Research

When NOT to use

• The user wants a quick opinion or guess — research means delegation to a background agent with source verification.
• The question is about code already in the workspace that can be answered by reading the source directly.

Workflow

Default: Use Task (general) when ≥2 independent sources; otherwise research inline via Read/Grep. Always save to scratch/research/<slug>.md (or docs/research/ if repo convention exists). OpenCode/Copilot runners differ — don't hard-require Task when it is unavailable.

Spin up the researcher to do the work, so you keep working while it reads.

Its job:

1. Investigate the question against primary sources — official docs, source code, specs, first-party APIs — not a secondary write-up of them. Follow every claim back to the source that owns it.
2. Write the findings to a single Markdown file, citing each claim's source.
3. Save it where the repo already keeps such notes; match the existing convention, and if there is none, put it somewhere sensible and say where.

Source trust

Primary (owns claim): official docs, source code, RFC/spec. Secondary: blog summarizing primary — follow link to primary. Tertiary (avoid): LLM summary, forum guess.

Completion criteria

• Question investigated against primary sources
• Each claim cites its source
• Findings saved to a Markdown file in the repo
• User informed of the file location

Related skills

• brainstorming — ideation before investigation.
• ml-paper-writing — academic source handling.
• domain-modeling — keep language consistent.