Was macht research?
Use this skill when the next step depends on something not yet known. Research answers questions; it does not implement.
This is the canonical research lane. deep-dive, sciomc, and autoresearch
route here.
Goal
Replace assumptions with evidence, and say plainly what remains uncertain.
Workflow
- State the question precisely enough to know when it is answered.
- Search the repo and its docs first — local evidence outranks recollection.
- For external SDKs, frameworks, or APIs, consult official documentation.
- Sweep more than one way when the answer could hide: by file, by symbol, by caller, by history.
- Synthesize into findings, each tied to where it came from.
Scale
- Narrow lookup — answer it directly.
- Multiple independent questions — investigate in parallel.
- Unknown-size discovery — keep going until additional passes surface nothing new.
Rules
- Cite the source: file and line, or the document consulted.
- Distinguish what was verified from what was inferred.
- Report contradicting evidence rather than picking the tidier story.
- Do not implement as a side effect of researching.
Output
- The question
- Findings, each with its source
- What remains unknown or unverifiable
- Recommended next step, if one follows