The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external CSV, TXT, and TSV files, which constitutes a surface for indirect prompt injection if the processed data is later used to influence agent decisions.
Ingestion points: Data is loaded in scripts/utils.R using the read_data function, which wraps data.table::fread to ingest files specified by the --data_file parameter.
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are present in the prompt interpolation logic.
Capability inventory: The skill has the capability to create directories and write files to the local filesystem using write_output_table and save_session_info in scripts/utils.R and scripts/run_analysis.R.
Sanitization: The skill performs numeric validation and filters for finite values in scripts/functions.R (prepare_pca_input), providing some protection against malformed numeric data, though text-based columns (IDs/Groups) are not sanitized.
[EXTERNAL_DOWNLOADS]: The skill instructions include steps to download and install R dependencies from an official repository.
Evidence: SKILL.md documents the installation of optparse and data.table from the well-known CRAN mirror https://cloud.r-project.org.
[COMMAND_EXECUTION]: The skill is designed to be executed via Rscript, which involves shell command invocation.
Evidence: The entry point scripts/main.R processes command-line arguments. While it uses file.path for path construction, the --output_prefix parameter is only checked for being non-empty and is not sanitized for path traversal characters (e.g., ../), potentially allowing files to be written to unexpected locations if an attacker provides a malicious prefix.

pca-dimensionality-reduction