Files
nixos-configs/home/roles/development/skills/micro/research-agents.md
John Ogle 7903b2dfd0
All checks were successful
CI / check (push) Successful in 3m16s
refactor(dev): restructure Claude commands/skills directories
Correct terminology mismatch:
- Rename skills/ to commands/ (these are user-invokable commands)
- Create new skills/ for reference materials
- Move bd_workflow.md to skills/ (it's reference material)
- Add micro-skills and formulas directories
- Update default.nix to install both commands and skills

Commands → ~/.claude/commands/ (invokable as /command-name)
Skills → ~/.claude/commands/skills/ (reference materials)
Formulas → ~/.beads/formulas/ (workflow templates)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-15 17:16:55 -08:00

1.8 KiB

description
description
How to spawn and coordinate research sub-agents

Research Agents

Use parallel sub-agents for efficient codebase research.

Available Agents

Agent Purpose
codebase-locator Find WHERE files and components live
codebase-analyzer Understand HOW specific code works
codebase-pattern-finder Find examples of existing patterns
thoughts-locator Discover relevant documents in thoughts/

Spawning Protocol

  1. Decompose - Break the research question into 3-5 specific questions
  2. Spawn parallel - Use one Task call with multiple agents
  3. Be specific - Include directories and file patterns in prompts
  4. Wait for all - Do not synthesize until ALL agents complete
  5. Synthesize - Combine findings into coherent summary with file:line references

Example

Task(codebase-locator, "Find all files related to authentication in src/")
Task(codebase-analyzer, "Explain how JWT tokens are validated in src/auth/")
Task(codebase-pattern-finder, "Find examples of middleware patterns in src/")
Task(thoughts-locator, "Find documents about auth design decisions in thoughts/")

Key Principles

  • Parallel when different - Run agents in parallel when searching for different things
  • WHAT not HOW - Each agent knows its job; tell it what you need, not how to search
  • Document, don't evaluate - Agents should describe what exists, not critique it
  • Specific directories - Always scope searches to relevant directories
  • File references - Include specific file:line references in synthesis

Agent Prompts

When spawning agents, include:

  • The specific question or goal
  • Relevant directories to search
  • Reminder to document (not evaluate) what they find
  • Request for file:line references in findings