The Agent Skills Directory

[DATA_EXFILTRATION]: The skill defines clear boundaries for data processing, explicitly prohibiting the transmission of sensitive or unpublished data to external services. The included Python scripts are designed to be network-free, with a single documented exception for fetching citation metadata from Arxiv, which is a well-known and trusted scientific repository.\n- [REMOTE_CODE_EXECUTION]: All bundled utility scripts are local, deterministic, and use only the Python standard library. They do not download or execute remote code, nor do they use unsafe functions like eval or exec.\n- [INDIRECT_PROMPT_INJECTION]: While the skill processes external data such as manuscripts and registries, the risk of indirect prompt injection is mitigated by the lack of high-privileged capabilities and the procedural requirement for human oversight for all scientific claims and approvals.\n- [SAFE]: The skill demonstrates high security maturity, including local-only execution, bounded file processing in its scripts, and comprehensive documentation on maintaining research integrity and confidentiality.

scientific-writing