gologin-local-agent-browser-skill

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill explicitly requires verifying a GOLOGIN_TOKEN and instructs the agent to stop and ask the user for the token if missing (accepting tokens "explicitly provided by the user in the conversation"), which encourages the LLM to receive and potentially echo or embed secret values in outputs, creating an exfiltration risk.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 1.00). This skill explicitly opens and interacts with arbitrary public URLs (see SKILL.md and the Command Map's open , and workflows/profile-warmup.md and login-and-cookie-capture.md which use external sites like https://www.reddit.com and runbook steps), captures snapshots and evaluates page content as the compact page model, and then uses those untrusted third-party pages to drive subsequent clicks/types/eval actions—so external web content can materially influence agent behavior.