ai-video-generation
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data via URLs for images and audio, which can serve as a vector for indirect prompt injection. \n
- Ingestion points: External URLs are accepted via
image_urlandaudio_urlfields in several model routes (e.g., HappyHorse I2V, Wan 2-7 lip-sync) within SKILL.md. \n - Boundary markers: The skill uses a structured JSON
--inputstring to pass parameters to the CLI, providing separation between instructions and data. \n - Capability inventory: The skill executes the
runcomfyCLI tool through the Bash tool (restricted viaallowed-tools). \n - Sanitization: The skill acknowledges the risk of embedded instructions in external media assets and instructs the agent to only use URLs explicitly provided by the user. \n- [COMMAND_EXECUTION]: The skill invokes the
runcomfyCLI tool to perform generation and management tasks. \n - Execution is scoped to the
runcomfycommand space as defined in theallowed-toolsmetadata. \n - Input is handled via JSON strings to mitigate potential shell injection vulnerabilities from user-supplied prompts.
Audit Metadata