The Agent Skills Directory

[PRIVILEGE_ESCALATION]: Multiple scripts, including builder-skill_analyze_cpu.py, builder-skill_analyze_thermal.py, and builder-skill_optimize_system.py, attempt to execute system commands with sudo. These are used for legitimate tasks such as reading thermal sensors via powermetrics and applying optimizations like purge or pmset.
[COMMAND_EXECUTION]: The skill makes extensive use of subprocess.run and asyncio.create_subprocess_shell to interact with the host operating system. Commands executed include system utilities like sysctl, system_profiler, ioreg, and various cleanup commands that modify system state.
[DATA_EXFILTRATION]: The system analysis tools collect and aggregate detailed telemetry about the host environment. This includes hardware specifications, resource utilization, disk partition layouts, and a full list of active network connections (sockets, local/remote addresses, and associated PIDs) via psutil.net_connections.
[INDIRECT_PROMPT_INJECTION]: The skill contains several components designed to process untrusted external data, such as builder-skill_debug_code.py (processing error messages and stack traces) and builder-skill_scaffold_test.py (parsing source code files). This ingestion of external data combined with the skill's high-privilege capabilities creates a surface for indirect prompt injection.
[DYNAMIC_EXECUTION]: The builder-skill_optimize_system.py script identifies performance issues and dynamically selects shell commands to remediate them. Although these commands are selected from predefined templates and typically include user confirmation prompts, the runtime generation and execution of shell commands is a high-capability feature.

builder-skill-uvscript