Drift, drawn
Drift
The build wanders from the spec
Every prompt nudges the product a little further from what was agreed. ntent keeps one product model in the repo, and product, design and code all read from it.
See the product modelA build kit for coding agents
What was agreed,
in your repo.
ntent puts what was agreed into your repo: one contract your coding agent builds from, a dated record of each decision, and checks it runs before each commit.
Plain files in your repo. Tested with Claude Code.
The problem
Three ways AI‑built products break.
One fix for each.
None of this is new. Agents just make it happen faster, across more files, with more confidence.
Hallucination, drawn
Hallucination
The agent invents what isn’t there
A field that doesn’t exist, a colour typed in by hand, a form that saves without checking. Checks refuse the commit and name the fix. Each one is tested by breaking it on purpose.
See the checksRot, drawn
Rot
Nobody remembers why
Six months on, a fresh agent undoes a rule because the reason lived in a chat thread. ntent keeps each decision beside its rule, with who agreed and when.
See the decision recordsSetup
One prompt. Your agent does the rest.
- 01Paste the prompt into Claude Code, Cursor or any coding agent.
- 02It reads your repo and picks the setup that fits. A weekend prototype gets 8 steps; a product built to last gets more.
- 03It adds the files, runs every check, and tells you what it did and what it skipped.
Set this repo up with ntent.
Read https://ntent.app/llms.txt. If this folder has no app in it yet, create one first on the stack llms.txt names for a new project. Look at this repo, then pick the tier that fits it and any capabilities that are true of the product. Tell me which you picked and why in one line each, then fetch that plan and follow it exactly. If this repo already had an app in it before you started, add existing=1 to the plan URL.
Write every file verbatim and check its sha256 first. Run each step's verify line. Finish by telling me what you installed, what you skipped, and why.Reads https://ntent.app/llms.txt
After setup
Guides for the everyday work
Short, step-by-step methods for the job in front of you. Hand one to your agent, or follow it yourself.
Agree on the product model
Bring requirements and call notes into one agreed source.
Create the screen list
Turn the agreed model into a list of screens and journeys.
Build a screen
Build a screen from the agreed fields, stages, and rules.
Handle a change during the build
Decide where a change belongs: the model, the design system, or the code.
Add an automated check
Turn an agreed rule into a check the whole team can run.
Report project progress
Prepare a clear progress update with coverage and named gaps.
Close out a piece of work
Show the work does what you say, then turn what slowed it down into fewer surprises next time.
The library
Every file, prompt and check, with what it catches and how to tell it works.
Free chat
Ok, how does this all fit together?
Book a free chat and I’ll walk you through it, using your own product as the example. Bring your questions, or just come to say hi.
Book a free chat
Built by rahulmax. Product designer, 20 years shipping software.
For agents
Files your agent can read
Agents read these addresses instead of the pages. Plans come as JSON and files as plain text, from the same files the tests run against.
- /llms.txtThe index. Start here if you are an agent.
- /r/registry.jsonEvery file, prompt and check, with a link to its content.
- /r/tier/2.jsonAn install plan: each step, the file it writes, and how to tell it works.
- /r/task/build-a-screen.jsonA task guide, as a plan.
- /r/f/check-envOne file, as plain text.