web_search

🔥 WEB SEARCH - MINIMUM 3 KEYWORDS, RECOMMENDED 5-7

This tool searches up to 100 keywords IN PARALLEL via Google. Using 1-2 keywords = wasting the tool's parallel search power!

Results Budget: 10 results per keyword, all searches run in parallel.

• 3 keywords = 30 results (minimum)

• 7 keywords = 70 results (RECOMMENDED)

• 100 keywords = 1000 results (comprehensive)

7-Perspective Keyword Formula - Each keyword targets a DIFFERENT angle:

1. Direct/Broad: "[topic]" Example: "React state management"

2. Specific/Technical: "[topic] [technical term]" Example: "React useReducer vs Redux"

3. Problem-Focused: "[topic] issues/debugging/problems" Example: "React state management performance issues"

4. Best Practices: "[topic] best practices [year]" Example: "React state management best practices 2024"

5. Comparison: "[A] vs [B]" Example: "React state management libraries comparison"

6. Tutorial/Guide: "[topic] tutorial/guide" Example: "React state management tutorial"

7. Advanced: "[topic] patterns/architecture large applications" Example: "React state management patterns large applications"

Search Operators with Examples:

• site:domain.com - Search within specific site Example: "React hooks" site:github.com → React hooks repos on GitHub

• "exact phrase" - Match exact phrase Example: "Docker OOM" site:stackoverflow.com → exact error discussions

• -exclude - Exclude term from results Example: React state management -Redux → find alternatives to Redux

• filetype:pdf - Find specific file types Example: React tutorial filetype:pdf → downloadable guides

• OR - Match either term Example: React OR Vue state management → compare frameworks

Keyword Patterns by Use Case:

Technology Research: ["PostgreSQL vs MySQL performance 2024", "PostgreSQL best practices production", "\"PostgreSQL\" site:github.com stars:>1000", "PostgreSQL connection pooling", "PostgreSQL vs MongoDB use cases"]

Problem Solving: ["Docker container memory leak debugging", "Docker memory limit not working", "\"Docker OOM\" site:stackoverflow.com", "Docker memory optimization best practices"]

Comparison Research: ["Next.js vs Remix performance", "Next.js 14 vs Remix 2024", "\"Next.js\" OR \"Remix\" benchmarks", "Next.js vs Remix developer experience"]

Example: ❌ BAD: {"keywords": ["React"]} → 1 vague keyword, no operators, no diversity

✅ GOOD: {"keywords": ["React state management best practices", "React useReducer vs Redux 2024", "React Context API performance", "Zustand React state library", "\"React state\" site:github.com", "React state management large applications", "React global state alternatives -Redux"]} → 7 diverse angles with operators

Pro Tips:

1. Use 5-7 keywords minimum - Each reveals different perspective

2. Add year numbers - "2024", "2025" for recent content

3. Use search operators - site:, "exact", -exclude, filetype:

4. Vary specificity - Mix broad + specific keywords

5. Include comparisons - "vs", "versus", "compared to", "OR"

6. Target sources - site:github.com, site:stackoverflow.com

7. Add context - "best practices", "tutorial", "production", "performance"

8. Think parallel - Each keyword searches independently

Workflow: web_search → sequentialthinking (evaluate which URLs look promising) → scrape_links (MUST scrape promising URLs - that's where real content is!) → sequentialthinking (evaluate scraped content) → OPTIONAL: web_search again if gaps found → synthesize

Why this workflow works:

• Search results reveal new keywords you didn't think of

• Scraped content shows what's actually useful vs what looked good

• Thinking between tool calls prevents tunnel vision

• Iterative refinement = comprehensive coverage

CRITICAL:

• ALWAYS scrape after web_search - that's where the real content is!

• ALWAYS think between tool calls - evaluate and refine!

• DON'T stop after one search - iterate based on learnings!

FOLLOW-UP: Use scrape_links to extract full content from promising URLs!