Build Project Memory for AI
Stop re-explaining your project in every new chat. Capture it once as a context profile the AI keeps across every conversation — stack-inferred conventions and all.
Tell AI your code style once. List the stack and the engine returns the naming rules and conventions it implies — plus your own house rules carried verbatim.
AI writes plausible code that quietly ignores your conventions — class components in a hooks codebase, hard-coded colors in a tokenized design system. This setup captures the conventions for Atlas UI, a shared React component library, where a breaking change ripples across every product team. You list the house rules (additive props, tokens not hard-coded values); the engine adds the React, TypeScript, and Vitest conventions and naming rules automatically. The AI stops guessing your style and starts following it.
List the stack
React, TypeScript, Vitest — each maps to known naming and convention rules.
Add your house rules
Additive props, tokens over hard-coded values; these travel verbatim.
Hand the AI the conventions
It writes code that matches the codebase instead of generic defaults.
Each stack name maps to its known conventions, so entering TypeScript pulls in 'No any — prefer unknown with narrowing; keep strict mode on' and the naming rule 'camelCase for variables and functions, PascalCase for types and components,' while React adds the function-components-with-hooks rule. The builder writes these into the profile text; you still paste it into your assistant, which does the actual coding.
The template's own note says to save it at your repo root as '.cursorrules' (or '.cursor/rules/project.mdc') so Cursor loads it for every request, which means the Atlas UI conventions travel with the code and every teammate inherits the same context. For ChatGPT or Claude, paste it into custom instructions or a project instead. It's project knowledge you re-update, not persistent memory.
They all sit under one Team Conventions section, so what tells them apart is the source tag. Your typed rules appear untagged and verbatim — 'Props are additive — never rename or remove a prop without a major version' and 'Design tokens come from the theme, never hard-coded values' — while stack-inferred rules carry a marker like '(React)' or '(Vitest)'. The builder never paraphrases the ones you wrote.
Stop re-explaining your project in every new chat. Capture it once as a context profile the AI keeps across every conversation — stack-inferred conventions and all.
A reusable profile that tells AI everything constant about your project — type, stack, conventions, glossary, architecture principles, and the rules it must never guess.
Onboard AI to a codebase the way you'd onboard a new hire: what it is, how it's built, the rules that aren't written down, and the things never to assume.
End a working session like a shift change, not an abandonment: state captured, decisions logged, next step named — ready for the next session to pick up.
Convert scattered bug notes, Slack messages, or user complaints into structured engineering tasks with reproduction steps, severity, and root cause hypothesis.
What a converged conversation contains: a rejected voice, an accepted framing, a hard constraint, a format decision, and a final-version moment — each mapped to its prompt section.
Define what AI should permanently know about your project — stack conventions, glossary, and never-assume rules in a profile you paste once.
The code compiles, the tests pass, and it doesn't look like anything else in the repo. "Follow our conventions" is a wish: yours aren't written down, they're spread across nine hundred files that disagree. Here's how to extract the real rules and test them.