Background Agent

Pattern

Delegate standalone tasks to background agents running in parallel while you stay focused on main work.

Workflow:

1. Collect: During main work, capture todos/ideas in markdown file (GTD-style)
2. Identify: Find tasks that are standalone, well-sized, clear enough to delegate
3. Spawn: Launch background agent per task (separate branch/context)
4. Continue: Stay focused on main thread while agents work in parallel
5. Integrate: Review and merge results when ready

What makes a task ready for background agent:

• Standalone: doesn't block your current work
• Well-sized: clear scope, not too big
• Clear enough: agent can execute without constant guidance

Human stays on main thread. Agent works asynchronously.

Example

Main work: Refactoring authentication flow

While working, you notice:

• Tests are missing for edge cases
• Documentation needs updating
• Dead code should be removed
• Logging format is inconsistent

Collect these in todo.md. Each is standalone and clear.

Spawn background agents:

• Agent 1: Add edge case tests (branch: add-edge-tests)
• Agent 2: Update auth docs (branch: update-auth-docs)
• Agent 3: Remove dead code (branch: cleanup-dead-code)

You continue refactoring. Agents work in parallel on their branches.

Tools that enable this:

• Claude Code GitHub app: runs agents in GitHub Actions, creates branches
• Multiple chat windows/sessions
• Separate working directories (git worktrees)