Cursor AI Prompts & Rules That Actually Work

Read the codebase and draft a project rules file for Cursor. Include: the tech stack and framework versions, the folder structure and where new files go, our naming and import conventions, the libraries to prefer (and any to avoid), and how we handle errors and tests. Output it as concise, imperative bullet rules I can paste into a .cursor/rules file.

Create a scoped rule that applies only to files under [path, e.g. src/api/**]. The rule should enforce: [specific conventions for that area, e.g. all handlers return a typed Result, validate input with zod, never log secrets]. Keep it under 15 bullet points and explain when each rule should trigger.

Generate an architecture overview I can keep as a rule for context: the main modules and what each does, how data flows through the app, the key entry points, and the external services we call. Use @[folder] to ground it in the real code and keep it to one page.

I'm using [library/framework] version [X]. Using @Docs for it, summarize the current recommended patterns for [task, e.g. data fetching] and note anything that changed from older versions. Then propose how we should standardize our usage and add it to our rules.

Implement [feature] end to end. First read @[relevant files/folders] to match existing patterns, then make a short plan listing the files you'll create or change and why. After I confirm, implement it, keeping each file consistent with our conventions. Don't add new dependencies without flagging them.

Here is the spec for [feature]: [paste requirements]. Implement it using the existing components in @[folder]. For each requirement, point to the exact code that satisfies it. If any requirement is ambiguous, ask me before coding rather than guessing.

Add a [METHOD] [route] endpoint that [does what]. Follow the pattern in @[existing handler file]: same validation, error handling, and response shape. Add the route, the handler, input validation, and any types. List the files you changed and show how to call the endpoint.

Create a [component name] component that [behavior and props]. Match the styling and structure of @[similar existing component], reuse our shared UI primitives, and handle loading, empty, and error states. Keep it accessible (labels, keyboard nav) and add the prop types.

Add [library] to handle [task]. First check @package.json for the current version (or tell me what to install), then reference @Docs for the correct setup. Wire it into @[target file] following the library's recommended pattern, and explain any config or env vars I need to add.

Refactor @[file or function] to improve readability and remove duplication. Do NOT change its external behavior or public interface. Explain each change briefly, keep the existing tests passing, and if a change is risky, flag it instead of applying it.

Search the codebase for code duplicated across @[folder]. List the duplicated patterns you find, then propose where to extract shared helpers. After I approve, do the extraction and update all call sites. Show the diff for each call site you changed.

Update @[file] to current best practices for [language/framework]: replace deprecated APIs, use modern syntax, and improve typing. Reference @Docs for the correct current patterns. Make changes incrementally and explain why each replacement is the recommended approach now.

Rename [old name] to [new name] across the codebase and update every import, reference, and string that depends on it. Use @[folder] to find all usages first and list them before changing anything. Verify nothing else breaks after the rename.

I'm getting this error: [paste error and stack trace]. It happens when [steps to reproduce]. Read @[relevant files] to trace the cause, explain the root cause before fixing, then apply the smallest fix that resolves it. Don't refactor unrelated code while you're in there.

I can't reproduce why [behavior] happens. Add temporary, clearly-labeled debug logging around @[function/area] to trace the values of [key variables] and the flow of execution. Tell me what to look for in the output, and remind me to remove the logs after.