deploying-machine-learning-models

Warn

Audited by Gen Agent Trust Hub on Mar 12, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill configuration allows unrestricted bash command execution via the Bash(cmd:*) tool. This capability is intended for CI/CD and infrastructure tasks but grants the agent broad control over the host environment.\n- [REMOTE_CODE_EXECUTION]: Documentation for scripts/deploy_model.py indicates it performs automated dependency installation, which involves downloading and executing third-party code at runtime.\n- [PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it processes user requirements to generate deployment code and infrastructure configurations. Mandatory Evidence: 1. Ingestion points: User deployment requests and specifications. 2. Boundary markers: Absent in the provided definition. 3. Capability inventory: Bash, Write, and Edit tools (SKILL.md). 4. Sanitization: No sanitization or validation of external context is documented.\n- [EXTERNAL_DOWNLOADS]: The deployment process involves interacting with cloud platforms and container registries to retrieve templates and configure serving environments.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 12, 2026, 01:36 AM