What problem does it solve? Browser automation for AI agents typically floods the context window with massive DOM dumps or accessibility trees, making long autonomous sessions expensive and slow. This Skill provides a CLI that uses a snapshot-plus-refs paradigm, reducing context usage by roughly 93% compared to Playwright MCP while supporting full browser control. ## Core Features & Use Cases - Snapshot + Refs Workflow: Capture interactive elements as compact refs (@e1, @e2) and interact with them via click, fill, type, and drag commands. - Session & State Management: Save and reload authenticated browser state, manage cookies and storage, and run named parallel sessions. - Recording & Cloud Browsers: Record session videos, export PDFs, and run on Browserbase cloud browsers for CI/CD pipelines. - Use Case: An AI agent needs to log into a web app, fill a multi-step form, and verify the result. It runs agent-browser open, takes a compact snapshot, fills fields by ref, and waits for the dashboard URL—all within a few hundred tokens of context. ## Quick Start Ask the agent to open a URL with agent-browser, take an interactive snapshot, and click or fill elements using the returned refs.
