "loop over all skills", "make all skills pass Waza", "repair all Chat Customizations Evaluation diagnostics", or asks to run the analyze-customization-evaluations and fix-customization-evaluation-diagnostics workflow for every Codex skill. Do not use for product code tests or non-skill repositories.4---56# Analyze And Fix All Skills78Coordinate the existing single-skill analysis and repair skills across a full local skills inventory.910**UTILITY SKILL. INVOKES:** `analyze-customization-evaluations`, `fix-customization-evaluation-diagnostics`, Waza CLI, `quick_validate.py`, local skill files.1112Read [batch repair workflow](references/batch-repair-workflow.md) for command details, result classification, patch rules, and the final report format.1314## Dependencies1516Read `analyze-customization-evaluations` for read-only Waza/skill-health analysis and `fix-customization-evaluation-diagnostics` for actionable repair policy. Do not create a separate repair policy here.1718## Target Discovery1920Use `scripts/list_skill_targets.py` to build the worklist:2122```bash23python3 <this-skill>/scripts/list_skill_targets.py --json24```2526Defaults include `${CODEX_HOME:-$HOME/.codex}/skills` and `$HOME/.agents/skills`, skip hidden/system directories, and skip plugin cache paths unless explicitly requested. If this coordinator appears in the worklist, process it last.2728## Batch Workflow29301. Discover targets and report the count before editing.312. Locate Waza, run `waza models` when available, and choose an available model.323. Run analyzer baselines per target root and `waza coverage` when Waza is available.334. For each skill, run `quick_validate.py`, run `waza check` when available, inspect relevant result JSON, and classify before editing.345. Record model/session/auth failures as runner blockers; patch only readiness, trigger, behavior, metadata, reference, or token-budget failures.356. Re-run validation after each patch, using at most two repair passes per skill unless the user asks for deeper iteration.367. Continue through the worklist when one skill is blocked.3738## Patch Rules3940- Preserve each skill's purpose and useful workflow details.41- Keep `SKILL.md` compact; move substantial detail into linked `references/*.md`.42- Keep reference files one hop from `SKILL.md`.43- Update `agents/openai.yaml` when UI metadata no longer matches.44- Do not add README, changelog, installation, or process-history files.45- Do not commit changes unless asked.
willakins