Браузерная автоматизация для AI-агентов

automation

vc-agent-browser is a Claude Code skill that provides a browser automation CLI built for AI agents: its "snapshot + refs" paradigm uses ~280 characters per snapshot compared to 8 000+ for Playwright MCP, cutting context consumption by 93%. It supports long autonomous sessions, self-verifying build loops, multi-tab workflows, video recording for debugging, and cloud browsers via Browserbase. The core four-step cycle is: open a URL, snapshot interactive elements with ref identifiers, interact using those refs (click, fill, drag, upload, etc.), then re-snapshot after page changes. The full command set covers navigation, form interaction, state checks, cookie and localStorage management, device emulation, network mocking, geolocation, and PDF export. Ideal for token-constrained agentic workflows where reproducibility and minimal context overhead matter.