file-share

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

• Third-party content exposure detected (high risk: 0.70). The public web UI (/ and /request-access) accepts untrusted user-provided data (emails and file names) which the server forwards into a prompt sent to the external LLM endpoint (https://api.zo.computer/zo/ask) in sendEmail (scripts/server.ts), meaning attacker-controlled content can be interpreted by that agent and influence tool actions (e.g., the instructed Gmail integration).

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

• Potentially malicious external URL detected (high risk: 1.00). The server makes runtime POST calls to https://api.zo.computer/zo/ask with a prompt that instructs the remote service to invoke Gmail tooling and send emails, meaning the skill relies on that external endpoint to execute remote actions driven by injected prompt instructions.