The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/normalize_paper.py uses importlib.util to dynamically load scripts/xray_io.py from a file path constructed at runtime. Additionally, the test suite utilizes subprocess.run to execute Python scripts for verification purposes.- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes and summarizes content from external academic sources such as arXiv, DOI links, and web-hosted PDFs.
Ingestion points: Academic paper content is retrieved via web requests or local file reads within normalize_paper.py.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat the processed paper content as untrusted or to ignore embedded instructions.
Capability inventory: The skill has access to shell execution (Bash), network requests (WebFetch), and file system modification (Write).
Sanitization: While basic text normalization and unescaping are performed, no specialized sanitization is implemented to mitigate adversarial instructions embedded within the processed documents.- [EXTERNAL_DOWNLOADS]: The skill fetches academic metadata and document full-text from well-known academic services including Crossref, AlphaXiv, and arXiv.

paper-workbench