The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The documentation file references/workflows.md provides code examples that demonstrate saving and loading models using the Python pickle module.
Evidence: The 'Saving and Loading Results' section includes code snippets using pickle.dump() and pickle.load().
Impact: The use of pickle.load() on data from potentially untrusted sources is unsafe because it can execute arbitrary code during the deserialization process.
[PROMPT_INJECTION]: The skill defines patterns for reading external data files, creating a surface for indirect prompt injection attacks where malicious instructions could be embedded in data.
Ingestion points: The assets/hierarchical_model_template.py and assets/linear_regression_template.py templates use pandas.read_csv() to load data into the modeling environment.
Boundary markers: The skill does not implement delimiters or specific instructions to the agent to ignore potential commands embedded within the input data.
Capability inventory: The skill has the capability to write files to the local system (plt.savefig, idata.to_netcdf), generate executable Python code, and perform complex statistical analysis.
Sanitization: There is no evidence of data sanitization or validation logic in the provided templates to prevent the processing of malicious content as instructions.

pymc