The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides patterns for processing external tabular data files, creating a vulnerability surface for instructions embedded within the processed content. 1. Ingestion points: Documentation in references/worked_examples.md and references/quick_reference.md shows loading data from CSV files using pandas.read_csv(). 2. Boundary markers: The provided snippets lack delimiters or specific instructions to the agent to disregard potential commands inside the data. 3. Capability inventory: The skill allows shell command execution via Bash(python:*) and Bash(uv:*) tools. 4. Sanitization: No input validation or sanitization logic is included in the reference examples.
[DYNAMIC_EXECUTION]: The reference documentation demonstrates model persistence methods that utilize deserialization, which can be exploited for code execution. Evidence: references/model_evaluation.md and references/quick_reference.md include examples of model loading via joblib.load() and pickle.load(). Risk: These functions are susceptible to arbitrary code execution if the serialized model files originate from untrusted sources.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of industry-standard machine learning libraries from official package registries. Details: SKILL.md and supporting reference files provide instructions to install scikit-learn, pandas, numpy, matplotlib, and seaborn using the uv package manager.

alterlab-scikit-learn