hand-tracking-toolkit

Warn

Audited by Gen Agent Trust Hub on Mar 15, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill retrieves its core source code from the official Meta Facebook Research repository on GitHub. This is a trusted and well-known service.
  • [COMMAND_EXECUTION]: The quick start and usage examples demonstrate the use of pickle.load() to read prediction data. This is a significant security risk because the pickle module is not secure and can execute arbitrary code during the deserialization of a maliciously crafted file.
  • [COMMAND_EXECUTION]: Installation instructions include running pip install -r requirements.txt, which triggers the execution of installation scripts for the required Python dependencies.
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by processing external files. Ingestion points: Data is loaded from user-provided .pkl files and HOT3D sequence directories. Boundary markers: The skill does not implement delimiters or ignore-instructions warnings for processed data. Capability inventory: The skill can perform file system operations and execute code through pickle deserialization. Sanitization: There is no evidence of input validation or sanitization for the data loaded from external files.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 15, 2026, 04:33 AM