Make Human Great Again.
Direct a coding agent on real repositories. Check its work. Correct it.
Token quotient: how smart you are with AI.
Low token quotient
you › fix the login bug
files the agent read140
extra worka .sha256 file no step reads
tests added+300, most redundant
when a test failedtest now copies production logic
fix loop“try again” × 14
tokens212k
resulthidden tests fail
High token quotient
you › in auth/form.ts, onSubmit drops the email. fix it, test onSubmit, touch no other file.
files the agent read3
extra worknone
tests added1
when a test failedroot cause fixed
fix loop1 message
tokens31k
resultverified
For teams & researchers
Every session is measured.
LeetVibeCode records the full trace: prompts, diffs, tool calls, test runs, tokens, and each human step. It scores the result and the process. Teams can measure agent skill. Researchers can study how people work with AI.
- Outcomehidden tests pass, nothing breaks
- Processhow often you verify, edit-test cycles, thrash
- Efficiencyturns, tokens, time
- Understandinga pop-up quiz