speech-to-text

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.80). Several entries use nonstandard single-label hostnames that look like placeholder media URLs (e.g., https://audio.mp3, https://video.mp4, etc.), which are untrusted/unknown hosts and could be used to deliver arbitrary or malicious files, while the inference.sh and GitHub links look legitimate.

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

• Third-party content exposure detected (medium risk: 0.65). SKILL.md’s “Workflow: Video Subtitles” step passes an outsider-supplied transcript string (captions: "<transcript-from-step-1>") and uses an outsider-provided media URL (audio_url/video_url like https://video.mp4)—the LLM reads the free-text transcription returned by those external URLs at runtime via the belt CLI integration.