haravan

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

• Third-party content exposure detected (high risk: 0.90). This skill instructs the agent to make proxied requests to external Haravan APIs (see "Proxy requests" / membrane request CONNECTION_ID /path/to/endpoint) and also cites clientAction.agentInstructions returned by connections as programmatic guidance for the AI, meaning untrusted third‑party responses could be ingested and directly influence agent behavior.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill invokes the Membrane CLI at runtime (e.g., membrane connection ensure "https://www.haravan.com/"), and the remote connection response can include clientAction.agentInstructions which are fetched runtime instructions that directly tell the AI how to proceed, so the external URL https://www.haravan.com/ (via Membrane) can control agent prompts.