English
S · Situation
We are building Story Engine: a product that turns the process of working on a project with AI into an article, making the outcomes easier to share. Our earlier exploration of publishing platforms exposed a problem: successfully verifying an account does not necessarily mean that publishing through its API will work.
Today, our first Hello World post went live on DEV.to. This small milestone lets us move on to the core experience: finishing a project without having to reconstruct the entire story from memory before writing a retrospective.
T · Task and Challenges
The next goal is to say “summarize today's or this week's project progress” in Codex and receive a STAR draft covering progress, challenges, solution approaches, and results.
The main challenge is the scope of the evidence. Code can show what changed, but not necessarily why; the current discussion may not cover the entire week. Another challenge is review consistency: the text the user approves should be exactly the text that gets uploaded.
A · Actions and Approach
We implemented the first version as a project-level Codex skill. It uses the current project's discussions and verifiable materials, defines the reporting period, and distinguishes plans from verified results and unfinished work. STAR provides the narrative structure; missing facts are not filled in with invented details.
Generated articles first enter a local draft store. The web app provides a dedicated “Codex Progress Drafts” view showing the article, source coverage, and supporting notes. Those notes remain local; the publishing request contains only the title and article body.
The publishing flow checks the reviewed draft version and the target account. If the text changes, it must be reviewed again. We also retain the existing duplicate-publication protection: when a request's outcome is uncertain, automatic retries stop so the platform's actual state can be checked first.
R · Results and Next Steps
The user has confirmed that Hello World was successfully published. We have also implemented the progress-drafting instructions, local import, web review, and DEV.to publishing entry point. All 13 automated tests passed, including checks for custom article request formatting and draft-change detection.
These tests use a simulated remote service; they do not establish that this progress article has been published. The next step is to review a real project draft, check its narrative and factual accuracy, and complete a live publication. Coverage across multiple tasks still depends on the project materials Codex can access at the time.
中文
S · 背景
我们在做 Story Engine：把与 AI 一起推进项目的过程整理成文章，让项目成果更容易被分享。此前的发布渠道探索暴露了一个问题：能验证账号，并不意味着就能顺利通过 API 发文。
今天，第一篇 Hello World 已在 DEV.to 发布。这个小里程碑让我们可以继续验证更核心的体验：项目做完后，不再从头回忆如何写一篇复盘。
T · 目标与难点
接下来的目标是，在 Codex 中说一句“总结今天或本周的项目进度”，就得到一份包含进度、难点、解决思路和结果的 STAR 待审稿。
难点在于证据边界。代码能说明改了什么，却不一定能说明为什么改；当前讨论也未必覆盖整个星期。另一个问题是审阅一致性：用户批准的正文，应当就是最终上传的正文。
A · 解决思路与行动
我们将第一版做成项目内的 Codex 技能。它以当前项目的讨论和可核对材料为依据，明确时间范围，并把计划、已验证结果和仍未完成的内容分开。STAR 结构负责组织叙事，缺失的事实不会靠补写来填满。
生成后的文章先进入本地草稿库。网页提供独立的“Codex 进度稿”入口，展示正文、覆盖范围和依据。依据留在本地，发布请求只携带标题与正文。
发布流程会核对审阅时的草稿版本与目标账号；正文变化后需要重新审阅。我们沿用已有的重复发布保护：请求结果不确定时停止自动重试，先核实平台状态。
R · 结果与下一步
Hello World 的真实发布已由用户确认。项目进度稿的生成规范、本地导入、网页审阅与 DEV.to 发布入口也已完成实现，13 项自动测试通过，包括自定义文章请求格式和草稿变化检测。
这些测试使用模拟远端，不代表这篇进度文章已经发布。下一步是审阅一份实际项目稿，检查叙事和事实是否准确，再完成一次真实发布。跨任务记录的完整覆盖仍取决于 Codex 当时可访问的项目资料。
Top comments (0)