kienbui1995 Dispatching Parallel Agents
Overview
Delegate tasks to specialized agents with isolated context. Each agent gets precisely crafted instructions — never your session history.
Core principle: One agent per independent problem domain. Let them work concurrently.
When to Use
Use when:
- 2+ independent tasks (different subsystems, different bugs)
- No shared state between tasks
- Each can be understood without context from others
Don't use when:
- Failures are related (fix one might fix others)
- Need to understand full system state
- Agents would interfere with each other (same files)
The Pattern
1. Identify Independent Domains
Group by what's broken:
- Domain A: Authentication flow
- Domain B: Data processing pipeline
- Domain C: UI rendering
Each domain is independent — fixing auth doesn't affect rendering.
2. Create Focused Agent Tasks
Each agent gets:
- Specific problem — one domain only
- Relevant files — only what they need
- Expected outcome — what "fixed" looks like
- Constraints — don't touch other domains
3. Dispatch with Model Routing
| Task Type | Agent | Why |
|---|---|---|
| Bug investigation | debugger (Sonnet) | Systematic debugging, full tools |
| Code review | reviewer (Haiku) | Fast, cheap, read-only |
| Architecture question | architect (Opus) | Deep reasoning |
| Implementation | (default subagent) | Sonnet, balanced |
4. Collect and Integrate Results
After all agents complete:
- Review each agent's changes
- Check for conflicts
- Run full test suite
- Use magic-powers:verification-before-completion
Anti-Patterns
| Pattern | Problem |
|---|---|
| Dumping full session context | Agents get confused, waste tokens |
| Overlapping file scope | Merge conflicts |
| Too many agents at once | Hard to coordinate |
| Not reviewing results | Agents can make mistakes |
Integration
- magic-powers:subagent-driven-development — sequential task execution with review
- magic-powers:systematic-debugging — each agent uses this for bug investigation