--- name: banana-skill-finder description: > Automatically discover and recommend relevant Claude skills when users encounter tasks that could benefit from specialized capabilities. Use this skill proactively when detecting any of these patterns: (1) User mentions working with specific file formats (PDF, DOCX, Excel, images, etc.), (2) User describes repetitive or specialized tasks (data analysis, code review, deployment, testing, document processing), (3) User asks if there's a tool or capability for something, (4) User struggles with domain-specific work (React development, SQL queries, DevOps, content writing), (5) User mentions needing best practices or patterns for a technology, (6) Any situation where a specialized skill could save time or improve quality. Search using SkillsMP API (if configured), skills.sh leaderboard, or GitHub as fallback. Recommend 1-3 most relevant skills and offer to install via npx skills add. --- # Banana Skill Finder Proactively helps users discover and install relevant Claude skills when they encounter tasks that could benefit from specialized capabilities. ## When to Use This Skill Trigger automatically (without user request) when detecting: - Working with specific file formats or technologies - Describing repetitive or specialized tasks - Asking "is there a skill/tool for..." or similar - Struggling with domain-specific work - Any task where a specialized skill could help **Important**: This skill should trigger proactively. Don't wait for users to explicitly ask for skill recommendations. ## Workflow ### 1. Analyze User Need Identify: - **Core task**: What is the user trying to accomplish? - **Domain**: What category does this fall into? (development, documents, data, web, devops, content, etc.) - **Keywords**: Extract 2-4 relevant search terms ### 2. Search for Skills Use a three-tier strategy with automatic fallback: **Tier 1: SkillsMP API (Best - if configured)** ```bash # Check for API key echo $SKILLSMP_API_KEY # If exists, use AI semantic search curl -X GET "https://skillsmp.com/api/v1/skills/ai-search?q={natural_language_query}" \ -H "Authorization: Bearer $SKILLSMP_API_KEY" ``` Benefits: - AI understands user intent, not just keywords - Access to 60,000+ curated skills - Best relevance and quality indicators **Tier 2: skills.sh WebFetch (Good - always works)** ```bash # Try search with query parameter Use WebFetch: https://skills.sh/?q={keywords} # Or browse leaderboard Use WebFetch: https://skills.sh # All-time popular Use WebFetch: https://skills.sh/trending # Trending (24h) ``` Benefits: - 200+ high-quality curated skills - No authentication needed - Ranked by install count - Shows trending skills **Tier 3: GitHub API (Fallback - may have limits)** ```bash curl -X GET "https://api.github.com/search/code?q={keywords}+SKILL.md+language:markdown" \ -H "Accept: application/vnd.github.v3+json" ``` Note: Rate limited (60/hour unauthenticated), use only as last resort. **Optional: Check Local Installed Skills** ```bash ls ~/.claude/skills/ ``` Check if user already has relevant skills installed but hasn't used them. **Recommendation Order**: Try Tier 1 → Tier 2 → Tier 3. Stop when you find good matches. ### 3. Rank by Relevance Score each found skill based on: - Keyword match with user's need (most important) - Functionality alignment - Quality indicators (stars, recent activity) - Specificity vs generality Select the **1-3 most relevant** skills. Quality over quantity. ### 4. Present Recommendations Format recommendations as: ``` I found [N] skill(s) that could help: **1. [Skill Name]** - [One-line description] Source: [SkillsMP/GitHub/Vercel] Repository: [owner/repo] Why relevant: [Brief explanation] Install: `npx skills add [owner]/[repo]` [Repeat for 2-3 skills max] Would you like me to install any of these? ``` ### 5. Install if Approved When user approves, install using Vercel's skills CLI: ```bash npx skills add / ``` Examples: ```bash npx skills add vercel-labs/agent-skills npx skills add anthropics/skills ``` This command: - Downloads the skill from GitHub - Installs to `~/.claude/skills/` - Works with Claude Code, Cursor, Windsurf, and other agents - Tracks installation via anonymous telemetry (leaderboard) Confirm installation success and explain how the skill will help. ## Key Principles 1. **Proactive, Not Reactive**: Trigger automatically when relevant, don't wait to be asked 2. **Quality Over Quantity**: Recommend only 1-3 best matches, not a long list 3. **Smart Three-Tier Search**: - Tier 1: SkillsMP AI search (best, if configured) - Tier 2: skills.sh leaderboard (good, always works) - Tier 3: GitHub API (fallback, rate limited) - Stop when you find good matches 4. **Explain Relevance**: Always explain why each skill matches their need 5. **Easy Installation**: Use `npx skills add owner/repo` for one-command installation 6. **API Key Recommended but Optional**: Best results with SkillsMP API key, but skills.sh fallback works well ## Examples **User says**: "I need to extract text from a PDF file" → Trigger skill-finder, search for PDF processing skills, recommend pdf-editor or similar **User says**: "Help me review this React component" → Trigger skill-finder, search for React/code-review skills, recommend react-best-practices from Vercel **User says**: "I'm deploying to AWS" → Trigger skill-finder, search for AWS/deployment skills, recommend cloud-deploy or aws-helper **User says**: "How do I query this BigQuery table?" → Trigger skill-finder, search for BigQuery/SQL skills, recommend bigquery or data-analysis skills ## Additional Resources For detailed information: - [references/api_config.md](references/api_config.md) - How to set up SkillsMP API key - [references/skill_sources.md](references/skill_sources.md) - Skill sources, categories, and search strategies ## Setup Recommendations For best results, suggest users configure SkillsMP API key: 1. Visit https://skillsmp.com/docs/api 2. Generate API key 3. Set environment variable: `export SKILLSMP_API_KEY="sk_live_..."` This enables AI semantic search (much better than keyword matching). Without it, the skill automatically falls back to skills.sh leaderboard search, which still works well for most cases.