The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data in the form of R package source code, metadata, and documentation, which could potentially contain malicious instructions intended to influence agent behavior.\n
Ingestion points: The agent reads and analyzes R/*.R files, the DESCRIPTION file, and README.md/README.Rmd files.\n
Boundary markers: There are no explicit delimiters or specific instructions for the agent to ignore instructions embedded within the reviewed files.\n
Capability inventory: The agent has the capability to execute R commands (e.g., usethis, devtools) and perform file-write operations to implement proposed fixes.\n
Sanitization: No specific sanitization or filtering of the content read from the package files is mentioned.\n
Mitigation: The workflow requires explicit user approval before the agent implements any suggested changes, serving as a human-in-the-loop safety checkpoint.\n- [COMMAND_EXECUTION]: The skill directs the agent to execute standard R development commands such as usethis::use_news_md() and devtools::build_readme(). These are legitimate operations for package maintenance and are used within the intended scope of the skill.\n- [EXTERNAL_DOWNLOADS]: The skill uses the urlchecker package to perform network requests for validating URLs within the R package documentation. This is a diagnostic function used to ensure link integrity for CRAN submission and follows standard R development practices.

r-cran-extrachecks