SuperAPI AI Directory

skill-search

Agents & models

View repository

Updated 10-15 08:04

About this skill

Skill Search and Installation Assistant (v2)

Search, filter, and install Claude Code Skills from multiple channels.Pick the best of the best, preferring high stars.

Search Strategy (executed by priority)

First layer: curated aggregation repositories (local cache preferred)

Search these top repositories first; they have the broadest coverage and highest quality:

Repository ⭐ Stars Description Search method
hesreallyhim/awesome-claude-code 24.8K Official-grade curated collection clone → find SKILL.md
sickn33/antigravity-awesome-skills 14.7K Collection of 900+ Skills clone → find SKILL.md
VoltAgent/awesome-agent-skills 7.8K 380+ community skills clone → find SKILL.md
VoltAgent/awesome-claude-code-subagents 11.2K 100+ sub-agents clone → find SKILL.md
travisvn/awesome-claude-skills 7.6K Curated resource list Extract links from README
jeremylongshore/claude-code-plugins-plus-skills 1.4K Categorized plugins + skills clone → find SKILL.md
heilcheng/awesome-agent-skills 2.3K Multi-agent tool tutorials Extract links from README
rohitg00/awesome-claude-code-toolkit 545 135 agents + 35 skills clone → find SKILL.md
menkesu/awesome-pm-skills 185 28 PM skills clone → find SKILL.md
ramihassen95/claude-builder-template — TDD/subagent/debug clone → find SKILL.md
aAAaqwq/AGI-Super-Skills — Own skill repository (preferred) Reference directly
VoltAgent/awesome-openclaw-skills — OpenClaw-exclusive skills clone → find SKILL.md
VoltAgent/awesome-claude-code-subagents 11.2K 100+ sub-agents (can be assigned to team agents) clone → find SKILL.md

Cache mechanism: on the first search, clone to /tmp/skill-cache/, reuse for subsequent searches (within the same session).

CACHE="/tmp/skill-cache"
mkdir -p "$CACHE"
# 只在不存在时 clone
[ -d "$CACHE/awesome-claude-code" ] || git clone --depth 1 https://github.com/hesreallyhim/awesome-claude-code.git "$CACHE/awesome-claude-code"
# 搜索
find "$CACHE" -name "SKILL.md" | xargs grep -li "{关键词}" 2>/dev/null

Second layer: web search platforms

When the aggregation repositories do not find it, search these skill marketplace websites:

Platform URL Search method
SkillsMP https://skillsmp.com web_fetch to fetch the search page
ClawdHub https://clawhub.com web_fetch to fetch
SkillKit https://skillkit.dev (if present) web_fetch
# 示例:搜索 SkillsMP
web_fetch("https://skillsmp.com/search?q={关键词}")

Third layer: GitHub global search

When neither aggregation repositories nor websites have it, the GitHub API is the fallback:

# 搜索关键词组合(按效果排序)
"claude skill {关键词}" sort:stars
"SKILL.md {关键词}" sort:stars  
"{关键词} agent skill" sort:stars

Filtering rules (strictly enforced):

  • ⭐ > 100 shown first
  • ⭐ > 10 second choice
  • ⭐ < 10 shown only when there is no better option
  • Must contain a SKILL.md file
  • For skills of the same type, show only the 1-2 with the most stars

Result display

Each result includes:

📦 {技能名称}
📝 {一句话描述}
⭐ {Stars} | 🔗 {仓库URL}
📂 {SKILL.md 路径}

Sorting: ⭐ Stars descending; for ties, look at the most recent update time.

Show the Top 5-10 results and let the user choose.

Installation process

  1. Duplicate check:ls ~/.claude/skills/ ~/clawd/skills/ whether it already exists
  2. clone + locate:git clone --depth 1 → find -name SKILL.md
  3. Dual-path installation:
    cp -r {skill_dir} ~/.claude/skills/{name}/     # 全局
    cp -r {skill_dir} ~/clawd/skills/{name}/       # workspace(团队共享)
    
  4. Cleanup:rm -rf {temp_dir}
  5. Confirmation: output the installation path + prompt to restart for it to take effect

Dependency handling

  • Python → prompt pip install
  • Node.js → prompt npm install
  • Requires API key → clearly inform

Search tips

User says Search keywords
"find one for writing code" coding, development, code-generation
"SEO-related" seo, content-writing, programmatic-seo
"UI design" ui-ux, frontend-design, web-design
"automate XX platform" {platform}-automation
"improve programming ability" tdd, code-review, debugging, subagent, coding-agent

Error handling

  • GitHub rate limit → switch to web_fetch search
  • clone failed → check the network, retry once
  • Invalid SKILL.md → skip, mark ⚠️
  • No results → suggest using skillforge to generate automatically