ProofShot
A local browser verification CLI for AI coding agents that records UI tests, captures errors, and saves review artifacts. Open source under MIT.
ProofShot is a local browser verification CLI for developers who want to inspect UI work produced by AI coding agents. It records browser sessions and packages screenshots with browser and development server errors, so reviewers can see what the agent tested rather than rely on its written account. It's open source under the MIT license.
Built on Vercel's agent-browser, it uses Chromium for browser testing and adds a record of the session for human review. It works with Claude Code, Cursor, Codex, OpenCode, Gemini CLI, Windsurf, and GitHub Copilot, as well as other agents that can run shell commands. Its agent skills teach the verification workflow across projects.
The standalone HTML viewer pairs video with an action timeline and markers. Console and server logs have highlighted errors and timestamps linked to the recording, so a reviewer can connect a failure to the interaction that triggered it. Sessions also include screenshots and a Markdown report. Screenshot comparisons against a baseline help check for visual regressions.
Server log error detection covers JavaScript/Node.js, Python, Ruby/Rails, Go, Java/Kotlin, Rust, PHP, C#/.NET, and Elixir/Phoenix. Recording and local review have no cloud dependency; the artifacts stay on your machine unless you choose to upload them. Optional GitHub integration uploads session media and posts a pull request comment with embedded screenshots.
