Research

Conduct structured multi-source research and deliver actionable findings.

Steps

1. Classify the Subtype

Identify which subtype applies:

• technical — How to implement something. Focus: APIs, patterns, constraints, examples.
• options — Compare alternatives. Focus: trade-offs, scoring, recommendation.
• feasibility — Can we do this? Focus: blockers, risks, prerequisites, viability verdict.

If unclear, infer from the request. State the subtype before proceeding.

2. Define Scope

State in one sentence what you are researching and what is out of scope.

3. Gather from All Applicable Sources

Run searches in parallel across every relevant source:

• Web search (WebSearch / mcp__claude_ai_exa__web_search_exa) — current docs, articles, changelogs
• Code search (mcp__claude_ai_exa__get_code_context_exa) — real usage examples, library internals
• Official docs (mcp__claude_ai_Context7__resolve-library-id + query-docs) — authoritative API reference
• Package registries — npm, PyPI, crates.io; check download counts, last publish, open issues
• GitHub (gh search repos, gh search code) — battle-tested implementations, community solutions

Skip sources that are clearly irrelevant (e.g. package registries for a pure architectural question).

4. Synthesize Findings

For technical:

• How it works (mechanism, key concepts)
• Implementation steps or pattern
• Known constraints, gotchas, version requirements
• Best available library or built-in approach

For options:

• List each option with: description, pros, cons, maturity, license
• Score each on the criteria that matter for the request (performance, DX, size, etc.)
• Clear recommendation with rationale

For feasibility:

• Viability verdict: FEASIBLE / FEASIBLE WITH CAVEATS / NOT FEASIBLE
• Prerequisites and blockers
• Estimated effort (HIGH / MEDIUM / LOW)
• Risks and mitigations
• Recommended path forward

5. Output Findings

Present results in this structure:

## Research: <topic>
Subtype: technical | options | feasibility

### Summary
<2-4 sentence executive summary with the key takeaway>

### Findings
<subtype-specific content from step 4>

### Sources
- <Title or description>: <URL>
- ...

### Recommendation
<Single clear recommendation or verdict. What should be done next.>

### Open Questions
<Anything that requires user input or further investigation to resolve. Omit if none.>

Notes

• Cite every claim that comes from an external source.
• Distinguish verified facts from inferences — label inferences explicitly.
• If sources conflict, surface the conflict and explain which to trust and why.
• Do not produce implementation code unless the user asks to proceed after reviewing findings.