model-versioning-manager

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [Indirect Prompt Injection] (HIGH): The skill is designed to process model versioning data and deployment configurations, which are often provided by external or untrusted sources. * Ingestion points: Metadata files, model versioning requests, and deployment manifests. * Boundary markers: No explicit delimiters or 'ignore' instructions are defined to separate data from instructions. * Capability inventory: Significant execution power via Bash, Write, and Edit tools. * Sanitization: No validation or filtering is specified for external content before tool usage.
  • [Command Execution] (MEDIUM): Requesting Bash access for 'automated assistance' allows for potential arbitrary command execution if the agent is influenced by malicious deployment patterns or instructions embedded in the data it processes.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 12:57 PM