The Agent Skills Directory

[COMMAND_EXECUTION]: The skill contains multiple bash command templates intended for use when launching coding agents.
[PRIVILEGE_ESCALATION]: The provided command templates include flags that intentionally bypass security controls and isolation mechanisms:
The Claude harness is configured with --dangerously-skip-permissions, which removes standard permission checks.
The Cursor Agent harness is configured with --sandbox disabled, which removes process isolation protections.
[PROMPT_INJECTION]: The skill provides instructions that technically implement safety filter bypasses by requiring the use of flags such as --yolo and --approve-for-me, which disable user oversight and auto-approve potentially destructive actions.

favorite-harness