Room 02 · Planning
The Planning Agent makes intent executable.
It grounds a note in the real tool and source, resolves material ambiguity, and produces work another agent can carry out and verify.
Input
The planner receives a Yesod note and the context attached to it: the registered tool, its source repository, prior discussion, constraints, and any dependencies already known.
Responsibilities
The planner inspects the actual source before prescribing changes and asks follow-up questions when an unresolved choice would materially change the result. It should:
- Restate the intended outcome in concrete terms.
- Resolve ambiguity that would materially change the implementation.
- Define acceptance criteria that can be checked later.
- Split the work into one root Bead and bounded child Beads.
- Express ordering within the feature and dependencies on other feature notes.
A useful plan identifies the files, systems, or interfaces likely to change without pretending that exploration has already happened. It makes the important decisions visible and leaves implementation detail to the coder where appropriate.
Output
Planning is complete when another agent can understand what to build, why it is bounded this way, what must happen first, and how success will be recognized. The note, its build plan, and its later complexity estimate form the grounded work the dispatcher can evaluate.
Authority boundary
The Planning Agent may shape and clarify work. It does not silently dispatch that work. A planned note remains ready but unarmed until a human or an explicitly authorized automatic flow moves it forward.
Related: Beads and Complexity Estimation.