wan-2-7
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute the
runcomfycommand-line interface to perform video generation tasks. - [EXTERNAL_DOWNLOADS]: The skill requires the global installation of the
@runcomfy/cliNode.js package and retrieves generated video content from the vendor's domains. - Fetches outputs from
*.runcomfy.netand*.runcomfy.com. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data which could contain malicious instructions for the underlying video model.
- Ingestion points: Untrusted data enters the agent context through the
promptinput and external audio files referenced byaudio_urlinSKILL.md. - Boundary markers: The CLI implementation uses JSON string encapsulation for input parameters to prevent direct shell command injection.
- Capability inventory: The skill utilizes shell command execution, local file system writes for output storage, and network operations conducted by the CLI tool.
- Sanitization: While inputs are passed as JSON strings, the skill documentation acknowledges that content processed by remote models poses an inherent injection risk.
Audit Metadata