🙌 DAG-Enabled Checklist Tool: Smarter, Faster Task Management

[arnm]

What is it?

This tool introduces DAG (Directed Acyclic Graph) checklists to CodeCompanion, letting you define tasks with explicit dependencies and enabling parallel execution of independent steps. It’s designed for complex coding workflows where simple linear checklists fall short.

GIST: https://gist.github.com/arnm/012bb0244e2d7731ac8db431ccfb3f65

codecompanion-checklist-dag.mp4

Main Benefits

• Keep Main Chat Context Small:
  Offload context-heavy or resource-intensive operations to dedicated task chats. These task chats have access to specifically defined tools. This keeps your main chat focused and responsive, while each task chat handles its own context and execution history.

• Parallelize Task Chats for Speed:
  Independent, read-only tasks are executed in parallel by autonomous LLM agents. This means you get results faster—analysis, code review, or research can all happen simultaneously, not sequentially. "write" or "writeread" tasks are not executed in parallel as tasks operate in brand new chats with no prior context other than what was provided in the task definition, so they are meant to be executed by main chat agent.

• Go Beyond Simple Checklists:
  Define explicit task dependencies using a DAG structure. This allows the agent to understand prerequisite relationships, optimize execution order, and avoid blocked or redundant work.

• Task Model/Adapter Selection:
  You can specify a different LLM adapter or model for the tasks. This enables you to leverage specialized models for code generation, analysis, or research as needed.

How does it work?

• Checklist Creation:
  Define a goal, tasks (with dependencies and access modes), and commit metadata. The tool validates the DAG and prepares an execution plan.

• Parallel Execution:
  Read-only, dependency-free tasks are dispatched to separate agent chats for autonomous completion.

• Dependency Management:
  The system tracks task status (pending, in_progress, completed, blocked) and automatically advances the workflow as dependencies are satisfied.

• Rich Output:
  Both agents and users receive structured data and formatted summaries, including dependency graphs and execution logs.

Example Use Cases

• Multi-step refactors with parallelizable analysis and code changes.
• Coordinating documentation, testing, and code updates with explicit dependencies.
• Managing onboarding or review checklists that require both automated and manual steps.
• Assigning specific tasks to different LLM models for optimal results.

[olimorris]

You are on 🔥 !!

When I merge #1665, your work with tooling will look very slick in the chat buffer!

[Davidyz]

This looks cool. I was experimenting with a very basic planning tool (purely prompt driven, no task management, etc.), but the results were... meh

Would you be interested in packaging this into an extension (or a PR)?

[jakeschurch]

@arnm just wanted to check - i didnt see storage.lua anywhere - could you direct me towards this?