@@ -11,6 +11,25 @@
 
 *From the first spark to the next chapter: every commit makes the novel more coherent, not the context more chaotic.*
 
+## The core workflow — make every chapter strengthen the next
+
+This is not a linear “generate another passage” chat. It is a production loop that calibrates, accumulates, and recovers after every accepted chapter:
+
+| Stage | What moves forward | What stays protected |
+| --- | --- | --- |
+| **01 · Find direction** | Reader promise, chart opportunities, or the real problem in an existing story | A vague idea is not prematurely frozen into an outline |
+| **02 · Build the engine** | Protagonist drive, rules, relationships, and volume-level promises | Every volume has a clear reason to keep reading |
+| **03 · Finish one chapter** | Chapter contract, bounded context, complete prose, humanization, review | Goals, character limits, turns, and the ending pull remain coherent |
+| **04 · Create the next foundation** | Facts, promises, resources, relationship changes, recovery checkpoints | Only accepted material becomes durable memory |
+
+```text
+Direction → story engine → chapter plan → complete prose → review and commit
+    ↑                                                       ↓
+    └──────────────── continuity state and next chapter ───┘
+```
+
+**The key rule: stabilize one chapter at a time.** No stitched parallel fragments, no unreviewed candidates becoming facts, and no need to load the entire novel into the next context.
+
 ## What it helps you do
 
 | Challenge | Workflow | Durable output |
@@ -49,16 +68,6 @@ Use $produce-long-form-novel to analyze recent xianxia chart composition and cre
 Use $produce-long-form-novel to deconstruct this authorized novel text; establish coverage, segment notes, and an overview first.
 ```
 
-## Built for the long run
-
-```text
-Idea → reader promise → world and cast → volume payoff
-     → chapter contract → complete prose → review
-     → continuity commit → next chapter
-```
-
-Migrated workspaces use YAML as the sole continuity authority for facts, promises, resources, dynamic character state, and relationships. Markdown ledgers are generated views; SQLite only accelerates retrieval and can be deleted or rebuilt without changing story truth. This keeps context bounded as the novel grows.
-
 ## Install and use local tools
 
 Install this repository as a personal Codex Skill, then invoke `$produce-long-form-novel`. See [SKILL.md](SKILL.md) for the complete workflow contract.
 
@@ -13,6 +13,25 @@
 
 *从创意火花到下一章：每次回灌都让小说更完整，而不是让上下文更失控。*
 
+## 核心工作流｜让小说越写越稳
+
+这不是“生成一段，再生成一段”的线性对话，而是一个每章都会校准、积累并恢复的创作闭环：
+
+| 阶段 | 你在推进什么 | 系统帮你守住什么 |
+| --- | --- | --- |
+| **01 · 找到方向** | 读者频道、阅读回报、榜单机会或已有故事的真正问题 | 不把模糊灵感过早固化成大纲 |
+| **02 · 搭好发动机** | 主角欲望、世界规则、角色关系、卷级承诺 | 每一卷都有明确的追读动力和回报窗口 |
+| **03 · 完成一章** | 章节合同、最小上下文、完整正文、人性化修订与审查 | 当前章目标、人物边界、转折与章末牵引不丢失 |
+| **04 · 变成下一章的底座** | 事实、伏笔、资源、关系变化和恢复检查点 | 只有验收过的内容进入长期记忆，下一章从可信状态开始 |
+
+```text
+方向判断 → 故事发动机 → 章节计划 → 完整正文 → 审查回灌
+    ↑                                                    ↓
+    └─────────────── 连续性状态与下一章 ───────────────┘
+```
+
+**关键原则：一次只稳定一章。** 不并行拼接同一章，不让未验收候选进入事实，也不把整本书塞入下一次上下文。
+
 ## 你会得到什么
 
 | 你正在面对的问题 | Ani Book Skill 的做法 | 最终留下什么 |
@@ -57,20 +76,6 @@
 使用 $produce-long-form-novel 拆解这份有权使用的小说文本，先建立覆盖范围、分段笔记和快速总览。
 ```
 
-## 一部小说如何在这里持续生长
-
-```text
-选题/灵感
-  → 阅读承诺与故事发动机
-  → 世界、角色与卷级回报
-  → 当前章节合同与最小上下文
-  → 完整正文与人性化修订
-  → 审查、连续性回灌、恢复检查点
-  → 下一章
-```
-
-每个箭头都留下可编辑工件。作者修改正文时，依赖产物会被标记为 stale，而不是悄悄用旧设定覆盖新创作。
-
 ## 为长篇而设计的连续性
 
 当小说进入长期续写阶段，可将连续性迁移为：
@@ -136,5 +141,6 @@ python scripts/trend_snapshot.py validate trends/<范围>/snapshots/<日期>/<pl
 ### 2026-07-18
 
 - 新增工作流宣传图，以可视化方式呈现选题、故事架构、章节审查和长期连续性之间的闭环。
+- 将四阶段工作流置于首页核心位置，明确展示“方向判断 → 故事发动机 → 单章生产 → 连续性回灌”的闭环与单章稳定原则。
 
 完整历史见 [更新记录](docs/releases/release-notes.md)。
@@ -6,6 +6,7 @@
 
 - 新增工作流宣传图，以可视化方式展示从选题、故事架构、正文审查到连续性回灌和下一章的完整创作闭环。
 - 重写项目首页，围绕作者在长篇创作中的实际问题、可获得的工件和长期连续性价值展示核心能力，方便快速理解并开始使用。
+- 将四阶段核心工作流前置为首页重点，清晰说明方向判断、故事发动机、单章生产和连续性回灌如何组成可持续创作闭环。
 
 ### 2026-07-17