One plain file of ground rules your agent reads before it designs a page or writes a line. So it makes real choices instead of reaching for the same gradients, templates, and invented numbers. It sits on top of the rules your agent already has. It never fights them.
A number implies evidence. If it is not evidence, it should not look like evidence.
Never invent evidence to make a page feel seamless finished.
Same law, formatted for how each agent loads its instructions. Pick yours.
CLAUDE.md
AI coding agents are incredibly good at making things that look finished. Give them an empty page and you get a gradient. Give them three features and you get three cards. Ask for credibility and suddenly there are 10,000 users, 42% faster workflows, and a testimonial from someone who may or may not exist.
That isn't intelligence. It's pattern completion.
Unslopify is a set of rules for making AI-generated software more deliberate. It teaches coding agents to question the obvious default, avoid invented claims, keep interfaces coherent, write like an actual product team, and stop when the problem is solved. It works alongside the agents you already use instead of trying to replace them.
The problem isn't that AI makes software. The problem is what happens when it has an unanswered design or engineering decision and fills the gap with the most familiar pattern it has seen.
Unslopify exists to interrupt that process.
A good interface doesn't need twelve sections. A good codebase doesn't need another abstraction. A good landing page doesn't need another testimonial, another statistic, another glowing orb, or another button telling you to get started.
Unslopify asks a simpler question:
Why does this exist?
If the answer is because it helps the user, solves a real problem, communicates something important, or makes the software meaningfully better, keep it.
If the answer is because the page looked empty, the AI thought it was expected, or the code felt like it needed to be "more scalable," reconsider it.
It's to make it look intentional.
That means real numbers instead of invented ones. Specific copy instead of "powerful, seamless, next-generation" filler. Consistent systems instead of individually polished components that don't belong together. Functional controls instead of screenshots pretending to be applications. Small, justified changes instead of endless refactors.
And sometimes the most correct decision is to do nothing. When the requested problem is solved, the tests pass, the interface works, the result is coherent, and there is nothing concrete left to fix, the agent should stop.
That's Unslopify.
Not anti-AI. Not anti-design. Not anti-automation.
Just anti-default.
The whole law, as one plain-text file. Read it, change it, or hand it straight to your agent.