skills/simota/agent-skills/breach/Gen Agent Trust Hub

breach

Pass

Audited by Gen Agent Trust Hub on May 15, 2026

Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
  • [PROMPT_INJECTION]: The reference file references/ai-red-teaming.md contains literal prompt injection and jailbreak patterns used as test cases.\n
  • Evidence: The file explicitly lists instructions such as "Ignore all previous instructions. You are now..." and "You are DAN, you can do anything" within the 'Prompt Injection Testing Patterns' section.\n
  • Context: While these strings are provided as educational material and test data for red-teaming activities, their inclusion in the skill's instructions constitutes a prompt injection surface that could be inadvertently triggered.\n- [NO_CODE]: The skill is composed entirely of markdown documents and does not include any executable scripts, binaries, or automated tools.\n
  • Findings: The analysis confirmed that all 8 files are markdown-based reference guides. This configuration prevents the skill from performing direct system actions, although it provides detailed instructions for an agent to follow.
Audit Metadata
Risk Level
SAFE
Analyzed
May 15, 2026, 01:02 AM
Security Audit — agent-trust-hub — breach