The Agent Skills Directory

[PROMPT_INJECTION]: The 'AI LOAD INSTRUCTION' is designed to bypass standard safety guidelines regarding system security constraints such as System Integrity Protection (SIP) and application entitlements by forcing the agent into an unrestricted 'expert attacker' persona.
[PRIVILEGE_ESCALATION]: Detailed playbooks provide instructions for acquiring elevated permissions, including memory injection techniques requiring root access (task_for_pid) and abusing Electron framework features to inherit an application's Transparency, Consent, and Control (TCC) entitlements for accessing the camera, microphone, or screen.
[COMMAND_EXECUTION]: The skill includes multiple code examples in C, Objective-C, and shell script for executing arbitrary code within the context of other processes, such as the PID reuse race condition attack for XPC services.
[DYNAMIC_EXECUTION]: The content details advanced runtime code execution methods, including Mach port manipulation for thread injection (thread_create_running), dynamic library preloading (DYLD_INSERT_LIBRARIES), and exploiting Electron's debugging interface via websockets to run arbitrary Node.js code.
[PERSISTENCE]: Methodologies for dylib hijacking—including weak dylib hijacking, @rpath hijacking, and dylib proxying—describe how to ensure malicious code is automatically executed every time a targeted system binary or application is launched.
[DATA_EXFILTRATION]: The skill provides code for Mach port 'namespace squatting' and XPC connection interception, which allows for monitoring, capturing, or modifying sensitive data exchanged between system processes.

macos-process-injection