search
Installation
SKILL.md
/search — Iterative Multi-Tool Research
Research the user's question by orchestrating every available search tool — WebSearch, WebFetch, and installed MCPs (Exa, Perplexity, Tavily, Context7, DeepWiki) — iterating with different tool families until the answer is satisfactory or three passes are exhausted.
Workflow
Step 1 — Classify the question
Pick the best first tool based on question type. Use the table below; if the chosen MCP is not installed, fall back to the next row.
| Question type | First tool | Why |
|---|---|---|
| Named library / framework / SDK API | mcp__context7__resolve-library-id → mcp__context7__query-docs |
Authoritative, version-correct |
| Specific GitHub repo internals | mcp__deepwiki__ask_question |
Indexed wiki + code |
| Known URL to read | WebFetch (or mcp__exa__web_fetch_exa) |
Direct content |
| Comparison / "which is better" / synthesis | mcp__perplexity__reason |
Built for reasoning across sources |
| Deep multi-source research | mcp__perplexity__deep_research or mcp__tavily__tavily_research |
Long-form, multi-citation |
| Recent events / news | WebSearch + mcp__tavily__tavily_search |
Fresh index |
| General factual lookup | WebSearch |
Fast default |