find-skill-dual（找到双技能）

find-skill-dual · 最全 skill 中英搜索工具

Searches OpenClaw skills across ClawHub, AgentSkill.work (330+ curated repos, bilingual), and GitHub; optionally Smithery MCP for MCP servers. Presents merged deduplicated results (sorted by score/stars), with related-skill suggestions and inspect-before-install.

When to Use

• "找 skill"、"搜索技能"、"搜索工具"、"安装 skill"
• "有什么技能可以 X"、"找一个技能"
• "find a skill for X", "search skills", "how do I do X", "extend capabilities"
• ClawHub results insufficient, need "全网搜索" or "GitHub 上的 skill"
• User wants to browse trending skills ("看看热门 skill"、"有什么新 skill")
• User asks for "MCP 工具" or "MCP server" (optional Smithery search)

Workflow

Step 0: Intent Check

• Browse/Explore → User says "热门"、"trending"、"有什么新"、"explore" → skip to Step 0b
• Search → User has a topic/keyword → proceed to Step 1

Step 0b: Browse Trending (optional)

When user wants to explore without a specific query:

clawhub explore --sort trending --limit 10 2>&1
# or: clawhub explore --sort downloads --limit 10

AgentSkill.work topics (discover by topic):

curl -s "https://agentskill.work/api/facets/topics?limit=20" 2>&1
# then: curl -s "https://agentskill.work/api/skills?topic=<topic>&limit=10&sort=stars"

Present results in table, then offer to search by keyword if they want more specific results.

Step 1: Get Search Query

Extract topic/keyword from user. Examples: "pdf", "github", "weather", "翻译", "笔记".

If not specified, use broad query like "openclaw skill".

Step 2: Search ClawHub

clawhub search "<query>" 2>&1

Note: ClawHub returns vector similarity scores. Limit to top 10–15 results.

Step 3a: Search AgentSkill.work

No API key needed. Returns 330+ curated GitHub repos with bilingual metadata (description_zh):

curl -s "https://agentskill.work/api/skills?q=<query>&limit=10&sort=stars" 2>&1

Parse JSON: full_name, description or description_zh, stars, html_url. Prefer description_zh when user speaks Chinese.

Step 3b: Search GitHub

gh search repos "openclaw skill <query>" --limit 15 --json name,fullName,description,url,stargazersCount 2>&1

If gh fails (not installed or not authenticated):

gh auth status 2>&1

Use web_search: "openclaw" "skill" "<query>" site:github.com — summarize top 5–10 repos.

Step 3c (Optional): Search Smithery MCP

When user asks for "MCP 工具" or "MCP server" or wants MCP discovery. Requires SMITHERY_API_KEY in env:

curl -s -H "Authorization: Bearer $SMITHERY_API_KEY" "https://registry.smithery.ai/servers?q=<query>&pageSize=10" 2>&1

If no key, skip and mention: "Smithery MCP 搜索需要 API key，可在 smithery.ai/account/api-keys 创建"

Step 4: Merge, Deduplicate, Sort, Present

• Deduplicate: Same skill in multiple sources (e.g. slug ≈ owner/repo) → merge into one row, mark "ClawHub + AgentSkill.work" or "ClawHub + GitHub" etc.
• Sort: ClawHub by score; AgentSkill.work/GitHub by stars. Present: ClawHub first, then AgentSkill.work/GitHub (merged, deduped by full_name).
• Format:

## ClawHub 结果 (按评分)
| slug | 评分 | 描述 |
|------|------|------|
| xxx | 3.6 | ... |

## AgentSkill.work / GitHub 结果 (按 Stars)
| 仓库 | Stars | 描述 |
|------|-------|------|
| owner/repo | ⭐N | ... (优先用 description_zh) |

安装 ClawHub: `clawhub install <slug> --workdir ~/.openclaw/workspace`
安装 GitHub: `git clone <html_url> ~/.openclaw/workspace/skills/<skill名>`
多 skill 仓库: `npx skills add <html_url> --skill <skill-folder>`

Step 5: Inspect Before Install

When user wants to install a specific skill, offer to inspect first:

# ClawHub
clawhub inspect <slug> 2>&1
clawhub inspect <slug> --files 2>&1

For GitHub repos: fetch README or SKILL.md URL for user to review.

Step 5b: Related Skills (AgentSkill.work)

When user shows interest in a skill that has owner/repo (from AgentSkill.work or GitHub), offer related skills:

curl -s "https://agentskill.work/api/skills/<owner>/<repo>/related" 2>&1

Parse JSON and present 3–6 related skills in a compact table. Use description_zh when user speaks Chinese.

Step 6: Install

ClawHub: clawhub install <slug> --workdir ~/.openclaw/workspace GitHub: git clone https://github.com/owner/repo ~/.openclaw/workspace/skills/<name> Multi-skill repo: npx skills add https://github.com/owner/repo --skill <folder>

When No Skills Found

If ClawHub, AgentSkill.work, and GitHub all return no relevant results:

1. Acknowledge — "没有找到与「X」相关的 skill"
2. Offer direct help — "我可以直接帮你解决这个问题"
3. Suggest creating — "如果经常用到，可以自己发布: clawhub publish"

Prerequisites

• ClawHub: clawhub (via npm i -g clawhub)
• AgentSkill.work: no key, public REST API
• GitHub CLI (optional): gh auth login; if unavailable, fallback to web_search
• Smithery MCP (optional): SMITHERY_API_KEY for MCP server search; create at smithery.ai/account/api-keys

Example

User: "帮我找 pdf 相关的 skill"

1. clawhub search pdf
2. curl -s "https://agentskill.work/api/skills?q=pdf&limit=10&sort=stars"
3. gh search repos "openclaw skill pdf" --limit 10 --json name,fullName,description,url,stargazersCount
4. Merge, deduplicate, sort; show tables (use description_zh for Chinese)
5. User picks one → offer inspect → optionally fetch related → provide install command