TSTypeScript · First PrinciplesBuild judgment for vibe coding
0%
Phase one is open · P00 + M00–M02

TYPE · RUNTIME · EVIDENCE

Do not just make AI code faster.
Judge it better.

A first-principles TypeScript course built around one AI coding assistant. Review generated code, modify React, and gradually build a reliable Node.js backend.

Driven by AI failure cases Streaming backend project Plain Node → NestJS
INPUTTRUSTED OUTPUT
01
RUNTIME

Predict real behavior first

JavaScript values, references, scheduling, and resources decide what the program actually does.

02
STATIC

Then use type constraints

TypeScript is early evidence, not runtime magic. Make mistakes visible before execution.

03
EVIDENCE

Finish with evidence

Failure paths, tests, and observability determine whether AI-generated code deserves trust.

0%
LOCAL PROGRESS

Start with the first runtime fact

Progress stays in this browser. No account required.

Start P00
PHASE 01 · RUNTIME FOUNDATIONS

Build judgment that frameworks cannot hide

4 lessons · 5 hr 15 min · P00–M02 open

TRUST
THE REVIEW FRAMEWORK

Check five layers before accepting AI-generated code.

Types & contracts · Runtime boundaries · Unhappy paths · Side effects & security · Tests & telemetry

COURSE PROJECT · CODEPILOT

Grow a 40-line mock CLI into a reliable AI backend.

P0 needs no model credentials. First make events, cancellation, and tests correct.

View project status
codepilot — mock review
$ npm run codepilot -- "审查这段任务队列代码"
◇ review.started
│ 检查异步边界与资源清理…
│ 发现:取消后 timer 仍在运行
◆ review.completed