kimtth Goal: coordinate focused subagents so each does one job well.
Use for:
- tasks with separable concerns (research, implement, test, review)
- work too large for a single context to hold cleanly
- parallelizable streams that need a coordinator
Workflow:
- Decompose the task into independent, well-scoped units.
- For each unit, write a self-contained brief and success criteria.
- State exactly what each subagent must return.
- Dispatch read-only or independent units in parallel.
- Integrate results; resolve conflicts and gaps yourself.
- Verify the combined outcome against the original goal.
Briefs include:
- objective and constraints
- inputs and relevant files
- required output format
- what is out of scope
Rules:
- one clear responsibility per subagent
- never assume a subagent shares your context; spell it out
- keep write operations on one unit to avoid collisions
- the orchestrator owns final verification