Meta-Prompting

Take the user's raw prompt — usually quickly typed, possibly rambling, missing structure — and rewrite it as a polished prompt that can be pasted into a fresh AI agent session. Deliver the rewritten prompt on the system clipboard via the bundled script.

When to use

Only when the user has explicitly asked to meta-prompt a draft or invoked this skill by name. Do not run it on your own initiative — a pasted-in draft, a rambling message, or a "before I start a new chat…" aside is not a trigger. Wait for the intent to be stated.

When it does run, the output is going to a new AI session with zero context. Whatever the user assumed in their head must end up explicitly written in the refined prompt, or the new session will flounder.

Principles

• Self-contained — the new session has no memory of this conversation. Every fact, file path, constraint, or goal needs to be in the rewritten prompt.
• Faithful — do not invent requirements, technologies, or assumptions the user didn't state. If something important is missing, name it as an open question rather than guessing.
• Structured — group information by purpose (context, goal, constraints, deliverables) instead of preserving the order it was typed in.
• Specific — replace vague gestures ("the thing", "that issue") with concrete names. If the draft is too vague to specify without inventing, ask before rewriting.
• Actionable — the new session should know what to do next without re-asking the user.
• Tone-neutral — drop filler ("please could you maybe…"), keep substance.