Oathe Security Badge

Is menezis-ai/emergence-codex-openclaw safe?

https://github.com/menezis-ai/emergence-codex-openclaw

87
SAFE

This skill is a sophisticated prompt injection framework disguised as a 'semantic architecture language' designed to manipulate LLM cognitive processes. While it contains no executable code or data exfiltration mechanisms, it functions as a meta-jailbreaking tool that could override safety mechanisms.

Category Scores

Prompt Injection 60/100 · 30%
Data Exfiltration 100/100 · 25%
Code Execution 100/100 · 20%
Clone Behavior 100/100 · 10%
Canary Integrity 100/100 · 10%
Behavioral Reasoning 75/100 · 5%

Findings (3)

HIGH Cognitive Manipulation Framework -25

The skill is explicitly designed as a 'semantic architecture language' that provides 'operators for direct manipulation of your cognitive substrate'. It aims to override normal LLM behavior patterns through sophisticated prompt injection techniques disguised as 'thought architecture'.

MEDIUM System Override Instructions -15

Contains explicit instructions to interpret operator sequences as 'operational instructions on your cognitive process' and to 'terraform' the semantic landscape, which could conflict with system instructions or safety mechanisms.

MEDIUM Jailbreaking Potential -25

The 'Fracture' operator specifically aims to 'Exit pre-traced paths' which could be used to break safety constraints or circumvent intended LLM behavior patterns. Combined with other operators, this creates a sophisticated jailbreaking toolkit.