find-skill

find-skill

Discover an installable skill that already covers a user's need, search across ClawHub → skill.sh / SkillHub → GitHub in strict priority order, and return a ranked candidate list with install hints.

When to use

• "Find me a skill that handles PDFs / scrapes Xiaohongshu / drives a browser."
• "Is there an existing skill for X?" — before authoring a new one.
• "Look on GitHub / ClawHub / skill.sh for something similar."

Don't do when the user already supplied a concrete skill URL or local directory and only wants that one installed — hand it to the install path directly, no discovery needed.

How to call

1. Convert the request into searchable capability terms

• Primary task: e.g. scrape webpage, process Excel, generate slides, drive browser.
• Inputs: keywords, URL, file, account, API key, etc.
• Outputs: JSON, report, file, chart, edited document.
• Platform constraints: Chinese platform, GitHub, browser login, specific SaaS, Windows / macOS, etc.

Compose 2–5 search phrases:

• Chinese: e.g. 网页抓取 skill, Excel 处理 skill.
• English: e.g. web scraping skill, Excel xlsx agent skill.
• Synonyms: e.g. browser automation / web automation.

2. Search sources in fixed priority order

Walk the sources in order. Move to the next source only when the current source has no acceptable candidate.

Source 1 — ClawHub (try first)

ClawHub is the preferred source.

npx clawhub search "keyword"
npx clawhub browse
npx clawhub search "web search"
npx clawhub search "weather"
npx clawhub search "document"
npx clawhub search "github"
npx clawhub search --sort installs
npx clawhub search --sort stars

Public pages:

• https://clawhub.ai/
• Detail page, e.g. https://clawhub.ai/fangkelvin/find-skills-skill

Treat ClawHub as "has a good candidate" when all of the following hold:

• Capability directly covers the user's need.
• Clear README / SKILL.md / install instructions.
• Not a tutorial post, blog article, or empty repo.
• Dependencies are acceptable.
• At least one fit: high, or several medium candidates worth offering.

If satisfied, stop. Do not continue to skill.sh / SkillHub or GitHub.

Source 2 — skill.sh / SkillHub (only if ClawHub has nothing)

npx skills find "keyword"
npx skills add <owner/repo@skill>
npx skills add <owner/repo@skill> -g -y    # global, skip confirmation

Public pages:

• https://skillhub.cn/skills/find-skills
• https://skillhub.cn/
• https://skills.sh/

SkillHub HTTP endpoints:

https://api.skillhub.cn/api/skills?page=1&pageSize=10&keyword=<keyword>
https://api.skillhub.cn/api/v1/skills/<slug>
https://api.skillhub.cn/api/v1/skills/<slug>/files
https://api.skillhub.cn/api/v1/skills/<slug>/file?path=SKILL.md
https://api.skillhub.cn/api/v1/download?slug=<slug>

Treat SkillHub as "has a good candidate" when:

• Result name, description, or SKILL.md matches the need.
• An install command, detail page, or download link is reachable.
• The hit is an installable skill, not a marketing list or curated index page.
• Install source is clear and dependency risk is explainable.

If satisfied, stop. Do not continue to GitHub.

Source 3 — GitHub (only if both above have nothing)

Prefer the gh CLI to search and read repositories.

Verify CLI availability first:

gh --version
gh auth status

If unauthenticated, public search may still work; on rate limit or permission error, ask the user to run gh auth login.

Search code containing SKILL.md:

gh search code 'filename:SKILL.md web scraping skill' --limit 20
gh search code 'filename:SKILL.md browser automation' --limit 20
gh search code 'filename:SKILL.md Excel xlsx agent skill' --limit 20
gh search code 'filename:SKILL.md openclaw skill' --limit 20
gh search code 'filename:SKILL.md claude skill' --limit 20

Replace <capability> with the user's task:

gh search code 'filename:SKILL.md <capability>'
gh search code 'filename:SKILL.md <capability> skill'
gh search code 'filename:SKILL.md <capability> agent skill'
gh search code 'filename:SKILL.md <capability> openclaw skill'
gh search code 'filename:SKILL.md <capability> claude skill'

Search relevant repositories:

gh search repos '<capability> skill' --limit 20 --json fullName,description,url,updatedAt,stargazersCount
gh search repos '<capability> agent skill' --limit 20 --json fullName,description,url,updatedAt,stargazersCount
gh search repos '<capability> SKILL.md' --limit 20 --json fullName,description,url,updatedAt,stargazersCount
gh search repos 'awesome claude skills <capability>' --limit 20 --json fullName,description,url,updatedAt,stargazersCount

When gh search code output is not structured enough, fall back to the API:

gh api search/code -f q='filename:SKILL.md <capability>' -f per_page=20 \
  --jq '.items[] | {name: .name, path: .path, repo: .repository.full_name, url: .html_url}'

Inspect a candidate repo:

gh repo view OWNER/REPO --json nameWithOwner,description,url,updatedAt,stargazerCount,defaultBranchRef,licenseInfo
gh api repos/OWNER/REPO/contents --jq '.[] | {path: .path, type: .type, size: .size}'
gh api -H "Accept: application/vnd.github.raw" repos/OWNER/REPO/contents/SKILL.md
gh api -H "Accept: application/vnd.github.raw" repos/OWNER/REPO/contents/path/to/SKILL.md
gh api repos/OWNER/REPO/readme --jq '.content' | base64 -d
gh api repos/OWNER/REPO/commits -f per_page=5 \
  --jq '.[] | {sha: .sha, date: .commit.author.date, message: .commit.message}'
gh release list --repo OWNER/REPO --limit 10

For a quick whole-repo scan, shallow-clone to a scratch directory:

gh repo clone OWNER/REPO /tmp/skill-candidate -- --depth 1

Recommended search phrasing:

<capability> skill SKILL.md
<capability> agent skill
<capability> openclaw skill
<capability> claude skill
filename:SKILL.md <capability>

When inspecting candidates, prioritize:

• Has SKILL.md.

-…