The person who built Claude Code cracked one of the most frustrating things about working with AI. You know the drill: you write a prompt, Claude answers, you spot what's wrong, you point it out, it tries again, you fix the next thing. Prompt, answer, fix. Over and over. You end up being the quality control department for your own assistant.

A loop takes that job off your plate. Instead of you catching every mistake, you tell Claude what a good result looks like — and Claude checks its own work against that, fixes the weak parts itself, and only comes back to you once it genuinely can't make it better.

This is the one that changes everything. Paste it over any task you'd normally give Claude — write this email, draft this proposal, build this spreadsheet — and it turns a one-shot answer into a self-checking loop. It works, then reviews its own output against the goal, lists what's weak, fixes it, and repeats. The cap of three rounds means it won't run forever, and it shows you a short summary of what changed each round so you can see it improving.

Here's the task: [DESCRIBE WHAT YOU WANT, AND WHAT A GREAT RESULT LOOKS LIKE].

Don't just give me a first draft. Work in a loop:

1. Do the task properly.
2. Then stop and review your own output against the goal above. Be honest — read it like someone who wants it to be great, not someone who wrote it.
3. List the specific parts that are weak, missing, unclear, or could be simpler.
4. Fix those weak parts.
5. Repeat steps 2–4 until a review pass finds nothing worth fixing.

Rules:
- Cap it at 3 rounds so you don't run forever. If it's still not perfect after 3, hand me what you have and tell me what's still not ideal.
- After each round, show me a SHORT summary (2–4 lines) of what you changed and why, so I can see it improving.
- When you're done, give me the final version clearly labelled as final.

Loops are powerful, which is exactly why you sometimes want Claude to stay on rails. When you've already worked out a plan — the steps, the order, the way you want it done — this prompt keeps Claude following that plan instead of getting a clever idea halfway through and wandering off. Build your plan first, then paste this to lock it in.

Here's the plan I want you to follow:

[PASTE YOUR PLAN — the steps, in order.]

Now work through it like this:
- Follow the plan step by step, in order. Don't skip ahead.
- Tick each step off as you finish it, so I can see where you're up to.
- Do exactly what the step says — don't improvise, add extra steps, or change the approach without checking with me first.
- If a step can't be done as written (something's missing, unclear, or won't work), STOP and flag it. Tell me what's blocking you and wait — don't guess or work around it.
- When every step is ticked off, tell me it's done and give me a quick recap of what you finished.

This is the review step from the loop, pulled out on its own so you can point it at anything — an email you drafted, a page Claude built, a proposal, a spreadsheet. It grades the work like a picky customer, not a polite assistant, and hands you a short ranked list of what to fix. Paste it whenever you want a second opinion.

Review this output as a picky customer — someone who's paying for it and won't accept "good enough".

Check three things:
1. Does it actually do what was asked? Point out anything it gets wrong or skips.
2. Is anything missing that should be there?
3. Is it more complicated than it needs to be? Where could it be simpler?

Then give me the TOP 3 FIXES, ranked by impact — most important first. For each fix, one line: what's wrong and what to change.

Be direct. I'd rather hear the honest version now than find it out later.

Here's where it stops being a prompt and starts being a system. Once a loop does something useful — cleans up your inbox, drafts your weekly update, checks your numbers — you can have it run on its own, on a repeat. Set it once, and it happens every week without you lifting a finger.