compass-ai

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

• Third-party content exposure detected (high risk: 0.90). The skill's SKILL.md instructs the agent to connect to public Compass AI endpoints (e.g., membrane connection ensure "https://www.getcompass.ai/"), to proxy arbitrary API requests (membrane request ...), and to consume connection responses including clientAction.agentInstructions, meaning the agent will fetch and interpret third-party content/instructions that can change subsequent actions.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill calls membrane connection ensure with the app URL "https://www.getcompass.ai/" at runtime, and Membrane can build or fetch a connector whose returned clientAction.agentInstructions are remote content that would directly control the agent's instructions—constituting a runtime external dependency that can inject prompts.