Browser Agent Harness

Browser Agent Harness is a local-first browser automation companion for agent clients. It enables authorized agents to observe and operate user-selected browser tabs through a constrained and auditable control layer. Supported workflows include creating and organizing task tabs, capturing accessibility-first page snapshots, clicking elements, typing into non-sensitive fields, scrolling, waiting for page state changes, running bounded page scripts, and taking screenshots. Key features: • Local-first architecture with no cloud relay • Native Messaging integration with a locally installed host • Task isolation using dedicated browser tab groups • Accessibility-first page observation • Controlled clicking, typing, scrolling, waiting, and screenshots • Password and one-time-code field blocking • Sensitive-field masking in screenshots • Local audit records with sensitive values redacted • Side panel showing connection status, controlled tabs, task state, and recent activity • Immediate task stop and debugger detachment Browser Agent Harness does not provide arbitrary Chrome DevTools Protocol access, does not contain advertising, and does not track tabs outside user-initiated automation tasks. A compatible Browser Agent Harness Native Host and agent client must be installed locally. The extension is intended only for user-directed workflows on websites and content the user is authorized to access.