tavus-replica-video
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of user-controlled strings (scripts, audio URLs, and system prompts) for use in API requests, creating a theoretical surface for indirect prompt injection. This is inherent to the skill's primary function and is accompanied by strong safety and ethical guidelines.
- Ingestion points: User scripts, audio URLs, and system prompts are processed in SKILL.md.
- Boundary markers: No specific structural delimiters for untrusted input are mandated in the instruction templates.
- Capability inventory: The skill enables network requests to the Tavus API (tavusapi.com) for media generation and real-time conversations.
- Sanitization: The instructions focus on manual QA and ethical review rather than automated input sanitization.
- [SAFE]: No malicious code, exfiltration patterns, or obfuscation were detected. The skill explicitly directs the agent to verify consent, respect biometric privacy, and disclose AI generation to end users.
Audit Metadata