ai-avatar-video
Pass
Audited by Gen Agent Trust Hub on Nov 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides numerous functional examples of shell commands utilizing the 'belt' CLI tool to interact with the inference.sh platform. Access is explicitly scoped via the 'allowed-tools' metadata.
- [EXTERNAL_DOWNLOADS]: The skill references external media assets (images, audio) and an installation guide hosted on GitHub (github.com/inference-sh/skills) for setting up the required CLI environment.
- [INDIRECT_PROMPT_INJECTION]: The skill defines an attack surface where untrusted data, such as external URLs and user-supplied text scripts, is ingested and passed as arguments to the 'belt' command-line interface. Evidence: (1) Ingestion points: JSON input fields like 'image', 'voice_script', and 'audio' in SKILL.md. (2) Boundary markers: Arguments are encapsulated within JSON strings. (3) Capability inventory: Executes subprocesses via the Bash tool limited to the 'belt' command. (4) Sanitization: Dependent on the underlying platform's handling of shell argument escaping.
Audit Metadata