What problem does it solve? Browser automation with AI agents risks leaking credentials, hijacking user profiles, and blurring the boundary between CLI-driven and harness-provided browsers. This Skill defines the consent, privacy, and session-ownership rules that must accompany every agent-browser session. ## Core Features & Use Cases - Consent and privacy rules: Governs when confirmation is required, how authorized test-persona logins and reauthentication are handled, and how credentials must stay out of logs, output, and artifacts. - Runtime boundaries: Separates CLI evidence from harness browser evidence, restricts CDP attachment to approved endpoints, and requires pinned tabs for shared sessions. - Profile and session management: Verifies private profile paths and ownership, handles profile locks safely, and mandates task-unique named sessions with scoped cleanup. - Use Case: When an agent needs to log into a staging app with a test persona via agent-browser, this policy ensures credentials are entered through a secret-safe helper, identity is verified after login, and only the task-owned session is closed afterward. ## Quick Start Load the agent-browser-policy skill alongside agent-browser before performing any browser automation that involves logins, profiles, or CDP endpoints.
