model-explainability-tool

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • COMMAND_EXECUTION (HIGH): The skill requests Bash(python:*) permissions, which allows for the execution of arbitrary Python scripts. This creates a high-risk execution environment if the logic is influenced by untrusted data.
  • EXTERNAL_DOWNLOADS (HIGH): The skill requests Bash(pip:*) permissions, enabling the download and installation of third-party packages from external registries. This introduces supply chain vulnerabilities and the potential for executing malicious dependencies.
  • PROMPT_INJECTION (HIGH): As a tool for model explainability, this skill is designed to ingest external models and datasets (Category 8). The lack of specified boundary markers or input sanitization, combined with the requested high-privilege tool access, creates a critical vulnerability to indirect prompt injection attacks where data content can control agent behavior.
  • METADATA_POISONING (MEDIUM): The trigger section contains redundant keywords ('model explainability tool' repeated), which is a deceptive metadata pattern likely intended to manipulate agent activation or selection priority.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 07:38 AM