playgroundlisted
Install: claude install-skill mayankmankhand/llm-peer-review
# Playground
**Use this when:** The user is making a decision that benefits from seeing options side by side, reordering items, toggling between variants, or tuning a value, and a flat markdown comparison would be hard to scan or interact with.
**Don't use this when:** A short markdown list or table is enough (most simple "pick one of two" decisions), the output needs to live in the repo (use markdown), or the user wants a published artifact (use `/codebase-to-course` for learning content, or a regular HTML file in `artifacts/html/` for cycle-bound documents).
## Hard Rules
<rules>
1. **Never modify any toolkit file.** No edits to commands, skills, rules, settings, or any prompt file. The export-loop button is the only way state flows back to the conversation.
2. **Static-default invocation.** Generate a static side-by-side comparison unless interactivity (drag, toggle, slider) genuinely changes the decision. If unsure, ask the user first.
3. **Disposable.** Always write to `/tmp/playground-{timestamp}.html`. Never to the repo. The `/tmp/` location is reboot-wiped on Linux, macOS, and WSL by default.
4. **Self-contained.** One HTML file, inline CSS and JS, no CDN, no external assets. Must work offline. Must work in any browser without dev tools open.
</rules>
## Behavior
### When invoked directly (`/playground` cold)
The user has not given context. Ask: "What are we comparing or tuning?" Wait for the answer before generating anything. Do not guess.
### When dispatche