planlisted

<!-- User-invoked, and in Claude Code it could not be otherwise: the name collides with the built-in /plan, so a Skill(plan) call comes back telling you to ask the human. Declaring it here makes the constraint visible in the file instead of discoverable by hitting it mid-task, and takes the description out of permanent context. Open this file and follow it. --> # Plan — before the first line of code > 🔴 **In Claude Code this file is read, not invoked.** `Skill(plan)` collides with the built-in > `/plan` and comes back asking you to have the user run that instead — so open this file and follow > it. The name is shared; the procedure below is the one that belongs to this repository. - **Can:** read anything, grep the blast radius, and ask the blocking questions — each with a default. - **Must:** declare assumptions that can be falsified, run the gates now rather than after coding, write the test plan before the code plan, and then STOP. - **Cannot:** write product code. `codereview` judges finished code. This skill is the **same discipline at the start**: finding out what the change really touches **before** a diff exists. A review finding costs one correction; a planning error costs the entire implementation. ## Golden rule — this skill does NOT write code It **reads, greps and writes a plan file**. No editing source, creating branches, committing or pushing. Code only starts with an **explicit OK from `{{OWNER}}`**. If during the survey you find the obvi