model-deployment
Pass
Audited by Gen Agent Trust Hub on Apr 4, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTIONCREDENTIALS_UNSAFEREMOTE_CODE_EXECUTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill includes code templates using
joblib.load()to load machine learning models from disk. This performs deserialization which can execute arbitrary code if the model files are untrusted. - [COMMAND_EXECUTION]: Provides instructions and shell scripts for container builds, Kubernetes orchestration, and dependency management using
docker,kubectl, andpip. - [EXTERNAL_DOWNLOADS]: Fetches base images (e.g.,
python:3.11-slim,redis:alpine) and third-party libraries from public registries during the build and deployment workflow. - [INDIRECT_PROMPT_INJECTION]: API endpoints ingest numerical features from external requests. The skill mitigates this attack surface by implementing Pydantic models with strict validation logic.
- [CREDENTIALS_UNSAFE]: Includes a default credential (
admin) in a Grafana configuration template intended for local development environments.
Audit Metadata