agent-browser

[Skill Scanner] Installation of third-party script detected Based on the provided documentation for agent-browser, the capabilities described are consistent with a headless browser automation CLI and match the stated purpose. There are no explicit malicious indicators in this documentation: no hardcoded credentials, obfuscated payloads, or suspicious download sources are present. However, high-risk features are documented (arbitrary JS eval in page context, network interception, custom proxy/cloud provider integrations, auth state persistence, and upload) which — if implemented insecurely or configured to use untrusted third parties — could enable credential harvesting or traffic interception. Final risk depends on the actual implementation and default provider endpoints; review of the codebase and runtime network endpoints is recommended before trusting with sensitive credentials or data. LLM verification: The skill's documented capabilities are consistent with a legitimate headless browser automation CLI: requested capabilities (network, filesystem, native binaries) match the purpose. There is no direct evidence of malicious code or obfuscation in this SKILL.md. However, the documentation omits details about the provenance and verification of downloaded native/browser binaries (agent-browser install), which is a supply-chain risk. Additionally, features like arbitrary JS evaluation in page contex