Cursor's autocomplete started suggesting the exact method name I'd been hunting for across three Stack Overflow tabs. That single moment—watching it fill in
getServerSideProps
before I'd finished typing getServer
—was when I stopped fighting the editor and started letting it drive.Here's what actually saves minutes every day, not the marketing screenshots.
The Dot Menu You're Missing #
Most people click the little sparkle icon. Wrong move. Hover your cursor in the gutter (the line number area) and hit .
— that's the contextual command palette.
Type: /fix → fixes lint errors inline
Type: /explain → explains the selected function
Type: /test → writes a test file for the current module
Type: /doc → generates JSDoc comments
It reads your selection first, so highlight a messy function then hit .
to get targeted options. The wild part is how much faster this feels than opening Copilot's sidebar.
Vim Mode That Doesn't Fight You #
Cursor's vim support works until it doesn't—specifically when the AI starts autocompleting mid-motion. The fix:
// .cursor/settings.json
{
"vim.useSystemClipboard": true,
"editor.cursorBlinking": "smooth",
"ai.composer.acceptOn": "tab",
"ai.composer.enabled": true
}
Set acceptOn
to tab
only. Using enter
breaks vim's paragraph motions. Trust me, I learned this after a week of muscle memory hell.
Prompt Tracking You Can Actually Use #
Every time Cursor generates code, it logs what prompt produced it. Cmd/Ctrl + Shift + P, then "Cursor: Open Conversation History." You'll see entries like:
Generated Button.jsx in 2.1s
Prompt: "Create a reusable button component with state"
Tokens: 420 in / 680 out
Cost: $0.0034
Bookmark this. When you're knee-deep in a 2000-line refactor and something breaks, you can trace exactly which AI suggestion landed the bug. The PromptCube homepage has a thread where devs share particularly gnarly prompt iterations—worth browsing when you hit a wall.
File Scoping (When Your Repo Gets Noisy) #
Working in a monorepo with 15 packages? Cursor indexes everything unless you tell it not to.
// .cursorignore
node_modules/
dist/
build/
*.log
.env.local
.studio/
packages/docs/
Pair this with AI Models research—some models handle large contexts better than others. If you're jumping between frontend and backend simultaneously, the gpt-4.5-turbo variant keeps both contexts coherent longer than anything shipped last year.
Hotkey Combos That Feel Illegal #
Cmd+K then / → quick switch model (no menu diving)
Cmd+Shift+L → AI: Apply all suggestions in current file
Alt+A → AI: Accept current suggestion (don't use Tab—too easy to misclick)
Cmd+Enter → Open composer panel with current selection pre-filled
The Apply all suggestions
shortcut is the closest thing to cheating. Working on a CRUD app? Generate the entire user service, then hit Cmd+Shift+L and watch it cascade across every file that imports it.
Debugging AI Hallucinations #
Last week Cursor insisted moment-timezone
exports utcToZonedTime
directly. It doesn't. The fix is embarrassingly simple:
-
Select the broken import
-
Hit
. -
Type
/explain -
Then
/fix
This forces a fresh analysis instead of trusting the cached suggestion. I keep a scratch file called ai-bugs.md
where I log these—turns out the same false imports show up across different projects.
Multi-File Edits Without Losing Your Mind #
The composer panel isn't just for new files. Need to rename a prop across six components?
// Instead of manual find-replace, paste:
"Rename the 'onSubmit' prop to 'handleSubmit' in all form-related components.
Update both the prop definition and every usage inside the component tree."
Watch it queue up each file change before applying. Yes, it sometimes misses one. But catching one missed file manually beats updating six by hand.
Terminal Integration That Actually Helps #
Cursor's terminal isn't just themed—it listens. After running tests:
// Terminal output:
FAIL: 3 assertions in UserProvider.test.js
Expected: user.email to equal '[email protected]'
Received: undefined
// In any editor tab, type:
/fix the test failure shown in terminal output above
It parses the active terminal buffer and fixes context-aware. Feels like witchcraft until you see it work three times in a row.
Custom Model Switching #
Not every task needs maximum intelligence. Set up quick switches:
// .cursor/settings.json
{
"model.default": "claude-sonnet-4",
"model.quickEdit": "gpt-4o-mini",
"model.reasoning": "o1-preview"
}
Then in the composer: @model quickEdit
for trivial refactors, @model reasoning
for complex algorithm work. Saves real money when you're shipping daily.
The One Setting Everyone Gets Wrong #
Default context window? Way too aggressive. Cursor pulls in every imported file.
// Disable this
"ai.context.autoImportTracking": false,
// Enable manual control
"ai.context.manualSelectionOnly": true
Now you choose what enters context. Hold Shift and click file tabs to add them selectively. This alone doubled my accuracy on TypeScript edge cases—turns out seeing unrelated CSS files in context was poisoning the well.
Quick Win: Tab Completion Shortcuts #
Stop accepting every suggestion. Use:
Tab → accept word-by-word
Ctrl+Right → accept entire line
Cmd+Right → accept full function/block
Shift+Tab → reject suggestion
The Cmd+Right
shortcut alone cuts down 30 percent of unnecessary backspacing. Try it once on a long React component—you'll wonder why every other editor feels like typing with mittens on.
These aren't hypothetical optimizations. They're the exact settings I run in production while shipping client work. Copy them, tweak them, then ignore half of them when you develop your own flow.
The point isn't to become a Cursor power user—it's to make the tool stop getting in your way. When the AI disappears into the background, that's when the real speed runs happen.
Next Alcatraz: Go-native PII detection that outpaces MS Presidio →
All Replies （0） #
No replies yet — be the first!