research

name	research
description	深度研究技术问题和代码库。使用 GitHub 代码搜索查找真实代码示例，使用 Exa 获取官方文档和最新网络内容。当需要查找代码模式、API 文档、最佳实践或解决技术问题时使用此技能。

Research Skill

深度研究技术问题的专业技能工具集。

可用工具

---

searchGitHub

Search GitHub repositories for actual code patterns.

Basic Usage

CRITICAL: Search for literal code, not keywords.

✅ Good queries:

// Finding React useState patterns
query: "useState("
language: ["TypeScript", "TSX"]

// Finding error handling in async functions
query: "(?s)try {.*await"
useRegexp: true
language: ["TypeScript"]

// Finding specific API usage
query: "betterAuth({"
language: ["TypeScript"]

❌ Bad queries:

// These are KEYWORDS, not code patterns
query: "react hooks tutorial"  // Won't find code!
query: "best practices"        // Won't find code!
query: "how to use useState"   // Won't find code!

Advanced Usage

1. Regex for flexible patterns

{
  query: "(?s)useState\\(.*loading",  // useState with loading variable
  useRegexp: true,
  language: ["TSX"]
}

Note: Prefix with (?s) to match across multiple lines.

2. Filter by repository

{
  query: "getServerSession",
  repo: "vercel/",  // All Vercel repos
  language: ["TypeScript"]
}

3. Filter by file path

{
  query: "export default {",
  path: "next.config.js",  // Only in Next.js config files
}

4. Case sensitivity

{
  query: "CORS(",
  matchCase: true,  // Exact case match
  language: ["Python"]
}

Hidden Tricks

Trick 1: Find specific library usage in popular projects

{
  query: "figma: {",
  repo: "better-auth/",
  language: ["TypeScript"]
}

Trick 2: Search for error handling patterns

{
  query: "(?s)try {.*catch.*error",
  useRegexp: true,
  language: ["TypeScript"]
}

Trick 3: Find configuration patterns

{
  query: "export default {",
  path: "*.config.ts",
}

Trick 4: Find how others handle edge cases

{
  query: "if (!data)",
  language: ["TypeScript"],
  repo: "vercel/"  // Learn from production code
}

Watch Out For

⚠️ Pitfall 1: Using keywords instead of code

• Don't search for "react hooks best practices"
• Search for "useState(" instead

⚠️ Pitfall 2: Forgetting regex anchors

• Use (?s) for multiline matching
• Escape special chars: \\{ not {

⚠️ Pitfall 3: Too broad search

• "export" will return millions of results
• Be specific: "export const auth ="

⚠️ Pitfall 4: Ignoring language filter

• Always specify language to reduce noise
• Use ["TypeScript", "TSX"] for React projects

---

get_code_context_exa

Get official documentation and high-quality examples for libraries/APIs.

Basic Usage

✅ Good queries:

{
  query: "Next.js 15 server actions best practices",
  tokensNum: 5000
}

{
  query: "Better Auth Figma OAuth integration guide",
  tokensNum: 3000
}

{
  query: "Drizzle ORM PostgreSQL schema migration",
  tokensNum: 8000
}

Advanced Usage

1. Adjust token count based on depth

// Quick reference - 1000-3000 tokens
{ query: "React useEffect cleanup", tokensNum: 2000 }

// Comprehensive guide - 5000-10000 tokens
{ query: "Next.js 15 full authentication flow", tokensNum: 8000 }

// Deep dive - 15000-50000 tokens
{ query: "Stripe webhook handling complete guide", tokensNum: 20000 }

2. Include version/year for latest docs

{
  query: "Next.js 15 2025 app router",  // Specify year!
  tokensNum: 5000
}

3. Focus on specific aspects

// Not: "React hooks"
// Better: "React hooks performance optimization patterns"

{
  query: "React hooks performance optimization patterns",
  tokensNum: 5000
}

Hidden Tricks

Trick 1: Compare official docs across versions

// Search 1
{ query: "Next.js 14 server components", tokensNum: 3000 }
// Search 2
{ query: "Next.js 15 server components", tokensNum: 3000 }
// Compare the differences

Trick 2: Get migration guides

{
  query: "migrate from Next.js 14 to 15 breaking changes",
  tokensNum: 8000
}

Trick 3: Find official examples

{
  query: "Vercel Next.js 15 official examples repository",
  tokensNum: 5000
}

Watch Out For

⚠️ Pitfall 1: Too generic queries

• "React" → Too broad
• "React Server Components rendering patterns 2025" → Specific

⚠️ Pitfall 2: Wrong token count

• Too low (< 2000): May miss critical details
• Too high (> 20000): Wastes time and tokens
• Default 5000 is usually good

⚠️ Pitfall 3: Outdated docs

• ALWAYS include year: "2025" or "latest"
• Check account for "Today's date" in context

---

web_search_exa

Real-time web search for articles, blog posts, discussions.

Basic Usage

✅ Good use cases:

// Find recent articles
{
  query: "Next.js 15 performance improvements 2025",
  numResults: 8
}

// Find comparisons
{
  query: "Drizzle vs Prisma 2025 comparison",
  numResults: 5
}

// Find official announcements
{
  query: "Better Auth 2.0 release notes",
  numResults: 3
}

Advanced Usage

1. Adjust result count

// Quick scan - 3-5 results
{ query: "React 19 new features", numResults: 3 }

// Comprehensive - 8-10 results
{ query: "Next.js deployment best practices", numResults: 10 }

2. Search type

// Quick search
{ query: "...", type: "fast" }

// Deep research
{ query: "...", type: "deep" }

// Balanced (default)
{ query: "...", type: "auto" }

3. Livecrawl for latest content

{
  query: "Next.js 15 RC release notes",
  livecrawl: "preferred",  // Get latest content
  numResults: 3
}

Hidden Tricks

Trick 1: Find GitHub discussions

{
  query: "site:github.com better-auth figma OAuth issue",
  numResults: 5
}

Trick 2: Official docs only

{
  query: "site:nextjs.org server actions",
  numResults: 3
}

Trick 3: Find real-world war stories

{
  query: "debugging Next.js server actions production",
  type: "deep",
  numResults: 8
}

Watch Out For

⚠️ Pitfall 1: Generic queries

• "Next.js tutorial" → Low quality
• "Next.js 15 server actions error handling patterns" → High quality

⚠️ Pitfall 2: Too many results

• 10 results is usually noise

• 3-5 results is often enough

⚠️ Pitfall 3: Not using site: filter

• Use site: to focus on authoritative sources

---

Token/Result Guidelines

searchGitHub

• No limit on results, but first 10-20 are usually enough
• Use specific queries to reduce noise

get_code_context_exa

Use Case	Tokens
Quick API reference	1000-3000
Comprehensive guide	5000-10000
Deep documentation	15000-30000
Exhaustive research	30000-50000

web_search_exa

Use Case	Results
Quick fact check	1-3
Standard research	5-8
Comprehensive	8-10
Avoid > 10	Too noisy

---

Common Patterns

Pattern 1: "How do I use X?"

1. get_code_context_exa: Get official docs (5000 tokens)
2. searchGitHub: Find real usage (3-5 examples)
3. web_search_exa: Find best practices (5 results)

Pattern 2: "Why is X not working?"

1. web_search_exa: Search exact error (deep, 5 results)
2. searchGitHub: Find how others fixed it (regex search)
3. get_code_context_exa: Check official troubleshooting (3000 tokens)

Pattern 3: "X vs Y?"

1. web_search_exa: Find comparisons (8 results)
2. searchGitHub: Check real adoption (compare result counts)
3. get_code_context_exa: Read both official docs (5000 tokens each)

Pattern 4: "Latest version of X?"

1. get_code_context_exa: "X 2025 latest docs" (5000 tokens)
2. web_search_exa: "X release notes" (livecrawl: preferred, 3 results)
3. searchGitHub: Find recent usage patterns

---

Quick Troubleshooting

No results from searchGitHub?

• Check: Are you using code, not keywords?
• Try: Simplify your query
• Example: "useState(" not "react useState hook"

Outdated docs from get_code_context_exa?

• Check: Did you include year in query?
• Try: Add "2025" or "latest" to query

Too many irrelevant results from web_search_exa?

• Check: Is query too generic?
• Try: Use site: filter for authoritative sources
• Reduce: numResults to 3-5