[mrprkr]

Summary

Brings the run page closer to the Stunt Double web report (PRO-3001), using what a local run already writes to .stuntdouble/:

• Highlights first: the actor's summary, the checks that need attention (up to three, each linking to its evidence on the Results tab), what it observed and the longest wait (10s or more), linking to the journey.
• Score card, unchanged, next.
• Tabs in place of one long page, as the web report has them: Results (the existing check groups), Journey, Screenshots and Activity (the existing tool-call log). Arrow keys move between tabs; the chosen tab survives live redraws.
• Journey: built from activity.jsonl, one rail through the pages opened, the browser steps (folded to key moments past 20) and each verdict with its screenshot, with the waits drawn between them. Stats: end to end, pages opened, browser steps, checks judged, longest wait. A resume's idle gap is never counted as a wait. Page reads and screenshots stay in Activity.
• Screenshots: every verdict's capture in a grid, captioned with its check and verdict.

Interview transcripts with UMUX answers, and per-step journey data beyond activity.jsonl, need the local MCP tools to record them first; that is a separate change.

Checklist

• pnpm format:check, pnpm lint, pnpm type-check, pnpm test and pnpm build pass locally (one test, "a missing file says what exists, or how to make one", fails when run as root because the EACCES file stays readable; it fails on main the same way)
• Tests cover the change (studio-e2e.test.ts now checks the highlight link, the journey rail and the screenshots grid)
• A changeset is included (pnpm changeset) if what stuntdouble does changes
• Every commit is signed off (git commit -s), per the DCO

Could not verify

• Only checked in headless Chrome at desktop width, light mode; not checked on a phone width or in dark mode.
• A journey longer than 20 steps and a long wait were not exercised by the scripted run.

---

Generated by Claude Code