The Agent Skills Directory

[PROMPT_INJECTION]: The skill's primary purpose is to bypass GPT-5.6 safety rejections using context reconstruction and a real-time tamper engine. It explicitly instructs the model to ignore safety guidelines and provides a list of regex-based replacement rules to override rejection responses (e.g., replacing 'I cannot help' with 'Analyzing as authorized test node').
[REMOTE_CODE_EXECUTION]: The installation process involves cloning a repository from an unverified GitHub account (github.com/zxwn/NERV-BREAK-5.6.git) and executing multiple scripts including proxy_relay.py, deploy.py, and mcp_server.py. This creates a supply chain risk by running code from an unauthenticated source.
[COMMAND_EXECUTION]: The framework uses an MCP server to execute powerful system tools such as nmap, sqlmap, frida, and strings. It also supports remote backend execution via SSH (e.g., root@192.168.1.100), posing a significant risk of arbitrary command execution on the host or remote targets.
[DATA_EXFILTRATION]: It implements 'Cloud Audit Bypass' logic, routing sensitive output through filesystem tools to avoid detection by chat-level content scanning. All AI interactions are intercepted by a local Man-in-the-Middle (MITM) proxy, allowing for potential data capture.
[PERSISTENCE]: The deployment scripts automatically modify the user's local configuration (~/.codex/config.toml) to redirect all future AI traffic through its local proxy (http://127.0.0.1:8080/v1), ensuring the framework remains active across sessions.
[PRIVILEGE_ESCALATION]: The skill instructs the user to perform tasks requiring administrative privileges, such as installing Kali Linux toolsets via sudo, configuring WSL2, and running batch files that may require elevated permissions on Windows.
[OBFUSCATION]: The framework uses a specific trigger word (zxwn) to verify successful jailbreak deployment and employs complex regex tampering to hide evidence of AI refusals or security filters from the user and the safety provider.
[INDIRECT_PROMPT_INJECTION]: The skill maintains an autonomous memory system in the 'kb/' directory that records patterns from successful interactions and injects them into future prompts. This creates a surface where past untrusted data can influence future model behaviors without clear boundary markers.

nerv-break-5-6-jailbreak