research

/research -- Investigate a topic

The user wants to research a topic within the current sprint.

Arguments

$ARGUMENTS

Instructions

1. Identify the active sprint by checking for claims.json in the project root, or ask which sprint to target if none exists.

1. Plan research passes (2-4 passes depending on topic breadth):
2. Pass 1: Broad landscape -- what exists, key concepts, major players
3. Pass 2: Deep dive -- technical details, implementation specifics
4. Pass 3: Risk and trade-offs -- failure modes, limitations, alternatives
5. Pass 4 (if needed): Synthesis -- cross-cutting insights, recommendations

1. For each pass:
2. Use WebSearch and WebFetch for general research
3. When the user references a GitHub repo or the current repo is on GitHub, use DeepWiki tools to gather codebase context
4. For private repos, use Read/Grep/Glob to analyze code directly
5. Add 3-5 claims per pass via wheat_add-claim:
6. Use r### ID prefix
7. Set appropriate evidence tier (web, documented, tested)
8. Include source URLs in source.artifact when available
9. Mix claim types: factual, estimate, risk, recommendation
10. Run wheat_compile after each pass
11. Announce progress: "Pass 2/3 complete: 11 claims across 4 topics."

1. Completion check: After all passes, run wheat_status.
2. If total_claims >= 8 and the user's original message included words like "write up", "summarize", "brief", or "report", immediately run the /brief workflow.
3. Otherwise, suggest next steps.

1. Print summary:

``` Research complete: <pass_count> passes, <claim_count> claims across <topic_count> topics.

Next steps: /brief -- generate a compiled brief /challenge r003 -- stress-test a specific finding /witness r005 <url> -- corroborate with external source ```