The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/convert_to_docx.py uses subprocess.run to interact with the system's winget package manager and the pandoc document converter.
Evidence: subprocess.run(["winget", "install", "--id", "JohnMacFarlane.Pandoc", ...]) in scripts/convert_to_docx.py is used to automate the installation of a legitimate document utility.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to fetch research data from PubMed and LitSense.
Evidence: Queries are sent to eutils.ncbi.nlm.nih.gov and www.ncbi.nlm.nih.gov in scripts/batch_search.py and scripts/remap_refs.py. These are well-known official services for medical research.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided Markdown files (academic papers) to extract knowledge points for search queries.
Ingestion points: The skill reads full paper text to identify sentences requiring citations, as described in SKILL.md Step 2.
Boundary markers: The workflow defines strict 'Citation zones' (e.g., Introduction and Discussion only for Research Articles) to limit the scope of processed text.
Capability inventory: The skill has file read/write access and network access to PubMed API endpoints.
Sanitization: The skill extracts core concepts and MeSH terms to build structured JSON queries, providing a layer of validation between user input and external searches.

find-paper-references