writing-assistant

Category: Content & Multimedia | Uploader: VegetaPn | Downloads: 0 | Version: v1.0（Latest）

Comprehensive writing workflow from ideation to publication. Guides users through creating polished articles by collecting ideas, asking clarifying questions, researching content, polishing drafts, adding images, and publishing to platforms like WeChat or X. Use when users want to write articles, blog posts, or long-form content, especially when starting from a topic idea, rough materials, or initial draft. Also use when users mention writing, publishing, or content creation workflows.

Changelog: Source: GitHub https://github.com/VegetaPn/writing-assistant-skill

Directory Structure

Current level: tree/main/

📁 assets/
- 📁 experiences/
  - 📁 cases/
    
    📄 .gitkeep 0 B
    
    📄 2026-02-16-dan-koe-style-mismatch.md 800 B
    
    📄 2026-02-16-follow-progress-steps.md 700 B
    
    📄 2026-02-16-manus-ai-case-unsuitable.md 808 B
    
    📄 2026-02-16-title-feedback.md 952 B
    
    📄 2026-02-16-vague-time-expression.md 826 B
    
    📄 2026-02-25-bold-spacing-and-density.md 1.3 KB
  - 📄 command-failures.md 706 B
  - 📄 lessons.md 2.2 KB
- 📁 topics/
  - 📁 benchmarks/
    
    📄 benchmarks-index.md 297 B
    
    📄 monitor-config.md 2.0 KB
  - 📁 developing/
    
    📄 .gitkeep 0 B
  - 📄 inbox.md 139 B
- 📄 progress-template.md 10.5 KB
📁 config/
- 📄 bird.json5 29 B
📁 dependencies/
- 📁 baoyu-post-to-wechat/
  - 📁 references/
    
    📄 article-posting.md 2.2 KB
    
    📄 image-text-posting.md 2.7 KB
  - 📁 scripts/
    
    📁 md/
    
    📁 extensions/
    
    📄 alert.ts 19.3 KB
    
    📄 footnotes.ts 2.4 KB
    
    📄 index.ts 280 B
    
    📄 infographic.ts 3.6 KB
    
    📄 katex.ts 4.6 KB
    
    📄 markup.ts 2.3 KB
    
    📄 plantuml.ts 8.1 KB
    
    📄 ruby.ts 3.9 KB
    
    📄 slider.ts 2.9 KB
    
    📄 toc.ts 2.1 KB
    
    📁 themes/
    
    📄 base.css 566 B
    
    📄 default.css 7.7 KB
    
    📄 grace.css 2.5 KB
    
    📄 simple.css 2.7 KB
    
    📁 utils/
    
    📄 languages.ts 8.1 KB
    
    📄 LICENSE 650 B
    
    📄 render.ts 17.6 KB
    
    📄 cdp.ts 10.7 KB
    
    📄 copy-to-clipboard.ts 11.9 KB
    
    📄 md-to-wechat.ts 8.1 KB
    
    📄 paste-from-clipboard.ts 5.5 KB
    
    📄 wechat-agent-browser.ts 10.4 KB
    
    📄 wechat-article.ts 19.0 KB
    
    📄 wechat-browser.ts 25.1 KB
  - 📄 SKILL.md 5.7 KB
- 📁 baoyu-post-to-x/
  - 📁 references/
    
    📄 articles.md 4.5 KB
    
    📄 regular-posts.md 3.4 KB
  - 📁 scripts/
    
    📄 copy-to-clipboard.ts 11.9 KB
    
    📄 md-to-html.ts 12.3 KB
    
    📄 paste-from-clipboard.ts 5.5 KB
    
    📄 x-article.ts 23.7 KB
    
    📄 x-browser.ts 7.6 KB
    
    📄 x-quote.ts 8.4 KB
    
    📄 x-utils.ts 8.5 KB
    
    📄 x-video.ts 8.9 KB
  - 📄 SKILL.md 6.8 KB
- 📁 baoyu-xhs-images/
  - 📁 references/
    
    📁 config/
    
    📄 first-time-setup.md 2.6 KB
    
    📄 preferences-schema.md 3.0 KB
    
    📄 watermark-guide.md 1.9 KB
    
    📁 elements/
    
    📄 canvas.md 3.2 KB
    
    📄 decorations.md 4.7 KB
    
    📄 image-effects.md 2.9 KB
    
    📄 typography.md 3.2 KB
    
    📁 layouts/
    
    📄 balanced.md 433 B
    
    📄 comparison.md 558 B
    
    📄 dense.md 551 B
    
    📄 flow.md 513 B
    
    📄 list.md 489 B
    
    📄 sparse.md 516 B
    
    📁 presets/
    
    📄 bold.md 1.5 KB
    
    📄 chalkboard.md 2.3 KB
    
    📄 cute.md 1.5 KB
    
    📄 fresh.md 1.5 KB
    
    📄 minimal.md 1.4 KB
    
    📄 notion.md 1.6 KB
    
    📄 pop.md 1.5 KB
    
    📄 retro.md 1.4 KB
    
    📄 warm.md 1.4 KB
    
    📁 styles/
    
    📄 bold.md 456 B
    
    📄 cute.md 465 B
    
    📄 fresh.md 469 B
    
    📄 minimal.md 437 B
    
    📄 notion.md 566 B
    
    📄 pop.md 474 B
    
    📄 retro.md 434 B
    
    📄 tech.md 465 B
    
    📄 warm.md 459 B
    
    📁 workflows/
    
    📄 analysis-framework.md 7.1 KB
    
    📄 outline-template.md 6.0 KB
    
    📄 prompt-assembly.md 6.1 KB
    
    📄 base-prompt.md 1.1 KB
  - 📄 SKILL.md 9.4 KB
- 📁 content-research-writer/
  - 📄 SKILL.md 13.9 KB
- 📁 generate-image/
  - 📁 scripts/
    
    📄 generate_image.py 8.7 KB
  - 📄 SKILL.md 6.9 KB
- 📁 wechat-article-search/
  - 📁 scripts/
    
    📄 search_wechat.js 22.7 KB
  - 📄 SKILL.md 2.2 KB
- 📁 xiaohongshu/
  - 📁 references/
    
    📄 content-guide.md 2.8 KB
    
    📄 cover-guide.md 3.7 KB
    
    📄 title-guide.md 2.1 KB
  - 📁 scripts/
    
    📄 cover.sh 16.2 KB
    
    📄 generate.sh 10.5 KB
  - 📄 check_env.sh 2.8 KB
  - 📄 SKILL.md 18.1 KB
📁 dev/
- 📁 archived/
  - 📄 iteration-plan-v1.md 40.4 KB
  - 📄 iteration-plan-v2.md 6.7 KB
- 📁 dev_reference_materials/
  - 📁 dontbesilent/
    
    📄 best_practice_reference.md 9.0 KB
    
    📄 Claude Code：我和 dontbesilent 相处的三个星期.md 10.7 KB
    
    📄 img.png 163.1 KB
    
    📄 img_1.png 315.9 KB
    
    📄 如果你想抄我，这篇就是说明书.md 10.8 KB
  - 📁 Yangyi/
    
    📄 growup_best_practice_reference.md 4.5 KB
    
    📄 img.png 199.1 KB
    
    📄 img_desc.md 4.6 KB
- 📄 iteration-plan.md 11.0 KB
- 📄 my_note.md 453 B
📁 references/
- 📁 authors/
  - 📁 dan-koe/
    
    📁 articles/
    
    📄 the-future-of-work-when-work-is-meaningless.md 26.8 KB
    
    📄 the-most-important-skill-to-learn-in-the-next-10-years.md 11.5 KB
    
    📄 profile.md 25.6 KB
  - 📁 dontbesilent/
    
    📁 articles/
    
    📄 001_出版社让我出书，我说今年戒掉了读书 AI 读书 AINative.md 5.3 KB
    
    📄 002_如何借助 AI 比任何人学得快 10 倍学习学习方法 AI Agent.md 7.6 KB
    
    📄 003_绝大多数人花钱上学读书，和花钱喝杯奶茶没有本质区别上学读书学历学历教育赚钱赚钱思维.md 5.9 KB
    
    📄 005_AI 不是新范式，是 1960s 分工模式的低成本实现 AIGC AI工作流自媒体创作 Agent.md 8.0 KB
    
    📄 006_一个人赚不到钱的核心原因就是：上班上多了上班思维赚钱思维赚钱商业思维.md 8.0 KB
    
    📄 007_小红书新规，逼我停更 11 万粉账号小红书小红书运营小红书开店小红书营销小红书规则.md 5.9 KB
    
    📄 009_一人公司只是表壳，你真正想要的是不那么累地赚钱一人公司裸辞不上班.md 9.4 KB
    
    📄 010_情绪上的无能，是我的第一生产力情绪创业.md 4.5 KB
    
    📄 011_善良的李亚鹏，30 年为什么赚不到钱李亚鹏赚钱商业模式.md 4.6 KB
    
    📄 012_AI 不会让你失业，但会让你的工作变 low AI失业 AI 失业.md 4.2 KB
    
    📄 013_3 个月，我设计出了一种不内耗的赚钱方式内耗第一桶金年入百万创业辞职.md 8.6 KB
    
    📄 014_读完几百本书，终于知道富人穷人到底差在哪富人富人思维商业.md 4.3 KB
    
    📄 015_小红书付费笔记功能，注定变现效率会极低小红书自媒体商业模式.md 5.6 KB
    
    📄 017_去年涨粉 70 万后，我所有方法压缩成了一句话自媒体自媒体IP.md 10.0 KB
    
    📄 018_旗帜鲜明地反对「精准获客」概念，这是商业误导精准获客 dontbesilent 语言的边界就是世界的边界.md 7.0 KB
    
    📄 019_垂直的意思是细分且深入，不是赚钱少😂 垂直生意.md 5.2 KB
    
    📄 020_当老板懂 AI 后：下班前给我 100 张图 AI 失业.md 6.3 KB
    
    📄 022_如何给 8 千人付费答疑，公开我的底层逻辑商业模式底层逻辑.md 10.9 KB
    
    📄 024_今年给超万人答过疑，聊一下如何设计引流款引流款知识付费.md 4.4 KB
    
    📄 025_副业思维锁死了一个人的财富上限副业商业思维.md 953 B
    
    📄 026_豆包 AI 手机，两句指令，写完爆款文案豆包豆包AI 豆包手机.md 690 B
    
    📄 027_不想做，就是不能做。还没做，就是做不了。执行力 dontbesilent.md 3.9 KB
    
    📄 028_40 万粉丝后，我发现 99% 的人不会看数据 dontbesilent 自媒体.md 7.3 KB
    
    📄 029_今年涨粉 70 万，一条视频讲清楚限流问题账号权重限流.md 11.1 KB
    
    📄 030_兴趣变现如何年入百万，5 分钟系统性学会商业模式自媒体兴趣变现爱好赚钱兴趣.md 11.1 KB
    
    📄 031_3 分钟，教你彻底摆脱 AI 味文案写作文案 AI.md 5.7 KB
    
    📄 032_为什么每次都那么凑巧错过发财的机会？ dontbesilent 商业商业思维.md 5.3 KB
    
    📄 033_做短视频的本质是黑掉人的认知系统短视频算法自媒体流量短视频开头.md 8.5 KB
    
    📄 034_小米股价暴跌后，华为一年前的预判应验了小米雷军华为华为刘亦菲.md 4.7 KB
    
    📄 035_知识付费低价引流直播课，玩不下去了干货都在这自媒体商业模式.md 5.4 KB
    
    📄 037_卖了几百万货我发现：人买的不是货，是幻想干货都在这独立思考电商干货自媒体 dontbesilent.md 5.7 KB
    
    📄 038_我用 AI 给算法做了一面镜子，流量百万社交媒体社交媒体营销.md 9.2 KB
    
    📄 039_如何成为更值钱的博主：流量变现底层逻辑流量变现.md 8.2 KB
    
    📄 040_赚钱的基本功是把引流款卖贵，而不是卖便宜干货都在这.md 4.8 KB
    
    📄 041_贩卖焦虑产业链：观点错误率和收入成正比干货都在这贩卖焦虑女性成长.md 3.5 KB
    
    📄 042_勾兑，就是让某些生意，越不谈钱，越赚钱干货都在这知识商业逻辑.md 3.5 KB
    
    📄 043_中国顾客在美国山姆，买中国产的欧盟标准奶知识分享.md 9.4 KB
    
    📄 044_不到1年涨粉10w，发现自己其实还是新兵蛋子干货分享短视频运营小红书小红书运营小红书营销.md 1.4 KB
    
    📄 045_没销售能力的人，都爱拿精准人群当遮羞布干货都在这电商干货知识分享商业知识商业知识分享.md 10.8 KB
    
    📄 046_和小红书官方聊了一小时，运营干货全公开小红书运营小红书小红书开店小红书营销 dontbesilent.md 17.5 KB
    
    📄 047_赚钱简单，成为想赚钱的人，很难赚钱思维 dontbesilent 商业思维商业知识赚钱.md 832 B
    
    📄 048_11.14～11.16 深圳线下课 dontbesilent 深圳.md 1.6 KB
    
    📄 050_赚钱，就是养一个 24 小时清醒的脑子赚钱赚钱思维思维训练思维认知 dontbesilent.md 6.0 KB
    
    📄 051_恢复主体性，是一个人开始赚钱的前提主体性主体性复苏创业心态主体性思维.md 6.3 KB
    
    📄 052_为什么企业买电脑可以抵税，我买电脑就不能商业认知商业思维知识知识分享 dontbesilent.md 4.4 KB
    
    📄 053_这世上所有的商品都只是给买家一个转账的理由而已干货都在这分享创业干货自媒体创业运营干货干货.md 5.7 KB
    
    📄 054_年入百万之后，发现最重要的赚钱技能是定价定价定价策略商业模式底层逻辑 dontbesilent.md 9.2 KB
    
    📄 055_什么赛道能赚钱？看定价，1 分钟搞清楚干货都在这分享创业干货独立思考.md 4.9 KB
    
    📄 057_我一直重复，但没人相信：赚钱和认知没关系分享创业干货独立思考干货.md 5.0 KB
    
    📄 058_AI短视频平台真的来了！社交媒体股价暴跌 OpenAI Sora dontbesilent samaltman AI.md 6.3 KB
    
    📄 059_扎心：让你爽的品牌标语，背后全是商业效率商业思维商业思维认知.md 427 B
    
    📄 060_有一种全世界最稳定的贫穷叫做：见路不走穷人富人富人思维穷人思维.md 5.9 KB
    
    📄 061_数字人颠覆了自媒体行业❗️一键完成全球化 AI 人工智能数字人分享标记我的生活.md 958 B
    
    📄 062_年收入低于 50 万的知识付费，都死在不会定价定价商业商业思维商业模式商业知识.md 11.6 KB
    
    📄 064_平台抽佣之下，商家是没有资格谈用户体验的干货都在这独立思考分享创业干货干货.md 651 B
    
    📄 065_果粉用了安卓折叠屏，iPhone成备用机玩转数码折叠屏显示器安卓折叠屏手机.md 9.2 KB
    
    📄 066_9.15 日起，不建议任何人绑定小红书个人店铺经验分享分享创业干货自媒体干货经验.md 4.2 KB
    
    📄 067_会发短信的 AI 火了🔥聊到一半竟然让我打钱 AI 人工智能 dontbesilent 科技 ChatGPT.md 438 B
    
    📄 068_打工思维 = 对任何事都一头雾水的大脑西贝罗永浩预制菜打工思维富人思维.md 2.5 KB
    
    📄 069_99% 的知识付费业务，第一天就注定赚不了钱创业思维商业知识产品定价知识付费定价 dontbesilent.md 4.7 KB
    
    📄 070_一年内，5 个平台粉丝过万，我的自媒体方法论自媒体自媒体干货自媒体运营 dontbesilent 底层逻辑.md 18.6 KB
    
    📄 071_放弃时间管理之后，我才知道什么是高效赚钱时间管理精力管理精力管理比时间管理更重要创业心法 dontbesilent.md 7.4 KB
    
    📄 072_小红书赚了 30 亿，商家们集体心悸小红书小红书盈利小红书上市小红书商家电商创业.md 4.6 KB
    
    📄 073_爆哭😭 JD 客服说，我不再是东哥的兄弟了京东客服东哥兄弟.md 3.8 KB
    
    📄 074_我用的 30 个超级 AI，一次性讲清楚 AI agent 人工智能 dontbesilent gemini.md 2.8 KB
    
    📄 075_重磅 AI 论文！AI 胡说八道的真相彻底揭开 AI 大模型幻觉 OpenAI dontbesilent 人工智能.md 3.7 KB
    
    📄 077_普通人第一个 100 万的底层逻辑商业逻辑白手起家从零到一白手起家创业 dontbesilent.md 20.0 KB
    
    📄 079_偷懒上瘾：一句话让 3 个 AI 一起给我打工 AI aiagent genspark dontbesilent 人工智能.md 1.4 KB
    
    📄 080_总是被白嫖，是因为你展示了你的工具性拒绝白嫖商业认知商业思维认知如何拒绝 dontbesilent.md 800 B
    
    📄 081_普通人如何建立自己的被动收入（喂饭版）被动收入睡后收入被动收入和主动收入被动收入的重要性 dontbesilent.md 2.1 KB
    
    📄 082_连续 8 个月每天涨粉 1 千，我是如何做到的自媒体干货短视频开头五秒完播 dontbesilent 商业思维.md 7.6 KB
    
    📄 083_真心建议普通人尽快让 AI 做人生导师 AI 提示词提示词工程师 AI人生教练 AI人生导师.md 496 B
    
    📄 084_分享创业干货干货运营干货自媒体创业新媒体运营.md 8.7 KB
    
    📄 085_我涨粉 40 万没花一分钱 AI 自媒体流量短视频 dontbesilent.md 8.0 KB
    
    📄 086_月薪是人生最大陷阱，稳定是穷人的精神鸦片月薪二千五命比咖啡苦反脆弱 dontbesilent 死工资反脆弱思维.md 6.8 KB
    
    📄 087_AI 不是用来赚钱的，是用来不被淘汰的 AI 人工智能赚钱思维商业思维第四次工业革命.md 297 B
    
    📄 088_高毛利生意可能是普通人唯一的活路毛利单量从0到1 生意 dontbesilent.md 956 B
    
    📄 089_99% 的私域流量都不是资产，而是负债私域私域流量 dontbesilent 公域公域流量和私域流量.md 604 B
    
    📄 090_不相信自己能赚到钱，是一种习得性无助 dontbesilent 习得性无助习得性无助如何解决第一桶金赚钱是一种能力.md 500 B
    
    📄 091_「一人公司」是一场违背商业规律的造富神话一人公司的优缺点科斯定律组织管理商业思维 dontbesilent.md 8.6 KB
    
    📄 092_总想解决问题的老实人心态，是赚钱最大障碍 dontbesilent 赚钱思维交易成本交易成本理论商业思维.md 1.5 KB
    
    📄 093_从用户需求出发，是你做不成生意的第一步用户需求需求需求痛点产品 dontbesilent 商业思维.md 2.2 KB
    
    📄 094_如何用 4 个智能体 5 分钟完成爆款视频创作智能体 AI 爆款短视频 AI爆款 dontbesilent.md 1.6 KB
    
    📄 095_普通人想赚钱，求你先干掉心中的那个“烂好人” 真实生活分享计划赚钱思维商业思维商业思维认知 dontbesilent.md 1.4 KB
    
    📄 096_赚第一桶金用了 1 天，破除执念花了 3 年真实生活分享计划 dontbesilent 第一桶金创业记录第一桶金的故事.md 12.2 KB
    
    📄 097_抄我的人很多，抄了我还要开课的这是头一个抄袭抄袭可耻尊重原创被抄袭抄袭事件福布斯中国u30.md 4.2 KB
    
    📄 098_在小红书卖了 5 千单，开店的坑一篇讲清楚小红书开店小红书运营小红书店铺小红书店铺开通 dontbesilent.md 1.1 KB
    
    📄 099_薄利多销只会坑死自己，虚拟资料如何科学定价虚拟资料定价策略定价 dontbesilent 定价技巧.md 11.3 KB
    
    📄 100_3 分钟学会，让「赚钱」变得清晰且有迹可循干货分享观点分享 dontbesilent dontbesilent聊赚钱赚钱有迹可循.md 9.5 KB
    
    📄 101_搞不清楚赚钱的这套系统，人就真的赚不到钱干货都在这商业思维系统性思维 dontbesilent dontbesilent聊赚钱.md 1.3 KB
    
    📄 102_独家四维模型，5 分钟理解做私域底层逻辑 dontbesilent 私域私域运营公域流量思维模型.md 1.6 KB
    
    📄 103_5 分钟教你系统性思维，养一个会赚钱的脑子系统性思维 dontbesilent 会赚钱思维训练商业认知.md 3.0 KB
    
    📄 104_搞懂平台想要啥，比你闷头发 100 条都有用流量思维平台思维自媒体底层逻辑 dontbesilent 自媒体干货.md 2.0 KB
    
    📄 105_如何只靠原创涨粉 20 万，我的爆款逻辑创业vlog 自媒体 dontbesilent 原创爆款内容逻辑.md 10.9 KB
    
    📄 106_想投流，请你一定要先搞清楚流量的基础逻辑 dontbesilent 干货自媒体投流流量逻辑.md 9.6 KB
    
    📄 107_突然意识到很多人做自媒体失败的真正原因 dontbesilent 自媒体创业自媒体干货思维训练思考本质.md 9.3 KB
    
    📄 108_赚钱的第一步，是解决掉自己逻辑混乱的问题 dontbesilent 逻辑思维底层逻辑深度思考思维训练.md 5.5 KB
    
    📄 109_攒够 20 万粉丝之后，发现粉丝量真的没用短视频涨粉自媒体创业自媒体干货自媒体运营 dontbesilent.md 7.8 KB
    
    📄 110_赚钱最大的误区是总想发挥自己的优势去变现从零到一赚钱思维商业思维创业 dontbesilent.md 6.9 KB
    
    📄 111_总想把产品做得太有用，是在耽误你赚钱｜商业反直觉赚钱思维 dontbesilent 商业思维商业反直觉商业逻辑.md 6.9 KB
    
    📄 112_在内容平台赚 100 万的思维：货只是内容的赠品内容电商兴趣电商赚钱思维商业思维 dontbesilent.md 628 B
    
    📄 113_做知识付费涨粉 30万，聊聊如何教人赚钱 dontbesilent 知识付费自媒体分享自媒体商业思维商业思维.md 1.3 KB
    
    📄 114_越研究人群画像，越赚不到钱｜流量底层逻辑 dontbesilent 自媒体干货内容思维流量逻辑人群画像.md 8.3 KB
    
    📄 115_用免费送资料做成的私域，不是资产，是负债 AI生成视频私域运营 dontbesilent.md 825 B
    
    📄 116_自媒体做垂直流量，账号必死自媒体创业干货垂直流量垂直内容 dontbesilent.md 763 B
    
    📄 117_为什么大佬永远精力充沛：建立自我觉知能力自我觉知创业心力 dontbesilent 心力.md 1.4 KB
    
    📄 118_赚不到钱是因为上班上多了，打工和创业的逻辑是相反的 dontbesilent 打工思维老板思维富人思维.md 1.3 KB
    
    📄 119_AI 浏览器真可以帮你赚钱：15s 完成对标账号分析 AI浏览器 dontbesilent 对标账号对标账号分析账号运营技巧.md 2.3 KB
    
    📄 120_人要有「老子时间很贵」的概念，摆脱时间换钱的枷锁独立思考赚钱思维赚到一百万 dontbesilent 商业思维.md 6.7 KB
    
    📄 121_用 AI 五分钟“复制”别人的文档，开启虚拟资料生意虚拟资料项目 DeepSeek dontbesilent AI技巧.md 2.9 KB
    
    📄 122_如果远期目标是财务自由，那当下应该做什么独立思考创业财务自由目标管理 dontbesilent.md 5.9 KB
    
    📄 123_内耗 = 心理资本匮乏，停止内耗就会开始赚钱内耗心理资本内耗焦虑 dontbesilent 创业.md 7.2 KB
    
    📄 124_能赚钱的知识付费从来不卖虚拟资料，卖情绪❗️ dontbesilent 知识付费创业干货情绪价值虚拟资料.md 4.3 KB
    
    📄 126_每天都在思考创新的聪明人，最不容易赚到钱商业思维创业思维用户需求 dontbesilent.md 8.4 KB
    
    📄 127_做自媒体的煎熬期，把 99% 的人都熬失败了 dontbesilent 自媒体干货涨粉百万自媒体心法.md 2.7 KB
    
    📄 128_京东：给兄弟们交社保，就是企业最好的广告京东外卖喊话竞争对手 dontbesilent 流量刘强东企业获客.md 4.2 KB
    
    📄 129_卖奶茶根本不挣钱！雪王做知识付费年入百亿 dontbesilent 商业思维蜜雪冰城上市知识付费.md 4.0 KB
    
    📄 130_所有教你「先做内容再变现」的人都是耍流氓 dontbesilent 商业思维内容变现起号实操.md 5.2 KB
    
    📄 131_3 分钟教会你从零到一找到自己的生意❗️公开我的表格工作法 dontbesilent 商业思维赚钱思维创业思维.md 8.4 KB
    
    📄 132_像素级复制，年入百万。搬运抄袭，颗粒无收自媒体运营 dontbesilent 像素级对标像素级模仿原创价值.md 9.6 KB
    
    📄 133_表面上是知识付费不挣钱，其实只是变相打工 dontbesilent 知识付费商业思维 IP生意.md 6.1 KB
    
    📄 134_做 1 v 1 咨询，越接单越穷，越努力越穷 dontbesilent 商业思维.md 9.4 KB
    
    📄 135_越有钱越艰难，过亿的老板们做短视频更痛苦创始人ip dontbesilent 短视频创业.md 3.5 KB
    
    📄 136_我全网 20 万粉丝，输出上百万字，还是被 AI 干掉了 AI ai取代人类自媒体创业 dontbesilent.md 8.7 KB
    
    📄 137_为什么有人上来就爆款，为什么你死活没流量 dontbesilent 流量思维商业思维变现思维.md 5.7 KB
    
    📄 138_逆商决定财富上限｜年入百万的决定要素分享创业干货财富思维赚钱思维 dontbesilent.md 4.9 KB
    
    📄 139_不要把正常的流量变现的摸索阶段，当成是绝望。 dontbesilent 流量变现从零到一商业思维自媒体运营.md 5.0 KB
    
    📄 140_连线几百人之后我发现：不挣钱的业务，都长一个样自媒体创业知识分享 dontbesilent.md 6.1 KB
    
    📄 141_越穷越要野路子，赚钱就靠信息差信息差赚钱思路尊重用户尊重市场 dontbesilent.md 5.4 KB
    
    📄 142_一个问题毁掉你的知识付费，我花了四年才搞清楚 dontbesilent 知识付费内容变现干货.md 4.0 KB
    
    📄 143_一条视频，播放千万，涨粉 6 万，流程拆解揭秘短视频运营自媒体干货 dontbesilent.md 4.5 KB
    
    📄 144_成本 ¥17 我卖 ¥18，同行卖 ¥3.9，怎么赚钱电商创业私域运营 dontbesilent 我在抖音涨知识一分钟干货教学.md 3.2 KB
    
    📄 145_普通人赚钱第一步：确定性提高执行力的逻辑执行力普通人赚钱 dontbesilent.md 7.1 KB
    
    📄 146_1000 粉丝月入 4 万：直播连麦遇到的真实案例低粉高变现直播连麦商业咨询.md 6.1 KB
    
    📄 147_赚钱靠生意不靠行业｜24h 找到盈利生意的正确姿势生意经从零到一 dontbesilent 商业思维.md 8.9 KB
    
    📄 148_每条抖音视频稳定涨粉 2000+，公开我的内容生产流程自媒体创业自媒体运营 dontbesilent 短视频生产.md 7.7 KB
    
    📄 149_同样是卖虚拟资料，为什么别人能卖你 100 倍虚拟资料项目虚拟资料商业思维私域运营.md 6.2 KB
    
    📄 150_饰品创业：流量生意抄面子，品牌生意抄里子商业思维饰品创业商业逻辑.md 6.2 KB
    
    📄 151_99%的人不知道：项目下滑期才是最佳变现期 dontbesilent 商业变现卖项目业务下滑.md 5.6 KB
    
    📄 152_有流量又无法变现的人，都是没搞清楚内容变现的规则是什么干货分享自媒体创业运营 dontbesilent.md 2.5 KB
    
    📄 153_一万粉丝如何月入十万：低粉高变现核心逻辑商业思维新知贺岁眼界大开 dontbesilent 低粉高变现 IP人设.md 6.0 KB
    
    📄 154_你的总结能力，能在毁掉你的赚钱机会新年拍出超好运新知贺岁眼界大开鸣潮分享创业干货商业认知.md 5.3 KB
    
    📄 155_赚不到钱的根源，是对商业信息的过度简化新知贺岁眼界大开干货都在这知识科普商业认知商业逻辑.md 5.4 KB
    
    📄 156_普通人想赚 100 万，别折腾产品，全靠内容 dontbesilent 商业思维内容电商新知贺岁眼界大开抖音知识年终大赏.md 6.4 KB
    
    📄 157_误打误撞，开直播连麦，月入 20万，案例全拆解分享创业干货.md 18.4 KB
    
    📄 158_赔钱是能力，赚钱是时机创业.md 2.2 KB
    
    📄 159_小红书投流卖首饰，一年私域变现200W，案例全拆解.md 8.3 KB
    
    📄 160_千万不要先涨粉后变现，拆解流量变现的底层逻辑直播回放录屏商业咨询.md 5.4 KB
    
    📄 161_白板聊商业：为什么你的账号光涨粉、不赚钱？赚钱思维流量变现.md 3.4 KB
    
    📄 162_卖项目教人赚钱，到底是不是割韭菜？赚钱思维创业干货分享.md 2.1 KB
    
    📄 163_人在没钱的时候，唯一的翻身机会就是愤怒赚钱思维没钱路难走.md 1.3 KB
    
    📄 164_如何只用一张图片，在两个小时内赚3000块钱赚钱思维.md 1.3 KB
    
    📄 165_成年人了，不要用性格测试来选择自己的行业｜年入百万认知篇赚钱思维行业选择.md 1.0 KB
    
    📄 articles-index.md 15.6 KB
    
    📄 dontbesilent-titles.txt 8.7 KB
    
    📄 profile.md 16.3 KB
- 📁 by-element/
  - 📁 hooks/
    
    📄 hook-examples.md 13.9 KB
  - 📁 openings/
    
    📄 openings-index.md 16.3 KB
  - 📁 structures/
    
    📄 structure-templates.md 11.3 KB
  - 📁 titles/
    
    📄 dontbesilent-collection.md 23.0 KB
    
    📄 titles-index.md 7.3 KB
- 📁 techniques/
  - 📁 ai-workflow/
    
    📄 ai-workflow-index.md 8.7 KB
    
    📄 extract-logic-not-content.md 8.6 KB
  - 📁 concept-deconstruction/
    
    📄 concept-deconstruction-index.md 6.9 KB
    
    📄 concept-deconstruction.md 10.1 KB
  - 📁 hooks/
    
    📄 4-factor-scoring.md 10.4 KB
    
    📄 hook-hierarchy.md 16.7 KB
    
    📄 hooks-index.md 6.6 KB
  - 📁 psychology/
    
    📄 content-funnel.md 4.6 KB
    
    📄 content-funnel.png 199.1 KB
    
    📄 content-funnel_desc.md 4.6 KB
    
    📄 emotion-first-principle.md 8.6 KB
    
    📄 psychology-index.md 9.0 KB
- 📄 scheduler-protocol.md 4.8 KB
- 📄 search-workflow.md 7.8 KB
- 📄 steps-polish-to-publish.md 18.4 KB
- 📄 three-level-protocol.md 3.0 KB
📁 scripts/
- 📄 check-env.sh 4.7 KB
- 📄 cron-runner.sh 25.2 KB
- 📄 install-schedule.sh 1.5 KB
- 📄 parse_title.py 2.5 KB
- 📄 uninstall-schedule.sh 706 B
📁 skills/
- 📄 content-adapter.md 13.4 KB
- 📄 experience-tracker.md 5.9 KB
- 📄 scheduler.md 11.0 KB
- 📄 title-generator.md 12.0 KB
- 📄 topic-manager.md 23.7 KB
📄 .birdrc.json5 29 B
📄 .gitignore 466 B
📄 CLAUDE.md 10.4 KB
📄 README.md 13.6 KB
📄 README.zh-CN.md 12.5 KB
📄 SKILL.md 41.8 KB

SKILL.md

---
name: writing-assistant
description: Comprehensive writing workflow from ideation to publication. Guides users through creating polished articles by collecting ideas, asking clarifying questions, researching content, polishing drafts, adding images, and publishing to platforms like WeChat or X. Use when users want to write articles, blog posts, or long-form content, especially when starting from a topic idea, rough materials, or initial draft. Also use when users mention writing, publishing, or content creation workflows.
---

# Writing Assistant

A complete end-to-end writing workflow that transforms ideas, materials, or rough drafts into polished, illustrated articles ready for publication.

## Overview

This skill orchestrates a multi-step writing process:
0. **Create progress tracker** — initialize per-session tracking file
1. **Choose starting mode + select platform** — load from topic pipeline or start fresh; determine target platform upfront
2. **Search references, benchmarks, and techniques** — find styles, patterns, viral cases, and writing methodologies
3. **Collect and clarify** — interactive questioning (Modes 1 & 2), apply selected techniques
4. **Element-level reference** — title (via title-generator), opening, structure, with technique cross-reference
5. **Process draft** — Mode 3 only, apply techniques throughout body
6. **Polish** — using content-research-writer, with technique-aware instructions
7. **Generate illustrations** — using baoyu-xhs-images
8. **Create final article** — combine content + images
9. **Review and platform adaptation** — review, optional multi-platform adaptation with technique re-application
10. **Publish** — optional, to WeChat or X

**Note:** This skill focuses on **writing**. For topic management (recording ideas, analyzing viral content, monitoring trends), use `{skill-dir}/skills/topic-manager.md`. For standalone title generation, use `{skill-dir}/skills/title-generator.md`.

## Execution Principles (MUST NOT skip)

1. **Progress file first**: After entering the writing workflow, the very first action must be creating the progress file (Step 0). No other steps may execute before the progress file is created. The progress file is the "roadmap" for this session — all subsequent operations strictly follow the step order in the progress file.
2. **Initialize only once**: Dependency checks, API key validation, and environment pre-checks are performed only once in Step 0, never repeated. Subsequent steps skip initialization when they see the `Initialization: ✅ completed` marker in the progress file.
3. **Output directory convention**: All output files are stored under `writtings/outputs/{topic-slug}/`.
4. **禁止擅自兜底**: 任何命令或工具调用失败时，必须如实报告给用户，由用户决定是否采用替代方案。不得未经允许自行使用 WebSearch 或其他方式作为降级兜底。
5. **搜索实时性**: 搜索热点/热门内容时，禁止以任何时间段（月/周/季度）为单位搜索。不搜"X月热点总结""本周趋势""近期回顾"等盘点类内容，只搜具体话题关键词获取当下实时内容。
6. **自主模式不中断原则**: 自主模式下，AI 全程自主执行，不在任何步骤等待用户。工具或技能调用失败时，记录失败详情到 Execution Log 和 Autonomous Decision Log，跳过该环节继续后续流程。在最终摘要中汇总所有失败项，供用户事后处理。此原则覆盖 Principle #4（禁止擅自兜底）在自主模式下的行为：自主模式下，失败 → 记录 + 跳过，而非报告等待。**自主模式必须执行到 Completion Gate 全部通过才能停止，见 Autonomous Mode 章节的"Completion Gate"。**
7. **Step Closure Gate（步骤关门协议）**: 每个 Step 的 Start 块包含对前一步的验证。开始 Step N 之前，必须确认 Step N-1 的三项已完成：(1) checkbox 全部标 `[x]`，(2) Execution Log 已填写（无 pending），(3) Corrections Log 已更新（如有纠错）。如果任何一项缺失，**立即补填后再继续**。这是自愈机制：即使步骤结束时忘了更新，下一步开始时必定补上。
8. **Session Close Protocol（会话关闭协议）**: 用户发出结束信号（"先到这里"/"暂停"/"够了"/"到这里"/"结束"等）时，不得立即停止。先检查是否有标记为"不可跳过"的未完成步骤（流程自检+复盘）。如有，先执行再结束。响应模板："好的，在结束前我先完成[步骤名]（不可跳过），完成后就到这里。"

## Three-Level Reference System

Assets and references follow a three-level hierarchy (system / user / project). Content merges on read; lower levels override higher on conflict.

**Full protocol:** Read `{skill-dir}/references/three-level-protocol.md` for level definitions, READ:3L merge rules, WRITE:user/project targets, and quick reference table.

**Shorthands used in this file:** `READ:3L` = read all three levels and merge. `WRITE:user` / `WRITE:project` = target a specific write level.

## Companion Skills (project-local)

以下 skills 位于 `{skill-dir}/skills/` 目录下（`{skill-dir}` 定义见 `{skill-dir}/references/three-level-protocol.md`）：

| Skill | 文件路径 | 调用时机 |
|-------|---------|---------|
| title-generator | `{skill-dir}/skills/title-generator.md` | Step 4 (标题生成) / 独立调用 |
| content-adapter | `{skill-dir}/skills/content-adapter.md` | Step 9b (平台适配) / 独立调用 |
| topic-manager | `{skill-dir}/skills/topic-manager.md` | 选题管理（独立调用） |
| experience-tracker | `{skill-dir}/skills/experience-tracker.md` | Experience Check（每步交互后） |
| scheduler | `{skill-dir}/skills/scheduler.md` | 定时任务管理（独立调用） |

### Project-Local Skill 执行协议（MUST）

调用 project-local skill 时，**必须**：

1. **使用 Read 工具读取完整文件**（不得凭记忆替代文件内容）
2. **严格按文件中的 Step 顺序执行**，不得跳过任何步骤
3. **每个 Step 的输出格式必须与文件要求一致**

❌ 禁止：看到 `skills/title-generator.md` 后凭记忆生成标题
✅ 正确：Read 工具读取文件 → 按 Step 0 → Step 1 → ... → Step 6 顺序执行

## Experience Check System

**Two mechanisms work together to ensure corrections are captured:**

### Mechanism 1: Pre-step lesson check
**Before every step**, check `writtings/assets/experiences/lessons.md` (`READ:3L`) if it exists. Apply any relevant lessons to avoid repeating past mistakes.

### Mechanism 2: Post-interaction correction checkpoint
**After every step that involves user interaction** (Steps 1, 3, 4, 5, 8, 9), perform this checkpoint:

> **Experience Checkpoint:**
> Review the user's response in this step. Did the user:
> - Directly negate your output? ("不是这样", "不对", "错了")
> - Point out style issues? ("太 AI 了", "爹味", "太正式了", "不自然")
> - Provide a corrected version? ("应该是...", "改成...", "我要的是...")
> - Express dissatisfaction? ("别这样写", "这不是我想要的")
> - Rewrite your output instead of accepting it?
>
> **If ANY of the above occurred:**
> 1. **Immediately** create a case file in `writtings/assets/experiences/cases/` (`WRITE:user`) (format: `{YYYY-MM-DD}-{slug}.md`, see `{skill-dir}/skills/experience-tracker.md`)
> 2. Log the correction in the progress tracker's Corrections Log with `Case Recorded? = Yes` and fill in the Case File path
> 3. Ask user: "这条经验是通用的还是仅针对本文？"
>    - 通用 → update `writtings/assets/experiences/lessons.md` (`WRITE:user`)
>    - 仅本文 → update `writtings/outputs/{topic-slug}/assets/experiences/lessons.md` (`WRITE:project`)
> 4. Confirm to user: "已记录这条经验。"
> 5. Then proceed to the next step
>
> **NEVER** leave a correction as "Pending" in the Corrections Log without creating a case file. Every correction must be closed out on the spot.
>
> **If none occurred:** Note "No corrections" in the progress tracker and proceed.

This checkpoint is marked as **"Experience Check"** in every applicable step below. Do NOT skip it.

### Autonomous Mode Behavior

自主模式下：
- Mechanism 1（步前读 lessons.md）：不变，照常执行
- Mechanism 2（步后纠错检测）：跳过。记录 "Autonomous mode — no interaction"

## Autonomous Mode（自主模式）

> **⚠️ 核心约束：自主模式必须执行到 Completion Gate 全部通过才能停止。不得在任何中间步骤停下。详见下方"Completion Gate"章节。**

自主模式下，用户发出指令后即可离开，AI 全程自动完成所有步骤，中间不会卡住。

### 开启方式

用户在初始消息中包含自主模式信号即可开启。例如：
- "自主模式，帮我写一篇关于 X 的小红书文章"
- "全自动写一篇公众号文章，主题是 Y，写完发到公众号"
- "autonomous mode, write about Z for XHS, just produce the final article"

AI 从用户的初始消息中理解以下信息（不强制格式，用自然语言即可）：
- **写什么**：主题/素材/草稿（对应 Mode 1/2/3）
- **发到哪**：目标平台（如未指定，AI 根据内容特征选择最合适的平台）
- **完成条件**：用户想要的最终结果是什么（见下方"完成条件"部分）

如果用户的初始消息信息充足，直接开始执行。
如果关键信息缺失（比如完全没提主题），在 Step 1 一次性追问所有缺失项——
这是自主模式下唯一允许的交互，且仅发生在执行开始前。

### 自主能力边界

自主模式只执行 AI 能完全自主完成的操作。判断标准：**该操作是否需要用户在过程中参与（如登录、扫码、手动确认）？**

- **可自主完成**：写文章、润色、配图、适配平台、通过 API 发布、读写本地文件
- **不可自主完成**：需要用户登录的发布流程、需要扫码授权的操作、需要用户手动确认的外部系统操作

自主模式下默认不执行发布（Step 10），除非该发布方式 AI 可以完全自主完成（如通过 API 提交、可用已有 cookie/token 的 CLI 工具等）。如果用户指令中包含发布要求，AI 评估发布路径是否可自主完成：
- 可以 → 执行并记录
- 不可以 → 跳过发布，在执行摘要中标注"发布需用户手动完成"并给出所需信息（文件路径、平台等）

### 完成条件（Completion Condition）

完成条件完全由用户自然语言定义，AI 负责理解并执行。不限于"写一篇"——支持任意任务组合。

**示例场景：**

| 用户说的 | AI 理解为 |
|---------|----------|
| "帮我写一篇关于 X 的小红书文章" | 单篇，写到最终文章 |
| "写完发到公众号" | 单篇，评估公众号发布是否可自主完成；可以则发布，否则写到最终文章并标注 |
| "写三篇不同角度的小红书文章" | 三篇循环，每篇独立走 Step 0-10，各自独立 progress tracker |
| "把 inbox 里的选题都写了" | 读取 inbox.md，逐个选题执行完整流程 |
| "一直写，把 developing/ 里的选题全部完成" | 循环处理 developing/ 中所有选题 |
| "写一篇小红书，再适配一版公众号" | 单篇 + 适配，写到最终文章 |
| "根据这份素材，写小红书和公众号两个版本" | 单次素材收集，两个平台各走一次 Step 4-10 |
| "每天从 inbox 里挑一个选题写" | AI 理解意图，本次执行一篇（无法跨会话持续） |

**设计原则：**
- 不定义固定格式，AI 从自然语言中理解
- 多篇场景：每篇文章独立走完整流程（独立 progress tracker、独立 outputs 目录）
- 批量场景：逐个执行，前一篇完成后自动开始下一篇
- 如果完成条件不明确，默认为"单篇，写到最终文章，不发布"
- 每篇文章完成后，在该篇的 progress tracker 中生成执行摘要；全部完成后，生成一份总体摘要

### 执行规则

1. **跳过所有用户交互**：所有 "ask user"/"present to user"/"wait for confirmation" 环节，AI 自主决策并记录理由
2. **失败不停**：任何工具/技能/搜索失败 → 记录到 Execution Log（失败原因 + 影响 + 跳过了什么）→ 继续下一步。不报错等待，不静默兜底
3. **卡住检测**：如果某个操作（工具调用、技能调用、搜索等）长时间无响应或反复重试无果，AI 应主动判断为卡住。处理方式：记录卡住情况（哪个操作、等了多久、尝试了什么）→ 尝试替代方案（如有）→ 替代方案也不行则跳过该环节继续后续流程。绝不在一个环节上无限等待。
4. **经验库照用**：步前读 lessons.md 不变，步后纠错检测跳过（无用户交互 = 无纠错）
5. **决策全记录**：每个自主决策记录到 Autonomous Decision Log（选了什么、为什么、还考虑了什么）
6. **灵活理解用户意图**：AI 根据用户的自然语言说明决定执行范围，不依赖固定的完成条件格式
7. **循环执行**：多篇/批量场景下，每篇独立执行完整工作流，前一篇的经验教训自动应用到后续篇章
8. **⚠️ 绝不提前停止（Completion Gate 强制）**：自主模式下，AI 不得在任何步骤中途停止、向用户汇报中间状态然后等待、或在未满足 Completion Gate 的情况下结束会话。每完成一个步骤后，必须立即读取 progress tracker 确认下一步是什么，然后继续执行。唯一允许停止的时刻是 Completion Gate 全部通过。

### ⚠️ Completion Gate（停止条件 — 强制）

**这是自主模式的核心约束。AI 不满足以下全部条件时，绝不允许停止执行。**

每完成一个步骤后，AI 必须对照此清单。如有任何一项未满足，必须继续执行对应步骤。

**单篇场景停止条件：**

| # | 条件 | 验证方式 |
|---|------|---------|
| G1 | `{topic-slug}-final.md` 文件已创建且非空 | 检查文件是否存在于 writtings/outputs/{topic-slug}/ |
| G2 | Progress tracker 中 Step 0-9 所有 checkbox 已标记 `[x]` | 读取 progress tracker 逐项核对 |
| G3 | Step 10 已执行或已明确标记为"跳过"（含跳过理由） | 检查 Step 10 Log |
| G4 | Execution Log 每个步骤都有记录（无空白步骤） | 检查 Step 1-10 Log 区域 |
| G5 | Autonomous Decision Log 至少有 3 条记录 | 检查 Autonomous Decision Log 表格 |
| G6 | 流程自检 + 复盘已完成 | 检查"流程自检 + 复盘"区域的 checkbox |
| G7 | 自主模式执行摘要已填写（非 pending） | 检查执行摘要区域 |

**多篇/批量场景额外停止条件：**

| # | 条件 | 验证方式 |
|---|------|---------|
| G8 | 所有计划文章都已完成（各自满足 G1-G7） | 检查每篇的 progress tracker |
| G9 | 总体摘要已生成 | 检查最后输出 |

**Completion Gate 执行协议：**
1. 每个 Step 完成后，读取 progress tracker，确认当前进度，然后立即执行下一个未完成的 Step
2. Step 9 完成后（或 Step 10 完成/跳过后），执行流程自检 + 复盘
3. 复盘完成后，逐项核对 G1-G7（多篇加 G8-G9）
4. **全部通过** → 填写执行摘要 → 向用户呈现最终结果 → 允许停止
5. **任何一项未通过** → 定位缺失项 → 回到对应步骤补完 → 重新核对，直到全部通过
6. **绝不允许**：在 Gate 未通过时向用户说"已完成"或停止执行

### 各步自主决策策略

每个步骤的 "If Autonomous Mode" 块定义了具体策略。总体原则：
- **有参考库内容时**：基于经验库 + 技巧匹配 + 平台规范做选择
- **无参考库内容时**：基于 AI 自身判断 + 平台搜索结果做选择
- **有多个选项时**：选择最契合平台规范和所选技巧的选项
- **失败时**：记录并跳过，在 Execution Log 标注影响
- **卡住时**：判断为卡住 → 记录 → 尝试替代方案 → 替代也不行则跳过并继续

### 最终输出

自主模式执行完毕后，呈现一份结构化的执行摘要：

**单篇场景**：写入该篇 progress tracker 的 "自主模式执行摘要" 部分。

**多篇/批量场景**：每篇各自有 progress tracker 中的执行摘要，另外在最后生成一份总体摘要，包含：
- 完成了几篇，每篇标题和平台
- 每篇的产出文件路径
- 全局跳过/失败项汇总
- 前一篇的经验是否影响了后续篇章的决策

摘要内容：
- 产出文件清单
- 关键决策 Top 3-5（选了什么、为什么）
- 跳过/失败项清单（如有）
- 建议复查点（AI 自评哪些决策最可能需要用户调整）

## Workflow

### Initial Setup: Initialize Workspace

Before starting the workflow, check the user's working directory and ensure the **user-level** directory structure exists. The system-level directories ship with the skill and are always available.

**User-level directory structure (create if absent):**
```
writtings/
├── assets/
│   ├── topics/
│   │   ├── inbox.md
│   │   ├── developing/
│   │   └── benchmarks/
│   │       ├── benchmarks-index.md
│   │       └── monitor-config.md
│   └── experiences/
│       ├── cases/
│       └── lessons.md
├── references/              # (optional — user-level references grow over time)
├── outputs/                 # (created per article)
└── schedules/               # (created when scheduling tasks)
```

**Process:**
1. Verify system-level assets exist (in skill directory)
2. Check if user-level directories and files exist in `writtings/`
3. Create any missing user-level directories and template files silently
4. Do NOT overwrite existing files — only create if absent
5. Proceed to dependency check

### Initial Setup: Check Dependencies

Run the environment check script:

```bash
bash scripts/check-env.sh
```

The script checks: OPENROUTER_API_KEY in `.env`, `bird` CLI availability, `config/bird.json5` propagation, installed skills in `.claude/skills/`, and baoyu-post-to-wechat npm dependencies.

**Handle results:**
- All checks pass → proceed to Step 0
- Required dependencies missing → install from bundled `dependencies/` directory:
  ```bash
  mkdir -p .claude/skills && cp -r dependencies/<skill-name> .claude/skills/
  ```
- User declines required dependencies → explain workflow cannot proceed, offer to pause

**Required dependencies:** content-research-writer, baoyu-xhs-images, xiaohongshu, wechat-article-search, generate-image, baoyu-post-to-wechat, baoyu-post-to-x (or x-article-publisher).

### Step 0: Create Progress Tracker

Before starting the writing workflow, create a session progress tracker file. This file serves as the "checklist brain" for the entire session, ensuring no step or sub-step is skipped.

**Process:**
1. Determine a topic slug (use a short English slug for the filename, e.g., "attention-economy", "ai-writing-tips")
2. Create the output directory: `writtings/outputs/{topic-slug}/`
3. Read `{skill-dir}/assets/progress-template.md` (`READ:3L`) and instantiate it — replace all `{topic-slug}` with the actual slug
4. Save as `writtings/outputs/{topic-slug}/{topic-slug}-progress.md`
5. **Environment pre-check (only in Step 0, once):** Run `bash scripts/check-env.sh` if not yet run in Initial Setup. Record results in the progress file's Session Metadata (Environment fields). If anything is missing, fix on the spot or prompt user — do not leave for later steps.
6. Proceed to Step 1

**Usage rules for the progress tracker:**
- **Before each step:** Read the progress tracker to see current status and what this step requires
- **After each step:** Update the checklist, marking items as `[x]` and adding notes. **Update the Execution Log** for this step (see below).
- **After each step (Autonomous Mode):** 更新完 progress tracker 后，**立即**确认下一个未完成的 Step 是什么，然后**立即**开始执行。不得停下来等待、汇报中间状态、或询问用户。直到 Completion Gate 全部通过。
- **On correction:** Immediately add a row to the Corrections Log
- **At session end:** Review the tracker for any missed items, then run the retrospective

### Execution Logging (每步必做)

**After completing each step**, update the Execution Log section in the progress tracker. Record:
- **What you did**: key actions and decisions made
- **What you used**: which references, techniques, tools were applied
- **What you skipped**: anything you could have done but didn't, and why
- **Friction**: any obstacles, tool failures, or awkward points in the process

This log is reviewed during the 流程复盘 (retrospective) at session end to identify process issues.

### Session Close Protocol（会话关闭协议）

当用户发出结束信号（"先到这里"/"暂停"/"够了"/"到这里"/"结束"/"今天就到这"等）时：

1. **不立即停止**。先检查当前状态：
   - 当前 Step 是否已完成 Closure？（checkbox + Execution Log + Corrections Log）
   - 是否有"不可跳过"的步骤尚未执行？（流程自检+复盘）
2. **如有未完成项**，告知用户并执行：
   > "好的，在结束前我先完成 [具体步骤]（流程要求不可跳过），完成后就到这里。"
3. **执行未完成的强制步骤**
4. **然后结束会话**

此协议覆盖所有结束信号，不论措辞。唯一例外：用户明确说"跳过复盘直接结束"等直接否定时，记录跳过原因后结束。

### Step 1: Choose Starting Mode + Select Platform

> **Start:** Read progress tracker. **⚠️ Closure Gate:** Verify Step 0 completed (progress file exists, environment checked). Update Step 1 status to in-progress.

**1a-0. Check for Autonomous Mode:**

检测用户初始消息是否包含自主模式信号（"自主模式"/"全自动"/"autonomous mode"/"自动完成" 等）。

如果检测到：
1. 从用户消息中理解：主题/素材、目标平台、完成条件（写几篇、是否发布、是否适配等）
2. 如果关键信息缺失，一次性追问（这是自主模式唯一允许的交互）
3. 在 progress tracker metadata 记录：`Autonomous Mode: enabled`，并记录解析出的执行范围
4. 多篇/批量场景：确定文章列表（从 inbox/developing/ 读取，或从用户消息解析）
5. 从此步起，所有后续步骤走自主路径，不再有任何用户交互
6. 多篇场景：每篇独立创建 progress tracker，循环执行完整 Step 0-10

**1a. Determine Target Platform:**

Before anything else, ask the user which platform they are writing for:
- 小红书 (Xiaohongshu)
- 微信公众号 (WeChat Official Account)
- 抖音 (Douyin)
- X/Twitter
- Other / Not sure yet

Record the platform in the progress tracker metadata. The platform choice influences:
- Title constraints and style (Step 4 via title-generator)
- Content length and structure (Steps 3-6)
- Tone and voice (Steps 3-6)
- Image style and quantity (Step 7)
- Publishing options (Step 10)

If user is unsure, default to the most common platform they use, or proceed without a platform constraint and determine it later. But always try to establish platform early.

> **If Autonomous Mode:** 从用户初始消息解析平台；未指定则根据内容特征自动选择最合适的平台（如生活类话题 → 小红书，深度观点 → 公众号），记录选择理由到 Autonomous Decision Log。跳过所有"ask user"环节，直接进入下一步。

**1b. Check for Developed Topics:**

- Look in `writtings/assets/topics/developing/` (`READ:user`) for existing topic files
- If topics exist, present them: "You have N developed topics. Would you like to continue with one?"
- If user picks a topic, load its file (outline, benchmark references, title candidates) and proceed to Step 2 with this context

**1c. Choose Starting Mode:**

If no developed topics, or user wants to start fresh, ask for one of three modes:

**Mode 1: Topic-Based**
- User provides a topic or theme they want to write about
- Most suitable when starting from scratch with just an idea

**Mode 2: Materials-Based**
- User provides loosely organized materials, notes, or reference content
- Can include rough notes, copied references, or miscellaneous content
- Ask for file paths if materials are in files

**Mode 3: Draft-Based**
- User provides an unpolished initial draft
- Suitable when the user has already written a rough version
- Ask for file path if draft is in a file

**1d. Update Progress Tracker:**

Update the progress tracker metadata (Platform, Mode, Topic) and mark Step 1 checklist items.

> **Experience Check:** Review user's responses. Did user provide any corrections? If yes, invoke `{skill-dir}/skills/experience-tracker.md` and log in Corrections Log. Then proceed.

> **If Autonomous Mode:** 跳过 Experience Check（无用户交互 = 无纠错）。记录 "Autonomous mode — no interaction"。

> **End — Step 1 Closure (complete ALL before proceeding):**
> 1. Mark all Step 1 checkboxes `[x]` in progress tracker
> 2. Fill Step 1 Execution Log (every field, no "pending")
> 3. Update Corrections Log if user corrected anything this step
> 4. **Verify**: re-read progress tracker, confirm no blanks for Step 1
> Then proceed to Step 2.

> **Start:** Read progress tracker. **⚠️ Closure Gate:** Verify Step 1: all checkboxes `[x]`? Execution Log filled (no "pending")? Corrections Log up to date? If anything missing → fix NOW before proceeding. Then update Step 2 status to in-progress.

After understanding the user's topic/theme and target platform, search all reference sources. **Full search workflow:** Read `{skill-dir}/references/search-workflow.md` for detailed process.

**Mandatory actions (do NOT skip):**

1. **Search references** (`READ:3L`): Check `writtings/references/authors/`, `writtings/references/by-element/`, `writtings/references/techniques/` for style matches, element patterns, and writing methodologies
2. **Search benchmarks** (`READ:3L`): Check `writtings/assets/topics/benchmarks/` for viral cases
3. **Select and record techniques** in progress tracker under "Applied References & Techniques"
4. **【必做】Search target platform for popular content** on the same topic using platform tools:
   - 小红书: Invoke xiaohongshu skill (MCP tool: `search_feeds`, keyword: "{keywords}") — returns interactInfo; sort by engagement priority (commentCount > likedCount = sharedCount > collectedCount)
   - 抖音: WebSearch
   - X/Twitter: `bird search "{keywords}" --cookie-source chrome`
5. **【强制】Record search results** in progress file Session Notes (platform, command, keywords, result count, top 3-5 high-engagement items, extracted patterns). See `{skill-dir}/references/search-workflow.md` for required format.
6. **Present reference summary** to user: author style, element patterns, techniques, benchmarks

> **If Autonomous Mode:** 有匹配作者风格则自动选用，无则跳过作者风格参考。搜索失败时记录失败原因到 Execution Log 并继续。不向用户呈现参考摘要，直接进入 Step 3。

> **End — Step 2 Closure (complete ALL before proceeding):**
> 1. Mark all Step 2 checkboxes `[x]` in progress tracker
> 2. Fill Step 2 Execution Log (every field, no "pending")
> 3. Update Corrections Log if user corrected anything this step
> 4. **Verify**: re-read progress tracker, confirm no blanks for Step 2
> Then proceed to Step 3.

### Step 3: Collect and Clarify (Modes 1 & 2 Only)

> **Start:** Read progress tracker. **⚠️ Closure Gate:** Verify Step 2: all checkboxes `[x]`? Execution Log filled (no "pending")? Corrections Log up to date? If anything missing → fix NOW before proceeding. Check which techniques were selected in Step 2. Then update Step 3 status to in-progress.

For Modes 1 and 2, use an interactive questioning approach:

1. **Analyze the provided content first**:
   - Read and understand what the user has already provided
   - Identify what's clear vs. what needs clarification
   - Note gaps, ambiguities, or areas that need expansion

2. **Ask tailored, content-specific questions**:
   - Formulate questions based on the specific topic and materials provided
   - Focus on filling identified gaps and resolving ambiguities
   - Ask about aspects that are unclear or need deeper exploration
   - Adapt questions to the user's context and writing goals

3. **Collect user responses** systematically

4. **Supplement with research** if context is insufficient:
   - Use WebSearch for relevant information
   - Use WebFetch if user provides URLs
   - Gather supporting materials from the internet

5. **Apply selected techniques to content organization**:
   - Review the techniques selected in Step 2 (recorded in progress tracker)
   - Apply them to how you organize and frame the content:
     - **If Content Funnel was selected**: Determine the TOFU/MOFU/BOFU positioning. For TOFU content: ensure the draft uses plain language, triggers emotion (curiosity/anxiety/empathy), and relates to the reader personally. Run the Practice Guide checklist from `writtings/references/techniques/psychology/psychology-index.md` against your draft outline.
     - **For each selected technique**: Review its Practice Guide and apply its principles to the content structure, framing, and language choices.
   - This is NOT just about elements (title, opening) — techniques should influence the entire article body: paragraph structure, argument flow, examples chosen, language register, and emotional arc.

6. **Organize into initial draft** based on:
   - User's answers to questions
   - Researched supplementary materials
   - Logical article structure
   - **Selected technique principles** (e.g., if TOFU: simple, emotional, relatable)
   - **Platform constraints** (e.g., 小红书: ≤1000 words, casual tone)

7. **Proceed to Step 4** for element-level refinement

> **If Autonomous Mode:** 跳过交互式提问。AI 自行生成澄清问题并自答（基于素材 + 搜索结果 + 选定技巧），将 Q&A 模拟记录到 Execution Log。如需补充研究（WebSearch/WebFetch），正常执行；搜索失败则记录并基于已有素材继续。直接组织初稿并进入 Step 4。

> **Experience Check:** Review user's responses during questioning. Did user provide any corrections? If yes, invoke `{skill-dir}/skills/experience-tracker.md` and log in Corrections Log. Then proceed.

> **If Autonomous Mode:** 跳过 Experience Check（无用户交互 = 无纠错）。记录 "Autonomous mode — no interaction"。

> **End — Step 3 Closure (complete ALL before proceeding):**
> 1. Mark all Step 3 checkboxes `[x]` in progress tracker
> 2. Fill Step 3 Execution Log (every field, no "pending")
> 3. Update Corrections Log if user corrected anything this step
> 4. **Verify**: re-read progress tracker, confirm no blanks for Step 3
> Then proceed to Step 4.
- Ask 2-4 questions at a time (avoid overwhelming the user)
- Tailor each question to the specific content provided - no fixed templates
- Let the content guide the questions - if something is already clear, don't ask about it

### Step 4: Element-Level Reference (Title, Opening, Structure)

> **Start:** Read progress tracker. **⚠️ Closure Gate:** Verify Step 3: all checkboxes `[x]`? Execution Log filled (no "pending")? Corrections Log up to date? If anything missing → fix NOW before proceeding. Review selected techniques and target platform. Then update Step 4 status to in-progress.

Before finalizing the initial draft, use the reference library AND selected writing techniques to refine key writing elements.

**This step applies to all modes** (Topic-Based, Materials-Based, and Draft-Based).

**Process:**

1. **Title Refinement**:
   - **【强制】使用 Read 工具读取 `{skill-dir}/skills/title-generator.md`，严格按照文件中的 Step 0-6 顺序执行。** 不得跳过任何步骤，不得凭记忆替代文件内容。
   - Pass the target platform (determined in Step 1) to the title-generator
   - The title-generator will search references, apply platform rules, and generate 5 candidates
   - **Technique cross-reference**: If Content Funnel technique was selected and article is TOFU, verify titles follow TOFU principles: broad appeal, emotional trigger, no jargon, relatable to wide audience
   - Let user choose or customize

> **If Autonomous Mode (Title):** 生成 5 个候选标题后，按平台规范 + 技巧匹配度自动选择最佳标题。记录全部候选和选择理由到 Autonomous Decision Log。

2. **Opening Paragraph Refinement**:
   - Read `writtings/references/by-element/openings/openings-index.md` (`READ:3L`) if available
   - Suggest an opening approach based on reference techniques
   - **Technique cross-reference**: Apply selected techniques to opening choice:
     - Content Funnel TOFU: opening should trigger universal emotions (anxiety, curiosity, empathy), be instantly understandable, and relate to the reader
     - Content Funnel MOFU: opening can assume some context, focus on credibility and depth
   - Common techniques to reference:
     - Anxiety resonance → contrarian pivot → cascade questions
     - Bold claim → self-correction → answer reveal
   - Draft 1-2 opening paragraph options for user to choose

> **If Autonomous Mode (Opening):** 生成 1-2 个开头选项后，按技巧原则 + 平台风格自动选择。记录选择理由到 Autonomous Decision Log。

3. **Structure Planning**:
   - Read `writtings/references/by-element/structures/structure-templates.md` (`READ:3L`) if available
   - Propose an article structure based on successful patterns
   - **Technique cross-reference**: Align structure with selected techniques:
     - Content Funnel: structure should match the funnel stage — TOFU favors shorter, punchier structures; MOFU favors deeper, more layered structures
     - Review the "Application Scenarios" table in the technique documentation for content type guidance
   - Adjust based on user's content, preferences, and **target platform** constraints

> **If Autonomous Mode (Structure):** 基于平台规范 + 所选技巧自动选择最合适的结构模板。记录到 Autonomous Decision Log。

4. **Hook Integration**:
   - Read `writtings/references/by-element/hooks/hook-examples.md` (`READ:3L`) if available
   - Note effective hook techniques to use within the article
   - Plan where to place engaging hooks in the draft
   - **Technique cross-reference**: Ensure hooks align with the article's funnel stage and platform expectations

> **Experience Check:** Review user's choices and feedback on proposed elements. Did user provide any corrections? If yes, invoke `{skill-dir}/skills/experience-tracker.md` and log in Corrections Log. Then proceed.

> **If Autonomous Mode:** 跳过 Experience Check（无用户交互 = 无纠错）。记录 "Autonomous mode — no interaction"。

> **End — Step 4 Closure (complete ALL before proceeding):**
> 1. Mark all Step 4 checkboxes `[x]` in progress tracker
> 2. Fill Step 4 Execution Log (every field, no "pending")
> 3. Update Corrections Log if user corrected anything this step
> 4. **Verify**: re-read progress tracker, confirm no blanks for Step 4
> Then proceed to Step 5 or Step 6 with the refined elements.

### Step 5: Process Draft (Mode 3 Only)

> **Start:** Read progress tracker. **⚠️ Closure Gate:** Verify Step 4: all checkboxes `[x]`? Execution Log filled (no "pending")? Corrections Log up to date? If anything missing → fix NOW before proceeding. Review selected techniques. Then update Step 5 status to in-progress.

For Mode 3 (Draft-Based):

1. **Analyze the user's draft**:
   - Read and understand the existing content
   - Identify the main theme and structure
   - Note areas that could be improved

2. **Apply Step 4 (Element-Level Reference)**:
   - Even with an existing draft, offer to refine title, opening, and structure
   - Suggest improvements based on reference library patterns
   - Let user decide what to keep vs. what to change

> **If Autonomous Mode (Step 5):** 保留用户原稿的核心结构和语气，基于选定技巧自动改进。不询问用户保留/修改决策，AI 自行判断并记录理由到 Autonomous Decision Log。

3. **Apply selected techniques throughout the article body**:
   - Do NOT limit technique application to just title and opening
   - Review the entire draft body against selected technique principles:
     - **Language register**: Does the body language match the technique's guidance? (e.g., TOFU = plain language, no jargon)
     - **Emotional arc**: Does the body maintain emotional engagement throughout? (e.g., hooks at section transitions, relatable examples)
     - **Reader relatability**: Can the target audience see themselves in the content? (e.g., "和我有关" principle from Content Funnel)
     - **Paragraph-level application**: Each paragraph should be checked against technique checklists, not just the overall structure
   - Suggest specific paragraph-level improvements based on techniques

> **Experience Check:** Review user's decisions on what to keep/change. Did user provide any corrections? If yes, invoke `{skill-dir}/skills/experience-tracker.md` and log in Corrections Log. Then proceed.

> **If Autonomous Mode:** 跳过 Experience Check（无用户交互 = 无纠错）。记录 "Autonomous mode — no interaction"。

> **End — Step 5 Closure (complete ALL before proceeding):**
> 1. Mark all Step 5 checkboxes `[x]` in progress tracker
> 2. Fill Step 5 Execution Log (every field, no "pending")
> 3. Update Corrections Log if user corrected anything this step
> 4. **Verify**: re-read progress tracker, confirm no blanks for Step 5
> Then proceed to Step 6 with the (optionally refined) draft.

### Steps 6-10: Polish to Publish

**Full details:** Read `{skill-dir}/references/steps-polish-to-publish.md` for complete instructions.

**Step 6: Polish** — **【强制】使用 Skill 工具调用 content-research-writer**，不得手动润色代替。Compile technique-aware instructions (platform, techniques, checklist, style, lessons) before invoking. Output: `{topic-slug}-polished.md`. Experience Check after. **Update Execution Log.**
> **If Autonomous Mode:** content-research-writer 失败 → 记录失败到 Execution Log 和 Autonomous Decision Log，AI 自行润色作为降级方案（这是自主模式的设计决策，非违反"禁止擅自兜底"）。跳过 Experience Check。

**Step 7: Illustrations** — **【强制】使用 Skill 工具调用 baoyu-xhs-images**，传入 polished.md。禁止手动编写 outline/prompt。Output: images in `xhs-images/`. Experience Check after. **Update Execution Log.**
> **If Autonomous Mode:** 默认跳过配图（Step 7 整步跳过），在 Execution Log 记录"自主模式 — 跳过配图"。如用户在初始指令中明确要求配图，则尝试调用 baoyu-xhs-images；调用失败 → 记录失败到 Execution Log 和 Autonomous Decision Log，继续。跳过 Experience Check。

**Step 8: Final Article** — Combine polished content + images. Create `{topic-slug}-final.md`. **呈现最终文章（不可省略）：** 告知用户文件路径 + 图片位置 + **等待用户确认**。Experience Check after. **Update Execution Log.** **⚠️ STOP: 不得直接跳到 Step 10，必须先执行 Step 9。**
> **If Autonomous Mode:** 跳过用户确认，创建 final.md 后直接继续到 Step 9。跳过 Experience Check。

**Step 9: Review & Adaptation** ← **不可跳过** — 即使用户已表达发布意图，仍需执行。
- 9a. Review: verbal summary + ask for revisions
- 9b. Platform Adaptation (optional): **【强制】使用 Read 工具读取 `{skill-dir}/skills/content-adapter.md`，严格按照文件中的 Step 顺序执行。** Extracts core message, searches target platform, restructures per platform spec, saves as `{topic-slug}-{platform}.md`
- 9c. Publishing Decision: ask user
- Experience Check after. **Update Execution Log.**
> **If Autonomous Mode:** 9a: AI 自行审阅，不询问修改意见。9b: 根据用户初始意图决定是否适配其他平台——用户要求了则执行，未要求则跳过。9c: 根据用户初始意图和自主能力边界决定是否发布。跳过 Experience Check。

**Step 10: Publish (Optional)** — Invoke platform-specific publishing skill (xiaohongshu / baoyu-post-to-wechat / baoyu-post-to-x). After publishing, remind user about data recording ("记录数据"). **Update Execution Log.** Then proceed to **流程自检 + 复盘**.
> **If Autonomous Mode:** 评估发布路径是否可自主完成：API/CLI 且已有凭证 → 执行并记录；需用户登录/扫码 → 跳过发布，在执行摘要中标注"发布需用户手动完成"并给出所需信息（文件路径、平台等）。

### 流程自检 + 复盘（不可跳过）

**Full protocol:** Read `{skill-dir}/references/steps-polish-to-publish.md` "流程自检 + 复盘" section for the complete 10-step self-check and retrospective process (checklist audit + execution review + experience recording).

> **If Autonomous Mode:** 标准自检 + 复盘照常执行。额外生成"自主模式执行摘要"填入 progress tracker 对应区域。复盘中发现的流程问题照常记录到经验系统。多篇/批量场景：每篇各自完成自检 + 复盘；全部完成后，生成总体摘要。**复盘完成后，必须执行 Completion Gate 核对（逐项检查 G1-G7），全部通过才能停止。**

## Best Practices

See `{skill-dir}/references/steps-polish-to-publish.md` "Best Practices" section for 11 guidelines covering user voice, reference usage, technique application, progress tracking, and write-level selection.

## Output Directory Convention

All output files are stored under `writtings/outputs/{topic-slug}/`:

```
writtings/outputs/{topic-slug}/
├── {topic-slug}-progress.md      # Progress tracker
├── {topic-slug}.md               # Initial draft
├── {topic-slug}-polished.md      # Polished draft
├── {topic-slug}-final.md         # Final version
├── {topic-slug}-{platform}.md    # Platform adaptation
├── metrics.md                    # Publication metrics
└── xhs-images/                   # Illustrations
    ├── outline.md
    ├── prompts/
    └── *.png
```

Use consistent naming throughout the workflow. All files use the topic slug as prefix and reside in the same output directory.

Comments 0

Latest | Hot

Please login before commenting.

No comments yet. Be the first one!

writing-assistant

Directory Structure

SKILL.md

Report

Notice