browsing-as-you

No direct malware is evidenced in this documentation fragment, but it describes a sensitive automation design: decrypting and injecting real browser authentication cookies using OS keychain material and controlling an authenticated browser via CDP/MCP. It also instructs runtime execution of an npm package using an unpinned @latest tag via npx, creating meaningful supply-chain risk. This should be treated as medium security risk and verified by inspecting the underlying code and pinning the MCP server dependency to known-good versions.

SUSPICIOUS: the skill is internally coherent and uses mostly official tooling, but it grants an AI agent broad power to operate the user's real authenticated browser sessions across many sites. There is no clear evidence of credential exfiltration or deceptive third-party routing, so this is not confirmed malware; the main concern is high-impact authenticated autonomy plus sensitive cookie/session access.

This script appears to be a legitimate agent-management utility for running a persistent Chrome instance with CDP enabled and a dedicated profile for durable logins. It contains no clear indicators of intentional malware (no exfiltration, no obfuscated execution, no hardcoded secrets). However, it is security-sensitive: it implements macOS persistence via LaunchAgents, relaxes CDP origin restrictions (--remote-allow-origins=*), and manages an authenticated browser session that could be high-impact if the local CDP endpoint/profile is accessed by an attacker.