scaffoldlisted

# code-idea Turns a plan into the actual files a coding agent needs to work correctly — not a lecture on best practices, the files themselves. ## When this runs - Right after a planning/design conversation, before the user moves to Claude Code (or another agent) to start building - When the user pastes or describes a new idea and wants it scaffolded from scratch - When updating or extending an existing project's agent docs as the plan evolves ## Core philosophy — bake this into every output - The root context file stays SHORT — start near 30 lines, treat ~150 as a hard ceiling, move anything longer into a linked doc. Which file that is depends on the target agent (Step 3); how short it should be does not. - **Nothing is settled until the user says so.** A fact that entered the conversation because *you* proposed it, or because it showed up in a prototype/demo, is not confirmed — see Step 1's source tagging and Step 2's confirmation rule. - **"Not yet decided" is a valid answer and a real output.** A deferred choice becomes a dated pending-decision rule at the top of the context file plus a `Status: Pending` entry in `decisions.md` — never a guess written as settled, and never a hollow doc. - Only write rules that fix an actual observed or clearly anticipated mistake. No vague advice, no restating what a linter/formatter already enforces, no pasted-in documentation, no large code samples. - Put the rules most likely to be violated at the TOP of the file — agents can lose in