gsplat-optimizer

Warn

Audited by Gen Agent Trust Hub on Feb 15, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS] (MEDIUM): The skill instructs the user to clone and execute code from a third-party GitHub repository (https://github.com/scier/MetalSplatter.git) and install the sogs Python package. Since these sources are not within the defined 'Trusted External Sources' list, they pose a medium risk of supply chain or untrusted code execution.
  • [COMMAND_EXECUTION] (LOW): The skill documentation includes instructions to run a local script (analyze_splat.py). While the script content is not provided in the input, the capability to execute local Python scripts is inherent to the skill's optimization purpose.
  • [PROMPT_INJECTION] (SAFE): No patterns of prompt injection, role-play bypass, or instruction overrides were detected in the markdown or metadata.
  • [DATA_EXFILTRATION] (SAFE): No evidence of hardcoded credentials or commands accessing sensitive directories (~/.ssh, ~/.aws) was found. Network operations are limited to standard package management and cloning.
  • [INDIRECT_PROMPT_INJECTION] (LOW): The skill processes external 3D data files (.ply, .splat). While this represents an attack surface (Category 8), the capabilities listed (math-heavy optimization and rendering advice) are low-privilege with minimal downstream impact on agent reasoning.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 15, 2026, 05:19 PM