Find Skills

Discover and install specialized agent skills from both the open ecosystem and local plugin catalogs.

The best skill is the one that already exists. Search before you build.

Guiding Principles

1. Local first — Check installed plugins before external search. Your ecosystem already has 400+ skills across SDF, MAO, PM, and SA.
2. Evidence over popularity — Install counts matter, but source reputation and MOAT depth matter more.
3. Security gate — Never auto-install unverified skills. Present evidence, let the user decide.

Procedure

Step 0 — Read References

Read references/skill-sources.md for the catalog of known sources and quality tiers.

Step 1 — Parse Intent

Extract from user request:

1. Domain — e.g., React, testing, deployment, data, security
2. Task — e.g., write tests, review PRs, scaffold project, optimize performance
3. Scope — local only, remote only, or both (default: both)

Tag intent: [INFERENCIA] if inferred from vague request, [DOC] if user stated explicitly.

Step 2 — Search Local Ecosystem

Search installed plugin skills first. They are pre-vetted and MOAT-certified:

# Search across all local plugin skill catalogs
grep -rl "$QUERY" /Users/deonto/skills/plugins/*/skills/*/SKILL.md 2>/dev/null
grep -rl "$QUERY" /Users/deonto/skills/plugins/*/references/ontology/skills-catalog.md 2>/dev/null

Check these catalogs in order:

1. SA (127 skills) — sa/references/ontology/skills-catalog.md
2. SDF (100+ skills) — sofka-discovery-framework/references/ontology/skills-catalog.md
3. MAO (100+ skills) — metodologia-discovery-framework/references/ontology/skills-catalog.md
4. PM (100+ skills) — pm-project-framework/references/ontology/skills-catalog.md

If a local match is found, skip to Step 5 with the local recommendation.

Step 3 — Search Remote Ecosystem

If no local match or user explicitly wants remote:

1. Check skills.sh leaderboard via WebFetch:

   https://skills.sh/?q=$QUERY

2. Run Skills CLI if available:

   npx skills find "$QUERY"

3. Search GitHub for known skill repositories:

   • vercel-labs/skills — Web dev, React, Next.js
   • anthropics/skills — Frontend design, document processing
   • ComposioHQ/awesome-claude-skills — Community curated

Step 4 — Evaluate Quality

Apply quality tiers to every candidate:

For local skills, apply MOAT depth assessment:

Step 5 — Present Recommendations

Format results as a comparison table:

## Skill Recommendations for: "{query}"

| # | Skill | Source | Type | Quality | Install Command |
|---|-------|--------|------|---------|-----------------|
| 1 | skill-name | SA (local) | MOAT-Full | S-tier | `/sa:skill-name` |
| 2 | skill-name | skills.sh | Remote | A-tier | `npx skills add owner/repo@skill -g -y` |

### Recommendation
[Your reasoned recommendation based on evidence]

Include for each skill:

• What it does (one sentence)
• Source and trust level
• Install/invoke command
• Link to learn more (skills.sh URL or local path)

Step 6 — Install (if approved)

Local skills: Already available. Provide the slash command.

Remote skills: Install with user confirmation only:

npx skills add <owner/repo@skill> -g -y

Never auto-install. Present the command and wait for approval.

Step 7 — No Match Protocol

If no relevant skill exists:

1. Acknowledge the gap explicitly
2. Offer to help with the task using general capabilities
3. If the task is recurring, suggest creating a custom skill:

   /pqa:create    # Full MOAT skill creation pipeline
   /sa:create     # Quick artifact creation
   npx skills init my-skill  # External skill scaffold

Common Search Categories

Edge Cases

• Ambiguous query: Ask one clarifying question maximum, then search broadly
• Multiple good matches: Present top 3 with comparison table
• Local + remote overlap: Prefer local (MOAT-certified), note remote alternative
• Skills CLI not installed: Fall back to WebFetch of skills.sh, suggest npm i -g skills
• Offline: Search local catalogs only, note that remote search was skipped