The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill's instructions and code examples for production deployment recommend using joblib.load() to load serialized models and SHAP explainer objects. Since joblib relies on the pickle module, loading data from untrusted sources could result in arbitrary code execution on the user's system. Evidence found in SKILL.md and references/workflows.md in production-related examples.
[EXTERNAL_DOWNLOADS]: The skill instructions provide commands to install the shap package and its dependencies via the uv package manager. Evidence found in the Installation section of SKILL.md. This is a routine operation for using a legitimate third-party library.
[INDIRECT_PROMPT_INJECTION]: The skill defines a processing flow for external models and datasets, creating a surface for indirect instruction injection. 1. Ingestion points: External model objects and dataset samples like X_test processed in SKILL.md and references/workflows.md. 2. Boundary markers: No explicit markers or safety instructions are used to separate user-provided data from agent logic. 3. Capability inventory: The skill utilizes file-writing capabilities through joblib.dump and matplotlib.pyplot.savefig. 4. Sanitization: No data validation or sanitization mechanisms are implemented for the ingested data.

shap