BrowserSkill

BrowserSkill connects an AI agent to Chrome or Microsoft Edge using the accounts already signed in there, so the agent can read pages, fill forms, work through websites, capture long screenshots, and investigate failing requests. Tasks run in a separate, visible Agent Window while you keep working, and an existing tab can be borrowed explicitly and returned when the task ends. You choose the agent and model, and BrowserSkill supplies the browser connection.

It has three parts: the bsk CLI with a background daemon, a browser extension, and a skill that teaches the agent how to use the CLI. Any shell-capable agent works, including Cursor, Claude Code, Codex, OpenClaw, CodeBuddy, WorkBuddy, Pi, and Hermes. DeepSeek Harness has a dedicated plugin that adds native browser tools, task previews, and screenshot results to its web UI.

For website debugging, capture links each operation to its requests, response bodies, console messages, and page changes, and summarizes page-load metrics, slow APIs, and suspected duplicate requests. Task-local request modification, blocking, mock responses, and same-origin replay test a hypothesis, and the evidence exports as JSON. An agent running on a server can pair with a browser on your computer over authenticated WebSocket, with the browser initiating the connection.

An Agent Window shares the selected profile's login state and is not a security sandbox, so agents act with the permissions of the signed-in websites. Two extension settings, both on by default, ask before borrowing a tab and let the agent request human help for logins or verification. There is no mandatory cloud service or product telemetry, and an optional operation audit records task metadata without input values, page bodies, or screenshots.