open-video
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user requests and reference media files, creating a potential surface for indirect prompt injection. * Ingestion points: User-supplied concepts and reference images, videos, or audio files are used to construct generation plans and prompts. * Boundary markers: The skill utilizes a strictly defined "3-field prompt" grammar to isolate and define the agent's output. * Capability inventory: The agent can execute local Python scripts (e.g., scripts/h3_agent.py), perform HTTP requests to the ComfyUI API, and use ffmpeg for media processing. * Sanitization: A mandatory validation step is enforced by core/validator.py to check all generated prompts against strict constraints before execution.
- [COMMAND_EXECUTION]: The instructions describe the use of shell commands to manage the ComfyUI server and execute internal helper scripts. These operations are documented and restricted to the local environment.
Audit Metadata