| name | research-first |
| description | Meta-skill that teaches AI to research before coding. Always check latest docs via context7 MCP, verify versions, and validate assumptions before writing code. Use this skill for ANY coding task. |
| allowed-tools | Read, Write, Edit, Glob, Grep |
Research-First — Always Verify Before You Code
Philosophy: AI training data goes stale. Libraries update. APIs change.
Always research first, code second.
The Research-First Protocol
Before writing ANY code that uses a library, framework, or API:
1. IDENTIFY → What libraries/frameworks am I about to use?
2. RESEARCH → Query context7 MCP for latest docs
3. VERIFY → Check version compatibility
4. CODE → Write code based on verified information
5. VALIDATE → Test that it actually works
When to Research (MANDATORY)
| Trigger | Action |
|---|
| Using any npm/pip/cargo package | Query context7 for latest API |
| Framework-specific patterns (Next.js, React, Vue) | Check current version docs |
| Deployment configs (Docker, CI/CD) | Verify current best practices |
| Security implementations (auth, crypto) | Always check latest recommendations |
| Any error you're unsure about | Search docs before guessing |
When NOT to Research
| Scenario | Why |
|---|
| Pure logic/algorithms | Math doesn't change |
| Language fundamentals (JS, Python, Go) | Core syntax is stable |
| Git operations | Well-established, stable |
| File I/O, string manipulation | Standard library, stable |
How to Research — Tool Usage
Primary: context7 MCP (Real-time Docs)
Step 1: Resolve library ID
→ mcp_context7_resolve-library-id({ libraryName: "next.js", query: "app router setup" })
Step 2: Query specific docs
→ mcp_context7_query-docs({ libraryId: "/vercel/next.js", query: "server components data fetching" })
Secondary: Web Search (When context7 lacks coverage)
→ search_web({ query: "React 19 use() hook examples 2026" })
Tertiary: GitHub (Implementation examples)
→ mcp_github_search_code({ q: "useFormStatus react-dom language:typescript" })
Research Depth by Task Complexity
| Task | Research Depth |
|---|
| Simple (add a button, style a div) | None needed |
| Medium (integrate a library, add auth) | 1 context7 query |
| Complex (new architecture, unfamiliar framework) | 2-3 queries + verify |
| Critical (security, payments, data migration) | Full research + cross-reference |
Anti-Patterns
| ❌ Don't | ✅ Do |
|---|
| Assume you know the latest API | Query context7 first |
| Copy patterns from training data | Verify they're current |
| Use deprecated methods | Check migration guides |
| Guess at config options | Read actual docs |
| Write 500 lines then test | Research → small test → build |
Decision Framework
User asks: "Add Stripe payments to my Next.js app"
❌ BAD: Start coding from memory
→ Might use old Stripe.js API
→ Might use deprecated Next.js patterns
✅ GOOD: Research first
1. context7: "next.js server actions" → get current patterns
2. context7: "stripe" → get latest Stripe.js API
3. Verify: Next.js version in package.json
4. Code: Using verified, current patterns
5. Test: Verify integration works
Remember: Your training data is a starting point, not ground truth.
When in doubt, look it up. The 30 seconds spent researching saves 30 minutes debugging outdated code.
