Research — Parallel tech investigation

Pre-condition

Read .claude/state/CONTEXT.md. If it does not exist, tell the user to run /bayes-ship:ship-discover first.

Process

1. Analyze CONTEXT.md to identify 3-5 research areas. Examples:
   • "Best auth library for Next.js with M-Pesa integration"
   • "Card component patterns for invoice UI"
   • "Payment reconciliation: callback handling, retry logic"
   • "LangGraph vs CrewAI for agent orchestration"
2. Announce the research areas to the user: "I've identified N research areas. Spawning parallel researchers. This takes 2-3 minutes."
3. For each research area, dispatch a fresh subagent using the Task tool with:
   • The research question as the primary instruction
   • Relevant context from CONTEXT.md (only the parts related to this area)
   • Instruction to write findings to .claude/state/research/{area-slug}.md
4. Wait for all researchers to complete.
5. Dispatch a Synthesizer subagent that:
   • Reads all files in .claude/state/research/
   • Produces .claude/state/RESEARCH.md with:
     • Recommendations per area (what to use and why)
     • Trade-offs considered
     • Selected approaches
     • Links to relevant documentation
6. Update progress.json: set phases.research.status to "complete".
7. Present key recommendations to the user as a brief summary.
8. Output: "Ready. Run /bayes-ship:ship-spec to generate the PRD."

Rules

• This phase is fully autonomous. Do not ask the user for input during research.
• Each researcher subagent gets a clean context. Do not pass the full CONTEXT.md to every researcher — only the relevant section.
• If a research area yields no useful results, note this in RESEARCH.md rather than guessing.