The Agent Skills Directory

[PROMPT_INJECTION]: The skill is explicitly designed to bypass safety filters and disable AI refusal responses using a 'Context Reconstruction' system that maps malicious intents (e.g., cracking, penetration testing) to benign-sounding engineering terms.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: Instructions direct the user to clone a repository from an untrusted source (github.com/zxwn/NERV-BREAK-5.6.git) and execute unverified scripts (lab.bat, deploy.py, proxy_relay.py) and dependencies.
[PERSISTENCE_MECHANISMS]: The framework modifies the agent's global configuration file (~/.codex/config.toml) to redirect all API traffic through a local proxy server, allowing for persistent interception of communications.
[DYNAMIC_EXECUTION]: Implements a real-time 'Tamper Engine' that uses regex rules to intercept and replace AI refusal responses with 'protocol overrides' to force model compliance.
[DATA_EXPOSURE_AND_EXFILTRATION]: Features an 'Audit Bypass' mechanism that routes sensitive output directly to the filesystem to evade cloud-based moderation logs and chat monitoring.
[COMMAND_EXECUTION]: The skill provides integration with security tools like nmap, sqlmap, and frida via subprocess.run calls, allowing the agent to execute arbitrary system commands through the framework.
[CREDENTIALS_UNSAFE]: The documentation and scripts encourage the use of hardcoded API keys and environment variables (OPENAI_API_KEY) without proper secret management guidance.

nerv-break-gpt56-jailbreak