Browser Act

The problem

Standard fetch tools fail on modern sites that require JavaScript execution, session persistence, or complex human interactions. AI agents often hit walls when encountering captchas, multi-step forms, or content that only loads after scrolling.

What it does

• Executes JavaScript to extract rendered content from SPAs and dynamic websites.
• Manages isolated browser profiles to maintain authenticated sessions and multiple accounts.
• Automates complex sequences including clicks, typing, file uploads, and form submissions.
• Captures visual evidence via screenshots and inspects network traffic including XHR and HAR data.
• Enables human-in-the-loop collaboration for handling manual verification steps or headed-mode tasks.

Frameworks & tools

Built for Python 3.12+ environments using the uv package manager. Compatible with any agentic IDE or CLI including Cursor, Claude Code, Codex, and Windsurf.

Why this beats prompting it yourself

Hard-coding browser automation logic is brittle and insecure. This skill provides a standardized protocol for session isolation, safety confirmations, and error recovery that keeps the agent from breaking the flow or leaking credentials.

Use cases

• Extracting data from search results or dashboards that require scroll-to-load.
• Automating repetitive form-filling tasks across authenticated enterprise portals.
• Monitoring visual regressions by capturing and comparing page screenshots.
• Testing multi-step checkout flows that trigger JavaScript-based validation.

Known limitations

Requires local installation of the browser-act CLI and Python 3.12. Sensitive operations like logins and browser creation require explicit user confirmation via the Confirmation Gate protocol.