菜单
Audit the skills library for duplicates, stale content, spec violations, and structural issues. Run periodically or before releases.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Regenerate catalog counts, layout claims, and README summaries from canonical sources. Run after adding, removing, or renaming skills, commands, or bundles.
Validate SKILL.md files against the Agent Skills spec and Claude Code extensions. Run on new or modified skills before committing.
Audit AI agent instruction files (AGENTS.override.md, AGENTS.md, configured fallbacks, CLAUDE.md, hooks, and settings) across workspaces in read-only report mode. Use when agent configs drift, rules duplicate, files go stale, or after workspace restructuring; apply fixes only when explicitly requested.
Exercises rejection of an unsupported Codex project command path.
Design and operate LLM-as-a-Judge evaluation systems using direct scoring, pairwise comparison, rubric calibration, evaluator bias mitigation, confidence scoring, and automated quality assessment. Use when building LLM-as-judge systems, comparing model responses, calibrating rubrics, debugging inconsistent evaluations, or designing A/B tests for prompt or model changes.
Expert guide on prompt engineering patterns, best practices, and optimization techniques. Use when user wants to improve prompts, learn prompting strategies, debug agent behavior, or design content generation prompts.
| name | skill-auditor |
| description | Audit the skills library for duplicates, stale content, spec violations, and structural issues. Run periodically or before releases. |
| metadata | {"internal":true,"version":"1.1.0","tags":"audit, skills, quality, maintenance"} |
Comprehensive audit of the skills library for quality and consistency.
| Mode | Scope | Command |
|---|---|---|
skills (default) | Public skills, commands, bundles, and catalog sync | Follow the categories below |
routines | Local Claude scheduled tasks and Codex automations | python3 scripts/audit-routines.py |
Routine mode follows
.agents/memory/system/routine-standards.md. It is read-only: detect normalized
duplicate bodies and prompt-level model, effort, schedule, cwd/workspace/worktree,
and environment leakage. Report source identifiers and field names only; never print
prompt bodies, environment values, or TOML values. Use --show-paths only when the
caller explicitly wants local paths in the report.
Find skills that overlap and should be consolidated:
when_to_use triggersreact-patterns vs react-refactor)For each SKILL.md, check:
.agents/memory/system/skill-standards.md)version/tags inside metadata: block, not top-levelauto_activate, auto_trigger, risk)metadata.tags is a comma-separated string, not YAML list.agents/memory/system/execution-boundary.md to skills, commands, and routine
templatesskills/<name>/<name>/)/workspace/, @genfeedai/)\```) instead of real fencesCompare skills/ directory listing against README.md skills table:
# Get skills from filesystem
find skills -maxdepth 1 -mindepth 1 -type d | sed 's|skills/||' | sort > /tmp/fs-skills.txt
# Get skills from README table
grep -oE '\[([a-z0-9-]+)\]\(https://skills\.sh' README.md | sed 's/\[//' | sed 's/\](https:\/\/skills\.sh//' | sort > /tmp/readme-skills.txt
# Diff
comm -23 /tmp/fs-skills.txt /tmp/readme-skills.txt # in fs, not README
comm -13 /tmp/fs-skills.txt /tmp/readme-skills.txt # in README, not fs
Report findings as a table:
| Skill | Issue | Severity | Action |
|---|---|---|---|
code-refactoring-refactor-clean | Duplicate of refactor-code | HIGH | Merge |
spec-to-code-compliance | Empty directory | HIGH | Delete |
For routine mode, report totals, leakage categories, anonymous source identifiers,
duplicate-family groups, and observed app-owned field names. Do not include the
matched text or configuration values.
| skill-capture | tags as YAML list | MEDIUM | Fix to string |