| name | long-running-goal |
| description | Run durable multi-phase work without relying on one context window. |
| trigger-condition | Use for large tasks that need research, plan, execution, verification, review, and resumability. |
| allowed-tools | ["read_file","apply_patch","bash","spawn_agent"] |
| required-context | ["goal","durable state","progress log","failure log","verification policy","worktree status"] |
| examples | ["resume goal workflow","coordinate agent team","recover after failed verification"] |
| verification-checklist | ["state updated","artifacts written","failures captured","reviewer or verifier separated from worker"] |
| progressive-disclosure | Always read the goal state.json and recent failure view from Wuu-managed goal storage before loading detailed artifacts. |
Long Running Workflow
Do not rely on model memory for long tasks.
- Initialize durable state.
- Run phases in order: research, plan, approval when needed, execution, verification, review, integration, summary.
- Write an artifact for each phase.
- Append progress, decisions, events, and failures after each agent round.
- Require reviewer or verifier evidence before final completion.
If blocked, persist the blocker and the next useful action.