Pipeline

---
name: pipeline
description: >
  Execute sequential, stage-based assembly line tasks where the output of one stage
  becomes the input of the next stage. Useful for progressive refinement workflows
  like Research -> Draft -> Review -> Format.
  Not for parallel execution (use divide-conquer) or consensus (use consensus-delegation).
  Triggers: "파이프라인", "순차 처리", "단계별 처리", "pipeline", "sequential processing".
---

# Pipeline

The Pipeline pattern organizes tasks into sequential stages, where the output (artifact) of a preceding stage serves as the input instruction/context for the next stage.

## 5-Stage Workflow

```
Input ──► [ Stage A ] ──► (Artifacts) ──► [ Stage B ] ──► (Artifacts) ──► Output
            Worker A                       Worker B
```

1. **Definition**: Outline the stages, assign specific assistant configurations, and define the handover criteria.
2. **Setup**: Provide a shared directory path via `workspaceOverride` so all stages modify the same project directory.
3. **Execution**:
   - Before each stage, inspect `agent__listAgents(type="sessions")` for an Idle child with the same assistant ID and a compatible workspace. Reuse it with `agent__messageToSession`; set `reset=true` when the stage is a fresh assignment and the prior conversation/runtime state must be discarded.
   - Start a new stage with `agent__spawnSession(..., waitForResult=true)` when no suitable child exists, the stage needs a different role or workspace, or the pipeline requires another parallel capacity slot.
   - Collect the output path and stage summary upon completion.
4. **Handover**: Bind the output paths and preceding summaries into the task prompt of the next stage session.
5. **Final Format**: Aggregate the final stage's output and present it to the user.

## 🛠️ MCP Tools Guide

- **Session Reuse**: Reusing a reset session preserves workspace files but closes its browser session. Treat each stage as runtime-isolated even when the session identity is reused, and do not reuse a child whose workspace contract does not match the stage.
- **Sequential Assignments**: Use `agent__messageToSession` with `waitForResponse=true` for a reused child, `agent__spawnSession` with `waitForResult=true` for a new child, or `agent__checkSession(sessionId, wait=true)` to synchronize.
- **Context Filtering**: Avoid passing the entire conversation logs of previous stages to prevent context bloat. Only pass the **summarized markdown text and file paths**.

See [pipeline-specs.md](references/pipeline-specs.md) for details.

## References

- [Pipeline specs](references/pipeline-specs.md)
- [Handover templates](references/handover-templates.md)
- [Failure recovery](references/failure-recovery.md)