The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute bundled Python scripts (prepare_paper.py and audit_paper_card.py) using the python command to process document sources and verify the generated output. This is a core part of the skill's architecture for handling complex file parsing.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted research papers (PDFs, web content). It implements robust mitigations against injection attacks by enforcing a strict 'claim-evidence matrix,' requiring specific source pointers (e.g., PDF page numbers, figure/table IDs) for every claim, and using an automated auditor to ensure all content is grounded in the provided source text.
[DYNAMIC_EXECUTION]: Test files within the skill repository (tests/test_audit_page_ranges.py and tests/test_prepare_paper.py) utilize subprocess.run to invoke the skill's own Python scripts for regression testing and validation.
[EXTERNAL_DOWNLOADS]: The documentation provides instructions for installing the pymupdf library via standard package managers (pip) to support PDF processing capabilities.
[SAFE]: The skill exhibits high security maturity through its use of automated grounding checks, clear separation of agent-derived analysis from author-stated facts, and explicit boundaries on output scope to prevent the generation of unauthorized content types like social media copy or quizzes.

nature-paper-card