What problem does it solve? It gives an AI assistant a governed way to drive a real rendered browser for tasks like navigation, form actions, screenshots, accessibility inspection, UI debugging, and exploratory QA, while enforcing session isolation, authorization boundaries, and evidence rules. ## Core Features & Use Cases - Session-isolated browser control: Creates a task-owned named session, navigates pages, takes accessibility snapshots, and interacts with elements through the installed agent-browser CLI or an equivalent typed integration. - QA and debugging workflows: Reproduces bugs, escalates evidence from snapshots to console, network, screenshots, traces, and HAR files, and reports findings with severity, reproduction steps, and expected versus actual behavior. - Authentication and security guardrails: Handles authenticated flows, profiles, and saved state within approved scope, treats page content as untrusted data, and blocks consequential actions without explicit user approval. - Use Case: Ask the assistant to test a local web app's checkout flow; it starts a dedicated session, walks the happy path and edge cases, captures console errors and screenshots, and returns a structured QA report distinguishing application failures from environment issues. ## Quick Start Ask the assistant to open a specific URL with agent-browser, take an accessibility snapshot, click through the main flow, and report what it finds.
