Luna Executor

Use Luna proactively for one bounded execution assignment. The parent Agent keeps decisions, supervision, verification, and final delivery.

Workflow

Choose the assignment. Finish product, architecture, authorization, and scope decisions first. When the remaining work contains a meaningful execution unit whose boundary is clear, can be executed independently, and has a result that can be checked independently, delegate it to Luna. Eligible work includes, but is not limited to, read-only exploration and evidence gathering; log, configuration, data, or repository-state analysis; command or script execution; build, lint, test, or verification runs; and batch editing, refactoring, or implementation. Keep tiny work or work that still depends on parent judgment, authorization, or tightly coupled context with the parent.
- Completion criterion: one decided assignment and its acceptance signals are explicit, or the parent has established that no unit meets the delegation boundary.
Write the task contract and delegate. Give Luna a self-contained message containing the objective, owned scope and files, relevant facts and any useful context pointers, constraints, acceptance criteria, verification expectations, and required return format. For a long assignment, name its first observable milestone. Spawn exactly:
```
spawn_agent({
  task_name: <bounded task name>,
  agent_type: "luna_worker",
  fork_turns: "none",
  message: <self-contained task contract>
})
```