A browser CLI that knows when to quit — and call you

What it does

BrowserAct Skills is a Python CLI that wraps browser automation for AI agents. It drives Chrome or stealth-mode browsers, handles CAPTCHAs, rotates fingerprints and proxies, and — unusually — generates a live remote-assist URL when the agent gets stuck so a human can take over from any device. The agent resumes when you’re done.

The interesting bit

The project treats LLM token efficiency as a first-class concern. Instead of dumping HTML or JSON, it returns an indexed text format where state lists clickable elements as numbered items — click 3, input 2 "..." — no DOM parsing required. It’s a small interface bet: agents reason better about indices than selectors.

Key highlights

• Three browser modes: reuse local Chrome profiles, ephemeral stealth scraping, or fixed-identity multi-account parallel sessions
• Concurrency isolation — cross-browser parallel with independent cookies/fingerprints, or same-browser multi-session with shared login state
• Confirmation gating for sensitive ops (browser create/delete, proxy changes, privacy toggles) — enforced at the Skill layer, not configurable
• Companion “Skill Forge” tool auto-generates repeatable scrapers by exploring a site once, then running stable paths at scale
• 30+ pre-built Skills for Amazon, Google Maps, Reddit, WeChat, Zhihu, etc.

Caveats

• The README is heavy on feature promises and light on implementation detail — no architecture overview, no test suite mention, no performance benchmarks
• “Auto-solves CAPTCHAs” and “stealth headless that isn’t detected” are claims without disclosed methods or success rates
• The star-for-credits Discord incentive is visible in the README

Verdict

Worth a look if you’re running Claude Code, Cursor, or similar agents and need browser automation that won’t leak state across parallel tasks. Skip if you want transparent anti-detection internals or a pure open-core tool without a credits-based service ecosystem attached.