Before a step
Pick the current wizard step, restate the done-when, and decide which command or test will prove the step is really finished.
Module 04.1
Codex manual and command reference
This companion page sits beside the Kanban wizard. Use it when you need the short manuals, prompt usage patterns, and commands worth remembering while building the Flask + HTMX + SQLite app.
Usage timing
Pick the current wizard step, restate the done-when, and decide which command or test will prove the step is really finished.
Stay in one thread, keep the stack constrained, and ask for a short plan if the repo has already drifted from the original wizard prompt.
Read the diff, run the highest-value checks, and do not move forward just because the code looks plausible.
使用时机
先确定当前是哪一步,重述完成标准,再提前决定用哪条命令或测试来证明这一步真的完成。
尽量保持在同一线程里,限制技术栈范围;如果仓库已经偏离原始向导提示词,就先要求一个短计划。
读 diff,跑最有价值的检查;不要因为代码“看起来像对的”就直接进入下一步。
Short manuals
Open the current wizard step, not the whole dream project. Paste the step prompt, then add the current repo facts before you let Codex edit.
Do not move from setup to schema, or schema to backend, until the present step actually runs and the verification path is clear.
Keep the wizard prompt as the base, then add current files, blockers, constraints, and a done-when statement that matches today’s repo state.
Read the diff, run or request checks, and ask for remaining risks. “Looks good” is weaker than “passed the intended checks.”
When commands and conventions stop changing every hour, move them into README or AGENTS.md so later sessions start from stronger defaults.
短手册
打开当前步骤,不要一上来就要求整个梦想项目。先贴向导提示词,再补充当前仓库事实,然后再让 Codex 修改。
从脚手架到 schema、从 schema 到后端之前,先确认当前这一步真的能跑,而且验证路径明确。
把向导提示词作为底稿,再补当前文件、阻塞点、限制条件,以及符合今天仓库状态的完成标准。
读 diff,跑或要求检查,并追问剩余风险。“看起来不错”远弱于“通过了本该通过的检查”。
当命令和约定不再每小时变化时,把它们沉淀进 README 或 AGENTS.md,让后续会话从更强默认值开始。
Prompt usage
I am on Module 04 Step __. Current repo state: - Existing files: ... - Relevant files: ... - Current blocker: ... Use the Step __ wizard goal as the base. Constraints: - Keep Flask + HTMX + SQLite. - No React and no frontend build step. - Reuse existing templates and routes where possible. Before editing: 1. Give me a short plan. 2. List key risks. 3. Tell me how you will verify the change. Then implement.
Continue this Module 04 Kanban project from its current state rather than regenerating it. First inspect the repo and tell me: 1. Which wizard step the project appears to be on. 2. What is missing to reach the next stable step. 3. Which files should change next. 4. How you would verify the next step cleanly. Then wait for confirmation before editing.
Act as a senior reviewer on this Module 04 Flask + HTMX + SQLite Kanban app. Inspect the current project and focus on: - broken CRUD flows - drag-drop and ordering safety - duplicated templates - missing tests - README or setup gaps Run the highest-value checks you can, make the necessary fixes, and summarize any remaining risks or tradeoffs.
提示词用法
我现在在模块 04 的第 __ 步。 当前仓库状态: - 已有文件:... - 相关文件:... - 当前阻塞点:... 请以第 __ 步的向导目标为基础。 限制条件: - 保持 Flask + HTMX + SQLite。 - 不要引入 React,也不要引入前端构建链。 - 尽量复用现有模板和路由。 编辑前先做三件事: 1. 给我一个短计划。 2. 列出关键风险。 3. 告诉我你会如何验证这次改动。 然后再实现。
请从这个模块 04 的 Kanban 项目当前状态继续,而不是重新生成整个项目。 先检查仓库,并告诉我: 1. 这个项目现在看起来处在哪个向导步骤。 2. 要进入下一个稳定步骤还缺什么。 3. 下一步应该改哪些文件。 4. 你会怎样干净地验证下一步。 然后先等待我确认,再开始编辑。
请以资深 reviewer 的视角检查这个模块 04 的 Flask + HTMX + SQLite Kanban 项目。 重点关注: - CRUD 流程是否有断点 - 拖拽与排序是否安全 - 模板是否重复 - 是否缺少关键测试 - README 或 setup 是否有缺口 运行你认为最高价值的检查,完成必要修复,并总结剩余风险或取舍。
Commands
npm i -g @openai/codex codex codex resume --last
/status /plan /diff /review /model
Use it when you want to continue the same Kanban feature line instead of rebuilding context from zero.
Use it when you are unsure what Codex thinks the task is, what it has already done, or what remains.
Use it before multi-file changes, risky refactors, or when the repo no longer matches the original wizard step cleanly.
Use it before you accept a step. The summary is not the proof; the actual diff is the proof.
Use it when the task has real risk and you want a deliberate bug, regression, or missing-test pass.
命令
npm i -g @openai/codex codex codex resume --last
/status /plan /diff /review /model
当你要继续同一条 Kanban 功能线,而不是从零重讲上下文时,用它恢复线程。
当你不确定 Codex 现在认为任务是什么、已经做了什么、还剩什么时,用它对齐状态。
在多文件修改、高风险重构,或者仓库已经不再干净对应原始向导步骤时,先用它拿短计划。
在接受当前步骤之前用它看实际改动。summary 不是证据,真正的 diff 才是证据。
当任务风险不低,而且你需要一次认真检查 bug、回归或缺失测试时,用它做审查。
Project runbook
python -m venv .venv source .venv/bin/activate pip install -r requirements.txt
flask --app app init-db flask --app app seed-demo flask --app app run --debug pytest
Before you leave a step, confirm the route or flow works locally, the UI matches the intended step, and the next prompt is now smaller than the previous one.
Write it when stack decisions, project commands, naming patterns, and verification expectations stop changing every session.
项目运行手册
python -m venv .venv source .venv/bin/activate pip install -r requirements.txt
flask --app app init-db flask --app app seed-demo flask --app app run --debug pytest
离开当前步骤前,先确认路由或流程能本地跑通,界面符合本步骤目标,而且下一条提示词比上一条更小、更明确。
当技术栈决策、项目命令、命名习惯和验证预期不再每次会话都变化时,就可以开始写它。
Common mistakes
Once the repo exists, ask Codex to continue from the current state. Full regeneration increases thrash and duplication.
Route behavior, form handling, and ordering logic need actual checks. Visual plausibility is not enough.
Parallel threads against the same templates, routes, or data layer multiply merge confusion instead of progress.
If the plan was Flask + HTMX + SQLite, treat new frameworks as a deliberate scope change, not a casual patch.
常见错误
仓库已经存在后,就该让 Codex 从当前状态继续。整仓重生只会增加混乱、重复和返工。
路由行为、表单处理和排序逻辑都需要真实检查。看起来合理,不等于真的可靠。
如果多个线程同时改模板、路由或数据层,得到的通常不是并行进展,而是合并混乱。
既然计划是 Flask + HTMX + SQLite,就把新框架当成明确的范围变更,而不是随手补丁。