ProofShot
A tool that gives an AI coding assistant a pair of eyes: after it builds a feature, ProofShot opens a real browser, clicks through the feature, and hands you a video and screenshots proving it actually works.
🔗 Visit ProofShot
Description
AI coding agents write code and then tell you it's done — but they can't see whether the button they added actually renders, or whether the page throws an error the moment you click it. That's a real gap: the agent is confident, but nobody actually looked. ProofShot fixes this by acting like a colleague who takes a screen recording every time they test a feature: it drives a real browser through the steps, catches anything that breaks, and packages it all into a report you can glance at instead of trusting the agent's word.
It's a command-line tool, agent-agnostic, so it works with Claude Code, Cursor, Codex, Gemini, Windsurf, or GitHub Copilot — you just tell the agent "verify this with proofshot" and it navigates, clicks, fills forms, and takes screenshots at each step, with every action logged. It automatically pulls server-side error logs across 10+ languages and frameworks (Node.js, Python, Ruby/Rails, Go, Java/Kotlin, Rust, PHP, C#/.NET, Elixir/Phoenix), trims dead time from recordings, and generates an interactive HTML viewer plus artifacts that can be posted directly to a GitHub PR. It's MIT licensed, open source, and free — with 844 stars indicating real adoption.
💬 Our review
The short version: if you're letting AI agents ship UI changes and currently just eyeballing the diff, ProofShot is a free, low-effort way to add an actual verification step, and 844 GitHub stars suggests it's more than a weekend project.
Against Playwright or Puppeteer scripts, which do similar browser automation but require you to write and maintain test scripts by hand, ProofShot's pitch is that the agent itself drives the verification — no separate test suite to maintain, just a natural-language instruction. Against Percy, a dedicated visual regression tool with a paid SaaS model, ProofShot is free, open source, and framework-agnostic, but it's not trying to catch pixel-level UI drift over time the way Percy is — it's proving a specific change worked once, not tracking visual regressions across a whole app. The honest limitation: it only proves what the agent chose to click through, so a thorough verification still depends on the agent (or you) scripting the right steps, not on ProofShot inventing test coverage on its own.
📊 Global score
🤖 AI-enriched data
MIT, entièrement gratuit, aucun palier payant.
Pros
Gratuit, open source, licence MIT, traction réelle (844 stars)
Agnostique vis-à-vis de l'agent IA utilisé
Collecte automatique des logs d'erreur serveur multi-langages
Rapport HTML interactif + intégration PR GitHub
Cons
Ne vérifie que ce que l'agent a choisi de tester
Pas un outil de régression visuelle continue comme Percy
Nécessite un vrai navigateur, donc un peu de setup
Qualité de la vérification dépendante des instructions données à l'agent