Apply loop-engineering controls to an explicitly goal-directed or skill-invoked request, classify execution risk, assign approval-preserving maker-checker roles, and turn the request into a ready-to-run Codex goal. Use when the user asks to use goal mode, create/start/shape a goal, turn a normal request into an executable loop, decide whether a task is ready for goal execution, or design a bounded delegation and independent-review policy. Do not silently start a goal for ordinary requests unless the user explicitly invokes goal mode or this skill for execution.
