browser-agent-safety

What problem does it solve?

Prevents unsafe or scope-expanding browser automation by forcing an explicit, auditable safety contract before any agent drives Playwright/DevTools/computer-use loops.

Core Features & Use Cases

• Browser surface scoping with domain allowlists and explicit forbidden actions, including redirect and origin-drift stop behavior.
• Profile isolation and sensitive-surface boundaries to protect accounts, credentials, PII, downloads/uploads, clipboard, and production risk.
• Risky-action point-of-risk approvals plus rollback/undo and clear stop conditions for submit/purchase/delete/permission changes.
• Prompt-injection defenses treating page text, DOM, screenshots, console, network, and downloaded content as untrusted evidence.
• Evidence capture and cleanup proof, producing a browser-agent-safety-report artifact for runtime validation and durable receipts.

Quick Start

Use the browser-agent-safety skill to run a scoped safety preflight for a browser automation task and output a browser-agent-safety-report before the agent executes any risky actions.