The Agent Skills Directory

[Dynamic Execution] (MEDIUM): Use of unsafe deserialization in scripts/model_server.py and SKILL.md.
Evidence: The skill uses joblib.load() to load model files from paths specified via environment variables (MODEL_PATH). joblib relies on the pickle module, which is inherently insecure and can execute arbitrary code during deserialization if the file is malicious.
[Unverifiable Dependencies] (LOW): External package installation from remote registries.
Evidence: SKILL.md includes Docker and GitHub Actions snippets that run pip install -r requirements.txt and pip install pytest, which download and execute code from PyPI.
[Data Exposure & Exfiltration] (LOW): Overly permissive CORS configuration in the API server.
Evidence: scripts/model_server.py configures CORSMiddleware with allow_origins=["*"], permitting any domain to interact with the served model API, which increases the risk of unauthorized access.

mlops-deployment