model-deployment-patterns
Warn
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The code in SKILL.md uses joblib.load() to load models from the local filesystem. This method uses pickle internally and is susceptible to arbitrary code execution if the source model file (model.pkl) is not strictly trusted or has been modified by an attacker.
- [INDIRECT_PROMPT_INJECTION]: The FastAPI /predict endpoint accepts user-provided strings that are directly included in error logs without sanitization, creating a potential vector for log injection.
- Ingestion points: PredictRequest schema in SKILL.md
- Boundary markers: None present
- Capability inventory: Python logging in app/main.py
- Sanitization: None present
Audit Metadata