What problem does it solve? AI agents struggle to operate real web pages because raw HTML is verbose and brittle selectors break easily. This Skill teaches agents to drive Chrome/Chromium through the agent-browser CLI using compact accessibility-tree snapshots with stable @eN refs, so clicking, filling forms, extracting data, and taking screenshots work reliably in a few hundred tokens. ## Core Features & Use Cases - Snapshot-and-act loop: Open a page, run agent-browser snapshot -i to get interactive element refs, then click, fill, select, or upload using those refs, re-snapshotting after every page change. - Session and auth management: Use named sessions, --restore persistence, the auth vault, and credential-provider plugins to log in once and stay logged in across runs without exposing passwords. - Advanced workflows: Mock network requests, record HAR files and videos, run axe-core accessibility audits, inspect React component trees, capture Web Vitals, and manage tabs, iframes, and dialogs. - Use Case: An agent needs to log into a web app, navigate to a dashboard, and extract a table of results. It opens the login page, snapshots the form, fills credentials via the auth vault, waits for the dashboard URL, then extracts the data with get text or eval. ## Quick Start Use the browser-usage skill to open https://example.com, take an interactive snapshot, and click the first link.
