research

name	research
description	Research a topic in parallel using multiple subagents with different approaches

Parallel Research

Launch multiple subagents in parallel to research a topic from different angles, then synthesize the results.

How to use

The user will provide a research question or topic. Parse it and decide on 2-4 distinct research angles.

Steps

1. Identify angles: Break the topic into 2-4 independent research approaches. Examples:

   • Different search queries or keywords
   • Codebase exploration vs. web search vs. documentation lookup
   • Different files, modules, or layers of the stack
   • Competing libraries, patterns, or solutions

2. Launch agents in parallel: Use the Agent tool to spawn all research agents in a single message (this is critical for true parallelism). Each agent should:

   • Have a clear, specific prompt describing what to find
   • Use subagent_type: "Explore" for codebase research, or subagent_type: "general-purpose" for web/docs research
   • Be named descriptively (e.g., "Research Redux approach", "Search codebase for auth patterns")

3. Synthesize: Once all agents return, combine their findings into a clear summary:

   • What each angle discovered
   • Where approaches agree or conflict
   • A recommendation or next steps if the user asked for one
   • Key file paths, URLs, or code snippets found

Rules

• Always launch agents in a single message for true parallelism — never sequentially.
• Keep each agent's prompt focused on one angle — don't give a single agent the whole question.
• Default to 3 agents unless the topic clearly needs fewer or more.
• Prefer model: "haiku" for simple search tasks to minimize cost. Use "sonnet" for tasks needing deeper analysis.
• Summarize the results yourself — don't ask the user to read raw agent output.