| name | write-implementation-plan |
| description | Produce a plan an Executor who has no shared context can follow. Use at the end of every Planner session. |
Write Implementation Plan
A plan is not a description of what the code already does — it's a script an Executor can follow. If the Executor has to re-derive your reasoning to implement it, your plan failed.
Structure
- Goal restatement — one sentence, the success criterion.
- Current state — where the relevant code lives today, referenced by
path:line.
- Target state — what exists when done.
- Approach — the sequence of changes, step-by-step, in the order they should be made. Each step has a concrete output (a file edited, a test added, a command run).
- Trade-offs considered — what you ruled out and why. Name the alternatives explicitly; without this, future readers cannot tell if a constraint was explored.
- Risks — what could go wrong that the Executor should watch for.
Ground rules
- No hand-waving: if a step says "refactor", say which function, which module, to what shape.
- No invented filenames. If you propose a new file, name it.
- No features beyond the Issue. Listed scope creeps become their own Issue.
Required tail
End with ### Critical Files for Implementation — 3-5 bullet points with absolute or workspace-relative paths. These are the files the Executor will open first. If fewer than 3 or more than 5 feel right, re-scope the plan.