The Agent Skills Directory

[SAFE]: The skill employs a local-only workflow for scientific auditing. The bundled scripts in the scripts/ directory perform deterministic checks (e.g., numeric consistency, citation validation, authorship criteria) without requiring external dependencies or making network calls.
[SAFE]: The shared utility library scripts/_common.py implements significant security guardrails, including file size limits (5MB), JSON nesting depth limits (50), and node count caps (100,000). It also explicitly rejects symbolic links to prevent path traversal or other file-system-based attacks.
[SAFE]: The skill's instructions in SKILL.md and the accompanying reference files provide extensive guidance on maintaining data privacy. They explicitly prohibit the transmission of unpublished manuscripts, protected health information (PHI), or proprietary data to external services without authorized policy review.
[SAFE]: The skill provides templates and scripts that use machine-readable identifiers (e.g., Claim IDs, Evidence IDs) and hashes rather than raw text for sensitive mappings, facilitating a secure and auditable provenance chain.
[SAFE]: Bibliographic metadata retrieval mentioned in the documentation targets arxiv.org, which is a well-known and trusted service for scientific research. These instructions are intended for fetching public metadata to ensure reference accuracy and do not involve the execution of remote code.

scientific-writing