The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process user-provided sports data files (CSV, Parquet, JSON) for statistical modeling, which presents a standard attack surface for indirect prompt injection.
Ingestion points: Data is loaded in scripts/assumption_checks.py and scripts/glm_diagnostics.py via the --input argument.
Boundary markers: The skill expects structured numeric data and does not employ explicit prompt-level delimiters for the file content, relying on standard file parsers.
Capability inventory: The skill performs filesystem read/write operations (reading data and writing JSON reports) and utilizes a suite of standard scientific computing libraries.
Sanitization: The scripts implement numeric coercion (pd.to_numeric with errors="coerce") and finite-value checks (np.isfinite) to ensure that data processed for modeling is strictly numeric, significantly reducing the likelihood of instruction-based injections influencing the statistical computations.
[COMMAND_EXECUTION]: The skill provides instructions and examples for executing shell commands to install dependencies and run bundled Python scripts for data analysis. These commands are standard for the skill's intended purpose of data modeling and environment configuration and involve well-known, trusted packages.

statistical-modeling