Multi-Model Discovery Skill

LIBRARY-FIRST PROTOCOL (MANDATORY)

Step 1: Library Catalog

• Location:

  .claude/library/catalog.json

• If match >70%: REUSE or ADAPT

Step 2: Patterns Guide

• Location:

  .claude/docs/inventories/LIBRARY-PATTERNS-GUIDE.md

• If pattern exists: FOLLOW documented approach

Step 3: Existing Projects

• Location:

  D:\Projects\*

• If found: EXTRACT and adapt

Decision Matrix

Kanitsal Cerceve (Evidential Frame Activation)

Purpose

When to Use This Skill

• Before implementing a new feature (find existing solutions first)
• When researching best practices for a technology
• When looking for code examples or patterns
• When evaluating libraries or frameworks
• When unsure if a problem has already been solved
• Before writing boilerplate code that might exist

When NOT to Use This Skill

• For implementation tasks (use codex-iterative-fix instead)
• When you already know the solution exists in the codebase
• For debugging existing code (use smart-bug-fix)
• For codebase analysis (use gemini-codebase-onboard)

Workflow

Phase 1: Research Query Formulation

1. Analyze the implementation goal
2. Formulate search queries for:
   • Existing libraries/packages
   • Code examples on GitHub
   • Best practice guides
   • Common patterns

Phase 2: Gemini Discovery Execution

# Execute via delegate.sh wrapper
./scripts/multi-model/delegate.sh gemini "Find existing solutions for: {goal}"

# Or via gemini-yolo.sh
./scripts/multi-model/gemini-yolo.sh "How do others implement {feature}? Find code examples and libraries." task-id research

Phase 3: Results Synthesis

1. Claude synthesizes Gemini's findings
2. Evaluate options:
   • Use existing library
   • Adapt existing pattern
   • Build from scratch (last resort)
3. Document decision rationale

Success Criteria

• Existing solution found and evaluated
• Build vs buy decision made with evidence
• Time saved by avoiding reinvention
• Quality improved by using proven patterns

Example Usage

Example 1: Auth Implementation

User: "Implement user authentication"

Discovery Process:
1. Gemini search: "What are best practices for auth in Node.js?"
2. Gemini search: "Find existing auth libraries: passport, next-auth, lucia"
3. Gemini search: "Code examples for JWT authentication Node.js"

Output:
- Recommended: next-auth (well-maintained, 40k+ stars)
- Alternative: lucia-auth (newer, type-safe)
- Pattern found: middleware-based validation

Example 2: PDF Generation

User: "Generate PDF reports from data"

Discovery Process:
1. Gemini search: "PDF generation libraries JavaScript 2024"
2. Gemini search: "Compare pdfkit vs puppeteer vs react-pdf"
3. Gemini search: "Production PDF generation best practices"

Output:
- Simple PDFs: pdfkit (lightweight)
- Complex layouts: puppeteer (HTML to PDF)
- React apps: react-pdf

Integration with Meta-Loop

META-LOOP PROPOSE PHASE:
    |
    +---> multi-model-discovery
    |         |
    |         +---> Gemini: Find existing solutions
    |         +---> Claude: Evaluate options
    |         +---> Decision: Build vs Adapt vs Use
    |
    +---> Continue to IMPLEMENT phase

Memory Integration

• Key:

  multi-model/discovery/{project}/{task_id}

• Tags: WHO=multi-model-discovery, WHY=avoid-reinvention

Invocation Pattern

# Via router (automatic detection)
./scripts/multi-model/multi-model-router.sh "Find existing solutions for X"

# Direct Gemini call
bash -lc "gemini 'How do others implement X? Find code examples and libraries.'"

Related Skills

• gemini-research

  : General research with search grounding

• gemini-megacontext

  : Full codebase analysis

• codex-iterative-fix

  : After discovery, for implementation

• literature-synthesis

  : Academic research synthesis

Verification Checklist

• Gemini search executed with clear queries
• Multiple solutions discovered and compared
• Build vs buy decision documented
• Memory-MCP updated with findings
• Decision rationale captured