provider-model-refresh
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the processing of untrusted user input and media data via external generation tools.
- Ingestion points: The skill accepts user-provided text descriptions for image/video generation, speech scripts, and reference media files (images/audio) via parameters such as
image_path,script, andaudio_url. - Boundary markers: No explicit instructions or delimiters are defined to prevent the agent from executing instructions potentially embedded within the user-supplied prompts.
- Capability inventory: The agent can call multiple external media generation APIs (including OpenAI, Google, ElevenLabs, and HeyGen) and write job checkpoints to the local file system via the
job_pathparameter. - Sanitization: The skill documentation lacks any mention of sanitization or validation for user-provided strings before they are passed to the model adapters.
Audit Metadata