learn-voice

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

• Third-party content exposure detected (high risk: 1.00). The skill explicitly searches and downloads public YouTube content (see SKILL.md pipeline and run.sh's download_and_separate which calls discover-music youtube-search and youtube-stems) and ingests those untrusted, user-generated audio tracks into the training pipeline used to build models, so third‑party content directly influences subsequent tool behavior.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill's run.sh invokes docker run with the external image "cherrymint/rvc_webui:rvc_boss", which will be pulled and executed at runtime (remote code execution) and is required for training, so this external image is a runtime dependency that executes remote code.