mermaid-to-imageslisted
Install: claude install-skill hcussi/claude-code-toolkit
# mermaid-to-images
> **Non-negotiable rules. Read these first; they override any instinct to be clever.**
>
> 1. **The bundled script is the ONLY renderer.** Every image is produced by
> running `scripts/mermaid_to_images.py`. You MUST NOT render diagrams any
> other way. Specifically forbidden: a headless browser or Playwright
> screenshot, a local HTTP server (`http.server`), converting SVG to PNG
> yourself, `@1x`/`@2x` intermediates, or drawing/editing an image by hand.
> 2. **If the script cannot render, STOP and report to the user.** A failure
> (no `mmdc`, no network to mermaid.ink) is a hard stop, not a problem to route
> around. Relay the script's error and its remediation; do not invent a
> fallback renderer. Producing the image by other means is a failure, not a
> save.
> 3. **Do not choose output names, formats, or folders.** The output is exactly
> `diagram-N.png` inside `<md-stem>-diagrams/`. Never rename to something
> "semantic" (e.g. `assets/big-picture.png`), never emit `.svg`, never save
> `.mmd` source, never write a `mermaid-config.json`.
> 4. **Your only edits to the doc** are: (a) turning an ASCII diagram fence into a
> ` ```mermaid ` fence (with a `%% alt:` comment), and (b) running the script,
> which does the fence-to-image rewrite. Nothing else.
>
> If any rule pushes you toward more work, you are misreading it: the goal is a
> boring, identical result every run. When blocked, stopping is the correct
> outcom