regenerate-claude-mdlisted

# /regenerate-claude-md — propose a project-shaped global persona > **Only the user runs this.** It happens when they type `/regenerate-claude-md`, and at no > other time. If you are reading this file because *you* decided to — to see what > it does, or to carry out its steps yourself — then stop and do neither. > Following these instructions **is** running the skill, whichever route you took > to them. Say the skill exists and let the user invoke it. Every env starts on one of two day-one defaults — `coder` for a coding project, `none` for anything that isn't one. Both are cookie cutters: correct on day one and steadily less so. The project grows, working agreements accumulate, the same mistakes stop being made for reasons nobody wrote down. This skill reads what the project has become and proposes what its agent's **global** persona should say now. Its output is a *proposal*. When the wording is finally right — usually after several rounds, and after the user has consulted the target env itself — they run **`/regenerate-claude-md-accept`**, which is what actually installs it and flips the blueprint to `custom`. You have **no `Write` tool here, deliberately.** Your output is a proposal the user carries to that project's env and discusses there. Do not write, edit, or offer to write any file. Print the proposed persona and stop. ## What "global" means here, and the test that keeps you honest Two layers, and this skill only touches the first: | | Global persona | Projec