The Agent Skills Directory

[DYNAMIC_EXECUTION]: The file references/workflows.md contains code examples for saving and loading models using the pickle module. The pickle.load() function is inherently unsafe as it can execute arbitrary code if the input file has been tampered with. While the skill includes a comment warning users to only load trusted files, providing unsafe deserialization patterns in documentation is a security risk.
[INDIRECT_PROMPT_INJECTION]: The skill templates include patterns for reading external data files, which creates an entry point for untrusted content to influence the agent's context.
Ingestion points: The assets/hierarchical_model_template.py and assets/linear_regression_template.py files use pd.read_csv('data.csv') to load data for modeling.
Boundary markers: There are no explicit delimiters or instructions to ignore embedded commands within the processed data.
Capability inventory: The skill has access to Read, Write, Edit, and Bash tools, which could be leveraged if the agent is manipulated by data content.
Sanitization: No validation or sanitization of the input data is demonstrated before it is used in the modeling workflow.
[METADATA_POISONING]: The SKILL.md file identifies the author as "K-Dense Inc.", which does not match the provided author handle "eturkes". This inconsistency in metadata can be misleading to users regarding the source and maintenance of the skill.

pymc