Shipping Checklist¶
- [ ] Success, failure, and stopping conditions are explicit.
- [ ] Tools and side effects are permissioned at the narrowest useful scope.
- [ ] Budgets, retries, checkpoints, and human escalation are enforced.
- [ ] State changes and decisions are traceable.
- [ ] Representative trajectories pass versioned evaluation gates.
- [ ] Cancellation, rollback, and incident ownership are tested.