| name | skill-finder |
| description | Search the installed skill library to find and fully activate the most relevant skills for any user task. Use when the user asks to find a skill, search for skills, says "find the best skill for this", "which skill should I use", "search skills", or when the assistant needs to identify which installed skills are most applicable to a complex or unfamiliar task before execution. Also use proactively when facing a task outside core capabilities to discover specialized workflows, scripts, and domain expertise from the skill library. |
| license | Unspecified |
Skill Finder
Identify, select, and fully activate the most relevant installed skills to execute any user task end-to-end — not summarize, but use them.
Workflow
-
Analyze the request — Extract the core task, domain, and key action verbs from the user's prompt. For multi-part tasks, decompose into distinct subtasks.
-
Search the library — Run the search script with a targeted query:
python3 /home/ubuntu/skills/skill-finder/scripts/find_skills.py "user task description" --top 5 --verbose
Query refinement: If the top result scores below 3.0 or seems off-target, rephrase using synonyms, domain-specific terms, or split into subtask queries. Retry up to 3 times with different formulations. For multi-part tasks, run separate searches per subtask.
-
Select skills — Pick the top 1-3 most relevant results. Prefer skills with executable scripts (more actionable). For guidance on combining multiple skills, see references/selection-strategy.md.
-
Read and activate — For each selected skill, sequentially:
a. Read the skill's full SKILL.md at /home/ubuntu/skills/{skill-dir-name}/SKILL.md
b. Run its scripts in their specified directories
c. Consult its references as directed
d. Apply its workflows to produce outputs
e. On errors, check the skill's troubleshooting guidance before falling back
-
Execute the task — Integrate outputs from all activated skills into a unified result. Produce concrete deliverables, not summaries.
-
Handle no-match — If no skills score above 3.0 after all retries, inform the user that no specialized skill was found and proceed with base model capabilities. Consider using internet-skill-finder to search GitHub for new skills to install.
Script Reference
| Command | Purpose |
|---|
find_skills.py "query" --top N | Search for top N matching skills |
find_skills.py "query" --verbose | Include full descriptions in output |
find_skills.py "query" --json | Output as JSON for programmatic use |
find_skills.py --rebuild-index | Force rebuild the cached skill index |
find_skills.py --list-all | List all indexed skills |
Example
User asks: "audit my Python project for security issues"
python3 /home/ubuntu/skills/skill-finder/scripts/find_skills.py "audit Python project security vulnerabilities" --top 3
Top result: multi-model-code-auditor → Read its SKILL.md → Run its audit scripts → Deliver full audit report.
Constraints
- Never merely summarize skills — always execute their full procedures.
- Limit query reformulations to 3 retries maximum.
- Run skill scripts in their specified directories to preserve environment integrity.
- For multi-part tasks, address each subtask sequentially, then integrate results.
- Only ask the user for clarification as a last resort after autonomous attempts.
