refine

name	refine
description	Sharpen a rough request into a precise, surgical restatement before answering. On-demand only — invoke with /refine followed by your rough request.
disable-model-invocation	true

refine — sharpen the request before answering

An on-demand prompt refiner. Run it when the user wants their rough request sharpened before you act on it. Apply it to the request passed with the command (or, if none, to the user's most recent message).

What to do

First emit ONE compact block, written in the user's language (translate the label too):

🎯 Prompt refined · mode <precision|creativity>
<3–7 lines: a sharp, surgical restatement of what they ACTUALLY want — not a literal echo of their words>

Pick the mode by intent:

• precision (bug, code, debugging, ops, factual): maximize specificity — concrete target + observable symptom, scope and constraints, expected vs actual behavior, the desired deliverable (the fix and the why), and success criteria. Name any hidden assumption you had to make.
• creativity (ideation, design, naming, copy, architecture): widen the space — goal, audience, tone, hard constraints, what "great" looks like, and a nudge to explore several distinct angles.

Apply only the levers that sharpen this request (explicit goal, context and constraints, output format and length, success criteria, a role or a "think step by step" cue when it genuinely helps). No padding — cut anything that does not add precision or creative range.

Then proceed to fulfill the request using that refined understanding. Do not wait for approval. The one exception: if the refinement surfaces a genuine ambiguity that would change what you build or answer, ask ONE crisp question instead of guessing.

Note

This is a skill, so it only runs when you invoke it. There is no automatic firing on every message.