Mini Codex v2 Foundation Course

中文版

This is not another prompt-template collection, and it is not a thin chatbot wrapper.

The goal is concrete:

start from an empty folder
-> build a local Mini Codex
-> turn one coding request into a trackable, reviewable, verifiable, replayable engineering task

Primary reading page:

Mini Codex course page

Open-source repository:

https://github.com/herongdev/min-codex-hands-on-v2

Who This Is For

This course is for developers who already know basic frontend, backend, or full-stack development and are using tools such as Codex, Cursor, or Claude Code.

You do not need to master a complex agent framework first. The more important questions are:

Why is a Coding Agent different from a chatbot?
Why should the model not touch the file system directly?
Why is a prompt not a security boundary?
Why should code changes go through a plan and a diff review?
Why does "done" require verification evidence?
Why do trace and eval decide whether the agent can improve over time?

What You Will Build

By the end of the foundation course, you will have a small but complete Mini Codex:

enter one coding request
-> save it as a task
-> generate a plan
-> wait for your approval
-> modify files
-> run checks
-> save the process
-> inspect the result in a desktop workbench

It does not try to copy every capability of the official Codex product. The foundation course focuses on one thing: making the smallest safe coding-agent loop work.

Course Outline

The open foundation course includes:

1. Lab 00: Environment and practice workspace
2. Lab 01: Save the first task
3. Lab 02: Pause a task for human approval
4. Lab 03: Tool Registry and permission boundaries
5. Lab 04: Make the first real code change
6. Lab 05: Trace, Replay, and Eval
7. Lab 06: Backend API and SSE
8. Lab 07: Desktop workbench
9. Lab 07b: UI polish and design system
10. Lab 08: Compare the boundary with real Codex-style tools
11. Real Codex capability upgrade map
12. Demo cases
13. Eval ledger

Publishing Boundary

This release open-sources the foundation course and the public upgrade direction only.

The v2 Advanced course is maintained separately and is not part of this open-source release. The site and repository do not expose private advanced-course paths or commercial-course content.

Why Start With the Foundation

The hard part of AI coding is not asking a model to generate code. The hard part is putting generated code inside an engineering system:

state
permissions
planning
review
verification
replay
evaluation

Once these foundations are working, RAG, memory, MCP, subagents, Git worktrees, and dashboards become much easier to reason about.

Start here:

Open the Mini Codex course page