# Flowing Docs

> Walter’s notes on work, AI tools, self-hosted services, technical writing, projects, and everyday life.

The links below provide Markdown copies. Each copy names its original page; examples retain their source syntax. Interactive markup exercises also have a full text reference.

## Chinese documentation

- [技术文档实践](https://flowingdocs.com/llms/zh/docs/index.md): 围绕标记语言、信息表达和文档发布，记录具体问题、处理方式与适用边界。
- [文档标记语言：Markdown、AsciiDoc 与 DITA 的能力与取舍](https://flowingdocs.com/llms/zh/docs/markup.md): Markdown、AsciiDoc 与 DITA 标记语言的对照实践：剖析跨行跨列单元格合并、条件属性包含与片段复用、DITA Topic/Map 模块化组织逻辑与选型取舍。
- [信息表达：把内容写清楚，让读者用得上](https://flowingdocs.com/llms/zh/docs/information-design.md): 从读者疑问找内容缺口，对照标签页、折叠和表格，提供作者自查与自动检查示例。
- [概念说明：把原理和边界讲清楚](https://flowingdocs.com/llms/zh/docs/information-design/concept.md): 讨论怎样解释原理、逐步补充细节并交代结论的条件。
- [操作步骤：让读者做完，也知道是否成功](https://flowingdocs.com/llms/zh/docs/information-design/task.md): 按读者的实际任务组织条件、步骤和验证，附完整 JDBC 驱动加载示例。
- [参数参考：把读者需要查的事实写全](https://flowingdocs.com/llms/zh/docs/information-design/reference.md): 用虚构参数对照单位、范围、特殊值与生效方式的写法。
- [文档发布：怎样选择静态发布工具与方案](https://flowingdocs.com/llms/zh/docs/publishing.md): 比较常见发布工具，结合本站说明内容、导航、主题与维护成本。

## Chinese articles

- [技术文档 AI 审核实践：从本地轻量校验到 Codex Code Review](https://flowingdocs.com/llms/zh/blog/ai-content-review-with-codex.md): 结合本地小模型 LLM 轻量校验、Codex Local 与 Codex Cloud，分享一套适用于 Doc-as-Code 技术文档的 AI 审校流程，以及 AGENTS.md 规则设计的落地经验。
- [别让 AI 写的文档误导用户：从单次 Prompt 到高可信文档工程化实践](https://flowingdocs.com/llms/zh/blog/ai-docs-engineering-practice.md): AI 能快速生成文档初稿，但如何保证其准确性与边界可控？本文分享了从单次 Prompt 试水、结构化受控生成，再到自动化 Harness 验证的文档工程化实践，探讨大模型时代下文档工程师的价值演进。
- [Docusaurus 个人博客搭建完整指南（2026）：从配置到上线](https://flowingdocs.com/llms/zh/blog/building-a-docusaurus-blog.md): 从框架选择、Docusaurus 配置、搜索、评论到 Netlify 部署，记录 Flowing Docs 个人博客搭建过程中的关键设置和实践经验。
- [从零构建本地 AI 内容审校系统：小模型推理到工程化落地](https://flowingdocs.com/llms/zh/blog/building-a-local-ai-content-review-system.md): 从技术文档审校中的错别字、术语误写和误报问题出发，分享一套本地中文 AI 文档纠错系统的设计思路，包括规则引擎、小模型、保护机制、Web 管理界面和持续反馈。
- [Codex Desktop 接入 DeepSeek：用 opencodex 切模型，不中断当前任务](https://flowingdocs.com/llms/zh/blog/codex-desktop-deepseek-opencodex.md): 记录我如何用 opencodex 在 Codex Desktop 的同一个任务中切换 DeepSeek，包括安装配置、模型选择、启动日志排查和恢复原生 Codex。
- [DocSifter 开源：一个面向中文技术文档的本地 AI 审校小工具](https://flowingdocs.com/llms/zh/blog/docsifter-open-source.md): 分享我们在中文技术文档场景下折腾的一个本地审校小工具 DocSifter。侧重代码结构保护、离线小模型基础纠错与误报沉淀，算是一点工程探索，现已开源供大家参考。
- [告别全局代理：为 DeepSeek Harness 实现多模型 API 域名精准分流](https://flowingdocs.com/llms/zh/blog/dsh-domain-proxy-open-source.md): 我在用 DeepSeek Harness 时，需要让特定模型 API 走本地代理，而其他直连端点保持原生链路。写了个小插件 dsh-domain-proxy，直接按域名做分流，不用改 Provider 配置或重新填 Key，顺便记录真实的 TTFT 首字延迟与生成速率。
- [那些生活中有意思的信息设计陷阱，你中招了吗](https://flowingdocs.com/llms/zh/blog/interesting-information-design-traps.md): 通过苹果官网、高铁显示屏和商业招牌三个真实案例，深入分析信息设计中的常见问题，提供基于设计原则的改进方案，帮助提升用户体验和信息的清晰传达。
- [用旧主机搭私有云：Nextcloud 部署、版本溯源与 RAG 知识库实践](https://flowingdocs.com/llms/zh/blog/nextcloud-self-hosted-cloud-storage.md): 商业云盘放不了敏感内容、本地存储又有限，用旧主机跑 Nextcloud 是个出路：低成本扩容、VPN 远程连回家、文件改坏了拉时间轴找——后来顺手接进 RAG，工作文档变成可查询的 AI 知识库。
- [写在前面：为什么我决定搭一个属于自己的博客？](https://flowingdocs.com/llms/zh/blog/why-i-started-this-blog.md): 记录我为什么开始搭建 Flowing Docs / 知流小记，以及这个博客未来会关注的技术写作、开发者体验、AI 工具和生活实践方向。

## English documentation

- [Technical Documentation Practices](https://flowingdocs.com/llms/en/docs/index.md): Practical problems, approaches and limits in markup, information design and publishing.
- [Document markup: Capabilities and tradeoffs across Markdown, AsciiDoc, and DITA](https://flowingdocs.com/llms/en/docs/markup.md): Empirical comparison of Markdown, AsciiDoc, and DITA: dissecting complex table spans, conditional attribute inclusion, single-sourcing modular reuse, and DITA topic/map architectures.
- [Information design: make documentation clear and useful](https://flowingdocs.com/llms/en/docs/information-design.md): Find missing facts and compare tabs, disclosures and tables, with review questions and build checks.
- [Concepts: explain the mechanism and its limits](https://flowingdocs.com/llms/en/docs/information-design/concept.md): Explain a mechanism, add detail gradually and state its limits.
- [Procedures: actions and verifiable results](https://flowingdocs.com/llms/en/docs/information-design/task.md): Organize prerequisites, actions and verification, with a complete JDBC driver-loading example.
- [Reference: facts readers need to look up](https://flowingdocs.com/llms/en/docs/information-design/reference.md): Compare units, ranges, special values and timing using fictional parameters.
- [Documentation Publishing: Choosing the Right Static Tooling](https://flowingdocs.com/llms/en/docs/publishing.md): Compare publishing tools and understand content, navigation, themes and maintenance costs.

## English articles

- [How I Review Technical Docs with AI: From Lightweight Local Checks to Codex Code Review](https://flowingdocs.com/llms/en/blog/ai-content-review-with-codex.md): A layered AI review workflow for technical docs using local checks, Codex Local, Codex Cloud, and AGENTS.md to catch issues humans often miss.
- [Don't Let AI-Written Docs Mislead Users: From One-Off Prompts to Reliable Documentation Engineering](https://flowingdocs.com/llms/en/blog/ai-docs-engineering-practice.md): AI can generate polished documentation drafts quickly, but accuracy and boundaries still need engineering discipline. This post shares a documentation engineering workflow that moves from one-off prompts to structured generation and automated harness validation.
- [How to Build a Docusaurus Blog in 2026 — A Complete Setup Guide](https://flowingdocs.com/llms/en/blog/building-a-docusaurus-blog.md): A practical walkthrough of how I built this blog with Docusaurus, covering site configuration, search, comments, analytics, and Netlify deployment.
- [Building a Local AI Content Review System: From a Small Model to a Web Tool Your Team Will Actually Use](https://flowingdocs.com/llms/en/blog/building-a-local-ai-content-review-system.md): How I built a local AI proofreading system for technical docs, from rule design and model choice to structure protection, web UI, and feedback loops.
- [DeepSeek in Codex Desktop Without Losing Your Thread](https://flowingdocs.com/llms/en/blog/codex-desktop-deepseek-opencodex.md): How I use opencodex to switch to DeepSeek within an active Codex Desktop task, troubleshoot startup messages, and restore the native Codex setup.
- [DocSifter: A Lightweight, Local-First AI Reviewer for Technical Docs](https://flowingdocs.com/llms/en/blog/docsifter-open-source.md): DocSifter is a lightweight, local-first review tool we built for technical docs. Focuses on code protection, offline small-model correction, and false-positive filtering.
- [Beyond Global Proxies: Domain-Selective Routing and Telemetry in DeepSeek Harness](https://flowingdocs.com/llms/en/blog/dsh-domain-proxy-open-source.md): When using DeepSeek Harness across multiple providers, I needed specific model APIs to route through a local proxy while direct endpoints stayed untouched. I wrote dsh-domain-proxy to split traffic by domain without rewriting provider configs or API keys, while logging real TTFT and streaming tokens/s.
- [Everyday Information Design Traps That Sneak Up on You – Have You Been Caught Out?](https://flowingdocs.com/llms/en/blog/interesting-information-design-traps.md): Real examples from Apple’s website, train displays, and storefront signs show how information design can confuse users and how small fixes improve clarity.
- [Running Nextcloud on a Spare Machine: Self-Hosted Storage, File Versioning, and a Local RAG Knowledge Base](https://flowingdocs.com/llms/en/blog/nextcloud-self-hosted-cloud-storage.md): Commercial cloud storage isn't always right for sensitive work files — and local storage has limits. Running Nextcloud on a spare machine turned out to cover a lot at once: cheap storage expansion, VPN access from anywhere, file version history on a timeline, and a local RAG knowledge base that's actually useful.
- [Preface: Why I Finally Decided to Build My Own Blog](https://flowingdocs.com/llms/en/blog/why-i-started-this-blog.md): Why I started Flowing Docs, what I plan to write about, and how this blog connects technical writing, developer experience, AI tools, and everyday experiments.

## Combined text

- [All included documentation and articles](https://flowingdocs.com/llms-full.txt): Chinese and English text in one file.