The Reverse Tornado - running a goal as a self-correcting loop

This skill is a workflow. Give it any goal. It sets the goal up so an LLM can do most of the work while a human keeps the direction.

Picture a reverse tornado. On day one you guess wide. Each loop narrows the guessing into known work. A wall you cannot cross bounds every loop - that wall is the anti-goal. The loop stops when the metric hits its target.

Apply the steps below in order. Do not skip the anti-goal. It is what makes the rest safe.

Choose the operating mode

Default to planning mode. Draft the frame and loop in the requested output only. Do not run helpers, create .okra, spawn workers, or call another model CLI merely because the skill loaded.

Use execution mode only when the user asks to run, delegate, automate, or resume the loop, or when a repository's own instructions require the operating workflow for the requested change. Before acting, keep the footprint clear: