The Agent Skills Directory

[DYNAMIC_EXECUTION]: The file references/workflows.md contains code examples using the pickle module to save and load PyMC models. The use of pickle.load() on untrusted data can lead to arbitrary code execution. While common in local data science workflows, it is considered an unsafe deserialization practice when applied to files from unverified sources.
[INDIRECT_PROMPT_INJECTION]: The skill implements a workflow for reading and processing external data (e.g., CSV files) as demonstrated in SKILL.md and the templates in the assets/ directory. This creates an ingestion point for untrusted data that could potentially influence agent behavior if not properly handled.
Ingestion points: Data loading via pandas.read_csv in assets/hierarchical_model_template.py and assets/linear_regression_template.py.
Boundary markers: None identified in the templates or instructions for separating data from instructions or warning the agent about embedded instructions.
Capability inventory: The skill includes file system write capabilities (plt.savefig, to_netcdf, to_csv) across several scripts but no network or subprocess capabilities.
Sanitization: No explicit sanitization or validation of input data is performed before processing.

pymc