nature-shared

Pass

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill provides a script (scripts/check_consistency.py) and instructions (core/consistency-sweep.md) that ingest and process external manuscript data to detect terminology and numeric drift.
  • Ingestion points: The check_consistency.py script reads text from files provided via command-line arguments (e.g., manuscript.tex).
  • Boundary markers: The script partially sanitizes input by ignoring content inside Markdown fences and unescaped LaTeX comments, but does not use explicit delimiters for the primary text blocks.
  • Capability inventory: The script is limited to read-only file access and standard library regex operations; it does not perform network operations or write to the filesystem.
  • Sanitization: The script performs raw text extraction without specialized sanitization, but its output is restricted to diagnostic warnings and evidence snippets.
  • [SAFE]: The skill contains informative links to the official Nature Portfolio website (nature.com) for editorial policy references. These references target a well-known and trusted academic publishing service.
  • [SAFE]: The Python script scripts/check_consistency.py uses only Python standard libraries (argparse, json, re, pathlib, decimal) and performs no dangerous system calls or remote executions.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 28, 2026, 06:24 AM
Security Audit — agent-trust-hub — nature-shared