regex-vs-llm-structured-text

Pass

Audited by Gen Agent Trust Hub on Mar 24, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill provides educational content and Python implementation examples for text processing. No external network requests (outside of the conceptual LLM integration), unauthorized file system modifications, or credential harvesting patterns were identified.
  • [PROMPT_INJECTION]: The validate_with_llm implementation demonstrates an architectural surface for indirect prompt injection by interpolating untrusted input text directly into an LLM prompt. As this is presented in an educational context for document parsing, it is noted as a risk factor rather than a malicious intent.
  • Ingestion points: Untrusted source text enters the system via the content parameter in the process_document function within SKILL.md.
  • Boundary markers: The prompt template in validate_with_llm does not utilize boundary delimiters (e.g., XML tags) or instructions to ignore embedded commands.
  • Capability inventory: The skill's scripts are restricted to data parsing and dataclass manipulation; no subprocess, shell, or sensitive API capabilities are exposed.
  • Sanitization: The code does not implement sanitization or escaping of the input text before prompt interpolation.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 24, 2026, 07:43 AM
Security Audit — agent-trust-hub — regex-vs-llm-structured-text