First paragraph: ProofShot, a new open-source tool from Vercel Labs, lets AI coding agents verify their UI changes by recording browser interactions. The tool addresses a key limitation: agents can’t see if their code breaks layouts or triggers errors without human browser checks. Using headless Chromium, it captures video, screenshots, and logs, bundling them into a reviewable HTML file. Developers can run commands like proofshot start --run "npm run dev" to automate testing workflows.

Second paragraph: The tool operates via three CLI commands: start (launches dev server and recording), exec (captures actions like clicks and form fills), and stop (generates artifacts). Unlike testing frameworks, ProofShot doesn’t assign pass/fail statuses—it provides raw evidence. Features include error detection scanning console logs and server outputs, an interactive timeline syncing video with code actions, and PR-ready summaries. It works with any AI agent (Claude Code, Cursor, etc.) through a one-time installation.

Third paragraph: The standalone HTML viewer allows developers to scrub through sessions, see element labels, and compare visual diffs against baselines. By wrapping dev servers, ProofShot automatically trims dead time from recordings, focusing on active agent behavior. The tool emphasizes transparency, letting teams audit AI decisions without manual browser reviews. With MIT licensing, it’s free and agent-agnostic, requiring minimal setup to integrate.

Fourth paragraph: ProofShot’s significance lies in bridging the gap between AI-generated code and real-world validation. By making browser verification a first-class citizen for agents, it reduces debugging overhead and improves reliability. The tool’s simplicity—three commands, one proof artifact—makes it a practical addition to AI-driven development pipelines. Developers can access it at [github.com/AmElmo/proofshot](https://github.com/AmElmo/proofshot).