mia-cx Refine plan
A plan is a step document executed by an agent that was not in the planning conversation. Apply agent-copy; Read it first if not already loaded. Checks specific to plans:
- Self-contained. The executor has none of the planning context: every fact the steps rely on is in the plan or behind an explicit pointer. Read it as the executor; anything you would have to guess is a gap.
- Observable done-conditions. Files changed, tests green, behavior demonstrated. Per step, plus acceptance criteria closing the whole plan.
- Constraints live inside the step they bind. A global "notes" section scatters meaning the executor will not carry to step 7.
- Decisions, not directions. Record what was decided and why. Code-level detail (paths, snippets) goes stale; include a snippet only when it encodes a decision more precisely than prose (schema, state machine, type shape).
- Bound the legwork. An explicit out-of-scope section stops the executor doing more than asked.
- Split across issues or handoffs when later steps tempt rushing earlier ones; those are the real context boundaries a plan can use.