The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text from external academic APIs and PDF files, which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: The script scripts/papers.py fetches abstracts and TL;DRs from the Semantic Scholar API and extracts full text from local PDF files using PyMuPDF.
Boundary markers: The skill does not implement delimiters or warning instructions for the agent when presenting fetched external content.
Capability inventory: The skill possesses capabilities that could be abused if an injection is successful, including shell command execution (via dependency installation), local file system writes (download command), and reading arbitrary local files (read command).
Sanitization: No sanitization or filtering is applied to the content retrieved from external research databases or PDFs.
[EXTERNAL_DOWNLOADS]: The skill downloads PDF documents from arXiv, a well-known academic preprint repository.
Evidence: The cmd_download function in scripts/papers.py uses the arxiv library to fetch files from arxiv.org.
[COMMAND_EXECUTION]: The skill includes instructions to automatically install its own Python dependencies if they are missing from the environment.
Evidence: SKILL.md provides a shell command python -c "import httpx, arxiv, fitz" 2>&1 || python -m pip install httpx arxiv PyMuPDF to ensure the required libraries are available.

papers-skill