Safe Browser

Safe Browser is an agent skill for creating constrained web automation with a clearly defined network boundary. It guides the coding agent toward a local runtime where browser actions pass through a dedicated interface, while request interception prevents navigation or resource loading outside an approved host list. The included example makes the security model observable instead of treating containment as an assumption.

Highlights

• Creates a focused browser capability instead of exposing unrestricted browser control
• Applies domain policy to page navigation and secondary network requests
• Produces an audit trail showing which requests were permitted or stopped
• Includes a runnable Claude Agent SDK template and a Hacker News demonstration
• Verifies containment through expected output, blocked navigation, and saved artifacts

This skill is useful for developers prototyping research agents, controlled scrapers, or security demonstrations where web access must remain predictable. It requires Node.js, npm, local Chromium, and an Anthropic API key for the supplied runtime example. Install it from the Browserbase skills repository with the standard skills CLI command.