Optimize a rough prompt using a real structure
Most rough prompts are missing role, constraints, or a clear success criterion — not clever wording.
- Use it for
- Anyone with a rough draft of a prompt who wants it restructured to actually perform better, not just reworded.
A rough prompt usually underperforms not because the wording is clumsy, but because it's missing a structural piece — no clear role for the model to take, no real constraint on the output, no way to tell if the response actually succeeded. Rewriting it with those pieces explicit tends to matter more than polishing the phrasing of what's already there.
When not to use this
If the prompt already has a clear role, constraints, and success criterion and just needs a wording pass, this full structural rework is unnecessary — a lighter edit will do.
Fill in
Here's my rough draft of a prompt: {{rough_prompt}}
Rewrite this using a real structure — a clear role for the model, the
actual constraints that matter, and (for anything non-trivial) a
step-by-step approach the model should follow. Tell me specifically
what structural piece was missing from my draft, not just that the
new version reads better.Why it works
Requiring the specific missing structural piece to be named, not just a rewritten prompt, is what makes this a repeatable skill instead of a one-off fix — knowing that your draft was missing a real constraint (target audience, length, tone) is something you can apply to the next rough prompt yourself, not just this one.
Common mistakes
- ×Accepting a rewritten prompt without understanding what structural piece was actually missing.
- ×Adding structure the task doesn't need, over-engineering a genuinely simple request.
Follow-ups
- This works better — now optimize this related prompt the same way: {{related_prompt}}.
- The output still isn't quite right — help me figure out if it's the prompt or something I need to clarify about what I actually want.
Did this work?
You can name specifically what structural piece was missing from the original (role, constraints, a success criterion), not just that the new version "sounds better."
Tested on claude-opus-5. Evidence status is draft; it moves to battle-tested only on recorded runs, never by hand.