elevenlabs-voice-isolator

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (medium risk: 0.30). The skill is a plausible voice-isolation wrapper, but its top-line install command pulls a CLI from a differently named org (belt-sh) than the inference-sh skills it points to, which is a mild lookalike-source concern that warrants a cautious rating under the recall-first override.

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

• Third-party content exposure detected (low risk: 0.10). The required workflow runs belt app run elevenlabs/voice-isolator on a user-supplied arbitrary audio URL, so outsider-authored content is reached only after the agent selects a specific item, and the only text returned to the agent is the transcript from the optional STT step.

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

• Potentially malicious external URL detected (medium risk: 0.30). Setup-time instructions and the npx skills add belt-sh/cli installer come from a separate, unverified org, and the file doesn't establish that belt-sh is inference.sh's first-party publisher, so the tier is medium under recall-first handling; the raw inference-sh install-instructions URL is likely first-party but is still an inbound setup instruction source.