draft-foundry
/draft-foundry
Outline-first drafting for Chinese WeChat Official Account articles, taking a confirmed topic (usually chosen from topic-radar's five candidates) and turning it into a load-bearing skeleton and a publishable draft. Builds a load-bearing skeleton (5-9 nodes, each with an explicit 'what the reader takes away' line), validates it with a script, then writes the draft against five reusable structural patterns and a library of opening moves. Enforces write-rough-not-full rules so the draft does not self-summarize. Use when producing an article from a topic, when an article keeps collapsing mid-way,
- Author
- newtonoooo ↗
- Repository
- czl-experts ↗
- Licence
- MIT
- Last repository change
- 01.11.2026
Assessment and security
- Files in this package
- 7
- Scripts
- 1
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 01.11.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
- 1 scripts in this package. Review them before installing, as Claude may execute them.
Scripts: scripts/outline_check.py
What it does and when to use it
Outline-first drafting for Chinese WeChat Official Account articles, taking a confirmed topic (usually chosen from topic-radar's five candidates) and turning it into a load-bearing skeleton and a publishable draft. Builds a load-bearing skeleton (5-9 nodes, each with an explicit 'what the reader takes away' line), validates it with a script, then writes the draft against five reusable structural patterns and a library of opening moves. Enforces write-rough-not-full rules so the draft does not self-summarize. Use when producing an article from a topic, when an article keeps collapsing mid-way,
Triggers from the description
Original author description
Outline-first drafting for Chinese WeChat Official Account articles, taking a confirmed topic (usually chosen from topic-radar's five candidates) and turning it into a load-bearing skeleton and a publishable draft. Builds a load-bearing skeleton (5-9 nodes, each with an explicit 'what the reader takes away' line), validates it with a script, then writes the draft against five reusable structural patterns and a library of opening moves. Enforces write-rough-not-full rules so the draft does not self-summarize. Use when producing an article from a topic, when an article keeps collapsing mid-way,
How it works
- 这个技能解决什么
- 什么时候用
- 执行步骤
- 输出规范
- 依赖
- 异常处理
- References
SKILL.md preview
--- name: draft-foundry description: "Outline-first drafting for Chinese WeChat Official Account articles, taking a confirmed topic (usually chosen from topic-radar's five candidates) and turning it into a load-bearing skeleton and a publishable draft. Builds a load-bearing skeleton (5-9 nodes, each with an explicit 'what the reader takes away' line), validates it with a script, then writes the draft against five reusable structural patterns and a library of opening moves. Enforces write-rough-not-full rules so the draft does not self-summarize. Use when producing an article from a topic, when an article keeps collapsing mid-way, when an opening is weak, or when a writer needs a reusable structure instead of a blank page." version: 1.0.0 --- # 初稿工坊(draft-foundry) ## 这个技能解决什么 写不出来的最大原因不是没灵感,是**没有骨架就开始写**。写到第三段发现和第五段说的是同一件事,然后开始糊。 这座工坊的顺序是死的:**先骨架,验过再写**。骨架的唯一标准是每个节点都能用一句话说清"这一段读者会得到什么"。说不出来,这个节点就不该存在。 ## 什么时候用 - 用户给了选题,要一篇稿子 - 用户写了一半卡住 - 用户的开头总是留不住人 - 用户要一套能反复用的结构,不想每次从零想 **不适用**:还没选题 → 先走 `topic-radar`;稿子已经写完要体检 → `quality-gate`。 ## 执行步骤 ### 第零步:先确认账号,再接选题 动笔前必须知道自己**替谁写**:公众号名称(落款用)+ 账号方向(写给谁、写什么领域)。 这两个都只能来自用户,**没有任何默认值**——不许默认教育、不许默认某个城市、不许默认某类读者。 用户没说方向就先问一句;没说账号名就让落款留空(渲染时整块不输出),**不要编一个像真的号名**。 ### 第一步:接选题 上游通常来自 `topic-radar` 的 5 选 1。接过来先确认三件事,不要重新讨论选题: - 角度是什么(事实/影响/反常识/方法/判断)——它决定骨架长相 - 类型是什么(热点/常销/争议/骨架)——它决定寿命与语气 - 素材够了没有——不够就先列缺什么 如果用户是直接给题(没走过选题),补一句"这篇从哪个角度切",不要默认他会说。 ### 第二步:确认三件事(一句话,别展开) 1. 这篇给谁看(具体到一个人) 2. 他看完会做什么(一个动作或一个判断) 3. 篇幅多少(决定节点数:1500 字约 4—5 节点,2500 字约 6—7 节点) 三件事没确认就开写,写出来的东西八成要重来。 ### 第三步:选骨架模式 从 `references/skeletons.md` 选一个,不要混用两种。五种常用模式: | 模式 | 适合 | 骨架长相 | |---|---|---| | **现场递进** | 观察、访谈、亲历 | 场面 → 反常 → 追因 → 另一面 → 收在动作 | | **判断+三证** | 观点、评论 | 判断 → 证据1 → 证据2 → 反例 → 适用边界 | | **流程拆解** | 方法、攻略 | 错误做法 → 为什么错 → 正确做法 → 步骤 → 常见卡点 | | **对比双轨** | 选择困难类 | A 做法 → B 做法 → 分水岭 → 你属于哪种 → 试错成本 | | **时间线** | 事件复盘、政策变化 | 起点 → 转折 → 当下 → 三种走向 → 现在该做什么 | ### 第四步:写骨架并跑检查 ```bash python scripts/outline_check.py 大纲.md ``` 检查三件事:每个节点有没有「读者得到」、节点数是否在 5—9、结尾是不是"总结"。 **不通过不许开写。** 脚本给出的每条"必须修"都要处理掉。 ### 第五步:写初稿(三条硬规则) 1. **允许写糙,不允许写满。** 每段先把事实、场面、对话写出来,不写总结句。 2. **不许边写边收。** 段落结尾不要"这就是为什么……",让下一段接上。 3. **每段至少一个可核对的东西**:一个数、一个时间、一句原话、一个动作。三个都没有,这段是空的。 ### 第六步:自查后交付 写完先自己过一遍,再交给用户: ```bash # 用 quality-gate 的脚本自查,别用感觉判断 python ../quality-gate/scripts/quality_gate.py 初稿.md ``` 把脚本数字和初稿一起交付。有硬伤就自己改完再交,不要交付"我知道有问题但先给你看"的稿子。 ## 输出规范 - **骨架和初稿分开交付**,不要一次给一整篇。骨架确认后再写,能省掉一半返工。 - 初稿附一行说明:字数、节点数、用了哪个骨架模式。 - 缺素材的地方写 `【待补:需要什么】`,不要用臆造细节填空。用户在交付时能一眼看到你还缺什么。 - 保留用户的原话。用户口述里的土话、比喻、骂人的话,是这篇稿子最值钱的部分,不要"优化"掉。 ## 依赖 - Python 3.8+,仅标准库 - `quality-gate` 技能(初稿自查),可选 `voiceprint-lab`(按作者指纹写) ## 异常处理 | 情况 | 处理 | |---|---| | 用户催着要整篇,不肯确认骨架 | 先给骨架(5 分钟的事),说明"骨架你看一眼,不对现在改成本最低" | | 素材明显不够 | 列出缺的 3 样东西,先写能写的部分,缺的段落留占位 | | 用户要写的题自己不熟 | 要求提供事实来源;没有来源就改成"我不懂这块,只能帮你组织,内容得你给" | | 写着写着发现骨架不对 | 停下来改骨架,不要在错的骨架上继续写。改骨架 10 分钟,改稿子 2 小时 | ## References - `references/skeletons.md` — 五种骨架模式的完整展开与选用判断 - `references/opening-patterns.md` — 开头写法库(含反例) - `templates/outline-template.md` — 骨架文件模板(脚本可直接检查)
Files in this package (7)
- SKILL.md 6 KB
- references/opening-patterns.md 3 KB
- references/skeletons.md 4 KB
- samples/outline_bad.md 1 KB
- samples/outline_ok.md 2 KB
- scripts/outline_check.py 6 KB
- templates/outline-template.md 2 KB
Versions
The database has no other skills named draft-foundry.
Database snapshot: 13.11.2026 · package version 6dda63bb194f (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 95wechat-article
生成可直接发布的微信公众号软文。当用户要求写公众号推文、软文、绩效文章(如"写本月公众号推文""写一篇XX主题的软文")时使用。覆盖账号建档、选题对齐、场景化写作、严谨性审查、文风去 AI 化、配图、标题打磨的完整七阶段流程,产出可直接排版发布的 .md 文件与配图。 - 100wechat-article-compose
把一个已确认的选题写成微信公众号长文。锁定复刻锚与用户价值、写出 article.md / wechat.yml / _review.md,并通过公众号通用合规检查和人工预览前自检。准备、改写或核验公众号文章时使用;不负责发布,也不产出小红书卡片。 - 96wechat-article-generator
交互式生成公众号文章。当用户需要创作公众号推文时调用,通过步骤化引导收集文章要素,生成符合公众号风格的文章。 - 95wechat-article
Create, format, and publish WeChat Official Account (微信公众号) articles to draft box with stable quality gates (config bootstrap, metadata validation, cover style×palette presets, and publish preflight). Use when user asks to write/format/publish a WeChat article, generate cover… - 100topic-radar
Turns a single keyword into five genuinely different WeChat Official Account topic candidates. Searches the live web for recent developments, separates time-sensitive news from evergreen questions, generates five candidates that each use a distinct angle (fact / impact /… - 100wechat-article-workflow
End-to-end WeChat Official Account (微信公众号) article workflow. Find a topic from live trending lists, write the article in a voice that reads well on a phone, make one cover image, lay it out with an inline-style theme, preview it at phone width, and save it to the drafts box…