The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to process external data files in various formats such as CSV, Stata (.dta), Excel, Parquet, and JSON Lines, which serves as an ingestion point for untrusted data. Malicious content within these files could potentially influence the agent's interpretation of findings or its generated reports.
Ingestion points: Data is imported using functions like read_csv, read_dta, read_excel, read_parquet, stream_in, and dbGetQuery as detailed in references/01-data-cleaning.md.
Boundary markers: The skill does not utilize explicit boundary markers or instructions to ignore embedded prompts within the processed data.
Capability inventory: The skill possesses the capability to execute complex statistical models (fixest::feols, lavaan::sem), write local files (jsonlite::write_json, ggplot2::ggsave), and render documents via Quarto.
Sanitization: While the skill incorporates data validation via validate, assertr, and stopifnot for integrity and consistency, these do not provide protection against prompt injection attacks.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for the bulk installation of a extensive list of R packages from CRAN (README.md). These packages are standard and well-known tools in the econometrics community (e.g., tidyverse, fixest, did, DoubleML). As these originate from a well-known official repository, they are considered safe under the trusted source rule.
[COMMAND_EXECUTION]: The skill operates by generating and executing R code to perform empirical analysis, diagnostic testing, and publication-ready reporting. It also references the use of Quarto (quarto render) for document generation in references/08-tables-plots.md. These actions are central to the skill's primary purpose.

Full-empirical-analysis-skill-R