| name | research |
| description | Investigate a question against high-trust primary sources and capture the findings as a cited Markdown file. Use when the user wants a topic researched, docs or API facts gathered, or reading legwork delegated to a background sub-agent. |
Research — Background Documentation Investigation
Delegate investigation to a background sub-agent (task tool), so you keep working while it reads.
Process
1. Identify the research question
Clarify the scope with the user: what to investigate, which sources to prioritize (official docs, source code, specs, first-party APIs), and the expected output format.
2. Dispatch a background sub-agent
Use the task tool to spawn a sub-agent with this instruction:
Investigate [question] against primary sources only — official docs, source code, specs, first-party APIs — not secondary blog posts or tutorials. Follow every claim back to the source that owns it. Write findings to a single Markdown file, citing each claim's source inline. Save it where the repo keeps research notes; if no convention exists, create docs/research/<topic>.md.
3. Continue other work
While the sub-agent investigates, proceed with unrelated tasks. The research agent works independently.
4. Review results
When the sub-agent completes, verify:
- Every claim has a citation back to its primary source
- Findings are relevant and actionable
- File location follows repo conventions
Sources hierarchy (highest trust first)
- Official documentation — framework/library docs, language specs, RFCs
- Source code — the actual implementation, type definitions, tests
- First-party APIs — SDK references, API schemas, changelogs
- Specifications — W3C, ECMA, IETF standards
Anti-pattern: Blindly trusting Stack Overflow, Medium posts, or LLM-generated summaries as primary sources. These are secondary at best.
Output format
# Research: [Topic]
> Investigated by sub-agent on [date]
## Key findings
- [Finding 1] — Source: [link/commit/line]
- [Finding 2] — Source: [link/commit/line]
## Detailed findings
### [Subtopic 1]
[Explanation with inline citations]
### [Subtopic 2]
[Explanation with inline citations]
## Sources
1. [Source name](url) — relevance note
2. [Source name](url) — relevance note
## Recommendations
[Actionable recommendations based on findings]
When to use
- "How does [library X] handle [feature Y] in version Z?"
- "What are the breaking changes in [framework] v[N]?"
- "Investigate the performance characteristics of [approach A] vs [approach B]"
- "Find the official recommended pattern for [scenario]"
When NOT to use
- The answer is trivially available (reading one known file)
- The user needs an immediate answer (sub-agent takes time)
- The question is about project-internal logic (use
grep/read directly)