Stunt Double Studio: run

/stuntdouble:run <checklist> [resume], one of the Studio skills (/stuntdouble:setup, /stuntdouble:run, /stuntdouble:open, /stuntdouble:cloud and /stuntdouble:doctor). Local runs use the person's own Claude plan: this session is the actor, and the plugin's local MCP server (npx stuntdouble@0.1.3 mcp) only drives a local Chrome, writes files and serves the viewer. Nothing else calls a model, and no account is needed. Say so before the first run in a session.

The local server ships only in the Claude Code plugin. Where its tools are missing, use the cloud skills. With no .stuntdouble/stuntdouble.json, offer /stuntdouble:setup first. To fix a field start_run names, the file format is in the setup skill's files.md (../setup/files.md from this skill's folder).

1. start_run(checklist, resume?), with resume: true when asked. Fix any field it names and call again.
2. Print only the viewer link from its output: the person can watch the run and stop it there.
3. Dispatch the stuntdouble:stuntdouble-actor agent with the full start_run output as its prompt. Never drive the browser from this thread: page text and screenshots stay in the actor's context.
4. Lead with the result from its reply: "Failed: 3 of 4 checks passed", then the score, or "no score" where score is null, explained in one clause (the checklist has no weighted checks, so it passes only when every pass/fail check passes). Then each failed and not-judged check with its evidence, and the viewer link. Say the verdicts are self-judged; a cloud run adds an independent judge. Offer to fix what failed and run again.

If the actor ends without finish_run (or was interrupted), start_run(checklist, resume: true) carries on with the run still open, and its output is the brief for a fresh dispatch: recorded verdicts stand. If the session closed mid-run, /stuntdouble:run <checklist> resume judges only the checks left.

---

Edit on GitHub