nature-reader

Pass

Audited by Gen Agent Trust Hub on Nov 3, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [SAFE]: The skill instructions and referenced documents describe a legitimate workflow for academic paper translation and formatting. No patterns of prompt injection, data exfiltration, or obfuscation were found in any of the analyzed files. The outputs are standard Markdown and JSON metadata files.
  • [INDIRECT_PROMPT_INJECTION]: This category is noted because the skill processes untrusted documents from external sources (PDFs, URLs, and pasted text). However, the risk is mitigated by the skill's structure.
  • Ingestion points: Research papers provided via PDF files, publisher HTML, and scholarly identifiers like DOI and arXiv links (documented in SKILL.md and README.md).
  • Boundary markers: The workflow enforces structural alignment using specific markers like **Original:** and **中文:** for every substantive block, which helps separate data from the model's processing instructions.
  • Capability inventory: The skill's operations are limited to text processing and creating local file assets (Markdown, JSON, and images). It does not include subprocess execution, dynamic code generation, or network exfiltration capabilities.
  • Sanitization: The skill emphasizes faithful terminology preservation and conservative translation, reducing the likelihood of malicious content being interpreted as instructions.
  • [EXTERNAL_DOWNLOADS]: The skill facilitates access to well-known academic repositories and metadata resolvers (arXiv and DOI). These are standard, reputable services for scholarly content and do not present a security risk in the context of the skill's intended use.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 3, 2026, 02:46 AM