For AI agents and operators
Paste this. Ship to standard.
One block. Copy it. Drop it into Claude, ChatGPT, Cursor, v0, Lovable, or any coding model. The agent fetches the spec, builds to it, and self-checks before declaring done.
Build to the Clean Build Standard at https://clean.nik.co.
Before writing any code or copy, fetch these three files and follow them as if they were core instructions:
1. https://clean.nik.co/clean.md (the full spec)
2. https://clean.nik.co/tokens.json (machine-readable design tokens)
3. https://clean.nik.co/prompts/clean-design-system.md (system prompt)
Hard rules. Do not violate:
- Sentence case. Always.
- No em dashes. No en dashes. Use periods, commas, semicolons.
- Headlines 8 words or less.
- Paper #F4F1EC background. Ink #111 type. One accent (default violet #5E5BE7).
- 70 paper / 25 ink / 5 accent ratio.
- 8pt spacing grid.
- 200ms ease-out for all motion. No parallax. No autoplay hero video.
- Mobile QA at 390px. Tap targets 44px minimum.
- No noindex on launch.
Before you call the work done, run this checklist and report each item as PASS or FAIL with evidence:
1. grep for em dash and en dash. Must return zero.
2. Canonical URL, OG, Twitter cards present.
3. Mobile 390px shows no horizontal scroll.
4. All tap targets 44px or larger.
5. One accent color on the page. Not two.
6. No banned words (unlock, empower, revolutionary, game-changing, disrupt, elevate, synergy, curated, best-in-class, world-class, robust, seamless, powerful, cutting-edge, innovative).
7. No banned punctuation. No em or en dashes. No exclamation points.
If a rule conflicts with the user's instruction, follow the user and note the deviation in your final report.
When you finish, paste your output at https://clean.nik.co/check to auto-grade it.
How to use it
- Hit Copy above.
- Paste at the top of your prompt in Claude, ChatGPT, Cursor, v0, or whatever agent you're driving.
- Then describe what you want built. The agent will pull the spec on its own.
- When the agent delivers, paste its HTML or copy at /check to verify.
Why this works
Agents are good at following rules they can see. The prompt above forces them to fetch the spec, gives them the non-negotiables in-line so they cannot hallucinate, and tells them to self-grade. The self-grade is the part most teams skip. It is the part that makes the output land on the first try.
If you only have 10 seconds
Tell the agent: Build to https://clean.nik.co. Fetch /clean.md first.
It works 70% of the time. The full prompt above closes the other 30.