hands-3d-pose
Warn
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the user to clone a third-party repository from 'https://github.com/ap229997/hands.git' and suggests downloading model weights via wget.
- [COMMAND_EXECUTION]: Several shell commands are provided for setup and execution, including 'git clone', 'git checkout', and 'pip install -r requirements.txt'.
- [REMOTE_CODE_EXECUTION]: The workflow includes executing a shell script ('scripts/download_models.sh') and a Python script ('demo.py') originating from the external repository.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external media (videos and images) and writes output data to the local system, creating an attack surface for indirect prompt injection. • Ingestion points: Media files provided via '--video_path' or loaded through 'cv2.imread'. • Boundary markers: No explicit delimiters or instructions to ignore embedded content are provided. • Capability inventory: The skill uses 'cv2.imwrite', 'np.save', 'json.dump', and 'cv2.VideoWriter' to create or modify local files. • Sanitization: No sanitization or validation of the input media files is mentioned.
Audit Metadata