zero2codex
OpenAI Codex workflow course

Course goal: ship your personal brand site

Build a personal brand site with Codex

Keep improving one project from an empty folder to a responsive one-page site with an introduction, projects, and contact details. Every lesson adds a visible result or a check toward that finish line.

Curriculum

Personal brand site build path

01

Start With Codex

Set up Codex and build the hero section of your personal brand site.

0/4
08

Finish And Deliver The Site

Polish independent tasks in parallel, then complete one final acceptance review.

0/3
Complete The Site GoalTurn the finished personal site into one measurable goal with content, layout, interaction, and evidence checks.Polish Copy And Accessibility In ParallelUse isolated worktrees only when copy polish and accessibility review can avoid editing the same lines.Run The Final Three ReviewsRun separate content, link, and accessibility reviews, then compare all evidence against the same launch checklist.
12 minReady

Run The Final Three Reviews

Finish my-brand-site with three focused reviews and one final human acceptance decision.

content reviewlink reviewaccessibility review

Watch, then try

24|完成三項獨立驗收(CLI 版・35 秒)

真實 CLI 輸出檢視器與本機網站錄影,非 Codex 原生介面;繁中字幕、無旁白,等待區段已剪接。依序執行三類驗收,不啟動 subagent 或排程。

zero2codex
MP4

Step-by-step

  1. 沿用前一課的 my-brand-site,或從下方練習檔取用本課 before 起點。
  2. 在專案目錄執行本片 CLI 指令;本課權限為 read-only。
  3. 核對實際輸出與結束碼,不把文字回報當成瀏覽器操作證據。
  4. 三項驗收都有證據;不把依序檢查冒充 subagent 或自動排程。
跟著本片操作:CLI 指令與練習檔(繁體中文)

這是本課的 CLI 替代實作路徑;下方概念課程仍保留其他介面的說明。先安裝並登入 Codex CLI,在 my-brand-site 資料夾內執行下列 macOS/Linux 指令。影片將真實 CLI JSON 事件排版成易讀畫面,下列指令使用一般 CLI 輸出;不是 Codex 原生介面,等待區段已剪接。

codex exec --ephemeral --ignore-user-config --sandbox read-only '本片是依序執行的 CLI 驗收,不啟動 subagent 或排程。執行 python3 check_site.py、git diff --check,再讀 index.html。分別檢查文案、錨點與互動可辨識名稱,每類附實際證據;指出瀏覽器操作仍由錄製者驗證。限 200 字。'

下載逐課練習檔與實際結果(ZIP)

Plain-language explanation

Finish with three separate questions: Is the site copy clear? Do all links work? Can the page be understood and used accessibly? Each review can report evidence without changing the site on its own.

After the three reports return, compare them with the same launch checklist and make one final acceptance decision.

Why it matters

Focused reviews catch different kinds of mistakes while one final reviewer keeps the site coherent.

Common beginner mistake

A common mistake is letting every reviewer edit index.html independently, then trusting their summaries without checking the page.

Good prompt

Codex, run three independent evidence-only reviews of my-brand-site: copy clarity, broken links, and accessibility. Each review must cite exact text, links, or elements and must not edit files. Then compare all findings with the launch checklist and identify what still needs a human decision.

Try this

Write the exact evidence format for the copy, link, and accessibility reports.

Completion checklist

  • Each review reports exact evidence without independently rewriting the site.
  • The final reviewer checks the original findings and every launch item.

Sources

Practice lab

Copyable Codex prompts

Plan before coding

Codex, inspect the relevant files first, then propose a plan. Do not edit until the plan names tests and verification steps.

Verification checklist

  • Relevant files are named before edits begin.
  • The plan includes tests and verification commands.

Verify a UI change

Codex, start the dev server, open the route in the browser, check desktop and mobile layouts, then report the exact evidence.

Verification checklist

  • Desktop and mobile states are checked in the browser.
  • The final note includes exact evidence.

Review focused risk

Codex, review the uncommitted diff. Focus on behavioral regressions, missing tests, and accessibility issues. Do not modify files.

Verification checklist

  • Findings lead with severity and file references.
  • Unchanged files are left untouched.
Every good Codex request says what to change, what to preserve, and how to prove it worked.

Reference

Codex concepts to keep nearby