Maker here. I kept approving AI-written PRs and noticed my review instincts getting lazier while the diffs got more confident. lgtmxp is my attempt to train that muscle deliberately. Every exercise is a PR with a realistic diff. Most hide the mistakes LLMs actually make — off-by-one pagination, verify_exp=False in a JWT refresh, runaway agent tool loops, prompt injection through RAG. Some are clean on purpose: approving a clean diff correctly pays full XP, because knowing when NOT to comment is the rarest review skill. Miss a bug and it "ships" — your dashboard collects the incidents, 3am pages, and postmortem hours you failed to prevent. Nothing is permanent: retake the exercise, catch everything, and the incident moves to the prevented column. 244 exercises across Python/FastAPI, React/TS, Go, SQL, and an agentic/LLM track. Free, no signup, progress lives in your browser. Would love to hear: do the bugs feel like real review finds or toy puzzles, and what stack should come next?

lgtmxp

Duolingo, but to train the code review muscle

Duolingo, but to train the code review muscle