The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides instructions and code for ingesting data from external files (CSV, Excel, JSON, Parquet) and web scraping, which are potential surfaces for indirect prompt injection if the source data contains malicious instructions intended to influence the AI agent.\n
Ingestion points: Multiple functions in references/import-io.md (e.g., read_csv, read_excel, rvest::read_html) and references/data-table.md (fread).\n
Boundary markers: As a reference skill, it does not implement runtime boundary markers, but it recommends defensive programming practices.\n
Capability inventory: The skill is configured with Read, Write, and Bash tool permissions in the frontmatter.\n
Sanitization: The documentation advises using explicit schema definitions (col_types) and data cleaning utilities like janitor::clean_names().\n- [EXTERNAL_DOWNLOADS]: The scripts/check-r-facts.py utility script performs network requests to verify the availability of recommended R packages on an external registry.\n
Evidence: The script uses urllib.request to query https://crandb.r-pkg.org/, which is a well-known service for R package metadata.\n- [COMMAND_EXECUTION]: The skill includes a shell script for integrity testing and a Python script for fact-checking, both of which execute standard system commands and interpret project metadata.\n
Evidence: tests/run.sh executes grep, sed, and python to run structural and content-currency checks.

r-ops