The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several command-line utilities to perform its verification logic. It uses rg and rga for searching local files and PDF contents, gws to interact with Google Drive, and pixi to dynamically run a Python script for extracting footnotes from DOCX documents.
[EXTERNAL_DOWNLOADS]: Retrieves a BibTeX reference file (paperpile.bib) and source PDF documents from Google Drive using the gws tool.
[REMOTE_CODE_EXECUTION]: Executes SQL queries on a remote database server via an SSH tunnel (ssh wrds) to the Wharton Research Data Services (WRDS) platform at wharton.upenn.edu. This is a legitimate research database connection.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface (Category 8) because it extracts text from user-provided manuscripts and uses it in RAG queries and search parameters. Ingestion points: Text content from manuscript.docx or markdown files. Boundary markers: No delimiters or warnings are used in the prompt templates sent to RAG systems like NLM or Readwise. Capability inventory: Subprocess execution for tools like ssh and gws, local file read/write, and network access. Sanitization: There is no evidence of sanitization or escaping of the extracted footnote text before it is used in CLI commands or LLM prompts.

source-verify