19 lines
891 B
Markdown
19 lines
891 B
Markdown
# Subagent Deployment Rules
|
|
|
|
## When to Use Subagent vs. Main Agent
|
|
|
|
| Situation | Approach | Why |
|
|
|-----------|----------|-----|
|
|
| Reading/analyzing documents | Subagent | Keeps source content out of main context |
|
|
| Research and competitive analysis | Subagent | Heavy reading, return summary only |
|
|
| Writing deliverables | Main agent | Needs full decision-making context |
|
|
| Schema/architecture design | Main agent | Needs holistic project understanding |
|
|
| Code generation | Subagent | Isolated implementation, return result |
|
|
| Review and QA | Subagent | Fresh perspective, no bias from writing |
|
|
|
|
## Output Requirements
|
|
|
|
Subagent output must conform to the Output Contracts in `templates/claude-templates.md`. No free-form prose returns.
|
|
|
|
Optimal subagent return size: 1,000-2,000 tokens of structured summary. Longer returns consume main agent context without proportional benefit.
|