narrative-voice-coach

Pass

Audited by Gen Agent Trust Hub on Aug 10, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill ingests untrusted text data for analysis through various commands, creating a potential surface for indirect prompt injection attacks.
  • Ingestion points: The commands voice match, voice consistency, voice compare, and voice experiment specifically ingest user-supplied text or read external files (SKILL.md).
  • Boundary markers: The skill instructions do not define clear delimiters or specific instructions to ensure the agent ignores embedded commands within the analyzed prose.
  • Capability inventory: The skill is restricted to file:read and file:write permissions; no network access, shell command execution, or other high-privilege capabilities were identified (SKILL.md).
  • Sanitization: The skill does not describe any mechanisms for filtering, sanitizing, or validating the input text before processing.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 10, 2026, 07:41 PM
Security Audit — agent-trust-hub — narrative-voice-coach