The Agent Skills Directory

[SAFE]: The skill implements a standard machine learning workflow using R and the well-known LightGBM library. The scripts follow secure coding practices, utilizing explicit error handling (tryCatch) and custom error codes (e.g., SKILL_MISSING_INPUT, SKILL_INVALID_DATA) to prevent stack trace leakage and handle failures gracefully.
[DATA_EXPOSURE]: The skill includes proactive safety features in scripts/functions.R, specifically the warn_on_suspicious_features function. This function identifies and warns the user if identifier-like columns (such as patient_id, sample_id, or accession numbers) are included in the predictors, reducing the risk of accidental PII leakage in exported artifacts.
[COMMAND_EXECUTION]: Shell execution is limited to standard Rscript invocations of the provided local scripts. Parameters passed to these scripts are validated for type and range (e.g., test_size must be between 0 and 0.5) before processing.
[INDIRECT_PROMPT_INJECTION]: While the skill ingests untrusted data via --data_file, it treats the content strictly as tabular data for modeling. It does not perform any dynamic execution or prompt interpolation based on the file's content. The skill further provides a 'remediation table' and 'quality flags' to ensure the agent and user can distinguish between valid models and degenerate results caused by poor data quality.

lightgbm-analysis