dean-large-scale-systems
Fail
Audited by Gen Agent Trust Hub on Mar 15, 2026
Risk Level: HIGHPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Detection of extensive zero-width Unicode characters used for obfuscation in the titles of
SKILL.md,philosophy.md, andreferences.md.\n - Evidence: Large blocks of invisible characters (U+200B, U+200C, U+200D, and U+2060) are appended to the markdown headers in all three files. The sequences are delineated by Word Joiners (U+2060) and Zero-Width Joiners (U+200D), forming a binary-like encoding scheme.\n
- Risk: This type of obfuscation is a primary vector for hidden prompt injection attacks, designed to deliver instructions to the AI agent that bypass safety guardrails or override established system prompts without the user's knowledge.
Recommendations
- AI detected serious security threats
Audit Metadata