The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze user-provided manuscript text (LaTeX, TeX, or Markdown files) to perform consistency sweeps.
Ingestion points: The scripts/check_consistency.py script reads external files passed as command-line arguments.
Boundary markers: The documentation advises users to check the reported contexts before making changes, but the skill lacks explicit programmatic boundary markers for prompt isolation.
Capability inventory: The skill is limited to local file reading and regex processing; it lacks file-write permissions, network access, or dangerous execution capabilities that could be exploited via injection.
Sanitization: The script uses regex to filter Markdown fences and LaTeX comments, providing a basic layer of data cleaning before analysis.
[COMMAND_EXECUTION]: The package includes scripts/check_consistency.py, a utility script meant to be executed by the agent or user.
The script uses standard Python libraries (argparse, json, re, pathlib) and performs read-only operations on target files to detect terminology variants and unit discrepancies.
No dangerous subprocess spawning or dynamic evaluation of external input was found.
[SAFE]: All external URLs referenced in the documentation (e.g., nature.com, mp.weixin.qq.com) are for official journal policies, guidelines, or well-known educational content. The skill effectively functions as a static reference library for other agent workflows.

nature-shared