agentleFS
lifedeverlifedever/skills-plugin/skills/skill-scan/SKILL.md

skill-scan

Skill14 starsChanged 6 months agoMITView on GitHub

Browse and recommend across all locally installed Claude Code skills (personal + plugin + project). Use when the user wants to see what skills are available, has forgotten what they have, or asks which skill best fits a specific scenario. List-mode triggers: "list my skills", "what skills do I have", "show installed skills", "browse skills", "有哪些 skill", "本地 skill", "列一下 skill", "看下我装了什么 skill", "skill 清单", "/skill-scan". Recommend-mode triggers: "which skill should I use for X", "recommend a skill for X", "is there a skill that does Y", "我想做 X 用哪个 skill", "推荐一个 skill", "哪个 skill 能搞定 Y", "有没有处理 X 的 skill".

3 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 8 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈8

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Run the scan script to get the current inventory.

    line 23
  2. Very specific:

    If the summary says scanned 0 skills, tell the user and stop — likely no install.

    line 31
  3. Distinctive:

    No preamble: Don't say "Here are your skills" — go straight into categories.

    line 45
  4. Distinctive:

    Read CLAUDE.md (if present) plus the manifest filenames you found.

    line 56
  5. Very specific:

    Source tag: only when ambiguity matters.

    line 95
  6. Distinctive:

    Read the user's scenario. Score each scanned skill internally (don't show scores).

    line 107
  7. Distinctive:

    If the top candidate is clearly dominant (only one skill matches the platform/domain), skip to output.

    line 108
  8. Very specific:

    If top 2–3 candidates are plausible and disambiguating questions would help, ask at most one narrowing question (e.g., "is this for macOS …

    line 109
  9. Distinctive:

    Cap at 2 rounds total: If still unclear, present the 2–3 candidates with their differences and let the user pick.

    line 110
  10. Distinctive:

    If no skill matches well, say so plainly:

    line 121
  11. Distinctive:

    Do not pad with low-confidence suggestions.

    line 125

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

skill-scan

Discover and recommend across the user's locally installed skills. Solves the "I have too many skills installed and don't remember which one to use" problem.

Step 1 — Always scan first

Run the scan script to get the current inventory. Pass through the working directory so project-level skills are included:

bash "${CLAUDE_SKILL_DIR}/scripts/scan.sh"

Output is TSV (name\tsource\tpath\tdescription) on stdout, plus a one-line summary on stderr. Sources: project, personal, or plugin:<plugin-name>. Same-named skills are deduplicated by priority (project > personal > plugin).

If the summary says scanned 0 skills, tell the user and stop — likely no install. Otherwise, continue.

Step 2 — Decide mode from the user's phrasing

User saidMode
"list / browse / 有哪些 / 清单" typeList
Describes a scenario / asks "which skill for X"Recommend
AmbiguousAsk once: list, or scenario-based recommendation?

List mode

Emit a categorized inventory. Rules:

  1. No preamble. Don't say "Here are your skills" — go straight into categories.
  2. Detect current project context, then order categories by relevance. In ${PROJECT_DIR:-$PWD}, check for:

    • Claude config: CLAUDE.md, AGENTS.md, .claude/, .claude-plugin/
    • Repo marker: .git/
    • Language manifests: package.json, Package.swift, Cargo.toml, go.mod, pyproject.toml, requirements.txt, Gemfile, pom.xml, build.gradle, composer.json
    • Container: Dockerfile, docker-compose.yml

    Project mode — any of the above exist:

    • Read CLAUDE.md (if present) plus the manifest filenames you found; infer the specific project type (e.g. "macOS Swift app", "Vue frontend", "Claude Code skills plugin", "Rust CLI", "Node monorepo").
    • Order: categories most-aligned with that project type first → general/utility categories middle → unrelated categories last.
    • Stat-line suffix: · 已按当前项目(<inferred-type>)相关性排序.

    Environment mode — NONE of the above exist (home dir, ~/Downloads, scratch, etc.):

    • Order: system / maintenance / utility first (macOS 维护, 剪贴板 / 图床) → skill-mgmt and doc-sync middle → all dev workflows (Superpowers, Git/CR, Apple/Swift, frontend, build/release, project-specific) at the bottom.
    • Stat-line suffix: · 未检测到项目上下文,按系统 / 通用工具优先排序.

    Within a category, keep skill order stable (alphabetical or as scanned). Reordering is at the category level only.

  3. Categories are LLM-inferred from the descriptions. Typical buckets: macOS 维护 / 本机工具, Apple / Swift 开发, 前端 / 开发脚手架, 调试, 构建 / 发版 / CI, Git / Code review, 实施工作流(Superpowers), Skill / Plugin 自动化管理, 剪贴板 / 图床, 多语言 / 文档同步, 客户 / 项目专属. Skip empty buckets. Don't invent a bucket for a single skill — group it with the closest neighbor.
  4. Each category header carries one emoji chosen for fit; the skill lines under it stay emoji-free (emoji on every line creates visual clutter). Suggested mapping:

    • 🧹 macOS 维护 / 本机工具
    • 🍎 Apple / Swift 开发
    • 🎨 前端 / 开发脚手架
    • 🐛 调试
    • 🚀 构建 / 发版 / CI
    • 🔀 Git / Code review
    • 🧭 实施工作流(Superpowers)
    • 🧩 Skill / Plugin 自动化管理
    • 📎 剪贴板 / 图床
    • 🔁 多语言 / 文档同步
    • 🏢 客户 / 项目专属
  5. One line per skill, format:

    · <name> — <core use case in one sentence>
  6. The one-liner must answer "when do I use this?" — distill from the description, do NOT copy it verbatim. Drop trigger-word lists, emoji, marketing language. Keep concrete signals (platform, file type, "before X / after Y").

    • ❌ Vague: debug-mode — 调试工具
    • ❌ Lazy verbatim: debug-mode — Runtime debug mode - insert log probes, collect runtime data...
    • ✅ Useful: debug-mode — 运行时插探针定位 race condition / 内存泄漏 / 偶发 bug,静态分析搞不定时用
  7. Source tag only when ambiguity matters. Add (project) or (personal) suffix only if a name collision was hidden by dedup, or if the user explicitly asks about source. Plugin skills get no tag by default.
  8. End with one stat line, terse: <N> skills · <K> categories + the context-aware suffix from rule 2 (only when context was detected — in the indeterminate case, just <N> skills · <K> categories).

That's it. No "let me know if..." trailing question.

Recommend mode

Goal: surface 1–3 best-fit skills for the user's scenario, with enough info that they can invoke directly next time without going through skill-scan.

Round 1:

  1. Read the user's scenario. Score each scanned skill internally (don't show scores).
  2. If the top candidate is clearly dominant (only one skill matches the platform/domain), skip to output.
  3. If top 2–3 candidates are plausible and disambiguating questions would help, ask at most one narrowing question (e.g., "is this for macOS or Linux?", "CLI or in-code?", "do you already have X installed?"). Use AskUserQuestion with 2–4 concrete options.
  4. Cap at 2 rounds total. If still unclear, present the 2–3 candidates with their differences and let the user pick.

Output format per recommended skill:

<name>  ← why this fits
  • 触发词: <one or two phrases that auto-activate it in future conversations>
  • 前置: <prerequisites if any, else "无">
  • 适用: <the scenario sentence that matched>

If no skill matches well, say so plainly:

本地装的 skill 里没有专门处理这个的。可以用 /skill-creator 创建一个,或者直接告诉我具体步骤我手动做。

Do not pad with low-confidence suggestions.

Edge cases

  • 0 skills scanned → check ~/.claude/skills and ~/.claude/plugins/cache exist; report which is missing.
  • Skill name collisions across sources → dedup keeps highest-priority; mention in list mode only if the user asks about sources.
  • Missing frontmatter → script skips silently, stderr reports count. Don't surface this unless the count is large.
  • Description in a language the user isn't using → translate the one-liner to the user's language, don't quote the original.

Why scan every time (no cache)

Skills get added, removed, and bumped (plugin versions) frequently. A stale cache would silently miss new skills or recommend a version that doesn't exist anymore. The scan takes well under a second for typical installs.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.