step 1/3
Read the PR
The requirements it claims to meet, then the diff — like a normal review.
✓ imports every user
reports success only when done
$ ai-era code review practice
AI writes the code.
You catch what it got wrong.
One realistic AI pull request a day. Leave inline comments, give a verdict, get graded against a canonical review.
One challenge free daily. No signup to try it.
ai-agent wants to merge ·
checkout/discounts.py
+3 −1
There's a second bug you haven't found.
01
The judgment loop
Daily code review practice built on the loop reviewers run at work — without waiting for a teammate's PR.
step 2/3
Comment & decide
Flag the broken lines, then approve or request changes. Clean code exists — approving it counts.
Request changes
Approve
step 3/3
See what you missed
Rubric-based grading and the canonical review, side by side with yours.
✓ caught the race✕ missed the zero floor
02
What you'll train
The failure modes of code that looks plausible but hides a real defect — hand-authored PRs across React, Angular, Node, and Python, each a self-contained code review exercise with a canonical review to compare against.
Security
The leaked secret, the unchecked lookup, the injected bio, the SQL string
→Requirement fidelity
Plausibly similar to the spec — but not the spec
→Edge cases & error handling
The malformed row, the empty list, the silent reject
→Hallucinated APIs
Methods and arguments that read perfectly and don't exist
→State & data flow
Stale closures, effects that never clean up, leaked subscriptions
→Performance
Correct-but-quadratic code, N+1 queries, defeated memoization
→Data correctness
Money as floats, dates that drift a timezone, lost pennies
→Deceptive tests
Green suites that assert nothing, or mock away the rule
→Over-engineering
The registry for two constants, abstractions nobody asked for
→03
Why now
Fixing is increasingly what AI does for you. Judgment — spotting the broken assumption and deciding whether it ships — is the skill that stays yours.
DiffDojo makes that judgment repeatable: small PRs, clear requirements, feedback that shows exactly what a strong review would have caught.
pricing.md
Daily challenge
freeFull library
free · early accessEvery challenge, any time. Free while in early access; Pro comes later — the waitlist hears first.
04
FAQ
The short answers before you click.
What is DiffDojo?
DiffDojo is a code review trainer: you review realistic AI-generated pull requests with planted bugs, leave inline comments, give a verdict, and get graded against a canonical review. Like LeetCode, but for reading code instead of writing it.
Is DiffDojo free?
The daily challenge is free forever, no signup needed. The full library needs a free account and is free during early access; Pro comes later.
How is my review graded?
Every PR ships with planted bugs and a rubric. Your inline comments are matched against the rubric, your verdict is checked, and you see the canonical review next to yours.
What stacks are the PRs in?
React, Angular, Node (JavaScript/TypeScript), and Python today. More stacks join the library as it grows.
How is this different from LeetCode?
LeetCode trains writing code. DiffDojo trains reviewing it: reading a diff, catching the planted bug, and defending a verdict. Judgment, not typing.
Is the code really AI-written?
Yes. Each PR is generated by an AI assistant, then curated by hand so the planted bug is realistic and the canonical review is defensible.
How long does a review take?
About five minutes for the daily PR. It is deliberate code review practice, not a time sink: one small diff, one verdict, instant feedback.
● today's PR is open · claims: "handles expiry, minimum totals, and the zero floor"
The AI says it's clean. Prove it wrong.
Review today's PR — free~5 minutes. No signup. Tomorrow there's another.