| name | web-researcher |
| description | Conduct thorough web-based research using fetch_content and web_search tools with delegation. Use when you need external info not in the local codebase -- docs, API references, current practices, error resolution. |
Web Research
When to Skip
- Local codebase answers --
grep, rg, ast_query first
- Well-known language features or stdlib
- Context already provided by the user or gathered this session
Tool Selection
fetch_content is the primary research tool. Use it when you know or can construct the URL:
- Official docs:
https://docs.example.com/api/...
- GitHub raw content:
https://raw.githubusercontent.com/...
- Package registries, changelogs, API references
web_search is for discovery only. Use it when you genuinely do not know where to look:
- Unfamiliar library or tool with no known docs URL
- Error messages with no obvious cause
- Comparing options you have not encountered before
Quick Research
- Construct the URL if possible →
fetch_content directly
- Cannot construct URL → single
web_search → fetch_content top 1-2 results
- Summarize with source URLs
Thorough Research
For complex questions, comparisons, or critical decisions.
- Formulate 2-4 distinct queries targeting different angles (see
references/query-tips.md)
- Spawn one minion per query following the
delegation skill pattern:
Search for "{query}". Fetch top 2-3 results. Return: summary + source URLs.
- Cross-reference findings across minions, flag contradictions, produce consolidated summary with citations
Rules
- ALWAYS cite source URLs for every claim from web research
- ALWAYS cross-reference multiple sources for critical decisions
- NEVER dump full pages into context -- extract relevant sections only
- Prefer official docs and primary sources over blog posts
- Reformulate queries at least once before giving up on no results
