Find Skills

find-skills

Checks whether somebody has already solved the problem before you spend three hours solving it again. Searches nearest-first, and tells you honestly how much to trust each result.

When to use this

Fires when the user asks:

• "find me a skill for "
• "is there already a skill that does ?"
• "where can I get a skill that ?"
• "I'm about to build a skill for " — check first, this is the best moment
• when a task comes up repeatedly and no installed skill covers it

Does NOT fire on:

• writing or editing a skill they already have — that is authoring, not discovery
• reviewing a skill's quality — use skill-reviewer
• an installed skill failing to load — that is a client configuration problem
• finding a Python package, VS Code extension, or MCP server. Those are not skills

Steps

1. Turn the request into English keywords. The indexes are English-only. If the user writes in Arabic or another language, translate the intent — do not pass the raw string.

2. Run the search:

   python scripts/search_skills.py "<keywords>" --limit 5
   

   It queries three sources in this order, and a dead source degrades the result rather than ending the run:

   Order	Source	Why it is in this position
   1	The Skill Foundry	Already reviewed, already has evidence, and a classmate can explain it
   2	skillspool.org	Curated directory with real metadata
   3	GitHub	The long tail — comprehensive but unreviewed

3. Read the degradation note. If the script reports sources down, say so in your answer. A partial search that looks complete is worse than one that admits its gaps.

4. Assess safety before recommending. Apply resources/trust-signals.md to every candidate. Never present an unreviewed result as if it were vetted.

5. Recommend at most three. For each: one line on why it fits, the trust signal, and the link. More than three and the user just picks the first anyway.

6. If nothing fits, say so and offer the alternative: open a skill request on the Foundry, or build it — the request template is at .github/ISSUE_TEMPLATE/skill-request.yml.

7. Before they install anything from outside the Foundry, read its SKILL.md and check it against the red flags in resources/trust-signals.md.

Rules

• Never auto-install. Recommend, link, and let the user decide. Installing a skill grants it the same tool access you have.
• Always check The Skill Foundry first. A reviewed skill from a classmate beats a 40k-star repo nobody here has read.
• Never present a GitHub result without the word unreviewed attached.
• Star count is a popularity signal, not a safety signal. A skill can be popular and still be wrong for this user.
• If every source is down, say so plainly. Do not answer from memory — skill directories change weekly and a remembered answer will be stale.

Resources

Scripts