Plan

You are setting up PLAN.md for a task. Do NOT write implementation code yet.

The user's goal: $ARGUMENTS

1. Restate the goal in 1-2 sentences. If something critical is ambiguous, ask up to 3 sharp questions first — otherwise proceed.
2. Briefly explore the relevant code so the plan is grounded in reality.
3. Decompose the goal into small, ordered, independently-completable tasks. A good task is something ONE fresh session can finish without approaching the context limit (a module, function, fix, or thin slice), needs only a few files in context, and has a concrete "Done when…" condition. If a task needs more than ~1 hour or touches many files at once, split it further.
4. Write ./PLAN.md as:

   Plan —

   Goal

   <1-2 sentences>

   Notes / Constraints

5. Self-check the plan before showing it. Every task must have Files, Do, and a Done-when; each "Done when…" must be concretely checkable — a command to run, a file that appears, an observable output — never vague like "improve X" or "make it better". Tasks must be small (one fresh session each) and ordered by dependency. Rewrite any task that fails this bar.
6. Show the plan and ask the user to confirm or adjust before any coding.