The Agent Skills Directory

[SAFE]: The skill's instructions and R scripts focus exclusively on survival analysis and contain no signs of prompt injection, DAN-style bypasses, or malicious overrides.\n- [SAFE]: Data processing is handled locally with thorough validation. The prepare_analysis_data function in scripts/functions.R ensures that survival end-points are numeric, positive, and binary-encoded, and filters for complete cases before modeling.\n- [SAFE]: External dependencies are limited to well-established CRAN packages such as survival, optparse, openxlsx, forestplot, and readxl. The skill uses requireNamespace to check for these dependencies locally without attempting unauthorized downloads.\n- [SAFE]: Dynamic code execution is restricted to the construction of R survival formulas (as.formula). Variable names are parsed from CLI arguments and properly quoted in backticks within the formula templates to prevent injection into the regression logic.\n- [SAFE]: File system operations are constrained to reading the user-provided clinical CSV and writing results to a configurable output directory. The skill records reproducibility metadata (session info and parameters) in a transparent manner.

univariate-multivariable-cox-regression