Your project already has a button.
AI just wrote another one.
Keeps the UI consistent
Finds the duplicate, names the component you already have and drafts the swap.
Keeps your type scale intact
Catches a stray size or font and gives the step from your scale instead.
Keeps the phone layout working
Catches overflow, broken stacking and hidden buttons, with before and after shots.
Keeps the site from getting slower
Measures what each change adds: dependencies, bytes, load order and rerenders.
What lands on the PR.
A score, the findings, two real screenshots each, and the patch.
Design review
0 critical · 3 serious · 0 minor
1. Apply button is off the button scale
| Current | Fixed | |
|---|---|---|
 | → |  |
Fix: Increase the compact action to h-8, px-3 and text-sm.
Evidence
Compact action renders 20px tall next to the 40px primary.
plan.cta === "Apply" ? "h-5 px-2 text-[11px]" : "h-10 px-4 text-sm" → plan.cta === "Apply" ? "h-8 px-3 text-sm" : "h-10 px-4 text-sm"
2. Plan description hardcodes a colour instead of the muted token
| Current | Fixed | |
|---|---|---|
 | → |  |
Fix: Use the existing muted-foreground design token.
Evidence
#9a9a9a against the token's #8b8b8b, one step off the scale.
className="mt-3 min-h-14 text-sm leading-6"
style={{ color: "#9a9a9a" }}
→ className="mt-3 min-h-14 text-sm leading-6 text-muted-foreground"3. Secondary action competes with the primary action
| Current | Fixed | |
|---|---|---|
 | → |  |
Fix: Render Learn more as a bordered secondary action using existing surface tokens.
Evidence
Two filled black actions in the same card, 12px apart.
bg-black text-white → border border-black/10 bg-background text-foreground
Three ways it runs.
Per repository, any mix of them.
Every pull request
One comment with the score, findings and patch, rewritten on every push. Never blocks the merge.
Every few commits on main
You pick how many, from 2 to 100. Fixes come back as one pull request.
On a schedule
Every few days, as often as you like. Whatever landed on main since the last review, from everyone or the contributors you pick.
Connect once. Reviews run automatically.
- Connect GitHub
- Install the app on the account that owns the repository.
- Choose triggers
- Enable pull request reviews, scheduled maintenance, or both.
- Run a review
- Start now from recent commits, or wait for the next trigger.
- Fix the issues
- Read the review here or on GitHub, then open a fix pull request.
Reviews draw on your AI credits, the same balance 21st AI uses. No seats and nothing per repository: you pay the model cost of each review.
See plans