commbox

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

• Third-party content exposure detected (high risk: 0.80). The SKILL.md instructs the agent to connect to CommBox via the Membrane CLI (e.g., membrane connect, membrane action run) to read and manage customer messages and social-media/chat content—user-generated, third-party data the agent would parse and act on—so untrusted external content can influence agent behavior.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The skill requires installing and running the Membrane CLI (e.g., via "npm install -g @membranehq/cli@latest" and subsequent "membrane" commands) which fetches remote code and is executed at runtime, so this runtime external dependency can execute code and is required by the skill.