Browser Automation with agent-browser

Two Pathways

Page control — use the agent-browser tool directly for all browser interaction: navigating, clicking, filling forms, taking screenshots, reading snapshots. This is the primary pathway for driving the browser.

Code execution — use exec:agent-browser via Bash when you need to run JavaScript in the page context. The body is piped to eval --stdin. Use this for DOM inspection, custom extraction logic, or anything requiring programmatic page access.

exec:agent-browser
document.title

Multi-line exec:agent-browser blocks where lines are recognized CLI commands run as a batch instead of eval — useful for sequencing page control steps when the agent-browser tool isn't available: