voice-extractor
Pass
Audited by Gen Agent Trust Hub on Aug 26, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill possesses an indirect prompt injection attack surface because it ingests untrusted text samples to generate voice fingerprints.\n
- Ingestion points: User-provided writing samples via pasted text or file paths (documented in SKILL.md under Mode: Extract).\n
- Boundary markers: Instructions specify the use of <voice_fingerprint> tags to delimit fingerprint data passed to other skills.\n
- Capability inventory: The skill reads from and writes to the local file system (~/.voice/) and influences agent drafting behavior.\n
- Sanitization: Includes a triage step to detect AI-generated patterns in input but lacks explicit sanitization for embedded prompt injection commands.
Audit Metadata