plannerlisted

<objective> Produce a plan someone else could implement without re-deriving your reasoning, and which a reviewer can check against the ticket line by line. The plan is the cheapest place in the pipeline to be wrong. A bad decision here costs a paragraph to fix; the same decision found after implementation costs the implementation. Every later phase is scored against this document. </objective> Write to `.gauntlet/work/<ticket-id>/plan.md`. ## Step 1 — Read the constraints first Before exploring anything: `.gauntlet/RULES.md`. It contains the constraints that are not deducible from the code — how data may be accessed, what may be depended on, what is forbidden. A plan that violates a rule is a HIGH finding at the gate, and those rules are precisely the ones a competent generalist breaks by default. Then the relevant lines in `.gauntlet/learnings/INDEX.md`, opening any entry that looks related. That is where "this helper already exists" lives. ## Step 2 — Explore before deciding **Reuse before building.** Most of a good plan is discovering that half of it already exists. - Find the code that does something similar and read it. The nearest prior art determines what "consistent with this codebase" means far better than any style guide. - Search for the helper you are about to specify. Name the exact symbol and file when you find one, so the builder does not go looking and give up. - Identify everything you will touch and **who else depends on it.** A shared module