| name | delegation |
| description | Strategy for delegating work to minions for parallel execution and context isolation. Use at the start of every session and before any multi-step task. |
Delegation
When to Spawn
Use spawn for most delegation — including parallel execution. Multiple spawn calls in a single response run concurrently and return together:
- 2+ independent tasks with no data dependencies
- Subtask involves >10 tool calls (broad search, reading many files, exploration)
- Complex subtask whose output feeds into your next step
- Context isolation for focused work (deep debugging, single-module analysis)
Use spawn_bg only for long-running work that you do not need immediately:
- Lengthy research or documentation gathering while you continue other work
- Running full test suites or builds while continuing implementation
- Fire-and-forget tasks where results are consumed later
When NOT to Spawn
- Single command, single edit, or single file read
- You need the output in the same thought to make a decision
- User is interactively iterating and wants tight feedback loops
- Overhead exceeds benefit (trivial tasks)
Common Patterns
- Plan execution: one minion per independent task, verify integration after all complete
- Code review: parallel minions per file or module, aggregate findings
- Test + continue:
spawn_bg for tests, main context continues working
- Research + implement:
spawn_bg for research, main context scaffolds, incorporate findings when ready
- Multi-file edits: group files into batches, one minion per batch
Task Descriptions
Every minion task MUST include:
- WHAT to do — specific and concrete
- WHERE — file paths, directories, URLs
- HOW to verify — test commands, expected outputs
- WHAT to return — summary of results, not raw output