菜单
Audit skill tree for overlap, split-pressure, and consolidation candidates. Use when finding confusing skill clusters or surfacing REFERENCE.md extraction candidates.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/laurigates/claude-plugins --skill health-skill-audit命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Install, configure and troubleshoot MCP servers. Use when adding/enabling servers, editing .mcp.json, fixing OAuth, or when a server runs stale code after an upstream fix.
FinOps snapshot — org billing, workflow stats, cache usage. Use when you want a high-level view of CI spending or workflow health before diving deeper.
GitHub Actions billing, workflow efficiency, and waste analysis at org or repo level. Use when investigating CI/CD costs, wasted runs, or optimizing triggers.
正在显示 SKILL.md
基于 SOC 职业分类
| created | "2026-04-24T00:00:00.000Z" |
| modified | "2026-07-18T00:00:00.000Z" |
| compatibility | claude-code |
| reviewed | "2026-04-24T00:00:00.000Z" |
| description | Audit skill tree for overlap, split-pressure, and consolidation candidates. Use when finding confusing skill clusters or surfacing REFERENCE.md extraction candidates. |
| allowed-tools | Bash(python3 *), Read, TodoWrite |
| args | [--plugin <name>] [--strict] |
| argument-hint | [--plugin <name>] [--strict] |
| name | health-skill-audit |
Produce triageable evidence for skill-to-skill quality: overlap clusters, split-pressure inside a skill body, and consolidation candidates. The analyzer reports only — every recommendation is editorial.
Complements the other skill audits — this skill assumes frontmatter hygiene is already green.
| Use this skill when... | Use another approach when... |
|---|---|
| Finding overlapping skills ambiguous to an invoking agent | Frontmatter / size / body corruption (use scripts/plugin-compliance-check.sh) |
| Scoring split-pressure in an oversized skill body | Missing "Use when…" triggers (use scripts/audit-skill-descriptions.py) |
| Locating name-prefix or name-suffix clusters | CLI-flag compactness (use /health:check --scope=agentic) |
| Surfacing small sibling skills that could merge | Plugin ↔ stack relevance (use /health:check --scope=stack) |
find . -path '*/scripts/*' -maxdepth 3 -name audit-skill-structure.py -type ffind . -path '*/tmp/skill-audit/*' -maxdepth 3 -type ffind . -path '*/scripts/*' -maxdepth 3 -name plugin-compliance-check.sh -type ffind . -path '*/scripts/*' -maxdepth 3 -name audit-skill-descriptions.py -type fParse these from $ARGUMENTS:
| Parameter | Description |
|---|---|
--plugin <name> | Restrict scan to a single plugin directory (e.g. configure-plugin) |
--strict | Exit non-zero on warn-tier findings in addition to error-tier |
Execute this skill-structure audit:
These audits own orthogonal concerns. Run them first and resolve their findings before triaging this skill's output:
scripts/plugin-compliance-check.sh — frontmatter, size budget, body corruptionpython3 scripts/audit-skill-descriptions.py --strict — description fields have trigger phrasesIf either reports errors, pause and ask the user whether to continue — a messy frontmatter baseline makes the overlap heuristics noisier.
python3 scripts/audit-skill-structure.py $ARGUMENTS
This writes five artefacts into tmp/skill-audit/:
| File | Contents |
|---|---|
report.json | Canonical machine-readable output (source of truth) |
summary.md | Top-N by severity, intended for triage |
overlap-clusters.md | Side-by-side description comparison per cluster |
split-candidates.md | Per-skill evidence (lines, tables, fenced code, examples) |
consolidation-candidates.md | Conservative merge suggestions |
The schema for report.json and the rationale for every threshold live in REFERENCE.md.
summary.md firstOpen tmp/skill-audit/summary.md. It lists:
REFERENCE.md)For each cluster in tmp/skill-audit/overlap-clusters.md, pick one:
| Disposition | When it fits |
|---|---|
| Merge | Two small skills describing the same intent → fold into one |
| Rename | Clear distinct intent but colliding names → disambiguate the slug |
| Rewrite descriptions | Intent is distinct but descriptions collide → sharpen trigger phrases |
| Leave | Cluster is a family by design (e.g. python-plugin ecosystem skills) — no change |
Do not auto-act. Each disposition lands in its own PR per the CLAUDE.md guidance.
For each row in tmp/skill-audit/split-candidates.md, pick one:
| Disposition | When it fits |
|---|---|
Extract to REFERENCE.md | Warn/error-tier size or contiguous table block |
Extract to scripts/ | Fenced-code aggregate exceeds threshold |
Extract to examples/ | Multiple lengthy example blocks |
| Leave | Content is genuinely part of the skill's core instruction |
The report's reason column carries the quantitative evidence — cite it when opening the split PR.
Consolidation suggestions are conservative — both skills must be small, share a plugin, share a name prefix, and have high description similarity. Still, merge is editorial:
File a GitHub issue per consolidation decision rather than bundling them.
report.json is the source of truth for any follow-up PRs — cite the relevant JSON entries in PR descriptionstmp/fixtures/ if a fix should be prevented from recurring (see .claude/rules/regression-testing.md)| Context | Command |
|---|---|
| Full scan | python3 scripts/audit-skill-structure.py |
| Single-plugin scan | python3 scripts/audit-skill-structure.py --plugin configure-plugin |
| CI-gating run | python3 scripts/audit-skill-structure.py --strict |
| Top splits only | jq '.split_candidates | sort_by(-.lines) | .[:10]' tmp/skill-audit/report.json |
| Top clusters only | jq '.overlap_clusters | sort_by(-(.pairs | length)) | .[:10]' tmp/skill-audit/report.json |
.claude/rules/skill-quality.md — size limits and required sectionsscripts/plugin-compliance-check.sh — frontmatter and body corruption ownerscripts/audit-skill-descriptions.py — trigger-phrase coverage owner/health:check --scope=agentic — CLI-flag compactness audit