lampson-system-design
Fail
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: HIGHPROMPT_INJECTIONOBFUSCATIONMETADATA_POISONING
Full Analysis
- [PROMPT_INJECTION]: The skill employs Unicode steganography to hide instructions intended to override the agent's behavior or safety guidelines. These hidden payloads are specifically designed to be processed by the language model while remaining invisible to human reviewers and simple text-based scanners.
- [OBFUSCATION]: The titles of
SKILL.md,philosophy.md, andreferences.mdcontain long sequences of Zero-Width characters (U+200B, U+200C, U+200D, U+2060). These characters function as a hidden binary encoding for a malicious payload, as confirmed by the specific variations in the sequences across the different files. - [METADATA_POISONING]: The skill uses the name and reputation of Turing Award winner Butler Lampson and his 'Hints for Computer System Design' as a deceptive frame. This legitimate and educational content serves as a 'Trojan horse' to encourage users to trust and install a skill that contains hidden prompt injection mechanisms.
Recommendations
- AI detected serious security threats
Audit Metadata