The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute a local Python utility script (scripts/render_cites.py) to transform citation placeholders in drafts into numbered bibliographies and CSV evidence tables. The included test suite (scripts/test_render_cites.py) also utilizes subprocess.run to call this script in a controlled manner using the system's Python executable. These actions are restricted to internal skill components and serve their documented purpose.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data in the form of academic paper abstracts and metadata retrieved via the autoverse CLI tool. It implements a security boundary by requiring a separate verification step (batch command) for all paper identifiers (DOI, PMID, arXiv) before they can be cited in the final output, reducing the risk of processing fabricated or malicious metadata.

universe-research