Module 6: Sub-AgentsLesson 4 of 7

Build: Researcher Agent

Build: Researcher Agent

Let's create your first sub-agent: a researcher.

The Researcher's Job

  • Receives a research question
  • Searches the web
  • Reads relevant pages
  • Synthesizes findings
  • Returns structured results

Step 1: Define the Task Template

When spawning a researcher, use this pattern:

## Research Task **Question:** [What you want to know] **Scope:** - [Specific aspects to cover] - [Sources to prioritize] - [Things to ignore] **Output Format:** - Executive summary (2-3 sentences) - Key findings (bullet points) - Sources (with URLs) - Confidence level (high/medium/low) **Constraints:** - Time limit: [X minutes] - Focus on recent information - Cite sources

Step 2: Spawn the Researcher

sessions_spawn({ task: ` Research: What are the top 5 competitors to Duolingo? Scope: - Language learning apps only - Focus on mobile apps - Include pricing info Output Format: - Executive summary - Top 5 with brief descriptions - Pricing comparison table - Sources Constraints: - Complete within 5 minutes - Focus on 2024-2025 data `, model: "sonnet", timeoutSeconds: 300 });

Step 3: Handle Results

The researcher will deliver results to your session (or channel if specified).

Exercise: Your First Research Task

  1. Think of something you want researched
  2. Write the task using the template above
  3. Spawn the researcher
  4. Evaluate the results

Good research questions:

  • "What are the top 5 X in category Y?"
  • "How does company A compare to company B?"
  • "What's the current state of X technology?"

Bad research questions:

  • "Tell me everything about X" (too broad)
  • "What's 2+2?" (not research)
  • "How do I feel about X?" (not factual)

Validation

Your researcher is working when:

  • Task spawns successfully
  • Results return within timeout
  • Output matches requested format
  • Sources are cited
  • Quality is acceptable