The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process machine learning datasets and model outputs which serve as ingestion points for untrusted data.
Ingestion points: Data enters the agent context through the load_breast_cancer dataset in scripts/tabular_report.py and via suggested file reads (e.g., CSV, DataFrames) in references/workflows.md and references/data-maskers.md.
Boundary markers: The instructions do not define explicit boundary markers or XML-style tags to isolate data from instructions.
Capability inventory: The skill includes the capability to write files (to_csv, savefig in scripts/tabular_report.py) and execute shell commands via uv (as seen in SKILL.md).
Sanitization: While the skill performs numeric validation (assertions and finiteness checks), it does not explicitly sanitize text-based data for potential prompt injection before processing.
[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute shell commands to set up environments and run reports.
SKILL.md and scripts/tabular_report.py document the use of uv venv, uv pip install, and uv run to execute local scripts and manage dependencies.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the shap package and its dependencies from PyPI, a well-known and trusted package registry.
Dependencies include standard libraries such as numpy, pandas, scikit-learn, and matplotlib, pinned to specific versions to ensure reproducibility and supply chain integrity.

shap