ace-step

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

• Third-party content exposure detected (high risk: 1.00). The skill's audio-inpaint and audio-outpaint routes explicitly accept arbitrary HTTPS audio URLs as input (see Route 2 and Route 3 schema "audio" field) and the Security & Privacy section even notes those source audio URLs are untrusted and can influence generation, so third-party audio could carry embedded instructions that affect model outputs.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill requires fetching user-provided audio URLs at runtime (e.g., https://your-cdn.example/original-track.mp3) for the audio-inpaint/outpaint endpoints, and that fetched content directly influences model generation and can carry steganographic or embedded instructions that effectively control outputs.