ProofShot packages an agent's browser run into reviewable UI evidence
At a glance
A developer lets a coding agent exercise a local page, then reviews one bundle of screenshots, video, and logs. This workflow is listed under UI verification, browser automation, coding agents.
- Agents / tools
- ProofShot, agent-browser, Claude Code, Cursor, Codex
SourceBuilding a tool for these workflows? See weekly rates.
What they did
1) Start the local UI server through ProofShot, for example with proofshot start --run "npm run dev" --port 3000. 2) Have the coding agent navigate the page, click through the feature, and take screenshots while ProofShot records the run. 3) Stop the recording with proofshot stop. 4) Open the generated self-contained HTML report and inspect the video, screenshots, and collected errors before deciding whether the UI is correct.
Worth knowing
ProofShot provides evidence but does not decide pass or fail; the developer reviews the report. The author says it works with Claude Code, Cursor, and Codex through shell commands.
