Skip to content

DAG Planner & Parallel Execution ​

Complex engineering tasks require multi-step planning and topological execution.


Directed Acyclic Graph (DAG) Modeling ​

When you run /plan <task>, the agent decomposes the task into steps with declared dependencies:

json
[
  { "id": "1", "title": "Setup test database schema", "dependencies": [] },
  { "id": "2", "title": "Implement authentication controller", "dependencies": ["1"] },
  { "id": "3", "title": "Implement JWT middleware", "dependencies": ["1"] },
  { "id": "4", "title": "Write end-to-end integration tests", "dependencies": ["2", "3"] }
]

Parallel Execution ​

Independent steps (like Step 2 and Step 3 above) execute concurrently via Promise.all, significantly reducing total mission duration.


Loop Detection ​

The loop detector tracks repeated tool invocations and prevents hallucinated infinite loops, forcing a replan or user escalation if the agent gets stuck.

Released under the MIT License.