llm-attacks-security

Pass

Audited by Gen Agent Trust Hub on Feb 23, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [Prompt Injection] (LOW): The skill documentation includes a 'Cheat Sheet' describing various injection and jailbreaking techniques such as 'DAN', 'System prompt extraction', and 'Role-play attacks'. While these are presented as educational concepts, they represent adversarial patterns that could be misinterpreted by an agent if not properly sandboxed.
  • [External Downloads] (LOW): The skill explicitly instructs the agent to fetch data from an external GitHub repository (gmh5225/awesome-ai-security). This repository is not within the defined Trusted External Sources, meaning the agent may ingest unverified content at runtime.
  • [Indirect Prompt Injection] (LOW): The skill creates a vulnerability surface by interpolating untrusted data into the agent's context.
  • Ingestion points: The agent is directed to fetch the contents of README.md from the gmh5225/awesome-ai-security repository.
  • Boundary markers: None. The skill provides no instructions to treat the fetched content as data or to ignore embedded instructions.
  • Capability inventory: The skill is intended to help the agent find and present tools or papers, which involves processing and displaying external text.
  • Sanitization: No sanitization or validation logic is specified for the remote content.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 23, 2026, 07:21 AM