qualaroo

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

• Third-party content exposure detected (high risk: 0.80). The skill instructs the agent to read and act on fields returned by third-party services (e.g., the clientAction.agentInstructions and clientAction.uiUrl in the output of membrane connection get and proxied responses via membrane request to the Qualaroo API), which are untrusted/third-party content (survey data and connector-provided instructions) that the agent is expected to interpret and that can change its next actions.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill requires the Membrane CLI (installed from npm @membranehq/cli) and at runtime calls membrane connection ensure "https://qualaroo.com", and the returned connection object can include clientAction.agentInstructions — remote content that directly instructs the AI agent — so the runtime fetch can control prompts.