BrowserSkill: Give AI Agents Full Browser Access
AI agents can now use your logged-in browser, BrowserSkill from Tencent lets Claude Code, Cursor, and Codex automate web tasks without hitting bot detection. BrowserSkill, open-sourced by Tencent in June 2026, hands AI coding agents like Claude Code, Cursor, and Codex direct access to your already-authenticated browser session instead of forcing them to spin up isolated test profiles. The tool runs as a local `bsk` CLI daemon paired with a browser extension, letting agents borrow your open tabs, complete with login cookies and session state, so they can read private dashboards, fill authenticated forms, and navigate behind-the-login pages that would otherwise block bot traffic. Setup takes one command; the extension installs manually (agents cannot self-install), and a `bsk doctor` diagnostic confirms readiness. This unlocks real capabilities: agents can now bypass bot detection by appearing as a real user, handle occasional captchas by handing control back to you, and work inside a visible Agent Window while you continue in other tabs. The trade-off is steep. One browser-extension permission grants access to every site you're logged into simultaneously, the only guard is an honor-system "borrow and return" rule. Security tools cannot see agent actions (they look like your clicks), making prompt-injection attacks easier and undetectable. Industry benchmarks show AI browser agents succeed 70, 85% of the time on novel tasks (roughly one failure per five attempts), and at $0.01, 0.05 per run, mistakes cost real money. BrowserSkill's own test suite covers only 6 stable cases across 28 operations, and live bug reports (Windows session hangs, form-submission connection drops) reveal silent failures, the diagnostic passes while the tool actually breaks. For builders deciding whether to hand an AI agent your authenticated session: the decision hinges on the cost of a wrong click. Safe for low-stakes automation (data reads, simple navigation); risky for anything destructive (account changes, financial transactions, message sending). Relevant for developers using Claude Code or Cursor who want to automate web workflows, and security teams evaluating browser-automation risk in AI-assisted coding. Chapters: 0:00 The authenticated browser revolution 0:55 Claude, Cursor, and Codex now on board 2:03 Why every other tool hit the login wall 3:04 Captchas just pause, don't stop 4:21 Every open tab is now reachable 5:31 One command, one daemon, one extension 6:27 The machine gets rules, not limits 7:38 One permission unlocks your entire session 8:41 Your security tools see nothing 9:36 Hidden text can rewrite agent orders 10:56 One in five attempts still fails 12:02 Only six paths were actually tested 12:56 Passed diagnostics, broken forms 14:20 What changed and what it still costs 15:26 The one click that decides everything Tools & resources mentioned: - BrowserSkill: https://github.com/Tencent/BrowserSkill - Chrome Web Store, BrowserSkill: https://chromewebstore.google.com/detail/browserskill/hhcmgoofomhgciiibhipgmgkgnoenaoi - Playwright: https://playwright.dev - Browser Use: https://github.com/browser-use/browser-use - Vercel agent-browser: https://github.com/vercel-labs/agent-browser - DeepSeek Harness BrowserSkill Plugin: https://deepseek1024.com/plugins/Tencent/BrowserSkill/packages/dsh-plugin-browserskill - Claude Code: https://claude.ai - Cursor: https://cursor.sh About The Stack The Stack helps you build with AI. Each video takes one tool, model, or workflow and shows how it works in a few focused minutes, with the real benchmarks and real costs. We go deep on Claude Code and Cursor for AI coding, AI agents and MCP servers, the open-source AI tools and GitHub repos most people miss, RAG and vector search, fine-tuning, and running local LLMs on your own machine with Ollama and LM Studio. We compare models like ChatGPT and Claude, test AI automation with Zapier, Make, and n8n, and flag the tools that actually ship. Subscribe for new breakdowns: https://www.youtube.com/@the-stack-ai?sub_confirmation=1 #AI automation #Claude Code #Browser agents
