comfyui-character-gen
Pass
Audited by Gen Agent Trust Hub on Mar 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The Python scripts located in 'references/talking-head-workflows.md' utilize 'subprocess.run' to interface with external tools like Wav2Lip and CodeFormer. While these are implemented using argument lists (reducing shell injection risk), they execute external processes based on file paths provided during runtime.
- [EXTERNAL_DOWNLOADS]: The skill provides extensive documentation and links for downloading AI models and custom nodes from platforms such as HuggingFace, Civitai, and GitHub. These include models from well-known organizations such as Black Forest Labs, Stability AI, and Alibaba.
- [REMOTE_CODE_EXECUTION]: Documentation in 'references/talking-head-workflows.md' and 'references/voice-synthesis.md' recommends cloning third-party GitHub repositories and installing their dependencies via 'pip install -r requirements.txt', which involves executing code provided by those external sources.
- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it is designed to ingest and process untrusted external data (user-provided images and audio) through scripts that possess shell execution capabilities.
- Ingestion points: 'references/talking-head-workflows.md' (source_image, driven_audio parameters)
- Boundary markers: None identified in the provided reference scripts
- Capability inventory: 'references/talking-head-workflows.md' (subprocess.run)
- Sanitization: No explicit sanitization of input paths or audio content is demonstrated in the reference scripts.
Audit Metadata