avatar-video
Fail
Audited by Gen Agent Trust Hub on Nov 14, 2026
Risk Level: CRITICALEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest external data and process user-provided scripts to generate video content. This creates an attack surface where maliciously crafted scripts or metadata in fetched assets could attempt to influence the agent's behavior.
- Ingestion points: Processes user-provided text as video scripts; fetches external assets (images, audio, video) via the
uploadFromUrlfunction and background URL parameters. - Boundary markers: The instructions do not define specific delimiters to isolate user-provided scripts from system instructions during processing.
- Capability inventory: The skill possesses network communication capabilities (fetching from arbitrary URLs and sending data to HeyGen APIs) and local file writing capabilities (downloading and saving generated videos).
- Sanitization: The provided
uploadFromUrlcode includes a basic check to ensure URLs use the HTTPS protocol. - [EXTERNAL_DOWNLOADS]: The skill includes documentation and code examples for fetching assets from external remote URLs.
- Evidence: The
uploadFromUrlfunction inreferences/assets.mddownloads binary data from a user-supplied URL to buffer it for API submission. - Mitigation: The implementation includes a check that restricts downloads to
https:protocols, reducing the risk of local file access viafile://or other unsafe schemes.
Recommendations
- CRITICAL: 1 file(s) identified as malware by FileRep - DO NOT USE
Audit Metadata