repo-diagramlisted

# repo-diagram Produces one self-contained HTML file: a zoomable map of a codebase, coloured by architectural layer, where every part says what it does, what it defines, and what it connects to. **Run this now, start to finish, without checking back.** The repository is whatever argument came with the invocation. If none came, ask for one, and stop; otherwise there is nothing to ask about. **You are the model in two of the pipeline's stages.** The driver runs everything deterministic and halts at those two gates with the prompt already written to disk. A halt is not a failure and not a question for the user: read the prompt, write the answer, validate, run the driver again. Both answers are cached per repository, so this only happens on a repo's first run. ## Do this 1. **Run the driver.** ```bash cd ~/.claude/skills/repo-diagram && ./run.sh <the repository you were given> ``` `--out FILE` chooses the output, `--force` re-fetches. Give the map a name from the repo, e.g. `--out ~/Desktop/<name>-map.html`. 2. **If it prints `NEEDS_GROUPING`, answer it and go back to step 1.** See below. 3. **If it prints `NEEDS_DESCRIPTIONS`, answer it and go back to step 1.** See below. Do not skip this because the map "already rendered" — unanswered, every box is labelled with its folder, which is the difference between a map and a directory listing. 4. **Read the audit.** Eleven structural checks print at the end. All must pass. A failure is a real defect i