cmux

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

• Third-party content exposure detected (high risk: 0.72). SKILL.md documents a browser automation workflow (cmux browser ... get text body, snapshot, etc.), which at runtime would fetch and ingest outsider-authored web page body text into the agent’s LLM context if the consuming agent uses those results as prompt content—i.e., public web content category via the browser text extraction path.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill includes an explicit "sudo ln -sf /Applications/... /usr/local/bin/cmux" install step which requires elevated privileges and modifies a system path, so it directs the agent to perform privileged system changes.