The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data files (expression matrices and group files) which could potentially contain malicious instructions meant to influence the agent's behavior during or after the analysis.
Ingestion points: Data is ingested through load_expression_data in scripts/functions.R and data.table::fread in scripts/run_analysis.R.
Boundary markers: No specific boundary markers or 'ignore embedded instructions' warnings are implemented to encapsulate the user-supplied CSV data.
Capability inventory: The skill has the capability to write files to the local file system (write_imputed_matrix in scripts/run_analysis.R) and create directories (ensure_output_dir in scripts/utils.R). It does not perform network operations.
Sanitization: The skill performs technical validation, such as checking that sample columns are numeric (normalize_sample_column) and verifying feature IDs, but it does not sanitize against natural language instructions embedded in the data.
[EXTERNAL_DOWNLOADS]: The skill documentation (SKILL.md) provides instructions for the user to install the DMwR2 package from the official CRAN mirror on GitHub using remotes::install_github("cran/DMwR2"). This is a standard procedure for R packages not currently hosted on the main CRAN repository.
[COMMAND_EXECUTION]: The skill is designed to be executed via Rscript with user-defined command-line arguments. These arguments are processed using the optparse library, which provides a layer of safety against basic shell injection by handling arguments through a structured parser.

knn-imputation