plan

# Plan ## Workflow 1. Read the source, repository instructions, and relevant code. 2. Stop and return to design if product or technical choices remain open. 3. Break the work into vertical, outcome-based tasks small enough for one agent run. Order them by dependency. Add milestones only when they create a useful delivery or review boundary. 4. Return the plan in chat by default. Create tracker tickets only when the user asks. Never write a plan document. 5. Stop after planning. Do not implement. Each task must stand alone: ```markdown ## <Task title> ### Outcome The working result. ### Context What a new agent needs with no session history. ### Constraints Decisions and behavior that must not change. ### Acceptance criteria - Testable condition. ### Checks Exact commands and any required manual verification. ### Out of scope Related work this task must not absorb. ``` ## Boundaries - Do not split by file or technical layer when one vertical slice can deliver working behavior. - Do not create scaffolding or cleanup tasks without a checked outcome. - Do not hide unresolved decisions inside implementation tickets.