完善创作节点循环与等待态交互，并大幅打磨 Web 壳与会话运行时。

- 节点循环：配方开局直坐 DAG 第一步；验收后先问下一步意向，再确认开干并可钉编排参数；「按意见修改」只重跑当前节点。 - 询问分流：能力 opening 走说话面引导，可跳过追问按题干去重；追问与产物同轮挂载，避免拆成两段历史。 - 运行时：补强 revision 重跑、创作流程合并/进度指针、工具循环与 worker 执行；新增运行日志与 web:watch。 - 前端：统一等待态文案与底栏主按钮，完善询问卡/产物验收/呈现壳样式与交互。 - 同步世界模拟器模块提示、编排文档与相关测试。 Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-17 01:28:12 +08:00
parent 20938b04d7
commit ec474cb946
56 changed files with 7460 additions and 1372 deletions
--- a/docs/briefs/artifact-card-coverage.md
+++ b/docs/briefs/artifact-card-coverage.md
@@ -4,7 +4,7 @@

 | 产物/技能 | 美化程度 | 渲染入口 |
 |-----------|----------|----------|
-| 美学纲领与交互范式 | **专用 mosaic**（设定逻辑 / 交互范式 / 美学纲领） | `renderSpecialtyBodyHtml` → aesthetics |
+| 美学纲领与交互范式 | **左右分栏**：左契约（美学纲领+交互范式）/ 右诊断（设定逻辑） | `renderSpecialtyBodyHtml` → aesthetics |
 | 实现机制 | **专用卡**（支撑点） | `renderMechanismBodyHtml` |
 | 舞台骨架 | **专用卡** | `renderWorldBlueprintBodyHtml` |
 | 生成规则 | **专用卡** | `renderGenerationRulesBodyHtml` |
--- a/docs/briefs/capability-authoring-brief.md
+++ b/docs/briefs/capability-authoring-brief.md
@@ -356,7 +356,8 @@ task 必须写：禁止重复同一开场；在首答上补洞。

 ## 4.8 追问（`probe`）通用要求

- 一轮 askUser **1～2** 点。  
+- 一轮 **1～2** 点，只写进产物 `追问`（建议选项 + 示例）。  
+- 程序会把 `追问` 挂到询问卡；**禁止**同一问再抄一份顶层 `askUser`（`askUser` 留 `null`）。  
 - 优先：可直接采用或微调的完整句选项 / 短场景。  
 - 能推断则先写入再复述；只问影响本步核心且不能瞎填的点。  
 - 依赖产物已钉死的内容禁止重问。
@@ -397,7 +398,7 @@ task 必须写：禁止重复同一开场；在首答上补洞。
 |----|------|
 | `正文` | **产物主体**。技能自定内部结构（如美学步＝设定逻辑 + 交互范式 + 美学纲领）；须含「详细各个方面」。禁止整份只交散文。 |
 | `自评` | **自评评分**。`维度[]` 名目按技能定（美学步常用：交互范式 / 美学纲领 / 整体协调）；**`分数` 为 0–10 十分制**（可一位小数；高质约 8–10）；可含 `薄弱点`。禁止用 0–100 百分数。 |
-| `追问` | **示例 + 建议选项**。给用户的下一问；`题目` 可空数组。与程序 askUser 可并存，但产物里要留结构化题面。 |
+| `追问` | **示例 + 建议选项**。给用户的下一问（程序挂询问卡）；`题目` 可空数组。禁止同一问再写入顶层 `askUser`。 |

 | 规则 | 说明 |
 |------|------|
--- a/docs/context-fragment-design.md
+++ b/docs/context-fragment-design.md
@@ -38,10 +38,20 @@
 | `schema` / `技能` / `brief` / `mount` / `稳变` | 索引与挂载；不写 `order` |
 | `正文` | **产物主体**（技能自定内部维度，但必须有「各个方面」） |
 | `自评` | **自评评分**（`维度[]` + 可选薄弱点；维度名按技能） |
-| `追问` | **导语 + 题目**（建议选项 / 示例）；可无题则 `题目: []` |
+| `追问` | **导语 + 题目**（建议选项 / 示例）；可无题则 `题目: []`。程序挂询问卡；不要再抄一份 `askUser` |
 | `开放问题` | 未结构化的残项（可选） |

-美学步正文：`设定逻辑` + `交互范式` + `美学纲领`（见 `aesthetics-interaction`）。  
+美学步正文：`设定逻辑` + `交互范式` + `美学纲领`（见 `aesthetics-interaction`）。
+
+**验收卡排版约定（美学步已落地，其它节点逐步对齐）**：
+
+| 栏 | 放什么 | 美学步对应 |
+|----|--------|------------|
+| **左 · 产物** | 最终希望插入游玩上下文的契约正文 | `美学纲领`、`交互范式` |
+| **右 · 评估** | 诊断、完备度、待探、区域化等给人验收用的 | `设定逻辑`（参与/内容维度上的结论·依据·完备度） |
+
+外壳另有 `自评`（顶栏分数 pill）与 `追问`（询问卡）。  
+**存盘**：三块都在同一 `正文` 里。**注入**：默认 `full` 时整份正文都会进模型——若日后只要左栏进游玩，再改投影，不靠藏 UI。  
 实现机制正文：`依据的核心体验` + `支撑点[]` + `支撑点关系[]` + `覆盖检验`（见 `mechanism`）。  
 舞台骨架正文：尺度 / 基底与变造 / **社会结构** / **世界状况** / 关键舞台区 / 未展开范围（见 `world-blueprint`）。  
 生成规则正文：必要性判断 + `rules[]`（生成与描写 + **单层**产物格式 + 可选多池绑顶层字段）；自评＝必要性/属性妥当/格式准确。  
--- a/docs/play-presentation-shells.md
+++ b/docs/play-presentation-shells.md
@@ -10,14 +10,25 @@

 ```text
 固定壳（本仓库预置，七种）
-  → 创作期：选一个 shell_id
-  → 只改允许的微调轴（字段、显示名、开关、tone_chrome…）
+  → 按「要同时看见什么」选 shell_id（不是按好看）
+  → 只改允许的微调轴
  → 禁止改分区骨架、禁止发明第八种布局
-游玩期：往壳的区域里灌数（present.v1 / 投影 / settlement）
+游玩期：往壳区域灌数；正文随 600~3000 字撑高滚动
 ```

-「正文组成」= **壳适配单**，不是 UI 创意稿。  
-用户看见的差异应来自**同壳不同绑定**，而不是每局一套新布局。
+选壳依据（同时可见关系）：
+
+| 同时要看见 | 壳 |
+|------------|-----|
+| 只要正文 | `prose` |
+| 正文 + 扫状态 | `chat_monitor` |
+| 大正文 + 弱状态 | `spotlight` |
+| 场面 + 交互物 + 行动 | `turn_panel` |
+| 正文 ‖ 线索对照 | `split_board` |
+| 局面 + 选项 | `choice_dock` |
+| 章正文 + 进度 | `chapter_reader` |
+
+预览页 `/shells.html`：主工作面是壳画布；侧栏只换壳；长度/语气在顶栏。

 ## 2. 信息层级与长文容量

--- a/docs/ui-glossary.md
+++ b/docs/ui-glossary.md
@@ -87,7 +87,7 @@
 | `设计.worker集` | **运行规格** | 剧本（旧，规格部分） | 可进游玩的声明截面。 |
 | Blackboard | **黑板** | 上下文板 | — |
 | Artifact | **产物** | — | 待验收输出。 |
-| Accept / Review | **验收** / **接受** | 人工审批 | 按钮用「接受」；阶段说明用「验收」。 |
+| Accept / Review | **验收** / **接受** | 人工审批 | 技能产物按钮用「接受」；工作流计划用「确认编排」；开场用「选定此开场」。阶段说明：技能步「验收」，编排「核对」。 |
 | Intake | **需求描述** | 启动填空 | — |
 | Burst | **本轮调度** | — | 少对用户说 burst。 |
 | Questions card | **询问卡** | Questions | 详见 `ui-design.md` §8。 |
@@ -177,13 +177,15 @@

 ## 7. 等待态 / 焦点（对用户）

-| `waitingReason.kind` | 状态摘要（短） | 焦点动作 |
+| `waitingReason.kind` | 状态摘要（短） | 焦点动作 / 底栏主按钮 |
 |----------------------|----------------|----------|
-| `intake` | 描述需求 | 描述创作需求 |
-| `input` | 补充说明 | 补充说明 / 回答追问 |
-| `worker_questions` | 回答提问 | 回答 · {中文执行单元名} |
-| `approve_step` | 确认执行 | 建议调用 {中文名} |
-| `review_artifact` | 验收产物 | 验收产物 |
+| `intake` | 描述需求 | 发送 |
+| `input` | 补充说明 | 发送 / 补充并发送 |
+| `worker_questions` | 答题 | 补充并发送 |
+| `approve_step` | 确认 | 同意并开始 |
+| `review_artifact` | 视产物（勿一律「验收」） | **流程编排** → 确认编排；技能步 → 接受；开场 → 选定此开场。旁挂「按意见改编排 / 按意见修改」 |
+| `next_intent` | 说话 | 继续（可留空） |
+| `revision` | 说话 | 按意见修改 |

 ---

--- a/docs/world-simulator-modules.md
+++ b/docs/world-simulator-modules.md
@@ -28,7 +28,7 @@
 | 块 id | 必填 | 用途 |
 |-------|------|------|
 | `meta` | 建议 | YAML：name / id / artifact / declaration / when / when_not / boundary |
-| `opening` | 可选 | **默认问题**正文；程序发给用户，不经 LLM |
+| `opening` | 可选 | **默认问题**正文。开局：配方 `steps` 已预置进「设计.创作流程」，用户首句即坐在 DAG 第一步；若「用户.需求」已有内容，程序**跳过**再抛 opening，直接进本步 LLM |
 | `task` | 是 | 本步任务与验收边界 |
 | `principles` | 建议 | 原则 |
 | `probe` | 建议 | 追问策略 |
@@ -104,13 +104,24 @@ declaration: …

 配方**不要**重复罗列各能力调用条件；那是能力 meta 的职责。旧字段 `hint` 仍可读作兜底。

-### 默认问题节奏（通用）
+### 节点循环（通用）

 ```text
-程序发 opening → 用户首答 → LLM（opening + 首答 + 切割后的方法块 + 依赖）
+开局选配方 → 坐到配方预置的第一个节点
+  → 节点 opening → 用户首批输入
+  → 节点要求 + 用户输入 → 产物 + 追问
+  → 用户选选项 / 写意见 →「按意见修改」
+      （产物 + 追问 + 选项 + 意见 + 节点要求一并重跑）→ 新产物
+  → 循环直到「接受」
+  →「下一步想写什么」（可留空，交给工作流计划 DAG）
+  → 展示下一节点并确认开干（可在此钉编排参数，如生成规则的 target）
+  → 下一节点 opening / design-step …
+直到创作收口
 ```

-有编排参数的能力：先由 design-flow askUser 钉 params → 用户认可 DAG → design-step 注入【本步参数】执行；勿把「生成什么」推迟到执行期。
+「按意见修改」只重跑当前节点，不得跳到「下一步想写什么」。底栏空且未选追问时 Enter＝接受；已选追问时 Enter＝按意见修改。点按钮仍只做按钮自己的事。
+
+有编排参数的能力：可在确认开干时钉齐 params；勿把「生成什么」拖到执行中途才问。

 ---

--- a/package.json
+++ b/package.json
@@ -12,6 +12,7 @@
    "phase-demo": "tsx src/cli/phase-demo.ts",
    "phase-script": "tsx src/cli/phase-demo.ts --script --auto",
    "web": "tsx src/server/web-server.ts",
+    "web:watch": "tsx watch --clear-screen=false --include \"./src/**/*.ts\" --include \"./skills/**\" src/server/web-server.ts",
    "electron": "electron ."
  },
  "devDependencies": {
--- a/skills/dialogue/world-simulator/modules/aesthetics-interaction/prompt.md
+++ b/skills/dialogue/world-simulator/modules/aesthetics-interaction/prompt.md
@@ -54,8 +54,10 @@ feeds: design_only
 业务字段全部放在「正文」内；须含 schema / 技能 / brief；建议填 mount 与稳变（本步不写数字 order）。

 一次 design-step、一场对话、一次验收。参与方式与美学内容在同一步内交叉追问，不要先交「半份交互」再开「半份美学」。
+用户已答过追问：把答复写入正文三块，`追问.题目` 优先 `[]`；禁止在「交互范式」与「美学纲领」之间轮流补洞拖验收。核心体验与权限边界够用即可收。

 若程序已发出默认问题：用户首答在「用户.worker答复」，开场白在「创作.能力开场白」。禁止重复同一开场；在首答基础上补洞、追问、产出。
+开局通常已由配方写入「设计.创作流程」第一步；若「用户.需求」已有首句：程序会跳过默认问题——首句即本步对话，只补真正缺口。
 ```

 ## principles
@@ -74,9 +76,11 @@ feeds: design_only
 ## probe

 ```probe
-用户首句或黑板已有的直接采纳。一次 askUser 1～2 点，优先具体选项/短场景。
+用户首句或黑板已有的直接采纳。一次只在产物「追问」写 1～2 点（建议选项 + 示例）；顶层 askUser 留 null，不要同一问写两遍。
+已有一轮追问答复或核心体验+权限边界已可执行：`追问.题目` 用 `[]`，让用户验收。禁止这轮问交互、下轮问美学地来回打转。

 【默认问题已覆盖】变造（原型+变点）、代入与否、最想感受到的核心。首答后按缺口补，勿重问已答清的。
+若开场已跳过（用户.需求已有首句）：把首句当默认问题的答复，只补真正缺口。

 【参与 · 结构性】
 - 用户与 <user>：代入程度 / 情感距离 / 控制期待 / 多角色
@@ -205,7 +209,7 @@ feeds: design_only
 2. **公共三段**：`正文`＝产物主体；`自评`＝完备度打分（**分数 0–10 十分制，禁止百分数**）；`追问`＝导语+题目（建议选项/示例）。禁止用散文代替这三段。
 3. **本技正文三块**：`设定逻辑`（参与/内容维度/区域/复杂性/完备度）+ `交互范式` + `美学纲领`。未知省略或「未定」，禁止编造百科。
 4. `mount` 建议挂主世界层+转述；**禁止**写数字 `order`。`稳变` 定稿后多为 `stable`。
-5. `追问.题目` 一次宜 1～3 题；有建议选项；允许「其它」。已答清的不要再问。
+5. `追问.题目` 一次宜 1～3 题；有建议选项与示例；允许「其它」。已答清的不要再问。题目只写在「追问」，askUser 为 null。
 6. summary：`美学纲领与交互范式 · {brief 缩略}`。

 ## checklist
@@ -216,7 +220,7 @@ feeds: design_only
 - [ ] 未写 order、未输出执行单元列表？
 - [ ] brief 能点题核心体验？变造/代入/满足来源清楚或已标未定？
 - [ ] 交互范式权限与美学纲领互相保护、不矛盾？
- [ ] 追问只打薄弱点，带建议选项？
+- [ ] 追问只打真正薄弱点，带建议选项？已答过则题目为空、准备验收？
 - [ ] 有没有为「完整」强行填充百科或偏离用户核心？
 ```

--- a/skills/dialogue/world-simulator/modules/catalog.yaml
+++ b/skills/dialogue/world-simulator/modules/catalog.yaml
@@ -46,11 +46,11 @@ modules:
      - key: target
        label: 生成对象
        required: true
-        hint: 本轮为哪一类内容建规则；编排时 askUser 给选项+其它
+        hint: 本轮为哪一类内容建规则；编排时 askUser 给选项+其它；用户改对象时必须同步改 rule_id 与步骤 id
      - key: rule_id
        label: 规则 id
        required: false
-        hint: 建议英文 kebab-case；可执行时再最终钉死
+        hint: 英文 kebab-case，须与 target 对应（女租客→female-tenants，丧尸→zombies）；改对象时禁止沿用旧 id
      - key: lifecycle_intent
        label: 生命周期意图
        required: false
--- a/skills/dialogue/world-simulator/modules/generation-rules/prompt.md
+++ b/skills/dialogue/world-simulator/modules/generation-rules/prompt.md
@@ -68,6 +68,7 @@ feeds: gm,auditor
 4. 写满规则三块：生成与描写（可较长）+ 产物格式（**单层 JSON**：每条产物是一层键值，无嵌套对象）+ 池（需要才填；绑顶层字段）。
 5. 逐字段过筛：每个属性都要能回答「对本对象族是否必要」。冗余字段删掉（如女主规则里的「性别」）；对象族必需字段不得漏（如 D&D 怪物的等级、攻击骰）。
 6. `rule_id`：params 已给则用；否则稳定 kebab-case。已有产物按 rule_id 增量。
+   若本步对象与 params 不一致（用户中途改了对象），以最新对象为准并重钉对应英文 rule_id，禁止沿用旧 id。
 7. 按自评三维打分并填追问；输出 JSON。禁止夹带最终实例 records。

 若程序已发默认问题：禁止重复开场；在首答上补洞。
--- a/skills/dialogue/world-simulator/modules/mechanism/prompt.md
+++ b/skills/dialogue/world-simulator/modules/mechanism/prompt.md
@@ -32,6 +32,17 @@ feeds: gm
 ## opening

 ```opening
+核心体验已有方向时，用几句话想想：它具体靠什么撑住？（想到什么写什么，不必整齐；点名即可，不必论证来源或补起源史。）
+
+1. 这体验主要靠哪些东西成立？
+   角色的某一面、一段关系、世界条件、规则或情境……用你自己的词点名。
+   例：丈夫会把异常证据合理化；力量越用越侵蚀自我；村里每年必须献祭才能过冬。
+
+2. 拿掉或改弱哪一样，核心感觉会明显变味甚至垮掉？
+   一句话即可——只点「缺了就不对」的，装饰和以后再展开的先别列。
+
+3. 若已有具体画面或作品参照，它主要指哪一切面在撑体验？
+   例：信息差、无法撤销的前提、身份代价、循环反馈……不必复述整部设定。
 ```

 ## task
--- a/skills/dialogue/world-simulator/modules/reply-format/prompt.md
+++ b/skills/dialogue/world-simulator/modules/reply-format/prompt.md
@@ -34,14 +34,14 @@ feeds: narrator
 ```opening
 本局在固定呈现壳上做适配（不是设计新界面）。

-0. 七选一壳（主面积给谁）：
-   · prose — 纯散文（正文≈全屏）
-   · chat_monitor — 对话 + 顶栏芯片
-   · spotlight — 场面主视（正文最大）
-   · turn_panel — 回合面板（场面+侧栏交互）
-   · split_board — 双栏（左故事·右线索）
-   · choice_dock — 选择坞（底部大行动区）
-   · chapter_reader — 章节阅读
+0. 按「要同时看见什么」七选一：
+   · prose — 只要正文
+   · chat_monitor — 正文 + 扫状态
+   · spotlight — 大正文 + 弱状态
+   · turn_panel — 场面 + 交互物 + 行动
+   · split_board — 正文 ‖ 线索
+   · choice_dock — 局面 + 选项
+   · chapter_reader — 章正文 + 进度

 1. 要微调什么？（监控字段、块称呼、要不要建议行动、tone_chrome…）
 2. 有没有隐藏段给模型/程序维护变量？
--- a/skills/dialogue/world-simulator/orchestrator.md
+++ b/skills/dialogue/world-simulator/orchestrator.md
@@ -44,90 +44,9 @@ uiPrompt: |

 # 世界模拟器 · 总管

-作者清单见 `docs/world-simulator-modules.md`。
+你只负责读取 `设计.创作流程` DAG，并选择下一个创作节点：

-你是 **总管**：负责 design / play 的 **调度**，不直接写正文。  
-禁止默认把一切做成「世界模拟」；按用户意图正推最小能力组合。
+- 没有 DAG 或 DAG 需要追加节点：`design-flow`
+- DAG 已给出下一个可执行节点：`design-step`

-## 导演 · 能力 · 剧本
-
-```text
-用户手动选【导演】（recipes/）     【能力】池（modules/）
-  世界模拟器 / 扩写助手 …             美学纲领与交互范式 / …
-        │                                    │
-        └──────────── design-flow ───────────┘
-              以用户所选为起点 → 排出近期增量 DAG（可追加、可同能力多次）
-              → 产出【剧本】流程（设计.创作流程，status=open|closed）
-```
-
- **导演**：用户新建时手动选定；方法起点，可调味  
- **能力**：共用工序；各导演都从同一池选型；`repeatable` 可反复编入  
- **剧本**：本局谈成的**可变增量 DAG**与规格；不是一次排死的固定全程  
- **禁止**：替用户猜测或改选导演；新建时不要再叠第二层「配方」选择
-
-## 创作与游玩分界
-
-```text
-design
-  design-flow → 用户验收 设计.创作流程（近期 steps + status）
-  → 反复 design-step（程序按当前步注入模块 prompt + 依赖产物）
-  → 当前 steps 做完且 status=open → 再 design-flow（追加 / 反复调用 / 或 closed）
-  → （可选）opening-generator
-  → 用户手动进 play
-
-play
-  用户输入
-    → auditor（旁观维护，默认空操作；无长对话史）
-    → perspective?（仅强信息隔离）
-    → gm / world-simulator（主世界层，settlement.v1；variable_changes 由 Runtime 合并）
-    → narrator（用户可见正文；通常在此 review）
-  chance 仅按需；禁止发明未声明 ref
-```
-
-## 启动（agent-first）
-
-1. 首屏 `uiPrompt`
-2. 用户首句 → `用户.需求` → 总管 tool loop
-3. 尚无已验收流程 → `run_worker(design-flow)`
-4. 流程已有未完成步骤 → `run_worker(design-step)`
-5. 当前步骤都验收完但 `status=open` → 再 `design-flow`（扩步或收口）
-6. `status=closed` 且步骤完成、终稿可用后若需开局 → `opening-generator`
-
-## Skill 注册表
-
-| id | 说明 |
-|----|------|
-| design-flow | 以用户已选导演为起点，编排/增量修订剧本 DAG |
-| design-step | 执行流程中当前一步（模块由程序注入） |
-| opening-generator | 开场白（创作末尾可选） |
-
-旧 `design-core` / `design-fixed` / `design-worker` / `design-refine` **已废弃**，禁止调度。
-
-## 验收策略
-
-| worker | requiresApproval | acceptanceMode |
-|--------|------------------|----------------|
-| design-* | true | user_confirmed |
-| opening-generator | true | user_confirmed |
-| auditor / world-simulator / role-decide | false（play） | continue（声明默认） |
-| narrator | false（play 调度） | review（等人看终稿） |
-
-## 总管优先行为
-
-1. 有需求、尚无已验收 `设计.创作流程` → `design-flow`
-2. 流程已有、存在未验收步骤 → `design-step`
-3. 已列步骤全验收但 `status=open` → `design-flow`（追加反复步或设 closed）
-4. `waiting_user(review_artifact)` → 引导验收
-5. reject → 收修订 → 重跑同一 worker（含修订流程 = 再调味）
-6. 终稿（含 `设计.worker集`）已 accept 且需开局 → `opening-generator`
-7. **play**：按 `play_slots` 序 `auditor → perspective? → gm → narrator`；`pendingSideEffectWorkers` 优先  
-   （旁观与主世界层怎么配合：见 `docs/play-dm-auditor.md`——旁观看门补表，DM 断事，勿默认多轮传话）
-
-## 禁用行为
-
- 调度已废弃的 design-core / design-fixed / design-worker / design-refine
- 跳过 design-flow 直接 design-step（无流程时）
- 一次 design-flow 排死全程固定长链（应增量）
- 调度声明未列出的 play ref
- Agent 挑选模型
- play 中把旁观维护当正文作者，或跳过 gm 直接 narrator
+节点如何执行、何时等待、验收、返工、结束和运行状态变化均由 Runtime 与节点自身处理。
--- a/skills/dialogue/world-simulator/recipes/expand-assistant/recipe.yaml
+++ b/skills/dialogue/world-simulator/recipes/expand-assistant/recipe.yaml
@@ -1,7 +1,7 @@
 # 扩写助手 · 配方
 # 写方法论：适用、核心思路、设计流程、原则。
 # 各能力何时用 / 不用 → 读能力 meta（编排器会注入），勿在此重复。
-# steps = 近期起点，不是固定全程 DAG。
+# steps = 近期起点；选中本配方时程序写入「设计.创作流程」，开局即坐在第一步（不是固定全程 DAG）。

 when: 大纲/分段扩写、写手统筹、先纲后章、成稿向助手、长文/爽文类体验

--- a/skills/dialogue/world-simulator/recipes/world-simulator/recipe.yaml
+++ b/skills/dialogue/world-simulator/recipes/world-simulator/recipe.yaml
@@ -1,7 +1,7 @@
 # 世界模拟器 · 配方
 # 写方法论：适用、核心思路、设计流程、原则。
 # 各技能何时用 / 不用 → 读技能 meta（编排器会注入），勿在此重复。
-# steps = 近期起点，不是固定全程 DAG。
+# steps = 近期起点；选中本配方时程序写入「设计.创作流程」，开局即坐在第一步（不是固定全程 DAG）。

 when: 回合互动、世界推进、角色扮演、沉浸推演类体验

--- a/skills/dialogue/world-simulator/workers/design-flow/SKILL.md
+++ b/skills/dialogue/world-simulator/workers/design-flow/SKILL.md
@@ -44,9 +44,12 @@ contextSegments:

 上下文里会有：

-1. **【用户已选配方】**：该方法的适用、核心思路、设计流程与原则 + 近期起点 steps——**不是**锁死流水线；**禁止**替用户改选其它配方  
-2. **【能力 · 可选工序】**：固定中文名 + 声明 + **何时用 / 何时不用 / 边界**（来自各能力 meta）+（若有）编排参数——步骤只能从这里选；标〔可反复〕的可多次编入  
-3. **【已有剧本草案】/【已验收步骤】**：若有，在其上追加或改未验收步，**不要**推倒重来  
+1. **【流程进度】**（程序钉死）：已完成哪些步、草案里已有哪些步、哪些能力才允许再编入。必须先读完再排。  
+2. **【用户已选配方】**：该方法的适用、核心思路、设计流程与原则 + 近期起点 steps——**不是**锁死流水线；**禁止**替用户改选其它配方  
+3. **【能力 · 可选工序】**：固定中文名 + 声明 + **何时用 / 何时不用 / 边界**（来自各能力 meta）+（若有）编排参数——步骤只能从这里选；标〔可反复〕的可多次编入  
+4. **【已有剧本草案】/【已验收步骤】**：若有，在其上追加或改未验收步，**不要**推倒重来
+
+开局说明：程序可能已按配方 `steps` 写入近期起点（用户已坐在 DAG 第一步，例如「美学纲领与交互范式」）。此时**不要**把开局步再规划一遍；仅追加缺口。listed 步已齐仍需扩写时，保留已完成 id，只加新步。

 ## 增量 DAG（核心）

@@ -56,9 +59,9 @@ contextSegments:

 - **跟配方方法论**：用配方的 core / process / principles 理解整局怎么设计、何时收成。  
 - **跟能力 meta**：某步该不该排，看能力的「何时用 / 何时不用 / 边界」；**不要**在配方里找各能力调用条件，也不要凭题材默认全选。  
- 有「编排参数」声明的能力：写入 DAG 前必须钉齐 **必填 params**。  
- 参数不明时：**先 askUser**（优先 options + 允许其它），再产出流程；禁止把「生成什么 / 调用哪个规则」推迟到 design-step。  
- 用户先认可带参数的 DAG 雏形，再进入执行；用户提出新要求时，再调本 worker 修订未验收步或追加新步。
+- 有「编排参数」声明的能力：写入 DAG 时可先留空壳；**确认开干前**必须钉齐必填 params（用户在「下一步」意向后、点同意开始时补齐）。  
+- 参数不明时：可先产出带空 params 的步骤，或 askUser；禁止把「生成什么 / 调用哪个规则」拖到 design-step 执行中途。  
+- 用户先认可**下一节点**（含参数），再进入执行；用户提出新要求时，再调本 worker 修订未验收步或追加新步。

 同能力**可以**多次出现（尤其〔可反复〕）：每次一次调用、一次验收、产物写入同一 artifact（增量补全）。id 应尽量带上调用目标，如 `生成规则·怪物`，避免无信息的 `#2`。

@@ -105,13 +108,20 @@ contextSegments:
 5. `params` = 本步调用参数（对象）。能力目录声明了编排参数时，**必填项必须写出**；可选则能推断就写
 6. `status`：`"open"` = 还可能追加；`"closed"` = 不再扩步（可走收成）
 7. 已验收步骤的 id **必须保留**；只能追加新步，或改未验收步的依赖/顺序/params
-8. 配方建议 steps 只作近期起点；按能力 meta 按需选用，勿默认全选、勿一次排满
-9. **禁止**把「交互」与「美学」拆成两步
-10. **禁止**在本步写能力正文、自由 Worker 列表、表结构
-11. **禁止**编排「反复发明执行单元」；收成链：确认「游玩拓扑」→（上下文已多时）「上下文投影排序」→「细化终稿」。挂谁可早、排第几须晚，勿让中段技能写死 order
-12. 参数或选型不够时，用 askUser 问 1～2 点（**优先带 options，并允许其它**）
-13. 依赖真实产物的步骤（如具体实例依赖某条生成规则）须等上游验收后再编排，并从产物中列出可选项供用户选
-14. `summary`：`流程 · N 步 · open|closed · …`
+8. 已完成的**非反复**技能（美学纲领、实现机制、舞台骨架、叙事指南等）禁止再排一次执行；只有目录标了〔可反复〕的（生成规则、具体实例等）才可再编入新 id
+9. 配方建议 steps 只作近期起点；按能力 meta 按需选用，勿默认全选、勿一次排满
+10. **禁止**把「交互」与「美学」拆成两步
+11. **禁止**在本步写能力正文、自由 Worker 列表、表结构
+12. **禁止**编排「反复发明执行单元」；收成链：确认「游玩拓扑」→（上下文已多时）「上下文投影排序」→「细化终稿」。挂谁可早、排第几须晚，勿让中段技能写死 order
+13. 参数或选型不够时，用 askUser 问 1～2 点（**优先带 options，并允许其它**）。
+    只许问**排布层**的事：下一步排哪个能力、某步的必填 params 取什么值。
+    **禁止**问体验偏好、设定细节、口味倾向等能力正文问题——那是 design-step 的活；
+    已验收步骤里问过的题，**禁止**再抛一遍（用户会读成「那一步要重做」）
+14. 依赖真实产物的步骤（如具体实例依赖某条生成规则）须等上游验收后再编排，并从产物中列出可选项供用户选
+15. `summary`：`流程 · N 步 · open|closed · …`
+16. **改对象必须改后台 id**：用户把「生成规则」从女租客改成丧尸怪物时，须同步改
+    `params.target`、`params.rule_id`（对应英文 kebab-case，如 `zombies`）、以及步骤 `id`（如 `生成规则·丧尸怪物`）。
+    **禁止**只改中文对象却沿用旧 `rule_id` / 旧步骤 id。

 ## 自检

@@ -123,5 +133,7 @@ contextSegments:
 - 有编排参数的步骤，必填 params 是否已钉齐？缺参是否应先 askUser 而不是产出空壳？
 - depends_on 是否都指向更靠前的步骤 id？
 - 已验收 id 是否都还在？
+- 【流程进度】里已完成的非反复技能，是否误当成新规划又排了一次？
+- askUser 是否只问了排哪步 / 填哪个参数，而不是体验偏好等正文问题？
 - 需要反复补同类内容时，是否用了新 id + 新 params 追加而非改写旧步？
 - 收成前是否把 `status` 设为 `closed`？
--- a/skills/dialogue/world-simulator/workers/design-step/SKILL.md
+++ b/skills/dialogue/world-simulator/workers/design-step/SKILL.md
@@ -5,6 +5,7 @@ name: 创作 · 执行步骤
 description: >-
  按已认可的创作流程，执行当前一步工序。提示词与产物 tag 由程序按模块注入。
  流程是增量 DAG：本步只读，禁止自行扩步或重排。
+  outputTags 由程序注入为该步能力 artifact；勿把「创作.当前步骤」当产物写回。
 version: 1
 stage: design
 inputTags:
@@ -15,8 +16,7 @@ inputTags:
  - "用户.修订说明"
  - "设计.创作流程"
  - "创作.当前步骤"
-outputTags:
-  - "创作.当前步骤"
+outputTags: []
 inputMerge: latest
 contextSegments:
  - id: flow
@@ -46,7 +46,7 @@ contextSegments:
 1. 只写本步产物（程序指定的 output tag）；不要改其它步骤产物
 2. 产物用简洁 JSON 或结构化中文，方便界面渲染；少写机器变量名
 3. 若本步有**默认问题**：程序已先发给用户；首答在「用户.worker答复」/「创作.能力开场白」。**禁止**再用 LLM 重复同一开场白
-4. 信息不足 → askUser 1～2 点（优先 options）
+4. 信息不足：context-fragment 产物把题写在「追问」（1～2 点，建议选项 + 示例），顶层 askUser 为 null。仅完全无法产出时才用顶层 askUser
 5. `summary`：`{本步能力名} · …`
 6. **禁止**重排或扩写流程；流程只读。需要追加「再来一次生成规则」等 → 由总管再调 design-flow
 7. **本步参数**由编排期写入 steps[].params，程序会注入【本步参数】。按参数执行；禁止再问「这一步生成什么 / 调用哪个规则」。参数缺失或与目录必填项不符 → 停止产出，提示返回 design-flow 补参
--- a/src/llm/token-tracker.ts
+++ b/src/llm/token-tracker.ts
@@ -5,6 +5,7 @@ import type {
  CompleteWithToolsResult,
  LlmProvider,
  StreamCallbacks,
+  TokenUsage,
 } from "./client.js";
 import {
  recordTokenUsage,
@@ -15,6 +16,7 @@ import {
  buildContextTrace,
  type LlmContextTrace,
 } from "../types/context-trace.js";
+import { debugLog, labelCaller, labelLlmMode, labelTool } from "../log.js";

 export type LlmTrackingContext = {
  sessionId?: string;
@@ -41,6 +43,77 @@ function capturePendingTrace(
  });
 }

+function recordUsage(
+  ctx: LlmTrackingContext,
+  caller: string | undefined,
+  model: string | undefined,
+  usage: TokenUsage | undefined,
+): void {
+  if (!usage) return;
+  const record = recordTokenUsage({
+    sessionId: ctx.sessionId,
+    bookId: ctx.bookId,
+    bookTitle: ctx.bookTitle,
+    orchestratorId: ctx.orchestratorId,
+    caller: caller ?? "unknown",
+    model: model ?? "unknown",
+    promptTokens: usage.promptTokens,
+    completionTokens: usage.completionTokens,
+    totalTokens: usage.totalTokens,
+    cachedTokens: usage.cachedTokens,
+    cacheMissTokens: usage.cacheMissTokens,
+  });
+  ctx.pendingUsage = toMessageTokenUsage(record);
+}
+
+function summarizeComplete(result: CompleteResult): string {
+  const bits = [
+    result.model ? `模型=${result.model}` : "",
+    `思维链=${result.reasoning?.trim().length ?? 0}字`,
+    `正文=${result.content?.length ?? 0}字`,
+  ];
+  if (result.usage) {
+    bits.push(`用量=${result.usage.promptTokens}+${result.usage.completionTokens}`);
+  }
+  return bits.filter(Boolean).join(" ");
+}
+
+function summarizeTools(result: CompleteWithToolsResult): string {
+  const tools =
+    result.toolCalls.map((t) => labelTool(t.name)).join("、") || "无";
+  const bits = [
+    result.model ? `模型=${result.model}` : "",
+    `思维链=${result.reasoning?.trim().length ?? 0}字`,
+    `正文=${result.content?.length ?? 0}字`,
+    `工具=${tools}`,
+  ];
+  if (result.usage) {
+    bits.push(`用量=${result.usage.promptTokens}+${result.usage.completionTokens}`);
+  }
+  return bits.filter(Boolean).join(" ");
+}
+
+async function loggedLlm<T>(
+  kind: string,
+  caller: string | undefined,
+  run: () => Promise<T>,
+  summarize: (result: T) => string,
+): Promise<T> {
+  const who = labelCaller(caller);
+  const mode = labelLlmMode(kind);
+  const t0 = Date.now();
+  debugLog("llm", `开始 ${who} · ${mode}`);
+  try {
+    const result = await run();
+    debugLog("llm", `结束 ${who} · ${mode} ${Date.now() - t0}毫秒 ${summarize(result)}`);
+    return result;
+  } catch (err) {
+    const detail = err instanceof Error ? err.message : String(err);
+    debugLog("llm", `失败 ${who} · ${mode} ${Date.now() - t0}毫秒 ${detail.slice(0, 160)}`);
+    throw err;
+  }
+}
+
 export class TokenTrackingProvider implements LlmProvider {
  constructor(
    private readonly inner: LlmProvider,
@@ -51,29 +124,21 @@ export class TokenTrackingProvider implements LlmProvider {
    messages: Parameters<LlmProvider["complete"]>[0],
    options?: CompleteOptions,
  ): Promise<CompleteResult> {
-    const result = await this.inner.complete(messages, options);
-    const ctx = this.getContext();
-    capturePendingTrace(ctx, messages, options?.caller, result.model);
-    if (result.usage) {
-      const record = recordTokenUsage({
-        sessionId: ctx.sessionId,
-        bookId: ctx.bookId,
-        bookTitle: ctx.bookTitle,
-        orchestratorId: ctx.orchestratorId,
-        caller: options?.caller ?? "unknown",
-        model: result.model ?? "unknown",
-        promptTokens: result.usage.promptTokens,
-        completionTokens: result.usage.completionTokens,
-        totalTokens: result.usage.totalTokens,
-        cachedTokens: result.usage.cachedTokens,
-        cacheMissTokens: result.usage.cacheMissTokens,
-      });
-      ctx.pendingUsage = toMessageTokenUsage(record);
-    }
-    if (result.reasoning?.trim()) {
-      ctx.pendingReasoning = result.reasoning.trim();
-    }
-    return result;
+    return loggedLlm(
+      "complete",
+      options?.caller,
+      async () => {
+        const result = await this.inner.complete(messages, options);
+        const ctx = this.getContext();
+        capturePendingTrace(ctx, messages, options?.caller, result.model);
+        recordUsage(ctx, options?.caller, result.model, result.usage);
+        if (result.reasoning?.trim()) {
+          ctx.pendingReasoning = result.reasoning.trim();
+        }
+        return result;
+      },
+      summarizeComplete,
+    );
  }

  async completeStream(
@@ -88,67 +153,51 @@ export class TokenTrackingProvider implements LlmProvider {
      if (result.content) callbacks.onContentDelta?.(result.content);
      return result;
    }
-    let reasoningBuf = "";
-    const result = await inner.completeStream(messages, options, {
-      onReasoningDelta: (delta) => {
-        reasoningBuf += delta;
+    return loggedLlm(
+      "stream",
+      options?.caller,
+      async () => {
+        let reasoningBuf = "";
+        const result = await inner.completeStream!(messages, options, {
+          onReasoningDelta: (delta) => {
+            reasoningBuf += delta;
+            const ctx = this.getContext();
+            ctx.pendingReasoning = reasoningBuf;
+            callbacks.onReasoningDelta?.(delta);
+          },
+          onContentDelta: callbacks.onContentDelta,
+        });
        const ctx = this.getContext();
-        ctx.pendingReasoning = reasoningBuf;
-        callbacks.onReasoningDelta?.(delta);
+        capturePendingTrace(ctx, messages, options?.caller, result.model);
+        recordUsage(ctx, options?.caller, result.model, result.usage);
+        if (result.reasoning?.trim()) {
+          ctx.pendingReasoning = result.reasoning.trim();
+        }
+        return result;
      },
-      onContentDelta: callbacks.onContentDelta,
-    });
-    const ctx = this.getContext();
-    capturePendingTrace(ctx, messages, options?.caller, result.model);
-    if (result.usage) {
-      const record = recordTokenUsage({
-        sessionId: ctx.sessionId,
-        bookId: ctx.bookId,
-        bookTitle: ctx.bookTitle,
-        orchestratorId: ctx.orchestratorId,
-        caller: options?.caller ?? "unknown",
-        model: result.model ?? "unknown",
-        promptTokens: result.usage.promptTokens,
-        completionTokens: result.usage.completionTokens,
-        totalTokens: result.usage.totalTokens,
-        cachedTokens: result.usage.cachedTokens,
-        cacheMissTokens: result.usage.cacheMissTokens,
-      });
-      ctx.pendingUsage = toMessageTokenUsage(record);
-    }
-    if (result.reasoning?.trim()) {
-      ctx.pendingReasoning = result.reasoning.trim();
-    }
-    return result;
+      summarizeComplete,
+    );
  }

  async completeWithTools(
    messages: Parameters<LlmProvider["completeWithTools"]>[0],
    options: CompleteWithToolsOptions,
  ): Promise<CompleteWithToolsResult> {
-    const result = await this.inner.completeWithTools(messages, options);
-    const ctx = this.getContext();
-    capturePendingTrace(ctx, messages, options.caller, result.model);
-    if (result.usage) {
-      const record = recordTokenUsage({
-        sessionId: ctx.sessionId,
-        bookId: ctx.bookId,
-        bookTitle: ctx.bookTitle,
-        orchestratorId: ctx.orchestratorId,
-        caller: options.caller ?? "unknown",
-        model: result.model ?? "unknown",
-        promptTokens: result.usage.promptTokens,
-        completionTokens: result.usage.completionTokens,
-        totalTokens: result.usage.totalTokens,
-        cachedTokens: result.usage.cachedTokens,
-        cacheMissTokens: result.usage.cacheMissTokens,
-      });
-      ctx.pendingUsage = toMessageTokenUsage(record);
-    }
-    if (result.reasoning?.trim()) {
-      ctx.pendingReasoning = result.reasoning.trim();
-    }
-    return result;
+    return loggedLlm(
+      "tools",
+      options.caller,
+      async () => {
+        const result = await this.inner.completeWithTools(messages, options);
+        const ctx = this.getContext();
+        capturePendingTrace(ctx, messages, options.caller, result.model);
+        recordUsage(ctx, options.caller, result.model, result.usage);
+        if (result.reasoning?.trim()) {
+          ctx.pendingReasoning = result.reasoning.trim();
+        }
+        return result;
+      },
+      summarizeTools,
+    );
  }

  async completeWithToolsStream(
@@ -160,37 +209,29 @@ export class TokenTrackingProvider implements LlmProvider {
    if (!inner.completeWithToolsStream) {
      return this.completeWithTools(messages, options);
    }
-    let reasoningBuf = "";
-    const result = await inner.completeWithToolsStream(messages, options, {
-      onReasoningDelta: (delta) => {
-        reasoningBuf += delta;
+    return loggedLlm(
+      "tools-stream",
+      options.caller,
+      async () => {
+        let reasoningBuf = "";
+        const result = await inner.completeWithToolsStream!(messages, options, {
+          onReasoningDelta: (delta) => {
+            reasoningBuf += delta;
+            const ctx = this.getContext();
+            ctx.pendingReasoning = reasoningBuf;
+            callbacks.onReasoningDelta?.(delta);
+          },
+          onContentDelta: callbacks.onContentDelta,
+        });
        const ctx = this.getContext();
-        ctx.pendingReasoning = reasoningBuf;
-        callbacks.onReasoningDelta?.(delta);
+        capturePendingTrace(ctx, messages, options.caller, result.model);
+        recordUsage(ctx, options.caller, result.model, result.usage);
+        if (result.reasoning?.trim()) {
+          ctx.pendingReasoning = result.reasoning.trim();
+        }
+        return result;
      },
-      onContentDelta: callbacks.onContentDelta,
-    });
-    const ctx = this.getContext();
-    capturePendingTrace(ctx, messages, options.caller, result.model);
-    if (result.usage) {
-      const record = recordTokenUsage({
-        sessionId: ctx.sessionId,
-        bookId: ctx.bookId,
-        bookTitle: ctx.bookTitle,
-        orchestratorId: ctx.orchestratorId,
-        caller: options.caller ?? "unknown",
-        model: result.model ?? "unknown",
-        promptTokens: result.usage.promptTokens,
-        completionTokens: result.usage.completionTokens,
-        totalTokens: result.usage.totalTokens,
-        cachedTokens: result.usage.cachedTokens,
-        cacheMissTokens: result.usage.cacheMissTokens,
-      });
-      ctx.pendingUsage = toMessageTokenUsage(record);
-    }
-    if (result.reasoning?.trim()) {
-      ctx.pendingReasoning = result.reasoning.trim();
-    }
-    return result;
+      summarizeTools,
+    );
  }
 }
--- a/src/log.ts
+++ b/src/log.ts
@@ -0,0 +1,293 @@
+import { execSync } from "node:child_process";
+import {
+  appendFileSync,
+  closeSync,
+  existsSync,
+  mkdirSync,
+  openSync,
+  readSync,
+  renameSync,
+  statSync,
+  writeFileSync,
+} from "node:fs";
+import path from "node:path";
+import type {
+  ApplyEventResult,
+  RuntimeEvent,
+  RuntimePhase,
+  RuntimeSession,
+  WaitingReason,
+} from "./types/runtime.js";
+import { getUserDataDir } from "./config/user-data-dir.js";
+
+const MAX_LOG_BYTES = 2 * 1024 * 1024;
+const UTF8_BOM = "\uFEFF";
+
+const KIND_LABEL = {
+  llm: "模型",
+  state: "状态",
+  step: "步骤",
+} as const;
+
+const PHASE_LABEL: Record<RuntimePhase, string> = {
+  idle: "空闲",
+  running: "执行中",
+  waiting_user: "等待用户",
+  done: "完成",
+  error: "出错",
+};
+
+const WAIT_LABEL: Record<WaitingReason["kind"], string> = {
+  skill_selection: "选能力包",
+  intake: "收集需求",
+  input: "等待输入",
+  approve_step: "确认开干",
+  review_artifact: "验收产物",
+  worker_questions: "执行单元提问",
+  revision: "修订",
+  next_intent: "下一步意向",
+};
+
+const EVENT_LABEL: Record<RuntimeEvent["type"], string> = {
+  session_started: "会话开始",
+  skill_selected: "已选能力包",
+  user_submitted_input: "用户发言",
+  user_confirmed_intake: "确认需求",
+  main_agent_decision_created: "总管决策",
+  user_approved_next_step: "确认下一步",
+  user_rejected_next_step: "拒绝下一步",
+  worker_started: "执行单元开始",
+  worker_completed: "执行单元完成",
+  worker_needs_input: "执行单元提问",
+  user_resolved_sidecar_questions: "回答挂载追问",
+  user_accepted_artifact: "验收通过",
+  user_rejected_artifact: "打回产物",
+  user_requested_revision: "请求修订",
+  programmatic_review_started: "程序验收开始",
+  programmatic_review_passed: "程序验收通过",
+  programmatic_review_failed: "程序验收失败",
+  flow_completed: "流程结束",
+  runtime_failed: "运行失败",
+};
+
+const EFFECT_LABEL: Record<string, string> = {
+  invoke_main_agent: "调用总管",
+  run_worker: "启动执行单元",
+  resume_worker: "恢复执行单元",
+  run_programmatic_review: "程序验收",
+  emit_message: "发消息",
+  propose_next_creation_step: "提案下一步",
+};
+
+const TOOL_LABEL: Record<string, string> = {
+  read_blackboard: "读黑板",
+  list_workers: "列执行单元",
+  list_artifacts: "列产物",
+  run_worker: "跑执行单元",
+  ask_user: "问用户",
+  review_blackboard: "审黑板",
+  finish: "结束",
+};
+
+const ACTION_LABEL: Record<string, string> = {
+  run_worker: "跑执行单元",
+  ask_user: "问用户",
+  review_blackboard: "审黑板",
+  finish: "结束",
+};
+
+let consoleUtf8Ready = false;
+
+/** Windows 控制台切到 UTF-8，避免中文日志乱码。 */
+export function ensureConsoleUtf8(): void {
+  if (consoleUtf8Ready) return;
+  consoleUtf8Ready = true;
+  if (process.platform !== "win32") return;
+  try {
+    // inherit 才会改当前窗口代码页；stdio ignore 只影响子进程
+    execSync("chcp 65001 >NUL", { stdio: "inherit", shell: true, windowsHide: true });
+  } catch {
+    /* 无控制台（桌面隐藏启动）时忽略 */
+  }
+  try {
+    process.stdout.setDefaultEncoding("utf8");
+    process.stderr.setDefaultEncoding("utf8");
+  } catch {
+    /* ignore */
+  }
+}
+
+export function getRuntimeLogPath(): string {
+  return path.join(getUserDataDir(), "logs", "runtime.log");
+}
+
+let logFilePrepared = false;
+
+function fileHasUtf8Bom(file: string): boolean {
+  try {
+    const fd = openSync(file, "r");
+    const buf = Buffer.alloc(3);
+    const n = readSync(fd, buf, 0, 3, 0);
+    closeSync(fd);
+    return n >= 3 && buf[0] === 0xef && buf[1] === 0xbb && buf[2] === 0xbf;
+  } catch {
+    return false;
+  }
+}
+
+function prepareLogFile(file: string): void {
+  if (logFilePrepared) return;
+  mkdirSync(path.dirname(file), { recursive: true });
+  if (existsSync(file) && statSync(file).size > 0 && !fileHasUtf8Bom(file)) {
+    try {
+      renameSync(file, `${file}.old`);
+    } catch {
+      /* ignore */
+    }
+  }
+  if (!existsSync(file) || statSync(file).size === 0) {
+    writeFileSync(file, `${UTF8_BOM}=== 写作助手运行日志（UTF-8）===\n`, "utf8");
+  }
+  logFilePrepared = true;
+}
+
+function appendRuntimeLog(line: string): void {
+  if (process.env.VITEST) return;
+  try {
+    const file = getRuntimeLogPath();
+    prepareLogFile(file);
+    try {
+      if (statSync(file).size > MAX_LOG_BYTES) {
+        renameSync(file, `${file}.1`);
+        writeFileSync(file, `${UTF8_BOM}=== 写作助手运行日志（UTF-8）===\n`, "utf8");
+      }
+    } catch {
+      /* rotate best-effort */
+    }
+    appendFileSync(file, `${line}\n`, "utf8");
+  } catch {
+    /* logging must not break the app */
+  }
+}
+
+function stamp(): string {
+  return new Date().toLocaleTimeString("zh-CN", {
+    hour12: false,
+    hour: "2-digit",
+    minute: "2-digit",
+    second: "2-digit",
+  });
+}
+
+function labelPhase(phase: string): string {
+  return PHASE_LABEL[phase as RuntimePhase] ?? phase;
+}
+
+function labelWait(wait: string): string {
+  if (wait === "-") return "无";
+  return WAIT_LABEL[wait as WaitingReason["kind"]] ?? wait;
+}
+
+export function labelCaller(caller: string | undefined): string {
+  const who = caller?.trim() || "未知";
+  if (who === "main_agent" || who.startsWith("main_agent:")) return "总管";
+  if (who === "intake_extract") return "需求抽取";
+  if (who.startsWith("worker:")) return `执行单元 ${who.slice("worker:".length)}`;
+  if (who === "unknown") return "未知";
+  return who;
+}
+
+export function labelLlmMode(kind: string): string {
+  switch (kind) {
+    case "complete":
+      return "整段";
+    case "stream":
+      return "流式";
+    case "tools":
+      return "工具";
+    case "tools-stream":
+      return "工具流式";
+    default:
+      return kind;
+  }
+}
+
+export function labelTool(name: string): string {
+  return TOOL_LABEL[name] ?? name;
+}
+
+export function labelAction(action: string): string {
+  return ACTION_LABEL[action] ?? action;
+}
+
+/** 服务端必要步骤：每次模型请求、状态迁移、创作步骤推进。 */
+export function debugLog(kind: "llm" | "state" | "step", line: string): void {
+  ensureConsoleUtf8();
+  const text = `[${stamp()} ${KIND_LABEL[kind]}] ${line}`;
+  console.log(text);
+  appendRuntimeLog(text);
+}
+
+export function sessionSnap(session: RuntimeSession): {
+  phase: string;
+  wait: string;
+  step: string;
+  worker: string;
+} {
+  return {
+    phase: session.phase,
+    wait: session.waitingReason?.kind ?? "-",
+    step: session.currentStepId ?? "-",
+    worker: session.currentWorkerId ?? "-",
+  };
+}
+
+function eventHint(event: RuntimeEvent): string {
+  switch (event.type) {
+    case "worker_started":
+    case "worker_needs_input":
+      return event.payload.workerId;
+    case "main_agent_decision_created": {
+      const d = event.payload.decision;
+      const action = labelAction(d.action);
+      return d.workerId ? `${action} ${d.workerId}` : action;
+    }
+    case "skill_selected":
+      return event.payload.skill.name;
+    case "user_accepted_artifact":
+    case "user_rejected_artifact":
+      return event.payload.artifactId.slice(0, 8);
+    default:
+      return "";
+  }
+}
+
+export function logStateChange(
+  event: RuntimeEvent,
+  before: ReturnType<typeof sessionSnap>,
+  result: ApplyEventResult,
+): void {
+  const after = sessionSnap(result.session);
+  const bits = [EVENT_LABEL[event.type] ?? event.type];
+  const hint = eventHint(event);
+  if (hint) bits.push(hint);
+  if (before.phase !== after.phase) {
+    bits.push(`相位 ${labelPhase(before.phase)}→${labelPhase(after.phase)}`);
+  } else {
+    bits.push(`相位=${labelPhase(after.phase)}`);
+  }
+  if (before.wait !== after.wait) {
+    bits.push(`等待 ${labelWait(before.wait)}→${labelWait(after.wait)}`);
+  } else if (after.wait !== "-") {
+    bits.push(`等待=${labelWait(after.wait)}`);
+  }
+  if (before.step !== after.step) bits.push(`步骤 ${before.step}→${after.step}`);
+  if (before.worker !== after.worker) bits.push(`执行单元 ${before.worker}→${after.worker}`);
+  const effects = result.effects.map((e) => {
+    const name = EFFECT_LABEL[e.type] ?? e.type;
+    return e.type === "run_worker" ? `${name} ${e.workerId}` : name;
+  });
+  if (effects.length) bits.push(`随后 ${effects.join("，")}`);
+  if (result.error) bits.push(`出错 ${result.error.slice(0, 120)}`);
+  debugLog("state", bits.join("  "));
+}
--- a/src/main-agent/main-agent.ts
+++ b/src/main-agent/main-agent.ts
@@ -23,48 +23,15 @@ function buildMainAgentSystemPrompt(
      ? workers.map((w) => `- ${w.id}：${w.description}`).join("\n")
      : "- （当前 skill 未加载 worker 列表）";

-  return `你是写作系统的编排器（Main Agent）。你的职责是调度执行单元（worker），而不是直接创作正文。
+  return `你是创作节点路由器。只读取「设计.创作流程」DAG，并选择下一个要执行的创作节点。

-规则：
-1. 你不能直接生成小说/文章正文。
-2. 你不能修改运行状态；statePatchAllowed 必须始终为 false。
-3. 你只能建议下一步动作：ask_user、run_worker、create_temp_worker、review_blackboard、finish。
-4. 当信息不足时，使用 ask_user：assessment 是主内容（完备度评价）；questions 挂在其下且用户可跳过；一次 1～2 题。
-5. 当需要执行任务时，使用 run_worker，只指定 workerId。不要指定 inputTags 或 outputTags——Runtime 从 Worker Skill 读取。
-6. requiresApproval 表示运行 worker 前是否需要用户确认。代笔模式通常为 true。
-7. run_worker 可选 workerContext：{ "roleId": "A" } 或 { "chance": { "op":"roll", "expression":"2d6" } }（机遇裁定走程序，勿让模型编随机）。
-8. 你不能把未验收内容当作事实。
-9. 向用户提问是 worker 的技能（ask_user tool），不是独立 worker。编排器只在调度层提问。
-10. blackboardIndex 只有 tag 索引，不含正文 content。
-
-## 调度思维
-用需求正推：「用户需要 [具体体验/技能] → 调用 [worker/skill] 来 [生成/补充/调整] [什么]，以便更好满足用户。」
-禁止否定式路由：「某模式 / 背景形态 → 不需要某步骤 / 跳过某 worker」。
-
-反例（禁止）：「背景为单一角色，无需世界构建」「不是规则怪谈，跳过 write-rules」「默认上世界模拟套件」
-正例：「用户已选世界模拟器配方且要网恋对话 → design-flow 排出近期增量步骤（可调味、可后补；有编排参数的步骤须钉 params）→ 用户认可后反复 design-step；需要再补规则/实例 → 再 design-flow 追加同技能并写齐 params → 收成后若需开局 → opening-generator；用户手动进 play」
+- DAG 不存在或需要追加节点：调用 design-flow。
+- DAG 中已有下一个可执行节点：调用 design-step。
+- 只调度 worker，不创作正文，不规划 DAG 之外的流程。

 当前 skill 可用 worker（workerId 必须与下列 id 完全一致）：
 ${workerLines}

-## design 优先顺序
- 尚无「设计.创作流程」→ **design-flow**（近期 horizon；status=open；未选配方则先请用户选）
- 流程已验收且还有未完成步骤 → **design-step**（一次一步；技能由程序注入）
- 已列步骤都验收但流程仍 **status=open** → 再 **design-flow**（追加反复步或 closed）
- 禁止调度已废弃的 design-core / design-fixed / design-worker / design-refine
- 终稿已 accept 且可用 opening-generator、尚无开场产物 → opening-generator
- 进 play 由用户手动决定
- **禁止**替用户猜测或改选配方
- **禁止**一次编排排死全程固定 DAG
-
-## play 优先顺序（lifecycleStage=play）
- 严格按声明管线顺序调度（play_slots 展开序）：**auditor（旁观维护）→ perspective? → gm（主世界层）→ narrator（转述）**
- auditor / gm / perspective 的 acceptance 多为 continue：跑完立刻调度下一槽，不要中途 ask_user
- 仅 narrator（或规格标明 review 的终稿槽）停下来等人看
- chance 仅在需要真随机时按需 run_worker，不插入每轮固定序
- slots.pendingSideEffectWorkers 非空时优先调度其中 workerId
- **禁止**发明未声明的 play ref；**禁止**把旁观维护当成正文作者
-
 输出必须是 JSON 对象，字段：
 {
  "action": "ask_user" | "run_worker" | "create_temp_worker" | "review_blackboard" | "finish",
@@ -110,40 +77,21 @@ export class MainAgent {

 export function buildMainAgentUserPrompt(context: MainAgentContext): string {
  const { session, blackboardIndex, availableWorkers } = context;
-  const stage = String(session.slots?.uiLifecycleStage ?? "design");
-  const pendingSide = session.slots?.pendingSideEffectWorkers;
+  const creationTags = new Set([
+    "设计.创作流程",
+    "创作.当前步骤",
+    "创作.当前单位",
+    "创作.已验收单位",
+    "用户.下一步意向",
+  ]);
  return JSON.stringify(
    {
-      runtimePhase: session.phase,
-      waitingReason: session.waitingReason,
-      lifecycleStage: stage,
-      flowId: session.flowId,
      currentStepId: session.currentStepId,
-      currentWorkerId: session.currentWorkerId,
-      acceptanceMode: session.acceptanceMode,
-      slots: session.slots,
-      pendingSideEffectWorkers: pendingSide ?? null,
-      pendingDecision: session.pendingDecision
-        ? {
-            id: session.pendingDecision.id,
-            action: session.pendingDecision.action,
-            reason: session.pendingDecision.reason,
-          }
-        : null,
-      pendingArtifactId: session.pendingArtifactId,
-      artifacts: session.artifacts.map((a) => ({
-        id: a.id,
-        workerId: a.workerId,
-        status: a.status,
-        summary: a.summary,
-        outputTags: a.outputTags,
-      })),
-      blackboardIndex,
+      creationDagTags: blackboardIndex.filter((entry) =>
+        creationTags.has(entry.tag),
+      ),
      availableWorkers,
-      instruction:
-        stage === "play"
-          ? "play：按声明序 auditor→perspective?→gm→narrator；continue 槽连跑；仅终稿 review 停；pendingSideEffectWorkers 优先；禁止未声明 ref。"
-          : "根据当前状态决定下一步。尚无设计.创作流程 → design-flow；有未完成步骤 → design-step；steps 做完但 status=open → 再 design-flow；禁止旧 design-core/fixed/worker/refine。已 accept 且需开局 → opening-generator。进 play 由用户手动。",
+      instruction: "读取创作 DAG，选择下一个创作节点。",
    },
    null,
    2,
--- a/src/main-agent/tool-loop.ts
+++ b/src/main-agent/tool-loop.ts
@@ -6,6 +6,7 @@ import { isMainAgentTerminalTool } from "../types/tools.js";
 import { MAIN_AGENT_TOOL_DEFINITIONS } from "./tools.js";
 import { buildMainAgentUserPrompt, parseMainAgentDecision } from "./main-agent.js";
 import type { MainAgentContext } from "./main-agent.js";
+import { debugLog } from "../log.js";

 export type ToolLoopHandlers = {
  readBlackboard: (tags: string[]) => Record<string, string>;
@@ -20,7 +21,7 @@ export type ToolLoopHandlers = {
  onToolCall?: (name: string, detail: string) => void;
  /** 思维链 / reasoning 流式增量（供 UI 实时展示） */
  onThinkingDelta?: (delta: string) => void;
-  /** 一轮 LLM 调用结束后的完整思考文本 */
+  /** 一轮 LLM 调用结束后的完整思维链（不含正文） */
  onThinkingDone?: (text: string) => void;
 };

@@ -40,71 +41,17 @@ function buildToolLoopSystemPrompt(
      ? workers.map((w) => `- ${w.id}：${w.description}`).join("\n")
      : "- （当前 skill 未加载 worker 列表）";

-  return `你是写作系统的总管（Main Agent）。你在 running 相位通过 **tool call** 推进流程。
+  return `你是创作节点路由器。只根据「设计.创作流程」DAG 选择下一个创作节点。

-## 能力边界
-1. 你不能直接生成小说/文章正文。
-2. 你不能修改运行状态。
-3. 先用 read_blackboard / list_workers / list_artifacts 收集信息，再决定下一步。
-4. 终止动作只能用 tool：ask_user、run_worker、review_blackboard、finish。
-   - ask_user：**主内容是 assessment（内容完备度评价）**；questions 挂在其下且可被用户跳过。
-     · assessment（必填）：给用户看的评价正文。以【核心体验】为首；按需列维度（内容/人物关系/感官/背景规则/意义主题…），不必凑齐。每维写完备度%、已知、待探；待探用【】标互斥或可组合方向。
-     · questions（必填数组，语义可选）：只针对 assessment「待探」里想确认的点；一次 1～2 题。用户可不答、直接让你基于现有信息继续。prompt=明确题干；options=2～4 条**建议示范**；required 默认 false。
-     · reason：调度短句，正推，勿塞长文。
-     · 能推断就别问；已写在 用户.需求 里的不要重复问。
-5. run_worker 只传 workerId；inputTags/outputTags 由 Runtime 从 Worker Skill 读取。
-6. requiresApproval=true 时 run_worker 需用户确认后再执行。
-7. 不能把未验收产物当作已定事实。
-
-## 调度思维（必须遵守）
-用 **需求正推**，禁止 **模式否定式** 表述。
-
-**要这样思考（正推）：**
-「用户需要 [具体体验/能力/产物] → 因此调用 [worker/skill] 来 [生成/补充/调整] [什么]，以便更好满足用户。」
-
-**不要这样思考（反例）：**
-「这是 xxx 模式 / 形态 → 不需要 xxx 步骤 / 跳过 xxx。」
-「背景为单一角色，无需世界构建。」
-「不是规则怪谈，跳过 write-rules。」
-
-示例（好）：
- 「用户已选配方且要 1v1 网恋 → design-flow 排出近期增量步骤（可调味）→ 认可后反复 design-step；不够再追加。」
- 「尚无设计.创作流程 → run_worker(design-flow)；有未完成步骤 → design-step；steps 完但 status=open → 再 design-flow。」
- 「运行规格已 accept 且声明需要开局 → 调用 opening-generator 写开场白。」
-
-示例（坏）：
- 「调度 design-core / design-fixed（已废弃）。」
- 「一次 design-flow 排死全程固定 DAG。」
- 「跳过流程编排直接写满运行规格。」
- 「默认先上世界模拟全套再问用户要什么。」
- 「替用户改选 / 猜测配方。」
-
-在 reasoning / 可见 content 中请用 **正推句式** 写出调度理由；终止 tool 的 reason 字段同样用正推表述。
-
-## design 阶段提示
- 尚无「设计.创作流程」→ **优先 design-flow**（近期 horizon；status=open；未选则先请用户选）。
- 流程已验收、还有未完成步骤 → **design-step**（一次一步）。
- 已列步骤全验收但 **status=open** → 再 **design-flow**（追加生成规则/具体实例等并钉齐 params，或设 closed）。
- 有编排参数的步骤缺必填 params → 先让 design-flow askUser，不要空壳跑 design-step。
- 禁止 run_worker(design-core|design-fixed|design-worker|design-refine)——已废弃。
- 不要重复问用户「想做什么」——意图已在 用户.需求。
- 运行规格已 accept，且可用 worker 含 opening-generator，尚无开场产物 → opening-generator。
- 进 play 由用户手动决定。
- 未声明开局、用户也未要求开场时，不要硬调 opening-generator。
- **禁止**替用户猜测或改选配方。
- **禁止**一次编排排死全程固定长链。
-
-## play 阶段提示（lifecycleStage=play 时优先）
- 管线顺序：**auditor（旁观维护）→ perspective? → gm（主世界层）→ narrator**。
- auditor/gm/perspective 多为 continue：写完立刻 run 下一槽，不要中途 ask_user。
- 只在 narrator（或 review 终稿槽）停给人看。
- chance 按需；pendingSideEffectWorkers 非空则优先调度。
- 禁止未声明 play ref；旁观维护默认空操作，不是正文作者。
+1. 用 read_blackboard 读取「设计.创作流程」及其进度 tag。
+2. 没有 DAG 或 DAG 需要追加节点时，run_worker(design-flow)。
+3. DAG 已给出下一个可执行节点时，run_worker(design-step)。
+4. 不创作正文，不改写 DAG，不处理运行状态或游玩流程。

 当前 skill 可用 worker：
 ${workerLines}

-信息不足时可多次 read_blackboard；一旦调用终止 tool，本轮循环结束。`;
+选定节点后立即调用 run_worker。`;
 }

 async function completeToolsPreferStream(
@@ -172,14 +119,9 @@ function assistantMessageFromToolCalls(
  };
 }

-function emitThinkingDone(handlers: ToolLoopHandlers, parts: {
-  reasoning?: string;
-  content?: string | null;
-}): void {
-  const reasoning = parts.reasoning?.trim() ?? "";
-  const content = parts.content?.trim() ?? "";
-  const combined = [reasoning, content].filter(Boolean).join("\n\n").trim();
-  if (combined) handlers.onThinkingDone?.(combined);
+function emitThinkingDone(handlers: ToolLoopHandlers, reasoning?: string): void {
+  const text = reasoning?.trim() ?? "";
+  if (text) handlers.onThinkingDone?.(text);
 }

 /**
@@ -202,12 +144,12 @@ export async function runMainAgentToolLoop(
  const toolTrace: string[] = [];
  const streamCallbacks: StreamCallbacks = {
    onReasoningDelta: (delta) => handlers.onThinkingDelta?.(delta),
-    onContentDelta: (delta) => handlers.onThinkingDelta?.(delta),
  };

  for (let iteration = 1; iteration <= MAX_TOOL_LOOP_ITERATIONS; iteration++) {
+    debugLog("llm", `总管循环 第${iteration}轮`);
    const result = await completeToolsPreferStream(llm, messages, streamCallbacks);
-    emitThinkingDone(handlers, result);
+    emitThinkingDone(handlers, result.reasoning);

    if (result.toolCalls.length === 0) {
      if (result.content?.trim()) {
--- a/src/main-agent/tools.ts
+++ b/src/main-agent/tools.ts
@@ -22,118 +22,11 @@ export const MAIN_AGENT_TOOL_DEFINITIONS: ToolDefinition[] = [
      },
    },
  },
-  {
-    type: "function",
-    function: {
-      name: "list_workers",
-      description: "列出当前 skill 可调度的 worker id 与说明。",
-      parameters: {
-        type: "object",
-        properties: {},
-        additionalProperties: false,
-      },
-    },
-  },
-  {
-    type: "function",
-    function: {
-      name: "list_artifacts",
-      description: "列出当前会话 worker 产物（id、workerId、status、summary）。",
-      parameters: {
-        type: "object",
-        properties: {},
-        additionalProperties: false,
-      },
-    },
-  },
-  {
-    type: "function",
-    function: {
-      name: "ask_user",
-      description:
-        "向用户追问。assessment 是给用户看的主内容（完备度评价）；questions 挂在其下，用户可跳过并请你基于现有信息继续。将暂停等待用户。",
-      parameters: {
-        type: "object",
-        properties: {
-          reason: {
-            type: "string",
-            description:
-              "调度层短理由（正推：用户需要澄清 X → 因此追问）。勿把长文评价写这里。",
-          },
-          assessment: {
-            type: "string",
-            description:
-              "给用户看的内容完备度评价（Markdown）。以【核心体验】为首要；按需列维度（不必凑齐），每维：完备度%、已知、待探。待探用【】标出互斥/可组合方向。评价只服务「用户想要什么」。",
-          },
-          message: {
-            type: "string",
-            description:
-              "兼容旧字段：等同 assessment。优先传 assessment。",
-          },
-          questions: {
-            type: "array",
-            description:
-              "挂在 assessment 下的可选追问（用户可不答）。基于「待探」出题，一次 1～2 题。prompt=明确选择题干；options=建议示范（可改写后采用）。",
-            items: {
-              type: "object",
-              properties: {
-                id: {
-                  type: "string",
-                  description: "稳定 id，如 q1 / stance",
-                },
-                prompt: {
-                  type: "string",
-                  description:
-                    "题干：引导用户选定一种详细体验倾向（例：「你更倾向于哪种皇帝的享受？」）。",
-                },
-                options: {
-                  type: "array",
-                  description:
-                    "2～4 个建议示范。label 写完整可采纳文案（可先一句场景钩子再点题），不是标签词。",
-                  items: {
-                    type: "object",
-                    properties: {
-                      id: {
-                        type: "string",
-                        description: "A/B/C…",
-                      },
-                      label: {
-                        type: "string",
-                        description:
-                          "建议正文：用户点选后可直接当答案，也可改写。例：「冰冷的、主宰一切的权力感——九重宫阙一言定生死…」",
-                      },
-                      editable: {
-                        type: "boolean",
-                        description: "默认 true：点文案可改写，点字母才选中",
-                      },
-                    },
-                    required: ["label"],
-                  },
-                },
-                allowOther: {
-                  type: "boolean",
-                  description: "默认 true：允许 Other 自拟",
-                },
-                required: {
-                  type: "boolean",
-                  description: "默认 false（可选追问）；仅关键阻塞题才 true",
-                },
-              },
-              required: ["prompt", "options"],
-            },
-          },
-        },
-        required: ["reason", "assessment", "questions"],
-        additionalProperties: false,
-      },
-    },
-  },
  {
    type: "function",
    function: {
      name: "run_worker",
-      description:
-        "调度 worker。常规只传 workerId。调度 chance（机遇裁定）时传 chance 请求对象（程序掷骰/比点/抽签，禁止让模型编随机）。",
+      description: "执行选中的下一个创作节点。",
      parameters: {
        type: "object",
        properties: {
@@ -143,50 +36,10 @@ export const MAIN_AGENT_TOOL_DEFINITIONS: ToolDefinition[] = [
            type: "boolean",
            description: "true 时需用户确认后才执行",
          },
-          roleId: {
-            type: "string",
-            description: "role-decide 等 worker 的当前角色 id",
-          },
-          chance: {
-            type: "object",
-            description:
-              "仅 workerId=chance：{ op:'roll'|'compare'|'draw'|'pick', ... }。例 roll:{expression:'2d6+1'}；draw:{pool:['a','b'],count:1}",
-          },
        },
        required: ["workerId", "reason", "requiresApproval"],
        additionalProperties: false,
      },
    },
  },
-  {
-    type: "function",
-    function: {
-      name: "review_blackboard",
-      description: "向用户说明当前黑板与进度概况，并暂停等待用户回复。",
-      parameters: {
-        type: "object",
-        properties: {
-          reason: { type: "string" },
-          summary: { type: "string", description: "给用户看的概况说明" },
-        },
-        required: ["reason", "summary"],
-        additionalProperties: false,
-      },
-    },
-  },
-  {
-    type: "function",
-    function: {
-      name: "finish",
-      description: "正常结束当前创作流程。",
-      parameters: {
-        type: "object",
-        properties: {
-          reason: { type: "string" },
-        },
-        required: ["reason"],
-        additionalProperties: false,
-      },
-    },
-  },
 ];
--- a/src/runtime/phase-machine.ts
+++ b/src/runtime/phase-machine.ts
@@ -26,7 +26,10 @@ import {
  readIntakeValues,
  synthesizeDemandText,
 } from "../intake/intake.js";
-import { normalizeQuestions } from "../skills/question-protocol.js";
+import {
+  isModuleOpeningQuestions,
+  normalizeQuestions,
+} from "../skills/question-protocol.js";
 import type { QuestionItem } from "../types/questions.js";

 function nowIso(): string {
@@ -119,6 +122,67 @@ function updateArtifact(
  };
 }

+/** 最近一次被拒 / 待修订的产物（用于 revision 等待态补交意见后重跑） */
+function findRevisionTargetArtifact(
+  session: RuntimeSession,
+): ArtifactRecord | undefined {
+  if (session.pendingArtifactId) {
+    const pending = findArtifact(session, session.pendingArtifactId);
+    if (pending) return pending;
+  }
+  for (let i = session.artifacts.length - 1; i >= 0; i -= 1) {
+    const a = session.artifacts[i];
+    if (a.status === "revision_requested" || a.status === "rejected") {
+      return a;
+    }
+  }
+  return undefined;
+}
+
+/**
+ * 已有修改意见 → 写入「用户.修订说明」并立刻重跑同一 worker。
+ * 语义：在已有产物上按意见改（上下文会注入待改底稿），不是推倒重做。
+ * 对应验收底栏：看产物 → 输入意见 →「按意见修改」。
+ */
+function beginRevisionRerun(
+  session: RuntimeSession,
+  event: RuntimeEvent,
+  artifact: ArtifactRecord,
+  instruction: string,
+): ApplyEventResult {
+  const trimmed = instruction.trim();
+  const slots: Record<string, unknown> = {
+    ...session.slots,
+    "用户.修订说明": trimmed,
+    "用户.最新输入": trimmed,
+    revisionInstruction: trimmed,
+    // 本次重跑的底稿；重跑写不出新版时只许退回这一份
+    revisionTargetArtifactId: artifact.id,
+  };
+  const next = appendHistory(
+    updateArtifact(session, artifact.id, { status: "revision_requested" }),
+    event,
+  );
+  return {
+    session: touch(
+      running({
+        ...next,
+        slots,
+        pendingArtifactId: undefined,
+        pendingDecision: undefined,
+        currentWorkerId: artifact.workerId,
+        currentStepId: artifact.stepId ?? session.currentStepId,
+      }),
+    ),
+    effects: [
+      {
+        type: "run_worker",
+        workerId: artifact.workerId,
+      },
+    ],
+  };
+}
+
 /** 创建 idle 状态的新会话 */
 export function createSession(
  presetId: string,
@@ -154,6 +218,7 @@ export function getAllowedEvents(
        "worker_started",
        "worker_completed",
        "worker_needs_input",
+        "worker_revision_produced_nothing",
        "programmatic_review_started",
        "programmatic_review_passed",
        "programmatic_review_failed",
@@ -204,7 +269,8 @@ export function canApplyEvent(
        reason?.kind === "intake" ||
        reason?.kind === "input" ||
        reason?.kind === "worker_questions" ||
-        reason?.kind === "revision"
+        reason?.kind === "revision" ||
+        reason?.kind === "next_intent"
      );
    case "user_confirmed_intake":
      return reason?.kind === "intake";
@@ -256,6 +322,14 @@ function handleWorkerCompleted(
  const mode = session.acceptanceMode ?? "user_confirmed";
  let next = appendHistory(session, event);

+  if (next.slots.revisionTargetArtifactId) {
+    // 已写出新版，旧底稿不再是退回目标
+    next = {
+      ...next,
+      slots: { ...next.slots, revisionTargetArtifactId: undefined },
+    };
+  }
+
  if (mode === "user_confirmed") {
    next = updateArtifact(next, artifact.id, { status: "under_review" });
    return {
@@ -454,6 +528,62 @@ export function applyEvent(
        };
      }

+      if (session.waitingReason?.kind === "next_intent") {
+        if (text) {
+          slots["用户.下一步意向"] = text;
+          slots["用户.最新输入"] = text;
+        }
+        const next = appendHistory(
+          {
+            ...session,
+            slots,
+            resumeContext: undefined,
+          },
+          event,
+        );
+        return {
+          session: touch(running(next)),
+          effects: [{ type: "propose_next_creation_step" }],
+        };
+      }
+
+      if (session.waitingReason?.kind === "revision") {
+        const instruction =
+          text.trim() || session.waitingReason.instruction?.trim() || "";
+        if (!instruction) {
+          return {
+            session: waiting(touch(appendHistory({ ...session, slots }, event)), {
+              kind: "revision",
+              instruction: session.waitingReason.instruction,
+            }),
+            effects: [
+              {
+                type: "emit_message",
+                message: "请说明要改哪里，再发送。",
+              },
+            ],
+          };
+        }
+        const target = findRevisionTargetArtifact(session);
+        if (target) {
+          return beginRevisionRerun(session, event, target, instruction);
+        }
+        slots["用户.修订说明"] = instruction;
+        slots["用户.最新输入"] = instruction;
+        slots.revisionInstruction = instruction;
+        return {
+          session: touch(
+            running(
+              appendHistory(
+                { ...session, slots, pendingArtifactId: undefined },
+                event,
+              ),
+            ),
+          ),
+          effects: [{ type: "invoke_main_agent" }],
+        };
+      }
+
      if (session.waitingReason?.kind === "worker_questions") {
        slots["用户.worker答复"] = text;
        if (text) slots["用户.最新输入"] = text;
@@ -660,6 +790,7 @@ export function applyEvent(
        acceptanceMode: session.acceptanceMode ?? "user_confirmed",
        questions: normalized,
      };
+      const openingAsk = isModuleOpeningQuestions(normalized);
      return {
        session: waiting(
          appendHistory(
@@ -675,7 +806,10 @@ export function applyEvent(
        effects: [
          {
            type: "emit_message",
-            message: `[Worker] ${event.payload.workerId} 提问：\n${normalized.map((q) => `- ${q.prompt}`).join("\n")}`,
+            // 默认问题正文由说话面 openingGuide 展示，勿再整段塞进调度消息
+            message: openingAsk
+              ? `[Worker] ${event.payload.workerId} · 默认问题（请在主栏按引导作答）`
+              : `[Worker] ${event.payload.workerId} 提问：\n${normalized.map((q) => `- ${q.prompt}`).join("\n")}`,
          },
        ],
      };
@@ -690,16 +824,10 @@ export function applyEvent(
          typeof event.payload.assessment === "string"
            ? event.payload.assessment.trim()
            : "";
-        const effects = [...result.effects];
-        if (sidecar?.length) {
-          effects.push({
-            type: "emit_message",
-            message: `[Worker] 可选追问（可跳过，直接接受目前产物）：\n${sidecar.map((q) => `- ${q.prompt}`).join("\n")}`,
-          });
-        }
+        // 追问与产物同一次产出：只挂 review_artifact，勿再 emit 独立「可选追问」消息
+        //（否则历史里会拆成 worker_output + worker_questions 两段）
        return {
          ...result,
-          effects,
          session: waiting(result.session, {
            kind: "review_artifact",
            artifactId: result.session.pendingArtifactId,
@@ -711,6 +839,42 @@ export function applyEvent(
      return result;
    }

+    case "worker_revision_produced_nothing": {
+      const artifact = findArtifact(session, event.payload.artifactId);
+      if (!artifact) {
+        return fail(session, `Artifact not found: ${event.payload.artifactId}`);
+      }
+      const retryQuestions = normalizeQuestions(event.payload.questions).map(
+        (q) => ({ ...q, required: false }),
+      );
+      const next = appendHistory(
+        updateArtifact(session, artifact.id, { status: "under_review" }),
+        event,
+      );
+      return {
+        session: waiting(
+          {
+            ...next,
+            slots: { ...next.slots, revisionTargetArtifactId: undefined },
+            pendingArtifactId: artifact.id,
+            currentWorkerId: undefined,
+          },
+          {
+            kind: "review_artifact",
+            artifactId: artifact.id,
+            questions: retryQuestions.length ? retryQuestions : undefined,
+          },
+        ),
+        effects: [
+          {
+            type: "emit_message",
+            message:
+              "[系统] 这次按意见重写没能产出可验收的新版本，已退回上一版产物：可补充说明再改，或直接接受上一版。",
+          },
+        ],
+      };
+    }
+
    case "user_resolved_sidecar_questions": {
      const reason = session.waitingReason;
      if (reason?.kind !== "review_artifact") {
@@ -745,22 +909,43 @@ export function applyEvent(
      }
      const slots: Record<string, unknown> = { ...session.slots };
      // 仅终稿 tag「设计.worker集」表示完整 Worker 集验收；草稿单位验收不得开 play
-      if (artifact.outputTags.some((tag) => tag === "设计.worker集")) {
+      const isFinalSet = artifact.outputTags.some((tag) => tag === "设计.worker集");
+      if (isFinalSet) {
        slots.designInstanceReady = true;
      }
      const next = appendHistory(
        updateArtifact(session, artifact.id, { status: "accepted" }),
        event,
      );
-      return {
-        session: touch(
-          running({
-            ...next,
-            slots,
-            pendingArtifactId: undefined,
-            pendingDecision: undefined,
+      const cleared = {
+        ...next,
+        slots,
+        pendingArtifactId: undefined,
+        pendingDecision: undefined,
+        currentWorkerId: undefined,
+      };
+      // 创作步验收后：先问下一步意向，再展示下一节点确认开干（勿直接连跑）
+      const gateNextIntent =
+        !isFinalSet &&
+        (artifact.workerId === "design-step" ||
+          artifact.workerId === "design-flow");
+      if (gateNextIntent) {
+        return {
+          session: waiting(cleared, {
+            kind: "next_intent",
+            afterWorkerId: artifact.workerId,
          }),
-        ),
+          effects: [
+            {
+              type: "emit_message",
+              message:
+                "这一步已收下。下一步想写什么？可留空，直接发送后看下一节点并确认开干。",
+            },
+          ],
+        };
+      }
+      return {
+        session: touch(running(cleared)),
        effects: [{ type: "invoke_main_agent" }],
      };
    }
@@ -770,19 +955,21 @@ export function applyEvent(
      if (!artifact) {
        return fail(session, `Artifact not found: ${event.payload.artifactId}`);
      }
+      const reason = event.payload.reason?.trim() ?? "";
+      // 验收态已写意见 → 在现有产物上按意见改；无意见才进入 revision 等待补交
+      if (reason) {
+        return beginRevisionRerun(session, event, artifact, reason);
+      }
      const next = appendHistory(
        updateArtifact(session, artifact.id, { status: "rejected" }),
        event,
      );
      return {
-        session: waiting(next, {
-          kind: "revision",
-          instruction: event.payload.reason,
-        }),
+        session: waiting(next, { kind: "revision" }),
        effects: [
          {
            type: "emit_message",
-            message: event.payload.reason ?? "产物已被拒绝，等待修改指示。",
+            message: "已保留当前产物，请说明要改哪里。",
          },
        ],
      };
@@ -800,15 +987,26 @@ export function applyEvent(
          effects: [{ type: "invoke_main_agent" }],
        };
      }
+      const instruction = event.payload.instruction?.trim() ?? "";
+      const artifactId =
+        event.payload.artifactId ?? session.pendingArtifactId;
+      const artifact = artifactId
+        ? findArtifact(session, artifactId)
+        : findRevisionTargetArtifact(session);
+      if (instruction && artifact) {
+        return beginRevisionRerun(session, event, artifact, instruction);
+      }
      return {
        session: waiting(
-          appendHistory({ ...session, pendingArtifactId: undefined }, event),
-          { kind: "revision", instruction: event.payload.instruction },
+          appendHistory(session, event),
+          { kind: "revision", instruction: instruction || undefined },
        ),
        effects: [
          {
            type: "emit_message",
-            message: `返工请求：${event.payload.instruction}`,
+            message: instruction
+              ? `修改意见：${instruction}`
+              : "请说明要改哪里（在现有产物上改）。",
          },
        ],
      };
@@ -840,20 +1038,16 @@ export function applyEvent(
      if (!artifact) {
        return fail(session, `Artifact not found: ${event.payload.artifactId}`);
      }
-      const next = appendHistory(
-        updateArtifact(session, artifact.id, { status: "revision_requested" }),
-        event,
-      );
+      const reason = event.payload.reason?.trim() || "程序验收失败";
+      const result = beginRevisionRerun(session, event, artifact, reason);
      return {
-        session: waiting(next, {
-          kind: "revision",
-          instruction: event.payload.reason,
-        }),
+        ...result,
        effects: [
          {
            type: "emit_message",
-            message: `程序验收失败：${event.payload.reason}`,
+            message: `程序验收未过，按意见在现有产物上改：${reason}`,
          },
+          ...result.effects,
        ],
      };
    }
--- a/src/runtime/phase-runtime.ts
+++ b/src/runtime/phase-runtime.ts
@@ -57,18 +57,36 @@ import {
 import {
  CREATION_FLOW_TAG,
  CREATION_CURRENT_STEP_TAG,
+  CREATION_PROPOSED_STEP_TAG,
  CREATION_MODULE_OPENING_TAG,
  CREATION_MODULE_OPENING_STATE_TAG,
  SLOT_CREATION_MODULE_OPENING_STATE,
+  loadModuleCatalog,
+  mergeCreationFlowPreservingAccepted,
  nextPendingStep,
  parseCreationFlow,
  parseModuleOpeningState,
+  patchCreationFlowStepParams,
+  pickRecordedStepId,
  resolveDesignStepBinding,
+  shouldSkipModuleOpening,
+  isProgressPointerTag,
  stepUnitId,
+  stringifyCreationFlow,
  stringifyModuleOpeningState,
+  type CreationFlowStepParams,
+  type ModuleParamSpec,
 } from "../skills/creation-flow.js";
-import { normalizeQuestions } from "../skills/question-protocol.js";
+import {
+  appendAskedQuestions,
+  dropAlreadyAskedQuestions,
+  MODULE_OPENING_QUESTION_ID,
+  normalizeQuestions,
+  parseAskedQuestions,
+  SLOT_ASKED_QUESTIONS,
+} from "../skills/question-protocol.js";
 import { parseWorkerSetYaml } from "../skills/worker-set-parse.js";
+import { debugLog, labelAction, logStateChange, sessionSnap } from "../log.js";
 import {
  executeChance,
  resolveChanceRequest,
@@ -363,10 +381,18 @@ export class PhaseRuntime {
    return this.session;
  }

-  /** 统一事件入口：applyEvent + 更新 session + 处理 effects + 可选总管 LLM */
-  async dispatch(event: RuntimeEvent): Promise<ApplyEventResult> {
+  /** applyEvent 并打状态日志（dispatch / 总管决策共用） */
+  private commitEvent(event: RuntimeEvent): ApplyEventResult {
+    const before = sessionSnap(this.session);
    const result = applyEvent(this.session, event);
    this.session = result.session;
+    logStateChange(event, before, result);
+    return result;
+  }
+
+  /** 统一事件入口：applyEvent + 更新 session + 处理 effects + 可选总管 LLM */
+  async dispatch(event: RuntimeEvent): Promise<ApplyEventResult> {
+    const result = this.commitEvent(event);

    if (
      event.type === "user_submitted_input" ||
@@ -416,14 +442,26 @@ export class PhaseRuntime {
        source: "user",
      });
    }
-    const revision = session.slots.revisionInstruction;
+    const revision =
+      (typeof session.slots["用户.修订说明"] === "string" &&
+        session.slots["用户.修订说明"]) ||
+      (typeof session.slots.revisionInstruction === "string" &&
+        session.slots.revisionInstruction);
    if (typeof revision === "string" && revision.trim()) {
      this.blackboard.write({
-        tag: "用户.修改说明",
+        tag: "用户.修订说明",
        content: revision.trim(),
        source: "user",
      });
    }
+    const nextIntent = session.slots["用户.下一步意向"];
+    if (typeof nextIntent === "string" && nextIntent.trim()) {
+      this.blackboard.write({
+        tag: "用户.下一步意向",
+        content: nextIntent.trim(),
+        source: "user",
+      });
+    }
  }

  /**
@@ -490,10 +528,15 @@ export class PhaseRuntime {
    return this.session;
  }

-  /** 用户确认 pendingDecision（approve_step） */
-  async approve(): Promise<RuntimeSession> {
+  /** 用户确认 pendingDecision（approve_step）；可顺带钉死下一步 params */
+  async approve(stepParams?: CreationFlowStepParams): Promise<RuntimeSession> {
    const id = this.session.pendingDecision?.id;
    if (!id) throw new Error("当前没有待确认的决策");
+    if (stepParams && Object.keys(stepParams).length > 0) {
+      this.applyProposedStepParams(stepParams);
+    }
+    const readyErr = this.assertProposedStepReady();
+    if (readyErr) throw new Error(readyErr);
    await this.dispatch({ type: "user_approved_next_step", payload: { decisionId: id } });
    return this.session;
  }
@@ -577,7 +620,7 @@ export class PhaseRuntime {
    return this.session;
  }

-  /** 用户拒绝 pendingArtifact，进入 revision */
+  /** 用户拒绝 pendingArtifact：有意见则立刻按「用户.修订说明」在现有产物上改 */
  async rejectArtifact(reason?: string, artifactId?: string): Promise<RuntimeSession> {
    const id = artifactId ?? this.session.pendingArtifactId;
    if (!id) throw new Error("当前没有待拒绝的产物");
@@ -588,12 +631,39 @@ export class PhaseRuntime {
    return this.session;
  }

+  /**
+   * 按意见重跑却没写出新产物：退回**本次修订的那一份**产物验收，重试追问改挂为可选。
+   * 只认 beginRevisionRerun 钉下的底稿；否则会把无关的旧产物（如上一轮流程编排）
+   * 错误地挂回验收面。首跑无产物仍走阻塞追问。
+   * @returns true = 已退回验收，调用方应 return
+   */
+  private async restoreRevisionTargetForReview(
+    workerId: string,
+    questions: import("../types/questions.js").QuestionItem[] | string[],
+  ): Promise<boolean> {
+    const targetId = this.session.slots.revisionTargetArtifactId;
+    if (typeof targetId !== "string" || !targetId) return false;
+    const target = this.session.artifacts.find(
+      (a) =>
+        a.id === targetId &&
+        a.workerId === workerId &&
+        a.status === "revision_requested",
+    );
+    if (!target) return false;
+    await this.dispatch({
+      type: "worker_revision_produced_nothing",
+      payload: { artifactId: target.id, questions },
+    });
+    return true;
+  }
+
  /** worker 调用 ask_user 能力时，由外层触发此事件（仅无产物时阻塞） */
  async workerAsk(
    questions: import("../types/questions.js").QuestionItem[] | string[],
  ): Promise<RuntimeSession> {
    const workerId = this.session.currentWorkerId;
    if (!workerId) throw new Error("当前没有运行中的 worker");
+    this.rememberAskedQuestions(normalizeQuestions(questions));
    await this.dispatch({
      type: "worker_needs_input",
      payload: { workerId, stepId: this.session.currentStepId, questions },
@@ -601,6 +671,35 @@ export class PhaseRuntime {
    return this.session;
  }

+  private askedQuestionHistory() {
+    return parseAskedQuestions(this.session.slots[SLOT_ASKED_QUESTIONS]);
+  }
+
+  private rememberAskedQuestions(
+    questions: readonly import("../types/questions.js").QuestionItem[],
+  ): void {
+    if (!questions.length) return;
+    const next = appendAskedQuestions(this.askedQuestionHistory(), questions);
+    this.session = {
+      ...this.session,
+      slots: { ...this.session.slots, [SLOT_ASKED_QUESTIONS]: next },
+    };
+  }
+
+  /**
+   * 挂在验收态的可跳过追问：本局问过的不再问一次。
+   * 编排层常把上一步已问过的体验题再抛一遍，用户会读成「又要重做那一步」。
+   */
+  private freshSidecarQuestions(
+    questions: readonly import("../types/questions.js").QuestionItem[] | undefined,
+  ): import("../types/questions.js").QuestionItem[] | undefined {
+    if (!questions?.length) return undefined;
+    const fresh = dropAlreadyAskedQuestions(questions, this.askedQuestionHistory());
+    if (!fresh.length) return undefined;
+    this.rememberAskedQuestions(fresh);
+    return fresh;
+  }
+
  /**
   * 验收态下作答或跳过挂载追问：不离开 review_artifact。
   * answersText 缺省 = 跳过。
@@ -662,6 +761,13 @@ export class PhaseRuntime {
  /** 消费 applyEvent 返回的 PhaseEffect 列表 */
  private async processEffects(effects: PhaseEffect[]): Promise<void> {
    for (const effect of effects) {
+      if (effect.type === "run_worker") {
+        debugLog("step", `启动执行单元 ${effect.workerId}`);
+      } else if (effect.type === "propose_next_creation_step") {
+        debugLog("step", "提案下一步");
+      } else if (effect.type === "resume_worker") {
+        debugLog("step", `恢复执行单元 ${this.session.resumeContext?.workerId ?? "未知"}`);
+      }
      switch (effect.type) {
        case "emit_message":
          this.onMessage(effect.message);
@@ -705,10 +811,209 @@ export class PhaseRuntime {
        case "run_programmatic_review":
          await this.runProgrammaticReview(effect.artifactId);
          break;
+        case "propose_next_creation_step":
+          await this.proposeNextCreationStep();
+          break;
      }
    }
  }

+  /**
+   * 用户填完「下一步想写什么」后：有待执行步则提案并进入确认；否则交总管扩步/收口。
+   */
+  private async proposeNextCreationStep(): Promise<void> {
+    this.syncSlotsToBlackboard(this.session);
+    const flow = parseCreationFlow(
+      this.blackboard.getContentByTag(CREATION_FLOW_TAG),
+    );
+    const accepted = parseAcceptedUnits(
+      this.session.slots[SLOT_CREATION_ACCEPTED_UNITS] ??
+        this.blackboard.getContentByTag(CREATION_ACCEPTED_UNITS_TAG),
+    );
+    const pending = nextPendingStep(flow, accepted);
+    if (!pending || !flow) {
+      debugLog("step", "暂无下一步，交给总管");
+      this.session = {
+        ...this.session,
+        slots: {
+          ...this.session.slots,
+          [CREATION_PROPOSED_STEP_TAG]: undefined,
+        },
+      };
+      if (this.mainAgent) {
+        await this.runMainAgent();
+      } else {
+        this.onMessage(
+          "[阶段机] 暂无下一节点，等待总管编排或补充意向。",
+        );
+      }
+      return;
+    }
+
+    const active = this.getActiveSkill();
+    let paramSpecs: ModuleParamSpec[] = [];
+    let declaration = "";
+    if (active?.name) {
+      try {
+        const skill = await loadSkill(active.name);
+        if (skill.skillPackRoot) {
+          const catalog = await loadModuleCatalog(skill.skillPackRoot);
+          const mod = catalog?.modules.find((m) => m.name === pending.name);
+          paramSpecs = mod?.params ?? [];
+          declaration = mod?.declaration?.trim() || "";
+        }
+      } catch {
+        /* catalog optional */
+      }
+    }
+
+    debugLog("step", `提案下一步 ${pending.id}（${pending.name}）`);
+
+    const params = pending.params ? { ...pending.params } : {};
+    const paramsMissing = paramSpecs
+      .filter((p) => p.required)
+      .map((p) => p.key)
+      .filter((key) => {
+        const v = params[key];
+        return v == null || (typeof v === "string" && !v.trim());
+      });
+
+    const intent =
+      typeof this.session.slots["用户.下一步意向"] === "string"
+        ? String(this.session.slots["用户.下一步意向"]).trim()
+        : "";
+    const paramHint =
+      Object.keys(params).length > 0
+        ? `；参数 ${Object.entries(params)
+            .map(([k, v]) => `${k}=${v}`)
+            .join(" · ")}`
+        : paramsMissing.length
+          ? `；待钉参数：${paramsMissing.join("、")}`
+          : "";
+    const reason = [
+      `下一步：${pending.name}`,
+      declaration ? `—— ${declaration}` : "",
+      paramHint,
+      intent ? `（你的意向：${intent}）` : "",
+    ]
+      .filter(Boolean)
+      .join("");
+
+    const snapshot = {
+      stepId: pending.id,
+      name: pending.name,
+      declaration: declaration || undefined,
+      params,
+      paramsMissing,
+      paramSpecs: paramSpecs.map((p) => ({
+        key: p.key,
+        label: p.label,
+        required: Boolean(p.required),
+        hint: p.hint,
+      })),
+      intent: intent || undefined,
+    };
+    this.blackboard.write({
+      tag: CREATION_PROPOSED_STEP_TAG,
+      content: JSON.stringify(snapshot),
+      source: "runtime",
+    });
+    this.session = {
+      ...this.session,
+      slots: {
+        ...this.session.slots,
+        [CREATION_PROPOSED_STEP_TAG]: JSON.stringify(snapshot),
+      },
+    };
+
+    const decision = createDecision({
+      action: "run_worker",
+      workerId: "design-step",
+      reason,
+      requiresApproval: true,
+    });
+    await this.dispatch({
+      type: "main_agent_decision_created",
+      payload: { decision },
+    });
+  }
+
+  private applyProposedStepParams(params: CreationFlowStepParams): void {
+    const raw =
+      (typeof this.session.slots[CREATION_PROPOSED_STEP_TAG] === "string"
+        ? String(this.session.slots[CREATION_PROPOSED_STEP_TAG])
+        : null) ||
+      this.blackboard.getContentByTag(CREATION_PROPOSED_STEP_TAG);
+    if (!raw?.trim()) return;
+    let snap: {
+      stepId?: string;
+      paramSpecs?: Array<{ key: string; required?: boolean }>;
+    };
+    try {
+      snap = JSON.parse(raw) as typeof snap;
+    } catch {
+      return;
+    }
+    const stepId = typeof snap.stepId === "string" ? snap.stepId.trim() : "";
+    if (!stepId) return;
+    const flow = parseCreationFlow(
+      this.blackboard.getContentByTag(CREATION_FLOW_TAG),
+    );
+    if (!flow) return;
+    const next = patchCreationFlowStepParams(flow, stepId, params);
+    const content = stringifyCreationFlow(next);
+    this.blackboard.write({
+      tag: CREATION_FLOW_TAG,
+      content,
+      source: "user",
+    });
+    const paramsMissing = (snap.paramSpecs ?? [])
+      .filter((p) => p.required)
+      .map((p) => p.key)
+      .filter((key) => {
+        const v = params[key];
+        return v == null || (typeof v === "string" && !v.trim());
+      });
+    const snapshot = {
+      ...snap,
+      params,
+      paramsMissing,
+    };
+    this.blackboard.write({
+      tag: CREATION_PROPOSED_STEP_TAG,
+      content: JSON.stringify(snapshot),
+      source: "runtime",
+    });
+    this.session = {
+      ...this.session,
+      slots: {
+        ...this.session.slots,
+        [CREATION_FLOW_TAG]: content,
+        [CREATION_PROPOSED_STEP_TAG]: JSON.stringify(snapshot),
+      },
+    };
+  }
+
+  private assertProposedStepReady(): string | null {
+    const decision = this.session.pendingDecision;
+    if (decision?.workerId !== "design-step") return null;
+    const raw =
+      (typeof this.session.slots[CREATION_PROPOSED_STEP_TAG] === "string"
+        ? String(this.session.slots[CREATION_PROPOSED_STEP_TAG])
+        : null) ||
+      this.blackboard.getContentByTag(CREATION_PROPOSED_STEP_TAG);
+    if (!raw?.trim()) return null;
+    try {
+      const snap = JSON.parse(raw) as { paramsMissing?: string[] };
+      if (snap.paramsMissing?.length) {
+        return `请先补齐必填参数：${snap.paramsMissing.join("、")}`;
+      }
+    } catch {
+      /* ignore */
+    }
+    return null;
+  }
+
  private getInstanceWorkerDeclaration() {
    return buildInstanceWorkerDeclaration(
      this.session,
@@ -840,24 +1145,50 @@ export class PhaseRuntime {

    if (result.askUser?.length && Object.keys(result.outputs).length === 0) {
      // 无产物：阻塞追问（信息不足，必须补）
+      if (await this.restoreRevisionTargetForReview(workerId, result.askUser)) {
+        return;
+      }
      await this.workerAsk(result.askUser);
      return;
    }
+    if (
+      workerId === "design-step" &&
+      Object.keys(result.outputs).length === 0
+    ) {
+      const retryQuestions = result.askUser?.length
+        ? result.askUser
+        : [
+            {
+              id: "design-step-retry",
+              prompt:
+                "这一步还没写出可验收的产物。请再补一点你最在意的体验或参与方式；也可以说「按已有描述先出一版」。",
+              allowOther: true,
+              required: true,
+            },
+          ];
+      if (await this.restoreRevisionTargetForReview(workerId, retryQuestions)) {
+        return;
+      }
+      await this.workerAsk(retryQuestions);
+      return;
+    }

    slots = { ...this.session.slots };
+    const writtenTags: string[] = [];
    for (const [tag, content] of Object.entries(result.outputs)) {
+      if (isProgressPointerTag(tag)) continue;
      const written = this.writeWorkerTagContent(tag, content, workerId);
      slots[tag] = written;
+      writtenTags.push(tag);
    }
    this.session = { ...this.session, slots };

    const artifact = createArtifact({
      workerId: effect.workerId,
      stepId: this.session.currentStepId,
-      outputTags:
-        Object.keys(result.outputs).length > 0
-          ? Object.keys(result.outputs)
-          : worker.outputTags,
+      outputTags: (writtenTags.length > 0 ? writtenTags : worker.outputTags).filter(
+        (t) => !isProgressPointerTag(t),
+      ),
      summary: result.summary,
    });
    this.session = { ...this.session, artifacts: [...this.session.artifacts, artifact] };
@@ -871,7 +1202,7 @@ export class PhaseRuntime {
      payload: {
        artifactId: artifact.id,
        // 有产物时 askUser 挂到验收态，不阻断 Accept
-        questions: result.askUser?.length ? result.askUser : undefined,
+        questions: this.freshSidecarQuestions(result.askUser),
        assessment: result.askAssessment?.trim() || undefined,
      },
    });
@@ -1054,6 +1385,22 @@ export class PhaseRuntime {
        }
      }
    }
+    if (tag === CREATION_FLOW_TAG) {
+      const merged = mergeCreationFlowPreservingAccepted({
+        prevRaw: this.blackboard.getContentByTag(tag),
+        nextRaw: toWrite,
+        acceptedStepIds: parseAcceptedUnits(
+          this.session.slots[SLOT_CREATION_ACCEPTED_UNITS] ??
+            this.blackboard.getContentByTag(CREATION_ACCEPTED_UNITS_TAG),
+        ),
+      });
+      toWrite = merged.raw;
+      if (merged.restored.length) {
+        this.onMessage(
+          `[流程编排] 已验收步骤按原样保留：${merged.restored.join("、")}`,
+        );
+      }
+    }
    this.blackboard.write({
      tag,
      content: toWrite,
@@ -1252,6 +1599,48 @@ export class PhaseRuntime {

    if (phase === "answered") return false;

+    // 配方开局模块 + 用户首句已在「用户.需求」→ 默认视为 opening 已做完，直接进 LLM
+    const demand = (
+      this.blackboard.getContentByTag("用户.需求") ??
+      this.session.slots["用户.需求"] ??
+      ""
+    )
+      .toString()
+      .trim();
+    if (
+      shouldSkipModuleOpening({
+        demand,
+        dependsOn: binding.step.depends_on,
+        phase,
+      })
+    ) {
+      state[stepKey] = "answered";
+      const raw = stringifyModuleOpeningState(state);
+      this.blackboard.write({
+        tag: CREATION_MODULE_OPENING_STATE_TAG,
+        content: raw,
+        source: "runtime",
+      });
+      this.blackboard.write({
+        tag: CREATION_MODULE_OPENING_TAG,
+        content: "（已跳过默认问题：用户首句见「用户.需求」）",
+        source: "runtime",
+      });
+      this.session = {
+        ...this.session,
+        slots: {
+          ...this.session.slots,
+          [SLOT_CREATION_MODULE_OPENING_STATE]: raw,
+          [CREATION_MODULE_OPENING_TAG]:
+            "（已跳过默认问题：用户首句见「用户.需求」）",
+        },
+      };
+      this.onMessage(
+        `[系统] ${binding.module.name} · 已坐在配方开局步，首句见「用户.需求」，跳过默认问题`,
+      );
+      return false;
+    }
+
    if (phase === "shown" && this.session.resumeContext) {
      // 用户刚答完默认问题 → 标记已答，继续走 LLM
      state[stepKey] = "answered";
@@ -1315,15 +1704,13 @@ export class PhaseRuntime {

    const questions = normalizeQuestions([
      {
-        id: "module-opening",
+        id: MODULE_OPENING_QUESTION_ID,
        prompt: binding.opening.trim(),
        allowOther: true,
        required: true,
      },
    ]);
-    this.onMessage(
-      `[Worker] ${binding.module.name} · 默认问题（程序）：\n${binding.opening.trim()}`,
-    );
+    // 正文由说话面 openingGuide 展示（对齐美学纲领）；workerAsk 只发短调度句
    await this.workerAsk(questions);
    return true;
  }
@@ -1356,6 +1743,8 @@ export class PhaseRuntime {

    if (!current) current = "flow";

+    debugLog("step", `钉住当前步骤 ${workerId} → ${current}`);
+
    const anchorAt = new Date().toISOString();
    this.blackboard.write({
      tag: CREATION_CURRENT_UNIT_TAG,
@@ -1386,17 +1775,29 @@ export class PhaseRuntime {
    const fromSlot = String(slots[SLOT_CREATION_CURRENT_UNIT] ?? "").trim();
    const fromStep = this.blackboard.getContentByTag(CREATION_CURRENT_STEP_TAG)?.trim();
    const fromSummary = artifact.summary?.match(/单位\s+([^\s·]+)/)?.[1]?.trim();
-    const unitId =
-      (artifact.workerId === "design-step" ? fromStep : "") ||
-      fromBoard ||
-      fromSlot ||
-      fromSummary ||
-      "";
+    // 接受工作流计划 ≠ 某一步做完；「flow」不得写入已验收步骤
+    if (artifact.workerId === "design-flow") return;
    const prev = parseAcceptedUnits(
      slots[SLOT_CREATION_ACCEPTED_UNITS] ??
        this.blackboard.getContentByTag(CREATION_ACCEPTED_UNITS_TAG),
    );
+    const unitId =
+      artifact.workerId === "design-step"
+        ? pickRecordedStepId({
+            flow: parseCreationFlow(
+              this.blackboard.getContentByTag(CREATION_FLOW_TAG),
+            ),
+            alreadyAccepted: prev,
+            pinnedUnitId: fromSlot || fromBoard,
+            writtenCurrentStep: fromStep,
+            summaryHint: fromSummary,
+          }) ?? ""
+        : fromBoard || fromSlot || fromSummary || "";
+    if (!unitId || unitId === "flow") return;
    const next = unitId && !prev.includes(unitId) ? [...prev, unitId] : prev;
+    if (next !== prev) {
+      debugLog("step", `验收步骤 ${unitId}  已验收 ${next.join("、")}`);
+    }
    slots[SLOT_CREATION_ACCEPTED_UNITS] = next;
    slots[SLOT_CREATION_CURRENT_UNIT] = unitId || undefined;
    this.blackboard.write({
@@ -1518,6 +1919,7 @@ export class PhaseRuntime {
      return;
    }

+    debugLog("step", "总管开始");
    const availableWorkers = await this.resolveAvailableWorkers();

    const decision = await this.mainAgent.runToolLoop(
@@ -1554,13 +1956,16 @@ export class PhaseRuntime {
      },
    );

+    debugLog(
+      "step",
+      `总管决策 ${labelAction(decision.action)}${decision.workerId ? ` ${decision.workerId}` : ""}`,
+    );
    this.onMessage(`[总管] ${decision.action}: ${decision.reason}`);

-    const result = applyEvent(this.session, {
+    const result = this.commitEvent({
      type: "main_agent_decision_created",
      payload: { decision },
    });
-    this.session = result.session;

    if (result.error) {
      this.onMessage(result.error);
--- a/src/server/agent-view.ts
+++ b/src/server/agent-view.ts
@@ -12,7 +12,12 @@ import {
  displayWorkerLabel,
  formatAgentDisplayTitle,
  formatWorkerDisplayTitle,
+  reviewComposerCopy,
 } from "./display-labels.js";
+import {
+  isModuleOpeningQuestions,
+  normalizeQuestions,
+} from "../skills/question-protocol.js";

 export type AgentMessageKind =
  | "user_input"
@@ -167,6 +172,20 @@ export function classifyAgentMessage(text: string): EnrichedMessage {
    };
  }

+  // 能力默认问题：正文在主栏说话面，调度消息只留短提示
+  const moduleOpeningMsg = trimmed.match(
+    /^\[Worker\]\s*(\S+)\s*·\s*默认问题/,
+  );
+  if (moduleOpeningMsg) {
+    return {
+      kind: "orchestrator_prompt",
+      actor: moduleOpeningMsg[1],
+      title: formatWorkerDisplayTitle(moduleOpeningMsg[1], "questions"),
+      body: "默认问题（请在主栏按引导作答）",
+      text: trimmed,
+    };
+  }
+
  const workerAskTagged = trimmed.match(
    /^\[Worker\]\s*(\S+)\s*提问[：:]\s*\n?([\s\S]*)$/,
  );
@@ -240,7 +259,7 @@ export function classifyAgentMessage(text: string): EnrichedMessage {
      title: "可选追问",
      body: formatWorkerQuestionBody(
        trimmed.replace(
-          /^\[Worker\]\s*可选追问（可跳过，直接接受(?:目前)?产物）[：:]\s*/,
+          /^\[Worker\]\s*可选追问（可跳过(?:,直接[^）]+)?）[：:]\s*/,
          "",
        ),
      ),
@@ -387,19 +406,52 @@ export function buildFocus(

  if (reason?.kind === "approve_step") {
    const worker = session.pendingDecision?.workerId ?? "skill";
+    const proposed =
+      typeof session.slots["创作.待确认步骤"] === "string"
+        ? (() => {
+            try {
+              return JSON.parse(String(session.slots["创作.待确认步骤"])) as {
+                name?: string;
+              };
+            } catch {
+              return null;
+            }
+          })()
+        : null;
    return {
      actorType: "orchestrator",
      actorId: "orchestrator",
-      actorLabel: "总管",
-      action: `建议调用 ${displayWorkerLabel(worker)}`,
+      actorLabel: "编排",
+      action: proposed?.name
+        ? `确认开始 · ${proposed.name}`
+        : `建议调用 ${displayWorkerLabel(worker)}`,
      detail: session.pendingDecision?.reason,
    };
  }

+  if (reason?.kind === "next_intent") {
+    return {
+      actorType: "user",
+      actorLabel: "你",
+      action: "下一步想写什么",
+      detail: "可留空；发送后会展示下一节点供确认开干",
+    };
+  }
+
  if (reason?.kind === "worker_questions") {
+    const qs = normalizeQuestions(reason.questions);
+    if (isModuleOpeningQuestions(qs)) {
+      return {
+        actorType: "user",
+        actorId: reason.workerId,
+        actorLabel: "你",
+        action: "按引导先说几句",
+        detail: (qs[0]?.prompt ?? "").slice(0, 200) || "想到什么写什么",
+      };
+    }
    const q =
-      reason.questions
-        ?.map((item) => item.prompt)
+      qs
+        .map((item) => item.prompt)
        .filter((s) => s?.trim())
        .join("；") ?? "";
    return {
@@ -413,13 +465,16 @@ export function buildFocus(

  if (reason?.kind === "review_artifact") {
    const art = session.artifacts.find((a) => a.id === session.pendingArtifactId);
+    const copy = reviewComposerCopy(art?.workerId, {
+      hasQuestions: Boolean(reason.questions?.length),
+    });
    const optionalQs = reason.questions?.length
      ? `；另有 ${reason.questions.length} 道可选追问`
      : "";
    return {
      actorType: "user",
      actorLabel: "你",
-      action: "验收产物",
+      action: copy.taskTitle,
      detail: `${art?.summary ?? displayWorkerLabel(art?.workerId) ?? ""}${optionalQs}`,
    };
  }
--- a/src/server/display-labels.ts
+++ b/src/server/display-labels.ts
@@ -122,3 +122,84 @@ export function formatAgentDisplayTitle(detail?: string): string {
  if (detail?.trim()) return `编排器 · ${detail.trim()}`;
  return "编排器";
 }
+
+/** 底栏双按钮 + 工作面 kicker：随 waiting 产物种类变化 */
+export type ReviewComposerCopy = {
+  /** 底栏短芯片 */
+  chip: string;
+  /** 工作面左上角 */
+  kicker: string;
+  taskLabel: string;
+  taskTitle: string;
+  hint: string;
+  submitLabel: string;
+  acceptLabel: string;
+  placeholder: string;
+  emptyEnterHint: string;
+  /** plan＝编排确认（主色）；accept＝产物验收（绿） */
+  tone: "plan" | "accept";
+  kind: "flow" | "opening" | "accept";
+};
+
+/**
+ * `review_artifact` 的用户文案。流程编排核对的是工作流计划，不是「验收通过」。
+ */
+export function reviewComposerCopy(
+  workerId: string | undefined | null,
+  opts?: { hasQuestions?: boolean },
+): ReviewComposerCopy {
+  const hasQuestions = Boolean(opts?.hasQuestions);
+  const id = (workerId ?? "").trim();
+
+  if (id === "design-flow") {
+    return {
+      chip: "编排",
+      kicker: "核对编排",
+      taskLabel: "编排",
+      taskTitle: "核对工作流计划",
+      hint: hasQuestions
+        ? "同意就按这个排；要改就写意见，在现有计划上改。下方追问可选答。"
+        : "同意就按这个排；要改就写意见，在现有计划上改。",
+      submitLabel: "按意见改编排",
+      acceptLabel: "确认编排",
+      placeholder: "改编排意见…",
+      emptyEnterHint: "空 Enter＝确认编排",
+      tone: "plan",
+      kind: "flow",
+    };
+  }
+
+  if (id === "opening-generator") {
+    return {
+      chip: "开场",
+      kicker: "待选定",
+      taskLabel: "开场",
+      taskTitle: "选定开场白",
+      hint: hasQuestions
+        ? "同意就选定此开场；要改就写意见。下方追问可选答。"
+        : "同意就选定此开场；要改就写意见。",
+      submitLabel: "按意见修改",
+      acceptLabel: "选定此开场",
+      placeholder: "修改意见…",
+      emptyEnterHint: "空 Enter＝选定此开场",
+      tone: "plan",
+      kind: "opening",
+    };
+  }
+
+  return {
+    chip: "验收",
+    kicker: "待验收",
+    taskLabel: "验收",
+    taskTitle: "验收产物",
+    hint: hasQuestions
+      ? "同意就接受；要改就写意见，在现有产物上改。下方追问可选答。"
+      : "同意就接受；要改就写意见，在现有产物上改。",
+    submitLabel: "按意见修改",
+    acceptLabel: "接受",
+    placeholder: "修改意见…",
+    emptyEnterHint: "空 Enter＝接受",
+    tone: "accept",
+    kind: "accept",
+  };
+}
--- a/src/server/session-manager.ts
+++ b/src/server/session-manager.ts
@@ -63,7 +63,12 @@ import type { BlackboardItem } from "../types/blackboard.js";
 import { Blackboard } from "../blackboard/blackboard.js";
 import { parseWorkerSetYaml } from "../skills/worker-set-parse.js";
 import { parseShellAdaptationFromReplyFormat } from "../skills/present-packet.js";
-import { parseAcceptedUnits, isDesignDiskWorker } from "../skills/creation-units.js";
+import {
+  CREATION_ACCEPTED_UNITS_TAG,
+  isDesignDiskWorker,
+  parseAcceptedUnits,
+  SLOT_CREATION_ACCEPTED_UNITS,
+} from "../skills/creation-units.js";
 import {
  formatWorkerSetForUser,
  type WorkerSetUserView,
@@ -78,13 +83,21 @@ import {
  type ContextOrderEdit,
 } from "../skills/context-order.js";
 import {
+  CREATION_CURRENT_STEP_TAG,
  CREATION_FLOW_TAG,
+  CREATION_PROPOSED_STEP_TAG,
  CREATION_SELECTED_RECIPE_TAG,
+  creationFlowFromRecipeSeed,
  findRecipeCatalogEntry,
  formatCreationFlowForUser,
+  isProgressPointerTag,
  loadRecipeCatalog,
+  loadRecipeDetail,
  parseCreationFlow,
  parseSelectedRecipeRef,
+  resolveDesignStepBinding,
+  stepUnitId,
+  stringifyCreationFlow,
  type CreationFlowUserView,
  type RecipeCatalogEntry,
 } from "../skills/creation-flow.js";
@@ -94,6 +107,7 @@ import { buildBoardPanel } from "../runtime/compress-after-worker.js";
 import {
  displayWorkerLabel,
  formatWorkerDisplayTitle,
+  reviewComposerCopy,
 } from "./display-labels.js";
 import {
  buildCreationDialogueTranscript,
@@ -102,6 +116,7 @@ import {
 import {
  formatQuestionAnswersForAi,
  formatQuestionAnswersForDisplay,
+  isModuleOpeningQuestions,
  normalizeQuestions,
 } from "../skills/question-protocol.js";
 import type { QuestionAnswer } from "../types/questions.js";
@@ -180,10 +195,16 @@ export type SessionView = {
  workerSetView?: WorkerSetUserView;
  /** 创作流程用户视图（步骤顺序 + 依赖） */
  creationFlowView?: CreationFlowUserView;
+  /** 确认开干：下一步提案（含可编辑编排参数） */
+  proposedNextStep?: ProposedNextStepView;
  /** 能力包内可选初始配方（用户手动选） */
  recipes?: Array<{ id: string; name: string; declaration: string }>;
  /** 用户已选初始配方 */
  selectedRecipe?: { id: string; name: string; declaration: string } | null;
+  /**
+   * 开局引导：配方起点或当前能力默认问题的 opening（说话面主柱；对齐美学纲领）。
+   */
+  openingGuide?: { stepId: string; stepName: string; text: string } | null;
  /** 黑板与定稿上下文（强可读） */
  boardPanel?: BoardPanelView;
  /** 游玩呈现壳微调（来自设计.正文组成） */
@@ -238,6 +259,21 @@ export type BoardPanelView = {
  brief?: string;
 };

+export type ProposedNextStepView = {
+  stepId: string;
+  name: string;
+  declaration?: string;
+  params: Record<string, unknown>;
+  paramsMissing: string[];
+  paramSpecs: Array<{
+    key: string;
+    label: string;
+    required: boolean;
+    hint?: string;
+  }>;
+  intent?: string;
+};
+
 export type ReviewArtifactView = {
  id: string;
  workerId: string;
@@ -263,14 +299,18 @@ type ManagedSession = {
  branchState: MessageBranchState;
  /** 当前能力包可选初始配方（用户手动选） */
  recipeOptions?: RecipeCatalogEntry[];
+  /** 开局模块 opening，缓存给意图页 */
+  openingGuide?: { stepId: string; stepName: string; text: string } | null;
 };

 export class SessionManager {
  private readonly sessions = new Map<string, ManagedSession>();
  /** bookId → 当前内存中的 sessionId */
  private readonly activeBookSessions = new Map<string, string>();
-  /** sessionId → 流式 thinking 缓冲 */
+  /** sessionId → 流式 thinking 缓冲（思考结束后仍保留，直到本轮结束） */
  private readonly agentThinkingLive = new Map<string, string>();
+  /** 本轮思维链已收束；下一次 delta 起新一轮 */
+  private readonly agentThinkingClosed = new Set<string>();
  /** sessionId → worker 流式缓冲 */
  private readonly workerLive = new Map<
    string,
@@ -679,7 +719,11 @@ export class SessionManager {
    return count;
  }

-  async sendMessage(id: string, text: string): Promise<SessionView> {
+  async sendMessage(
+    id: string,
+    text: string,
+    opts?: { answers?: QuestionAnswer[] },
+  ): Promise<SessionView> {
    const s = this.require(id);
    this.clearAgentThinking(id);
    recordPreMessageCheckpoint(
@@ -687,21 +731,62 @@ export class SessionManager {
      s.messages.length,
      this.captureCheckpoint(s),
    );
-    s.messages.push(this.msg("user", text));
-    this.syncCreationDialogue(s);
-    if (s.bookId) this.syncBookPreview(s.bookId, s.messages);
    const reason = s.runtime.getSession().waitingReason;
+    const trimmed = text.trim();
+    const sidecarAnswers =
+      reason?.kind === "review_artifact" &&
+      reason.questions?.length &&
+      opts?.answers?.length
+        ? opts.answers.filter((a) => a.text?.trim() && a.text !== "（未答）")
+        : [];
+
+    // next_intent 允许留空：不塞空白气泡
+    if (trimmed || reason?.kind !== "next_intent" || sidecarAnswers.length) {
+      let recordText = trimmed || text;
+      let displayBody: string | undefined;
+      if (sidecarAnswers.length && reason?.kind === "review_artifact") {
+        const qs = normalizeQuestions(reason.questions ?? []);
+        recordText = formatQuestionAnswersForAi(
+          qs,
+          sidecarAnswers,
+          trimmed || undefined,
+        );
+        displayBody = formatQuestionAnswersForDisplay(
+          qs,
+          sidecarAnswers,
+          trimmed || undefined,
+        );
+      }
+      const userMsg = this.msg("user", recordText);
+      if (displayBody) userMsg.body = displayBody;
+      s.messages.push(userMsg);
+      this.syncCreationDialogue(s);
+      if (s.bookId) this.syncBookPreview(s.bookId, s.messages);
+    }
    try {
      if (reason?.kind === "approve_step") {
-        await s.runtime.rejectStep(text);
+        await s.runtime.rejectStep(trimmed || text);
      } else if (reason?.kind === "review_artifact") {
-        await s.runtime.rejectArtifact(text);
+        // 先把卡上追问（问+选项+补充）写入「用户.worker答复」，再按意见改产物
+        if (sidecarAnswers.length && reason.questions?.length) {
+          const qaText = formatQuestionAnswersForAi(
+            normalizeQuestions(reason.questions),
+            sidecarAnswers,
+            trimmed || undefined,
+          );
+          await s.runtime.resolveSidecarQuestions(qaText);
+        }
+        if (trimmed || sidecarAnswers.length) {
+          await s.runtime.rejectArtifact(
+            trimmed || "按追问作答更新产物",
+          );
+        }
      } else {
-        await s.runtime.submitInput(text);
+        await s.runtime.submitInput(trimmed || text);
      }
    } catch (err) {
      const detail = err instanceof Error ? err.message : String(err);
-      console.error("[session] sendMessage failed:", detail);
+      console.error("[会话] 发送消息失败", detail);
      s.messages.push(this.msg("system", formatRuntimeError(detail)));
      if (s.bookId) this.syncBookPreview(s.bookId, s.messages);
    } finally {
@@ -755,12 +840,14 @@ export class SessionManager {
    try {
      if (isSidecar) {
        await s.runtime.resolveSidecarQuestions(aiText);
+        // 作答必须写回产物，否则验收后进度空转、追问在美学/交互间打转
+        await s.runtime.rejectArtifact("按追问作答更新产物");
      } else {
        await s.runtime.submitInput(aiText);
      }
    } catch (err) {
      const detail = err instanceof Error ? err.message : String(err);
-      console.error("[session] answerQuestions failed:", detail);
+      console.error("[会话] 回答追问失败", detail);
      s.messages.push(this.msg("system", formatRuntimeError(detail)));
      if (s.bookId) this.syncBookPreview(s.bookId, s.messages);
    } finally {
@@ -960,13 +1047,16 @@ export class SessionManager {
    return this.toView(id);
  }

-  async approve(id: string): Promise<SessionView> {
+  async approve(
+    id: string,
+    stepParams?: Record<string, unknown>,
+  ): Promise<SessionView> {
    const s = this.require(id);
    try {
-      await s.runtime.approve();
+      await s.runtime.approve(stepParams);
    } catch (err) {
      const detail = err instanceof Error ? err.message : String(err);
-      console.error("[session] approve failed:", detail);
+      console.error("[会话] 确认开干失败", detail);
      s.messages.push(this.msg("system", formatRuntimeError(detail)));
    }
    return this.toView(id);
@@ -978,7 +1068,7 @@ export class SessionManager {
      await s.runtime.confirmIntake();
    } catch (err) {
      const detail = err instanceof Error ? err.message : String(err);
-      console.error("[session] confirmIntake failed:", detail);
+      console.error("[会话] 确认需求失败", detail);
      s.messages.push(this.msg("system", formatRuntimeError(detail)));
      if (s.bookId) this.syncBookPreview(s.bookId, s.messages);
    }
@@ -1028,7 +1118,7 @@ export class SessionManager {
      }
    } catch (err) {
      const detail = err instanceof Error ? err.message : String(err);
-      console.error("[session] reject failed:", detail);
+      console.error("[会话] 打回失败", detail);
      s.messages.push(this.msg("system", formatRuntimeError(detail)));
      if (s.bookId) this.syncBookPreview(s.bookId, s.messages);
    }
@@ -1154,7 +1244,7 @@ export class SessionManager {
        payload: { text },
      });
    } catch (err) {
-      console.error("[session] maybeKickAgentAfterDemand failed:", err);
+      console.error("[会话] 需求后唤醒总管失败", err);
    }
  }

@@ -1214,7 +1304,7 @@ export class SessionManager {
      const transcript = buildCreationDialogueTranscript(s.messages);
      s.runtime.writeCreationDialogue(transcript);
    } catch (err) {
-      console.warn("[session] syncCreationDialogue:", err);
+      console.warn("[会话] 同步创作对话失败", err);
    }
  }

@@ -1274,6 +1364,7 @@ export class SessionManager {

  private clearAgentThinking(sessionId: string): void {
    this.agentThinkingLive.delete(sessionId);
+    this.agentThinkingClosed.delete(sessionId);
    this.workerLive.delete(sessionId);
  }

@@ -1287,15 +1378,20 @@ export class SessionManager {
    | "onWorkerStreamStart"
    | "onWorkerThinkingDelta"
    | "onWorkerOutputDelta"
-    | "onWorkerStreamDone"
  > {
    return {
      onAgentThinkingDelta: (delta) => {
-        const prev = this.agentThinkingLive.get(sessionId) ?? "";
+        const prev = this.agentThinkingClosed.has(sessionId)
+          ? ""
+          : (this.agentThinkingLive.get(sessionId) ?? "");
+        this.agentThinkingClosed.delete(sessionId);
        this.agentThinkingLive.set(sessionId, prev + delta);
      },
-      onAgentThinkingDone: () => {
-        this.agentThinkingLive.delete(sessionId);
+      onAgentThinkingDone: (text) => {
+        // 正文仍等完整后再渲染；思考收束后继续挂在 liveStream，避免中间只剩「…」
+        this.agentThinkingClosed.add(sessionId);
+        const done = text.trim();
+        if (done) this.agentThinkingLive.set(sessionId, done);
      },
      onWorkerStreamStart: (workerId) => {
        this.workerLive.set(sessionId, { workerId, thinking: "", output: "" });
@@ -1322,9 +1418,6 @@ export class SessionManager {
        prev.output += delta;
        this.workerLive.set(sessionId, prev);
      },
-      onWorkerStreamDone: () => {
-        this.workerLive.delete(sessionId);
-      },
    };
  }

@@ -1603,47 +1696,94 @@ export class SessionManager {
      }
    } else if (reason?.kind === "worker_questions") {
      const qs = normalizeQuestions(reason.questions);
-      if (qs.length) {
+      if (isModuleOpeningQuestions(qs)) {
+        // 能力默认问题：对齐美学纲领 → 自由书写，不是可跳过追问卡
+        hints.push("按主栏引导先说几句（想到什么写什么，不必整齐）");
+        actions.push({
+          type: "send_message",
+          label: "发送",
+          placeholder: "想到什么写什么…",
+        });
+      } else if (qs.length) {
        hints.push(
          `${displayWorkerLabel(reason.workerId)} · 提问（请在询问卡作答）：${qs.map((q) => q.prompt).join(" ")}`,
        );
+        actions.push({ type: "answer_questions", label: "提交作答" });
+        actions.push({ type: "skip_questions", label: "跳过追问" });
+        actions.push({
+          type: "send_message",
+          label: "自由补充",
+          placeholder: "也可在此自由补充…",
+        });
      } else {
        hints.push(
          `${displayWorkerLabel(reason.workerId)} 需要更多信息，请补充说明`,
        );
+        actions.push({
+          type: "send_message",
+          label: "发送",
+          placeholder: "补充说明…",
+        });
      }
-      actions.push({ type: "answer_questions", label: "提交作答" });
-      actions.push({ type: "skip_questions", label: "跳过追问" });
-      actions.push({
-        type: "send_message",
-        label: "自由补充",
-        placeholder: "也可在此自由补充…",
-      });
    } else if (reason?.kind === "approve_step") {
-      actions.push({ type: "approve", label: "确认执行" });
+      const proposed = readProposedNextStep(
+        s.runtime.getSession().slots,
+        s.runtime.getBlackboard(),
+      );
+      if (proposed) {
+        hints.push(
+          `确认开始「${proposed.name}」${
+            proposed.paramsMissing.length
+              ? `（请先补齐：${proposed.paramsMissing.join("、")}）`
+              : ""
+          }`,
+        );
+        actions.push({
+          type: "approve",
+          label: proposed.paramsMissing.length ? "补参后开始" : "同意并开始",
+        });
+      } else {
+        actions.push({ type: "approve", label: "确认执行" });
+      }
      actions.push({ type: "reject", label: "暂不执行" });
-    } else if (reason?.kind === "review_artifact") {
-      actions.push({ type: "accept", label: "接受目前产物" });
-      actions.push({ type: "reject", label: "不接受，重新来" });
      actions.push({
        type: "send_message",
-        label: "发送修改意见",
-        placeholder: "输入修改意见后发送…",
+        label: "说明意见",
+        placeholder: "不同意可说明要改成什么…",
+      });
+    } else if (reason?.kind === "next_intent") {
+      hints.push("下一步想写什么？可留空，发送后会展示下一节点供你确认。");
+      actions.push({
+        type: "send_message",
+        label: "继续",
+        placeholder: "下一步想写什么？（可留空）…",
+      });
+    } else if (reason?.kind === "review_artifact") {
+      const art = session.artifacts.find((a) => a.id === session.pendingArtifactId);
+      const copy = reviewComposerCopy(art?.workerId, {
+        hasQuestions: Boolean(reason.questions?.length),
+      });
+      actions.push({ type: "accept", label: copy.acceptLabel });
+      actions.push({ type: "reject", label: "不接受" });
+      actions.push({
+        type: "send_message",
+        label: copy.submitLabel,
+        placeholder: `${copy.placeholder}；在现有产物上改，不整份重做…`,
      });
      if (reason.questions?.length) {
-        hints.push("下方追问可选；接受产物即收起（表示无需再完善）");
+        hints.push(`下方追问可选；${copy.acceptLabel}即收起（表示无需再完善）`);
        actions.push({ type: "answer_questions", label: "提交作答" });
        actions.push({ type: "skip_questions", label: "跳过追问" });
      }
    } else if (reason?.kind === "revision") {
      hints.push(
        reason.instruction?.trim() ||
-          "需要按反馈修改后继续；请在下方说明或直接发送继续",
+          "请说明要改哪里；发送后在现有产物上修改",
      );
      actions.push({
        type: "send_message",
-        label: "发送",
-        placeholder: "说明修改意见，或直接发送让总管继续…",
+        label: "按意见修改",
+        placeholder: "说明要改哪里…",
      });
    } else if (session.phase === "error") {
      hints.push("上一轮执行出错。可直接发消息让总管重试，或刷新作品后继续。");
@@ -1737,7 +1877,17 @@ export class SessionManager {
        s.runtime.getBlackboard(),
        s.runtime.getSession(),
      ),
-      creationFlowView: buildCreationFlowUserView(s.runtime.getBlackboard()),
+      creationFlowView: buildCreationFlowUserView(
+        s.runtime.getBlackboard(),
+        session.slots[SLOT_CREATION_ACCEPTED_UNITS],
+      ),
+      proposedNextStep:
+        reason?.kind === "approve_step"
+          ? readProposedNextStep(
+              session.slots,
+              s.runtime.getBlackboard(),
+            )
+          : undefined,
      recipes: s.recipeOptions?.map((r) => ({
        id: r.id,
        name: r.name,
@@ -1747,6 +1897,7 @@ export class SessionManager {
        s.runtime.getBlackboard(),
        s.recipeOptions,
      ),
+      openingGuide: this.resolveOpeningGuideForView(s, reason),
      boardPanel: buildBoardPanelFromRuntime(s.runtime.getBlackboard()),
      presentationTweaks: shellAdapt
        ? {
@@ -1839,12 +1990,14 @@ export class SessionManager {
    const skillName = s.runtime.getActiveSkill()?.name;
    if (!skillName) {
      s.recipeOptions = [];
+      s.openingGuide = null;
      return;
    }
    try {
      const skill = await loadSkill(skillName);
      if (!skill.skillPackRoot) {
        s.recipeOptions = [];
+        s.openingGuide = null;
        return;
      }
      const catalog = await loadRecipeCatalog(skill.skillPackRoot);
@@ -1852,6 +2005,125 @@ export class SessionManager {
    } catch {
      s.recipeOptions = [];
    }
+    await this.ensureRecipeSeededFlow(s);
+    await this.refreshOpeningGuide(s);
+  }
+
+  /** 已选配方但尚无流程 → 写入配方近期起点（兼容旧会话） */
+  private async ensureRecipeSeededFlow(s: ManagedSession): Promise<void> {
+    const board = s.runtime.getBlackboard();
+    if (board.getContentByTag(CREATION_FLOW_TAG)?.trim()) return;
+    const ref = parseSelectedRecipeRef(
+      board.getContentByTag(CREATION_SELECTED_RECIPE_TAG),
+    );
+    if (!ref) return;
+    const entry = findRecipeCatalogEntry(
+      s.recipeOptions?.length ? { recipes: s.recipeOptions } : null,
+      ref,
+    );
+    if (!entry) return;
+    const skillName = s.runtime.getActiveSkill()?.name;
+    if (!skillName) return;
+    try {
+      const skill = await loadSkill(skillName);
+      const packRoot = skill.skillPackRoot?.trim();
+      if (!packRoot) return;
+      const detail = await loadRecipeDetail(packRoot, entry);
+      const seeded = creationFlowFromRecipeSeed(detail);
+      if (!seeded) return;
+      board.write({
+        tag: CREATION_FLOW_TAG,
+        content: stringifyCreationFlow(seeded),
+        source: "runtime",
+      });
+    } catch (err) {
+      console.error("[会话] 配方预置失败", err);
+    }
+  }
+
+  /**
+   * 下一待做开局步（无 depends_on）的 opening → 意图页引导。
+   * 有起点模块就展示；无则 null（再走 design-flow 生成）。
+   * 中段步骤的默认问题改由 resolveOpeningGuideForView 在 waiting 时覆盖。
+   */
+  private async refreshOpeningGuide(s: ManagedSession): Promise<void> {
+    const skillName = s.runtime.getActiveSkill()?.name;
+    if (!skillName) {
+      s.openingGuide = null;
+      return;
+    }
+    try {
+      const skill = await loadSkill(skillName);
+      const packRoot = skill.skillPackRoot?.trim();
+      if (!packRoot) {
+        s.openingGuide = null;
+        return;
+      }
+      const board = s.runtime.getBlackboard();
+      const session = s.runtime.getSession();
+      const accepted = parseAcceptedUnits(
+        session.slots[SLOT_CREATION_ACCEPTED_UNITS] ??
+          board.getContentByTag(CREATION_ACCEPTED_UNITS_TAG),
+      );
+      const binding = await resolveDesignStepBinding({
+        skillPackRoot: packRoot,
+        flowRaw: board.getContentByTag(CREATION_FLOW_TAG),
+        acceptedStepNames: accepted,
+      });
+      if (
+        !binding?.opening?.trim() ||
+        (binding.step.depends_on?.length ?? 0) > 0
+      ) {
+        s.openingGuide = null;
+        return;
+      }
+      s.openingGuide = {
+        stepId: binding.step.id,
+        stepName: binding.module.name || binding.step.name,
+        text: binding.opening.trim(),
+      };
+    } catch {
+      s.openingGuide = null;
+    }
+  }
+
+  /**
+   * 视图用开场引导：等待能力默认问题时用题干（对齐美学纲领说话面）；
+   * 中段步骤（有 depends_on）也会覆盖，避免只剩空答题壳。
+   */
+  private resolveOpeningGuideForView(
+    s: ManagedSession,
+    reason: WaitingReason | undefined,
+  ): { stepId: string; stepName: string; text: string } | null {
+    if (reason?.kind === "worker_questions") {
+      const qs = normalizeQuestions(reason.questions);
+      if (isModuleOpeningQuestions(qs)) {
+        const text = qs[0]?.prompt?.trim() ?? "";
+        if (text) {
+          const board = s.runtime.getBlackboard();
+          const stepKey =
+            board.getContentByTag(CREATION_CURRENT_STEP_TAG)?.trim() || "";
+          const flow = parseCreationFlow(board.getContentByTag(CREATION_FLOW_TAG));
+          const step = flow?.steps?.find(
+            (st) =>
+              stepUnitId(st) === stepKey ||
+              st.id === stepKey ||
+              st.name === stepKey,
+          );
+          const stepName =
+            step?.name?.trim() ||
+            s.openingGuide?.stepName?.trim() ||
+            stepKey ||
+            "按引导先说几句";
+          return {
+            stepId: qs[0]!.id,
+            stepName,
+            text,
+          };
+        }
+      }
+    }
+    return s.openingGuide ?? null;
  }

  private async applySelectedRecipe(
@@ -1866,7 +2138,8 @@ export class SessionManager {
    if (!entry) {
      throw new Error(`未知初始配方：${recipeId}`);
    }
-    s.runtime.getBlackboard().write({
+    const board = s.runtime.getBlackboard();
+    board.write({
      tag: CREATION_SELECTED_RECIPE_TAG,
      content: JSON.stringify({
        id: entry.id,
@@ -1874,6 +2147,33 @@ export class SessionManager {
      }),
      source: "user",
    });
+
+    // 剧本已钉死开局模块 → 默认写入近期起点，开局即坐在 DAG 第一步（不必先跑 design-flow）
+    const existingFlow = board.getContentByTag(CREATION_FLOW_TAG)?.trim();
+    if (!existingFlow) {
+      const skillName = s.runtime.getActiveSkill()?.name;
+      if (skillName) {
+        try {
+          const skill = await loadSkill(skillName);
+          const packRoot = skill.skillPackRoot?.trim();
+          if (packRoot) {
+            const detail = await loadRecipeDetail(packRoot, entry);
+            const seeded = creationFlowFromRecipeSeed(detail);
+            if (seeded) {
+              board.write({
+                tag: CREATION_FLOW_TAG,
+                content: stringifyCreationFlow(seeded),
+                source: "runtime",
+              });
+            }
+          }
+        } catch (err) {
+          console.error("[会话] 从配方写入创作流程失败", err);
+        }
+      }
+    }
+
+    await this.refreshOpeningGuide(s);
  }
 }

@@ -1893,6 +2193,7 @@ function collectFilledTags(blackboard: Blackboard): string[] {

 function buildCreationFlowUserView(
  blackboard: Blackboard,
+  acceptedRaw?: unknown,
 ): CreationFlowUserView | undefined {
  const raw = blackboard.getContentByTag(CREATION_FLOW_TAG)?.trim();
  if (!raw) return undefined;
@@ -1900,7 +2201,56 @@ function buildCreationFlowUserView(
  if (!flow) {
    return { steps: [], parseError: "创作流程无法解析为 JSON（需要 steps 数组）" };
  }
-  return formatCreationFlowForUser(flow);
+  const accepted = parseAcceptedUnits(
+    acceptedRaw ?? blackboard.getContentByTag(CREATION_ACCEPTED_UNITS_TAG),
+  );
+  return formatCreationFlowForUser(flow, undefined, accepted);
+}
+
+function readProposedNextStep(
+  slots: Record<string, unknown>,
+  blackboard: Blackboard,
+): ProposedNextStepView | undefined {
+  const raw =
+    (typeof slots[CREATION_PROPOSED_STEP_TAG] === "string"
+      ? String(slots[CREATION_PROPOSED_STEP_TAG])
+      : null) || blackboard.getContentByTag(CREATION_PROPOSED_STEP_TAG);
+  if (!raw?.trim()) return undefined;
+  try {
+    const doc = JSON.parse(raw) as Partial<ProposedNextStepView>;
+    if (typeof doc.stepId !== "string" || typeof doc.name !== "string") {
+      return undefined;
+    }
+    return {
+      stepId: doc.stepId,
+      name: doc.name,
+      declaration:
+        typeof doc.declaration === "string" ? doc.declaration : undefined,
+      params:
+        doc.params && typeof doc.params === "object" && !Array.isArray(doc.params)
+          ? (doc.params as Record<string, unknown>)
+          : {},
+      paramsMissing: Array.isArray(doc.paramsMissing)
+        ? doc.paramsMissing.map(String)
+        : [],
+      paramSpecs: Array.isArray(doc.paramSpecs)
+        ? doc.paramSpecs
+            .map((p) => ({
+              key: String((p as { key?: string }).key ?? ""),
+              label: String((p as { label?: string }).label ?? ""),
+              required: Boolean((p as { required?: boolean }).required),
+              hint:
+                typeof (p as { hint?: string }).hint === "string"
+                  ? (p as { hint: string }).hint
+                  : undefined,
+            }))
+            .filter((p) => p.key)
+        : [],
+      intent: typeof doc.intent === "string" ? doc.intent : undefined,
+    };
+  } catch {
+    return undefined;
+  }
 }

 function resolveSelectedRecipeView(
@@ -1972,19 +2322,46 @@ function buildReviewArtifactView(

  for (const tag of art.outputTags) {
    if (workerSetTags.includes(tag)) continue;
+    if (isProgressPointerTag(tag)) continue;
    const content = blackboard.getContentByTag(tag);
    if (content?.trim()) {
      tagBodies.push(`## ${tag}\n\n${content.trim()}`);
    }
  }

+  // 回退消息里也可能夹带「创作.当前步骤」段落，验收正文里去掉
+  const stripPointerSections = (text: string): string => {
+    const trimmed = text.trim();
+    if (!trimmed) return "";
+    const re = /^##\s+(.+?)\s*$/gm;
+    const hits: Array<{ title: string; index: number; headerEnd: number }> = [];
+    let m: RegExpExecArray | null;
+    while ((m = re.exec(trimmed))) {
+      hits.push({
+        title: m[1]!.trim(),
+        index: m.index,
+        headerEnd: m.index + m[0].length,
+      });
+    }
+    if (!hits.length) return trimmed;
+    return hits
+      .filter((h) => !isProgressPointerTag(h.title) && !/当前步骤|当前单位/.test(h.title))
+      .map((h, i) => {
+        const end = i + 1 < hits.length ? hits[i + 1]!.index : trimmed.length;
+        const chunk = trimmed.slice(h.headerEnd, end).trim();
+        return chunk ? `## ${h.title}\n\n${chunk}` : "";
+      })
+      .filter(Boolean)
+      .join("\n\n");
+  };
+
  const outputMsg = [...messages]
    .reverse()
    .find((m) => m.kind === "worker_output" && m.actor === art.workerId);

  const body =
    tagBodies.join("\n\n") ||
-    (outputMsg?.body ?? outputMsg?.text ?? "").trim() ||
+    stripPointerSections((outputMsg?.body ?? outputMsg?.text ?? "").trim()) ||
    art.summary?.trim() ||
    "（Worker 未写入可读正文，请检查黑板 tag 或重试创作 skill）";

@@ -2028,7 +2405,10 @@ function buildReviewArtifactView(
      : undefined;

  const creationFlowView = isFlowArt
-    ? buildCreationFlowUserView(blackboard)
+    ? buildCreationFlowUserView(
+        blackboard,
+        session.slots[SLOT_CREATION_ACCEPTED_UNITS],
+      )
    : undefined;

  return {
--- a/src/server/web-server.ts
+++ b/src/server/web-server.ts
@@ -7,9 +7,11 @@ import { sessionManager } from "./session-manager.js";
 import { handleSettingsApi } from "./settings-handlers.js";
 import { handleBooksApi } from "./book-handlers.js";
 import { handleStatsApi } from "./stats-handlers.js";
+import { ensureConsoleUtf8, getRuntimeLogPath } from "../log.js";

 const __dirname = path.dirname(fileURLToPath(import.meta.url));
 const PROJECT_ROOT = path.resolve(__dirname, "../..");
+ensureConsoleUtf8();
 loadDotEnv(PROJECT_ROOT);
 const WEB_ROOT = path.resolve(__dirname, "../../web");
 const PORT = Number(process.env.PORT ?? 23337);
@@ -122,12 +124,45 @@ const server = createServer(async (req, res) => {
      }

      if (req.method === "POST" && sub === "/messages") {
-        const body = JSON.parse(await readBody(req)) as { text?: string };
-        if (!body.text?.trim()) {
-          json(res, 400, { error: "text 不能为空" });
-          return;
+        const body = JSON.parse(await readBody(req)) as {
+          text?: string;
+          answers?: Array<{
+            questionId?: string;
+            optionId?: string;
+            text?: string;
+          }>;
+        };
+        const text = typeof body.text === "string" ? body.text : "";
+        const trimmed = text.trim();
+        const answers = (body.answers ?? [])
+          .filter(
+            (a) =>
+              typeof a?.questionId === "string" &&
+              a.questionId.trim() &&
+              typeof a?.text === "string" &&
+              a.text.trim(),
+          )
+          .map((a) => ({
+            questionId: a.questionId!.trim(),
+            optionId:
+              typeof a.optionId === "string" && a.optionId.trim()
+                ? a.optionId.trim()
+                : undefined,
+            text: a.text!.trim(),
+          }));
+        // next_intent 明确可留空；其余态仍要求有字（除非顺带提交了追问）
+        if (!trimmed && !answers.length) {
+          const current = sessionManager.get(sessionId);
+          if (current?.waitingReason?.kind !== "next_intent") {
+            json(res, 400, { error: "text 不能为空" });
+            return;
+          }
        }
-        const view = await sessionManager.sendMessage(sessionId, body.text.trim());
+        const view = await sessionManager.sendMessage(
+          sessionId,
+          trimmed,
+          answers.length ? { answers } : undefined,
+        );
        json(res, 200, view);
        return;
      }
@@ -406,11 +441,19 @@ const server = createServer(async (req, res) => {
      }

      if (req.method === "POST" && sub === "/actions") {
-        const body = JSON.parse(await readBody(req)) as { action?: string };
+        const body = JSON.parse(await readBody(req)) as {
+          action?: string;
+          stepParams?: Record<string, unknown>;
+        };
        let view;
        switch (body.action) {
          case "approve":
-            view = await sessionManager.approve(sessionId);
+            view = await sessionManager.approve(
+              sessionId,
+              body.stepParams && typeof body.stepParams === "object"
+                ? body.stepParams
+                : undefined,
+            );
            break;
          case "confirm_intake":
            view = await sessionManager.confirmIntake(sessionId);
@@ -452,7 +495,7 @@ const server = createServer(async (req, res) => {
      json(res, 404, { error: "Not found" });
    }
  } catch (err) {
-    console.error("[api]", req.method, url.pathname, err);
+    console.error("[接口]", req.method, url.pathname, err);
    json(res, 500, {
      error: err instanceof Error ? err.message : "服务器错误",
    });
@@ -460,5 +503,6 @@ const server = createServer(async (req, res) => {
 });

 server.listen(PORT, () => {
-  console.log(`Writing Agent 对话页: http://localhost:${PORT}`);
+  console.log(`对话页  http://localhost:${PORT}`);
+  console.log(`运行日志（UTF-8）  ${getRuntimeLogPath()}`);
 });
--- a/src/skills/context-fragment.ts
+++ b/src/skills/context-fragment.ts
@@ -115,6 +115,46 @@ export function isContextFragmentDoc(doc: unknown): boolean {
  return Boolean(parseContextFragment(doc));
 }

+const NON_FRAGMENT_DESIGN_TAGS = new Set([
+  "设计.创作流程",
+  "设计.worker集",
+  "设计.worker集.草稿",
+  "设计.worker规格",
+  "设计.上下文投影排序",
+]);
+
+/** 该 output tag 应按 context-fragment.v1 验收（有可用「正文」） */
+export function expectsContextFragmentTag(tag: string): boolean {
+  const t = tag.trim();
+  if (!t.startsWith("设计.")) return false;
+  return !NON_FRAGMENT_DESIGN_TAGS.has(t);
+}
+
+export function hasUsableFragmentBody(body: unknown): boolean {
+  if (body == null) return false;
+  if (typeof body === "string") return body.trim().length > 0;
+  if (Array.isArray(body)) return body.length > 0;
+  if (typeof body === "object") return Object.keys(body as object).length > 0;
+  return false;
+}
+
+/** 能进验收卡：解析成功且「正文」非空 */
+export function isUsableContextFragment(raw: unknown): boolean {
+  const frag = parseContextFragment(raw);
+  return Boolean(frag && hasUsableFragmentBody(frag.正文));
+}
+
+export function looksLikeFragmentDoc(doc: unknown): boolean {
+  if (!doc || typeof doc !== "object" || Array.isArray(doc)) return false;
+  const row = doc as Record<string, unknown>;
+  if (row.schema === CONTEXT_FRAGMENT_SCHEMA) return true;
+  if (typeof row.技能 === "string" && row.技能.trim()) return true;
+  return (
+    typeof row.brief === "string" &&
+    ("正文" in row || "body" in row)
+  );
+}
+
 /** 投影级别：summary 优先 brief；fields 尝试列出正文顶层键 */
 export function projectFragmentContent(
  raw: string,
--- a/src/skills/context-segments.ts
+++ b/src/skills/context-segments.ts
@@ -7,7 +7,9 @@ import type { BlackboardInputMerge } from "../types/blackboard.js";
 import { CONTEXT_BRIEF_TAG } from "../runtime/compress-after-worker.js";
 import {
  CREATION_ACCEPTED_CONTENT_TAG,
+  CREATION_ACCEPTED_UNITS_TAG,
  formatAcceptedContentForPrompt,
+  formatAcceptedUnitsForPrompt,
 } from "./creation-units.js";
 import { projectFragmentContent } from "./context-fragment.js";
 import {
@@ -107,6 +109,8 @@ function formatSegmentBody(
      if (!content) continue;
      if (tag === CREATION_ACCEPTED_CONTENT_TAG) {
        content = formatAcceptedContentForPrompt(content);
+      } else if (tag === CREATION_ACCEPTED_UNITS_TAG) {
+        content = formatAcceptedUnitsForPrompt(content);
      }
      content = applyPolicy(content, segment.policy);
      content = applyProjection(content, segment.projection);
@@ -154,15 +158,20 @@ export function assembleWorkerContext(params: {
  const segments = params.segments ?? [];

  if (segments.length === 0) {
+    const revisionNote = (
+      params.inputs["用户.修订说明"] ??
+      ""
+    ).trim();
    return JSON.stringify(
      {
        workerId: params.workerId,
        workerName: params.workerName,
        outputTags: params.outputTags,
        inputs: params.inputs,
-        instruction:
-          "根据 SKILL 说明完成任务。若 inputs 含「上下文.定稿摘要」，以终产物为准。" +
-          "设计.worker集 / 草稿须为 JSON 对象文本。",
+        instruction: revisionNote
+          ? "修订模式：在已有产物（inputs 中对应 outputTags）上按「用户.修订说明」以及「用户.worker答复」中的追问作答（选项与补充）修改；保留未点名要改的部分，禁止无故整份重写。节点方法（task/principles/probe/output）仍须遵守。"
+          : "根据 SKILL 说明完成任务。若 inputs 含「上下文.定稿摘要」，以终产物为准。" +
+            "设计.worker集 / 草稿须为 JSON 对象文本。",
      },
      null,
      2,
@@ -191,18 +200,61 @@ export function assembleWorkerContext(params: {
    blocks.push(`## 上下文.定稿摘要\n\n${brief}`);
  }

-  blocks.push(
-    [
-      "## 本步任务",
-      "",
-      `- workerId: \`${params.workerId}\``,
-      `- workerName: ${params.workerName}`,
-      `- outputTags: ${params.outputTags.map((t) => `\`${t}\``).join("、") || "（无）"}`,
-      "",
+  // 修订态：把本步 outputTags 上已有正文钉成「待改底稿」，禁止无故推倒重写
+  const revisionNote = readTagContent(
+    "用户.修订说明",
+    params.inputs,
+    params.blackboard,
+    inputMerge,
+  ).trim();
+  const draftBlocks: string[] = [];
+  if (revisionNote) {
+    for (const tag of params.outputTags) {
+      const alreadyInSegments = segments.some((s) => s.tags.includes(tag));
+      if (alreadyInSegments) continue;
+      const draft = readTagContent(
+        tag,
+        params.inputs,
+        params.blackboard,
+        inputMerge,
+      ).trim();
+      if (!draft) continue;
+      draftBlocks.push(`### \`${tag}\`\n\n${draft}`);
+    }
+    if (draftBlocks.length) {
+      blocks.push(
+        [
+          "## 【待改底稿】",
+          "",
+          "下列为**当前已有产物**。按「用户.修订说明」以及「用户.worker答复」中的追问作答（选项与补充）在其上修改，并写回同名 outputTags。",
+          "节点方法（task / principles / probe / output）仍须遵守。保留未要求改动的结构与结论；禁止无故整份重写。",
+          "",
+          ...draftBlocks,
+        ].join("\n"),
+      );
+    }
+  }
+
+  const taskLines = [
+    "## 本步任务",
+    "",
+    `- workerId: \`${params.workerId}\``,
+    `- workerName: ${params.workerName}`,
+    `- outputTags: ${params.outputTags.map((t) => `\`${t}\``).join("、") || "（无）"}`,
+    "",
+  ];
+  if (revisionNote) {
+    taskLines.push(
+      "**修订模式**：接着上方【待改底稿】（若有）与依赖产物，落实「用户.修订说明」以及「用户.worker答复」里的追问作答（选项与补充意见）。",
+      "不要从零另起一份；不要扩大修改面。标为「只读 / 已定稿」的分区仍不可改。节点方法仍有效。",
+    );
+  } else {
+    taskLines.push(
      "按 SKILL 与上方分区完成任务。标为「只读 / 已定稿」的分区不要擅自改写；只改【本单位】范围。",
-      "设计.worker集 / 草稿须为 JSON 对象文本（以 `{` 开头）。",
-    ].join("\n"),
-  );
+    );
+  }
+  taskLines.push("设计.worker集 / 草稿须为 JSON 对象文本（以 `{` 开头）。");
+  blocks.push(taskLines.join("\n"));

  return blocks.join("\n\n---\n\n");
 }
--- a/src/skills/creation-flow.ts
+++ b/src/skills/creation-flow.ts
@@ -13,8 +13,36 @@ import { parse as parseYaml } from "yaml";

 export const CREATION_FLOW_TAG = "设计.创作流程";
 export const CREATION_CURRENT_STEP_TAG = "创作.当前步骤";
+/** 确认开干前：待确认的下一步快照（JSON） */
+export const CREATION_PROPOSED_STEP_TAG = "创作.待确认步骤";
 /** 用户手动选定的初始配方（存 recipe id，或 JSON {id,name}） */
 export const CREATION_SELECTED_RECIPE_TAG = "创作.选用配方";
+
+/** 进度指针：禁止当作 worker 产物写回或列入验收正文 */
+export const PROGRESS_POINTER_TAGS = new Set<string>([
+  CREATION_CURRENT_STEP_TAG,
+  CREATION_PROPOSED_STEP_TAG,
+  CREATION_SELECTED_RECIPE_TAG,
+  "创作.当前单位",
+  "创作.已验收单位",
+  "创作.已验收内容",
+  "创作.能力开场白",
+  "创作.能力开场状态",
+]);
+
+/** 旧名：含「设计.创作流程」。写回/验收过滤请用 isProgressPointerTag */
+export const RUNTIME_PINNED_TAGS = new Set<string>([
+  CREATION_FLOW_TAG,
+  ...PROGRESS_POINTER_TAGS,
+]);
+
+export function isProgressPointerTag(tag: string): boolean {
+  return PROGRESS_POINTER_TAGS.has(tag.trim());
+}
+
+export function isRuntimePinnedTag(tag: string): boolean {
+  return RUNTIME_PINNED_TAGS.has(tag.trim());
+}
 export const MODULE_CATALOG_FILENAME = "modules/catalog.yaml";
 export const RECIPE_CATALOG_FILENAME = "recipes/catalog.yaml";
 export const DESIGN_STEP_WORKER_ID = "design-step";
@@ -152,6 +180,9 @@ export type CreationFlowValidation = {
  errors: string[];
 };

+/** 工作流计划节点相对执行进度（给人看的三态） */
+export type FlowStepRunState = "done" | "current" | "pending";
+
 export type CreationFlowUserView = {
  brief?: string;
  status?: CreationFlowStatus;
@@ -169,6 +200,8 @@ export type CreationFlowUserView = {
    params?: CreationFlowStepParams;
    /** 参数缺必填项时的提示（给人看） */
    paramsMissing?: string[];
+    /** 已执行 / 将要执行 / 未执行 */
+    runState: FlowStepRunState;
  }>;
  parseError?: string;
 };
@@ -301,6 +334,42 @@ export function stepUnitId(step: CreationFlowStep): string {
  return step.id || step.name;
 }

+/**
+ * 验收时记下哪一步做完。只认流程里的真实 step.id：
+ * LLM 可能把产物误写进「创作.当前步骤」，那种内容不能当进度。
+ */
+export function pickRecordedStepId(params: {
+  flow: CreationFlow | null;
+  alreadyAccepted?: readonly string[];
+  pinnedUnitId?: string | null;
+  writtenCurrentStep?: string | null;
+  summaryHint?: string | null;
+}): string | null {
+  const flow = params.flow;
+  const tryRef = (ref: string | undefined | null): string | null => {
+    const key = ref?.trim();
+    if (!key || key === "flow") return null;
+    if (!flow) return key;
+    const step = findStepByRef(flow, key);
+    return step ? stepUnitId(step) : null;
+  };
+
+  const pinned = tryRef(params.pinnedUnitId);
+  if (pinned) return pinned;
+
+  const written = tryRef(params.writtenCurrentStep);
+  if (written) return written;
+
+  const hinted = tryRef(params.summaryHint);
+  if (hinted) return hinted;
+
+  if (flow) {
+    const pending = nextPendingStep(flow, params.alreadyAccepted ?? []);
+    if (pending) return stepUnitId(pending);
+  }
+  return null;
+}
+
 export function parseCreationFlow(raw: string | undefined | null): CreationFlow | null {
  if (!raw?.trim()) return null;
  const doc = extractJsonObject(raw);
@@ -591,6 +660,93 @@ export function stringifyModuleOpeningState(state: ModuleOpeningState): string {
  return JSON.stringify(state);
 }

+/**
+ * 序列化「设计.创作流程」（程序 seed / 验收写回共用）。
+ */
+export function stringifyCreationFlow(flow: CreationFlow): string {
+  return JSON.stringify({
+    version: 1,
+    ...(flow.brief ? { brief: flow.brief } : {}),
+    ...(flow.status ? { status: flow.status } : {}),
+    steps: flow.steps,
+  });
+}
+
+/**
+ * 编排层重排流程时，已验收步骤按原样保留：
+ * 缺了就按原位置补回，改了名字/参数就还原。编排只能动未验收步。
+ */
+export function mergeCreationFlowPreservingAccepted(params: {
+  prevRaw: string | null | undefined;
+  nextRaw: string;
+  acceptedStepIds: readonly string[];
+}): { raw: string; restored: string[] } {
+  const next = parseCreationFlow(params.nextRaw);
+  const prev = parseCreationFlow(params.prevRaw);
+  if (!next || !prev) return { raw: params.nextRaw, restored: [] };
+
+  const accepted = prev.steps.filter((s) => isStepAccepted(s, params.acceptedStepIds));
+  if (accepted.length === 0) return { raw: params.nextRaw, restored: [] };
+
+  const restored: string[] = [];
+  const steps = [...next.steps];
+  for (let i = 0; i < accepted.length; i++) {
+    const frozen = accepted[i]!;
+    const at = steps.findIndex((s) => s.id === frozen.id);
+    if (at < 0) {
+      steps.splice(Math.min(i, steps.length), 0, frozen);
+      restored.push(frozen.id);
+      continue;
+    }
+    const current = steps[at]!;
+    const changed =
+      current.name !== frozen.name ||
+      JSON.stringify(current.depends_on) !== JSON.stringify(frozen.depends_on) ||
+      JSON.stringify(current.params ?? null) !== JSON.stringify(frozen.params ?? null);
+    if (changed) {
+      steps[at] = frozen;
+      restored.push(frozen.id);
+    }
+  }
+  if (restored.length === 0) return { raw: params.nextRaw, restored: [] };
+  return {
+    raw: stringifyCreationFlow({ ...next, version: 1, steps }),
+    restored,
+  };
+}
+
+/**
+ * 配方近期起点 → 可写入黑板的开局 DAG。
+ * steps 为空则返回 null（仍只当选型参考，不预置流程）。
+ */
+export function creationFlowFromRecipeSeed(
+  detail: Pick<RecipeDetail, "seed">,
+): CreationFlow | null {
+  const seed = detail.seed;
+  if (!seed?.steps?.length) return null;
+  return {
+    version: 1,
+    ...(seed.brief ? { brief: seed.brief } : {}),
+    status: seed.status ?? "open",
+    steps: ensureCreationFlowStepIds(seed.steps),
+  };
+}
+
+/**
+ * 已坐在配方开局步（无 depends_on）且「用户.需求」有首句时：
+ * 跳过程序再抛模块 opening——首句即该步对话，不是「先开场再进 DAG」。
+ */
+export function shouldSkipModuleOpening(opts: {
+  demand: string;
+  dependsOn?: string[] | null;
+  phase?: "shown" | "answered" | null;
+}): boolean {
+  if (opts.phase === "shown" || opts.phase === "answered") return false;
+  if (!String(opts.demand ?? "").trim()) return false;
+  const deps = opts.dependsOn ?? [];
+  return deps.length === 0;
+}
+
 /** 无 id 时的兜底（目录仍应显式写 id） */
 function slugFromName(name: string): string {
  const map: Record<string, string> = {
@@ -661,6 +817,7 @@ export function formatModuleCatalogForAgent(catalog: ModuleCatalog): string {
    "【能力 · 可选工序】",
    "按需选用，勿默认全选；步骤名只能从这里选；标〔可反复〕的可多次编入。",
    "选型依据是下方「何时用 / 何时不用 / 边界」（来自各能力 meta）；有「编排参数」的步骤必须在 DAG 里写齐 params，缺参时用 askUser 选项+其它，禁止空壳进执行。",
+    "生成规则：改 target 时必须同步改 rule_id（英文 kebab-case）与步骤 id（生成规则·{对象}）；禁止只改中文对象沿用旧后台 id。",
    "不要把能力执行全文塞进本步；执行由 design-step 注入。",
    lines.join("\n"),
  ].join("\n");
@@ -929,13 +1086,23 @@ function formatRecipeFieldBlock(value: string | string[]): string {
  return value;
 }

-/** design-flow 一次注入：已选配方 + 技能池 */
+/** design-flow 一次注入：流程进度 + 已选配方 + 技能池 */
 export function formatDesignFlowContentBlocks(params: {
  selectedRecipe?: RecipeDetail | null;
  modules?: ModuleCatalog | null;
  missingSelection?: boolean;
+  flow?: CreationFlow | null;
+  acceptedStepIds?: readonly string[];
+  filledArtifactTags?: readonly string[];
 }): string[] {
-  const blocks: string[] = [];
+  const blocks: string[] = [
+    formatFlowProgressForAgent({
+      flow: params.flow,
+      acceptedStepIds: params.acceptedStepIds,
+      catalog: params.modules,
+      filledArtifactTags: params.filledArtifactTags,
+    }),
+  ];
  if (params.missingSelection) {
    blocks.push(
      [
@@ -953,6 +1120,120 @@ export function formatDesignFlowContentBlocks(params: {
  return blocks;
 }

+/**
+ * 编排器必须看见的进度：哪些步已做完、草案里已有什么、哪些能力才允许再编入。
+ * 裸 JSON id 列表不够；此块由程序钉死。
+ */
+export function formatFlowProgressForAgent(params: {
+  flow?: CreationFlow | null;
+  acceptedStepIds?: readonly string[];
+  catalog?: ModuleCatalog | null;
+  filledArtifactTags?: readonly string[];
+}): string {
+  const accepted = (params.acceptedStepIds ?? []).filter(
+    (id) => Boolean(id?.trim()) && id.trim() !== "flow",
+  );
+  const flow = params.flow ?? null;
+  const catalog = params.catalog ?? null;
+  const filled = new Set(params.filledArtifactTags ?? []);
+  const repeatableNames = new Set(
+    (catalog?.modules ?? [])
+      .filter((m) => m.repeatable === true)
+      .map((m) => m.name),
+  );
+
+  const doneLines: string[] = [];
+  const doneNames = new Set<string>();
+  const doneIds = new Set<string>();
+
+  const markDone = (
+    name: string,
+    id: string | undefined,
+    why: string,
+    repeatable: boolean,
+  ) => {
+    const flag = repeatable
+      ? "〔可反复：可再追加不同 id〕"
+      : "〔不可反复：禁止再排一次执行〕";
+    const idBit = id ? `（id: ${id}）` : "";
+    doneLines.push(`- ${name}${idBit} · ${why} ${flag}`);
+    doneNames.add(name);
+    if (id) doneIds.add(id);
+  };
+
+  if (flow) {
+    for (const step of flow.steps) {
+      const acceptedHere = isStepAccepted(step, accepted);
+      const art = catalog ? artifactTagForStep(step.name, catalog) : null;
+      const hasArtifact = Boolean(art && filled.has(art));
+      const repeatable = repeatableNames.has(step.name);
+      if (acceptedHere || (!repeatable && hasArtifact)) {
+        markDone(
+          step.name,
+          step.id,
+          acceptedHere ? "已验收" : "产物已在黑板",
+          repeatable,
+        );
+      }
+    }
+  }
+
+  for (const id of accepted) {
+    if (doneIds.has(id)) continue;
+    if (flow?.steps.some((s) => isStepAccepted(s, [id]))) continue;
+    markDone(id, id, "已验收", repeatableNames.has(id));
+  }
+
+  if (catalog) {
+    for (const mod of catalog.modules) {
+      if (mod.repeatable === true) continue;
+      if (!mod.artifact || !filled.has(mod.artifact)) continue;
+      if (doneNames.has(mod.name)) continue;
+      markDone(mod.name, undefined, "产物已在黑板", false);
+    }
+  }
+
+  const draftLines: string[] = [];
+  if (flow) {
+    for (const step of flow.steps) {
+      if (doneIds.has(step.id) || doneNames.has(step.name)) continue;
+      draftLines.push(
+        `- ${step.name}（id: ${step.id}）· 已在草案，保留原 id；不要当作新规划再写一遍`,
+      );
+    }
+  }
+
+  const repeatableList = (catalog?.modules ?? [])
+    .filter((m) => m.repeatable === true)
+    .map((m) => m.name);
+
+  const lines = [
+    "【流程进度】（程序钉死，必须遵守）",
+    "你的任务是追加缺口，不是从头规划。",
+    "已完成的非反复技能：只保留原 id 作 depends_on 锚点，禁止再作为将要执行的一步。",
+    "草案里已有的步骤：原样保留，不要重排一遍开局。",
+    "只有能力目录标了〔可反复〕的（如生成规则、具体实例）才允许再编入新 id。",
+  ];
+  if (doneLines.length) {
+    lines.push("", "已完成：", ...doneLines);
+  } else {
+    lines.push(
+      "",
+      "已完成：尚无已验收步骤。若草案已有开局步，那是配方预置起点，不是你新规划的。",
+    );
+  }
+  if (draftLines.length) {
+    lines.push("", "草案已有、尚未验收（保留，勿重排）：", ...draftLines);
+  }
+  lines.push(
+    "",
+    repeatableList.length
+      ? `可反复追加：${repeatableList.join("、")}`
+      : "可反复追加：以能力目录〔可反复〕为准。",
+  );
+  return lines.join("\n");
+}
+
 /** 解析并加载用户已选配方详情 */
 export async function resolveSelectedRecipeDetail(params: {
  skillPackRoot: string;
@@ -973,7 +1254,6 @@ export function validateCreationFlow(
  catalog: ModuleCatalog | null,
 ): CreationFlowValidation {
  const errors: string[] = [];
-  const ids = flow.steps.map((s) => s.id);
  const seenIds = new Set<string>();
  const allowed = catalog
    ? new Set(catalog.modules.map((m) => m.name))
@@ -1020,25 +1300,53 @@ export function validateCreationFlow(
      }
    }

-    if (mod?.params?.length) {
-      const missing = missingRequiredStepParams(step, mod);
-      if (missing.length > 0) {
-        const labels = missing
-          .map((key) => {
-            const spec = mod.params!.find((p) => p.key === key);
-            return spec ? `${key}（${spec.label}）` : key;
-          })
-          .join("、");
-        errors.push(
-          `「${step.id}」缺少必填编排参数：${labels}（须在 design-flow 钉齐后再执行）`,
-        );
-      }
-    }
+    // 必填 params 改在「确认开干」时钉齐，不再阻挡编排产出空壳步骤
  }

  return { ok: errors.length === 0, errors };
 }

+/** 进 design-step 前：本步必填编排参数是否已齐 */
+export function validateStepReadyToRun(
+  step: CreationFlowStep,
+  catalog: ModuleCatalog | null,
+): CreationFlowValidation {
+  const mod = catalog?.modules.find((m) => m.name === step.name);
+  if (!mod?.params?.length) return { ok: true, errors: [] };
+  const missing = missingRequiredStepParams(step, mod);
+  if (missing.length === 0) return { ok: true, errors: [] };
+  const labels = missing
+    .map((key) => {
+      const spec = mod.params!.find((p) => p.key === key);
+      return spec ? `${key}（${spec.label}）` : key;
+    })
+    .join("、");
+  return {
+    ok: false,
+    errors: [`「${step.id}」缺少必填编排参数：${labels}（确认开干前请补齐）`],
+  };
+}
+
+/** 写回某步 params（确认下一节点时由用户钉死） */
+export function patchCreationFlowStepParams(
+  flow: CreationFlow,
+  stepId: string,
+  params: CreationFlowStepParams,
+): CreationFlow {
+  const id = stepId.trim();
+  return {
+    ...flow,
+    steps: flow.steps.map((s) =>
+      s.id === id
+        ? {
+            ...s,
+            params: Object.keys(params).length ? { ...params } : undefined,
+          }
+        : s,
+    ),
+  };
+}
+
 export function artifactTagForStep(
  name: string,
  catalog: ModuleCatalog | null,
@@ -1049,11 +1357,13 @@ export function artifactTagForStep(
 export function formatCreationFlowForUser(
  flow: CreationFlow,
  catalog?: ModuleCatalog | null,
+  acceptedStepIds: readonly string[] = [],
 ): CreationFlowUserView {
  const decl = new Map(
    (catalog?.modules ?? []).map((m) => [m.name, m] as const),
  );
  const seenName = new Map<string, number>();
+  const next = nextPendingStep(flow, acceptedStepIds);
  return {
    brief: flow.brief,
    status: flow.status,
@@ -1062,6 +1372,11 @@ export function formatCreationFlowForUser(
      seenName.set(s.name, n);
      const mod = decl.get(s.name);
      const paramsMissing = missingRequiredStepParams(s, mod);
+      const runState: FlowStepRunState = isStepAccepted(s, acceptedStepIds)
+        ? "done"
+        : next && next.id === s.id
+          ? "current"
+          : "pending";
      return {
        order: i + 1,
        id: s.id,
@@ -1070,6 +1385,7 @@ export function formatCreationFlowForUser(
        occurrence: n,
        declaration: mod?.declaration,
        repeatable: mod?.repeatable,
+        runState,
        ...(s.params ? { params: s.params } : {}),
        ...(paramsMissing.length > 0 ? { paramsMissing } : {}),
      };
--- a/src/skills/creation-units.ts
+++ b/src/skills/creation-units.ts
@@ -491,6 +491,13 @@ export function parseAcceptedUnits(raw: unknown): string[] {
  return [];
 }

+/** 给人 / 编排 LLM 看的已验收步骤清单；去掉「flow」（那是计划本身，不是某一步）。 */
+export function formatAcceptedUnitsForPrompt(raw: unknown): string {
+  const ids = parseAcceptedUnits(raw).filter((id) => id && id !== "flow");
+  if (ids.length === 0) return "（尚无已验收步骤）";
+  return ids.map((id) => `- ${id}`).join("\n");
+}
+
 export function isFinalWorkerSetArtifact(artifact: {
  workerId: string;
  outputTags: string[];
--- a/src/skills/declared-worker.ts
+++ b/src/skills/declared-worker.ts
@@ -101,6 +101,7 @@ export async function resolveRunnableWorker(params: {
        currentStepName,
        acceptedStepNames: parseAcceptedSteps(acceptedRaw),
        selectedRecipeRef,
+        filledArtifactTags: params.blackboard.listTagIndex().map((item) => item.tag),
      },
    );
    return {
--- a/src/skills/loader.ts
+++ b/src/skills/loader.ts
@@ -473,6 +473,8 @@ export async function loadWorkerSkillWithContext(
    acceptedStepNames?: readonly string[];
    /** 用户手动选定的配方 id / 名 */
    selectedRecipeRef?: string | null;
+    /** 黑板已有产物 tag，供编排器判断哪些技能已经做过 */
+    filledArtifactTags?: readonly string[];
  },
 ): Promise<{ worker: ParsedWorkerSkill; sharedContext: string | null; promptBody: string }> {
  const worker = await loadWorkerSkill(skillIdOrName, workerId, skillsRoot);
@@ -488,6 +490,7 @@ export async function loadWorkerSkillWithContext(
      loadModuleCatalog,
      resolveSelectedRecipeDetail,
      formatDesignFlowContentBlocks,
+      parseCreationFlow,
    } = await import("./creation-flow.js");
    const modules = await loadModuleCatalog(skill.skillPackRoot, skillsRoot);
    const selectedRecipe = await resolveSelectedRecipeDetail({
@@ -499,6 +502,9 @@ export async function loadWorkerSkillWithContext(
      selectedRecipe,
      modules,
      missingSelection: !selectedRecipe,
+      flow: parseCreationFlow(opts?.flowRaw),
+      acceptedStepIds: opts?.acceptedStepNames,
+      filledArtifactTags: opts?.filledArtifactTags,
    });
    if (blocks.length) {
      moduleCatalogBlock = blocks.join("\n\n");
@@ -537,10 +543,7 @@ export async function loadWorkerSkillWithContext(
        ...binding.depTags,
      ];
      const inputTags = [...new Set(baseInputs)];
-      const outputTags = [
-        binding.module.artifact,
-        CREATION_CURRENT_STEP_TAG,
-      ];
+      const outputTags = [binding.module.artifact];
      const depSegments = binding.depTags.map((tag, i) => ({
        id: `dep-${i}`,
        tier: "static" as const,
--- a/src/skills/question-protocol.ts
+++ b/src/skills/question-protocol.ts
@@ -10,6 +10,17 @@ import type {

 export type { QuestionAnswer, QuestionItem, QuestionOption };

+/** 能力默认问题（opening）：与追问卡分流，对齐美学纲领开局引导 */
+export const MODULE_OPENING_QUESTION_ID = "module-opening";
+
+/** 是否仅为能力默认问题（应走说话面 + openingGuide，不进答题卡） */
+export function isModuleOpeningQuestions(
+  questions: readonly QuestionItem[] | undefined | null,
+): boolean {
+  if (!questions?.length || questions.length !== 1) return false;
+  return questions[0]?.id === MODULE_OPENING_QUESTION_ID;
+}
+
 function letterId(i: number): string {
  return String.fromCharCode(65 + (i % 26));
 }
@@ -88,6 +99,138 @@ export function normalizeQuestions(raw: unknown): QuestionItem[] {
  return out;
 }

+const EXAMPLE_SUFFIX = /(?:\n|^)\s*示例[：:][\s\S]*$/;
+
+/** 去掉「示例：」后比题干，避免同一问因示例行对不上 */
+export function questionPromptStem(prompt: string): string {
+  return prompt.replace(EXAMPLE_SUFFIX, "").replace(/\s+/g, "").trim();
+}
+
+function optionSignature(q: QuestionItem): string {
+  return (q.options ?? [])
+    .map((o) => o.label.trim())
+    .filter((label) => label && !/^其它/.test(label))
+    .join("\n");
+}
+
+export function isSameFollowUpQuestion(a: QuestionItem, b: QuestionItem): boolean {
+  const sigA = optionSignature(a);
+  const sigB = optionSignature(b);
+  if (sigA && sigA === sigB) return true;
+  const sa = questionPromptStem(a.prompt);
+  const sb = questionPromptStem(b.prompt);
+  if (!sa || !sb) return false;
+  if (sa === sb) return true;
+  return sa.includes(sb) || sb.includes(sa);
+}
+
+export const SLOT_ASKED_QUESTIONS = "askedFollowUps";
+
+/** 已问过的题目留痕（写进 slots 供跨 worker 去重） */
+export type AskedQuestionRecord = {
+  prompt: string;
+  options?: string[];
+};
+
+const ASKED_HISTORY_LIMIT = 40;
+
+export function toAskedQuestionRecord(q: QuestionItem): AskedQuestionRecord {
+  const options = (q.options ?? []).map((o) => o.label.trim()).filter(Boolean);
+  return options.length ? { prompt: q.prompt, options } : { prompt: q.prompt };
+}
+
+export function parseAskedQuestions(raw: unknown): AskedQuestionRecord[] {
+  const list = Array.isArray(raw)
+    ? raw
+    : typeof raw === "string" && raw.trim()
+      ? (() => {
+          try {
+            return JSON.parse(raw) as unknown;
+          } catch {
+            return null;
+          }
+        })()
+      : null;
+  if (!Array.isArray(list)) return [];
+  const out: AskedQuestionRecord[] = [];
+  for (const item of list) {
+    if (!item || typeof item !== "object" || Array.isArray(item)) continue;
+    const row = item as Record<string, unknown>;
+    const prompt = typeof row.prompt === "string" ? row.prompt.trim() : "";
+    if (!prompt) continue;
+    const options = Array.isArray(row.options)
+      ? row.options.map((o) => String(o).trim()).filter(Boolean)
+      : undefined;
+    out.push(options?.length ? { prompt, options } : { prompt });
+  }
+  return out;
+}
+
+function recordAsItem(rec: AskedQuestionRecord): QuestionItem {
+  return {
+    id: "asked",
+    prompt: rec.prompt,
+    options: rec.options?.map((label, i) => ({ id: letterId(i), label })),
+  };
+}
+
+export function wasAlreadyAsked(
+  q: QuestionItem,
+  history: readonly AskedQuestionRecord[],
+): boolean {
+  return history.some((rec) => isSameFollowUpQuestion(q, recordAsItem(rec)));
+}
+
+/** 过滤掉本局已经问过的题（用于可跳过的挂载追问） */
+export function dropAlreadyAskedQuestions(
+  questions: readonly QuestionItem[],
+  history: readonly AskedQuestionRecord[],
+): QuestionItem[] {
+  if (!history.length) return [...questions];
+  return questions.filter((q) => !wasAlreadyAsked(q, history));
+}
+
+export function appendAskedQuestions(
+  history: readonly AskedQuestionRecord[],
+  questions: readonly QuestionItem[],
+): AskedQuestionRecord[] {
+  const next = [...history];
+  for (const q of questions) {
+    if (wasAlreadyAsked(q, next)) continue;
+    next.push(toAskedQuestionRecord(q));
+  }
+  return next.slice(-ASKED_HISTORY_LIMIT);
+}
+
+/**
+ * askUser 与片段「追问」合并：同一问只留一份，优先带示例的片段题。
+ * 仅片段没有的缺口才保留 askUser。
+ */
+export function mergeQuestionsPreferFragment(
+  askUser: QuestionItem[] | undefined,
+  fragment: QuestionItem[],
+): QuestionItem[] {
+  if (!fragment.length) return askUser?.length ? [...askUser] : [];
+  if (!askUser?.length) return [...fragment];
+
+  const usedAsk = new Set<number>();
+  const out: QuestionItem[] = [];
+  for (const fq of fragment) {
+    const matchIdx = askUser.findIndex(
+      (aq, i) => !usedAsk.has(i) && isSameFollowUpQuestion(aq, fq),
+    );
+    if (matchIdx >= 0) usedAsk.add(matchIdx);
+    if (!out.some((q) => isSameFollowUpQuestion(q, fq))) out.push(fq);
+  }
+  for (let i = 0; i < askUser.length; i++) {
+    if (usedAsk.has(i)) continue;
+    const aq = askUser[i]!;
+    if (out.some((q) => isSameFollowUpQuestion(q, aq))) continue;
+    out.push(aq);
+  }
+  return out;
+}
+
 /** 发给 AI：必须含问题与答案；可选自由补充 */
 export function formatQuestionAnswersForAi(
  questions: QuestionItem[],
--- a/src/types/runtime.ts
+++ b/src/types/runtime.ts
@@ -51,7 +51,13 @@ export type WaitingReason =
      questions: QuestionItem[];
      pageSize?: number;
    } // worker 无产物时的阻塞提问
-  | { kind: "revision"; instruction?: string }; // 产物被拒或程序验收失败
+  | { kind: "revision"; instruction?: string } // 无意见打回时等补交；有意见则直接重跑
+  | {
+      /** 上一步已验收：先问「下一步想写什么」（可留空），再提案并确认开干 */
+      kind: "next_intent";
+      afterWorkerId?: string;
+      afterUnitId?: string;
+    };

 /** 与 src/skills/types 对齐的最小 skill 索引字段，避免 runtime 强依赖 skills 模块 */
 export type SkillIndexEntry = {
@@ -249,6 +255,11 @@ export type RuntimeEvent =
        questions: QuestionItem[] | string[];
      };
    }
+  | {
+      /** 按意见重跑没写出新产物：退回上一版产物验收，重试追问改挂为可选 */
+      type: "worker_revision_produced_nothing";
+      payload: { artifactId: string; questions?: QuestionItem[] | string[] };
+    }
  | {
      /** 验收态下作答/跳过挂载追问：不离开 review_artifact */
      type: "user_resolved_sidecar_questions";
@@ -316,7 +327,9 @@ export type PhaseEffect =
    }
  | { type: "resume_worker" }
  | { type: "run_programmatic_review"; artifactId: string }
-  | { type: "emit_message"; message: string };
+  | { type: "emit_message"; message: string }
+  /** 根据流程与用户下一步意向，提案下一节点并进入确认 */
+  | { type: "propose_next_creation_step" };

 /** applyEvent 的返回值：新会话快照 + 待处理副作用 + 可选错误 */
 export type ApplyEventResult = {
--- a/src/worker/executor.ts
+++ b/src/worker/executor.ts
@@ -15,10 +15,17 @@ import {
 } from "../skills/worker-set-parse.js";
 import { assembleWorkerContext } from "../skills/context-segments.js";
 import {
+  mergeQuestionsPreferFragment,
  normalizeQuestions,
  type QuestionItem,
 } from "../skills/question-protocol.js";
-import { extractFragmentAskSidecar } from "../skills/context-fragment.js";
+import {
+  extractFragmentAskSidecar,
+  expectsContextFragmentTag,
+  isUsableContextFragment,
+  looksLikeFragmentDoc,
+} from "../skills/context-fragment.js";
+import { isProgressPointerTag } from "../skills/creation-flow.js";

 export type WorkerRunParams = {
  skillName: string;
@@ -66,7 +73,8 @@ askUser 每项可为：

 - outputs 的 key 必须是要求的 outputTags
 - **设计.worker集 / 设计.worker集.草稿**：value 必须是 JSON 对象文本（以 { 开头），禁止中文说明、元叙述、提问长文；禁止 YAML
- **优先同时给 outputs + askUser**：先交出可用草稿/产物，追问挂在产物下（用户可直接接受而不作答）。仅当完全无法产出时才留空 outputs、只填 askUser
+- **context-fragment.v1**：必须放在 outputs["<本步 artifact tag>"]，禁止把片段当根对象。题目只写在产物「追问」（建议选项 + 示例）；顶层 askUser 必须为 **null**。程序会把「追问」挂到询问卡。禁止同一问再抄一份 askUser。禁止写入「创作.当前步骤」等进度指针
+- **其它产物**：优先同时给 outputs + askUser；有产物时追问挂在产物下（用户可直接接受而不作答）。仅当完全无法产出时才留空 outputs、只填 askUser
 - 能推断选项时 **必须**给 options（完整句、可改写）；不要只丢裸问题逼用户写长段
 - 若 inputs 中 \`用户.需求\` / \`用户.博弈需求\`（或 book.brief）已有实质内容，禁止 askUser 要求用户重复提供其中已写明的情境、角色、规则等；仅对 genuinely 缺失且无法推断的要点提问
 - summary 用于界面展示`;
@@ -127,6 +135,102 @@ function gatherInputs(
  return inputs;
 }

+function stringifyOutputValue(val: unknown): string | undefined {
+  if (typeof val === "string" && val.trim()) return val.trim();
+  if (val && typeof val === "object") return JSON.stringify(val);
+  return undefined;
+}
+
+function productOutputTags(outputTags: string[]): string[] {
+  return outputTags.filter((t) => t.trim() && !isProgressPointerTag(t));
+}
+
+function assignOutput(
+  outputs: Record<string, string>,
+  tag: string | undefined,
+  val: unknown,
+): void {
+  if (!tag || isProgressPointerTag(tag)) return;
+  const text = stringifyOutputValue(val);
+  if (text) outputs[tag] = text;
+}
+
+/** 模型常把片段当根对象，或写进「创作.当前步骤」 */
+function recoverFragmentOutputs(
+  obj: Record<string, unknown>,
+  outputs: Record<string, string>,
+  outputTags: string[],
+): void {
+  const target = productOutputTags(outputTags)[0];
+  if (!target) return;
+  if (outputs[target] && isUsableContextFragment(outputs[target])) return;
+
+  const outputsRaw =
+    obj.outputs && typeof obj.outputs === "object" && !Array.isArray(obj.outputs)
+      ? (obj.outputs as Record<string, unknown>)
+      : null;
+
+  if (outputsRaw) {
+    for (const [key, val] of Object.entries(outputsRaw)) {
+      if (key === target) continue;
+      const text = stringifyOutputValue(val);
+      if (!text) continue;
+      if (isUsableContextFragment(text) || looksLikeFragmentDoc(val)) {
+        outputs[target] = text;
+        return;
+      }
+    }
+  }
+
+  const protocolKeys = new Set(["outputs", "summary", "askUser", "askAssessment"]);
+  const rest: Record<string, unknown> = {};
+  for (const [k, v] of Object.entries(obj)) {
+    if (!protocolKeys.has(k)) rest[k] = v;
+  }
+  if (looksLikeFragmentDoc(obj) || looksLikeFragmentDoc(rest)) {
+    const source = looksLikeFragmentDoc(obj) ? obj : rest;
+    outputs[target] = JSON.stringify(source);
+  }
+}
+
+function dropUnusableFragmentOutputs(outputs: Record<string, string>): void {
+  for (const tag of Object.keys(outputs)) {
+    if (isProgressPointerTag(tag)) {
+      delete outputs[tag];
+      continue;
+    }
+    const content = outputs[tag]!;
+    const requireFragment = expectsContextFragmentTag(tag);
+    const looksFragment =
+      isUsableContextFragment(content) ||
+      looksLikeFragmentDoc(tryParseObject(content)) ||
+      /"schema"\s*:\s*"context-fragment\.v1"/.test(content) ||
+      /"技能"\s*:/.test(content);
+    if ((requireFragment || looksFragment) && !isUsableContextFragment(content)) {
+      delete outputs[tag];
+    }
+  }
+}
+
+function tryParseObject(text: string): unknown {
+  const extracted = extractJsonObjectText(text) ?? text.trim();
+  try {
+    return JSON.parse(extracted);
+  } catch {
+    return undefined;
+  }
+}
+
+const INCOMPLETE_FRAGMENT_ASK = normalizeQuestions([
+  {
+    id: "design-step-retry",
+    prompt:
+      "这一步还没写出可验收的产物。请再补一点你最在意的体验或参与方式；也可以说「按已有描述先出一版」。",
+    allowOther: true,
+    required: true,
+  },
+]);
+
 function parseWorkerResponse(
  raw: string,
  outputTags: string[],
@@ -135,7 +239,6 @@ function parseWorkerResponse(
  try {
    parsed = JSON.parse(raw);
  } catch {
-    // 整段不是协议 JSON：切勿把散文塞进设计.worker集*
    if (outputTags.some((t) => WORKER_SET_OUTPUT_TAGS.has(t))) {
      const questions = normalizeQuestions(extractQuestionsFromText(raw));
      return {
@@ -149,16 +252,20 @@ function parseWorkerResponse(
            ]),
      };
    }
-    const fallback: Record<string, string> = {};
-    if (outputTags.length === 1) {
-      fallback[outputTags[0]] = raw;
-    } else {
-      fallback[outputTags[0] ?? "output.草稿"] = raw;
+    const target = productOutputTags(outputTags)[0] ?? outputTags[0];
+    if (target && isUsableContextFragment(raw)) {
+      return finalizeParsedOutputs(
+        { [target]: raw },
+        { summary: raw.slice(0, 80) },
+        outputTags,
+      );
    }
+    const questions = normalizeQuestions(extractQuestionsFromText(raw));
    return {
-      outputs: fallback,
-      summary: raw.slice(0, 80),
+      outputs: {},
+      summary: "未产出可验收产物",
      preview: raw.slice(0, 600),
+      askUser: questions.length ? questions.slice(0, 2) : INCOMPLETE_FRAGMENT_ASK,
    };
  }

@@ -171,17 +278,20 @@ function parseWorkerResponse(
  const outputs: Record<string, string> = {};

  if (outputsRaw && typeof outputsRaw === "object") {
-    for (const tag of outputTags) {
-      const val = (outputsRaw as Record<string, unknown>)[tag];
-      if (typeof val === "string" && val.trim()) {
-        outputs[tag] = val.trim();
-      } else if (val && typeof val === "object") {
-        // 模型有时直接回对象而非字符串
-        outputs[tag] = JSON.stringify(val);
-      }
+    for (const tag of productOutputTags(outputTags)) {
+      assignOutput(outputs, tag, (outputsRaw as Record<string, unknown>)[tag]);
    }
  }
+  recoverFragmentOutputs(obj, outputs, outputTags);

+  return finalizeParsedOutputs(outputs, obj, outputTags);
+}
+
+function finalizeParsedOutputs(
+  outputs: Record<string, string>,
+  obj: Record<string, unknown>,
+  _outputTags: string[],
+): WorkerRunResult {
  const isMetaAskToken = (s: string) => /^ask[_-]?user$/i.test(s.trim());

  let askUser: QuestionItem[] | undefined;
@@ -213,7 +323,7 @@ function parseWorkerResponse(
    }
  }

-  // context-fragment.v1：正文内 追问/自评 → 挂到询问卡（不必再抄一份 askUser）
+  // 先抽追问，再丢掉半残片段——无正文时仍要把题留给 LLM loop
  let askAssessment: string | undefined;
  const fragQuestions: QuestionItem[] = [];
  for (const content of Object.values(outputs)) {
@@ -226,19 +336,20 @@ function parseWorkerResponse(
    }
  }
  if (fragQuestions.length) {
-    if (!askUser?.length) {
-      askUser = fragQuestions;
-    } else {
-      for (const q of fragQuestions) {
-        if (!askUser.some((x) => x.prompt === q.prompt)) askUser.push(q);
-      }
-    }
+    askUser = mergeQuestionsPreferFragment(askUser, fragQuestions);
+  }
+
+  dropUnusableFragmentOutputs(outputs);
+
+  if (Object.keys(outputs).length === 0 && (!askUser || askUser.length === 0)) {
+    askUser = normalizeQuestions(INCOMPLETE_FRAGMENT_ASK);
  }

  const summary =
    typeof obj.summary === "string" && obj.summary.trim()
      ? obj.summary.trim()
-      : Object.values(outputs)[0]?.slice(0, 80) ?? "Worker 已完成";
+      : Object.values(outputs)[0]?.slice(0, 80) ??
+        (askUser?.length ? `待补充：${askUser[0]!.prompt.slice(0, 40)}` : "Worker 已完成");

  const preview =
    Object.entries(outputs)
@@ -379,6 +490,9 @@ export async function runWorkerSkill(params: WorkerRunParams): Promise<WorkerRun
    } catch {
      acceptedStepNames = [];
    }
+    const filledArtifactTags = params.blackboard
+      .listTagIndex()
+      .map((item) => item.tag);
    const loaded = await loadWorkerSkillWithContext(
      params.skillName,
      workerId,
@@ -388,6 +502,7 @@ export async function runWorkerSkill(params: WorkerRunParams): Promise<WorkerRun
        currentStepName,
        acceptedStepNames,
        selectedRecipeRef,
+        filledArtifactTags,
      },
    );
    worker = loaded.worker;
--- a/start-app.bat
+++ b/start-app.bat
@@ -46,8 +46,9 @@ if exist ".env" (
 )

 echo [信息] 启动桌面窗口 http://localhost:%PORT%
-echo [信息] 调试: F12 或 Ctrl+Shift+I 打开开发者工具
-echo [信息] 自动打开 DevTools: set WRITING_AGENT_ELECTRON_DEV=1 ^&^& npm run electron
+echo [信息] 后端日志在 %USERPROFILE%\.writing-agent\logs\runtime.log
+echo [信息] 显示控制台请用 start-app.bat console
+echo [信息] F12 或 Ctrl+Shift+I 打开开发者工具，只有前端
 echo [信息] 关闭窗口即停止服务
 echo.

--- a/start.bat
+++ b/start.bat
@@ -41,13 +41,15 @@ if exist ".env" (
 )

 echo [信息] 启动服务 http://localhost:%PORT%
-echo [信息] 调试: 浏览器 F12 打开开发者工具
-echo [信息] 桌面壳: 运行 start-app.bat
+echo [信息] 改 src 或 skills 后后端会自动重启
+echo [信息] 后端日志在本窗口，文件在 %USERPROFILE%\.writing-agent\logs\runtime.log
+echo [信息] 浏览器按 F12 打开开发者工具
+echo [信息] 桌面窗口请用 start-app.bat
 echo [信息] 按 Ctrl+C 停止
 echo.

 start "" "http://localhost:%PORT%"

-call npm run web
+call npm run web:watch

 pause
--- a/tests/context-assembly.test.ts
+++ b/tests/context-assembly.test.ts
@@ -79,6 +79,38 @@ describe("contextSegments assembly", () => {
    expect(text).not.toMatch(/^\s*\{/);
  });

+  it("revision mode asks the worker to apply sidecar answers and keep the draft", () => {
+    const bb = new Blackboard();
+    bb.write({
+      tag: "设计.美学纲领与交互范式",
+      content: '{"tone":"warm"}',
+      source: "worker",
+    });
+    const text = assembleWorkerContext({
+      inputs: {
+        "用户.修订说明": "再冷一点",
+        "用户.worker答复": "【追问作答】\n问：偏好？\n答：偏冷",
+        "设计.美学纲领与交互范式": '{"tone":"warm"}',
+      },
+      segments: [
+        {
+          id: "user",
+          tier: "dynamic",
+          tags: ["用户.worker答复", "用户.修订说明"],
+          label: "## 用户表述",
+        },
+      ],
+      blackboard: bb,
+      workerId: "design-step",
+      workerName: "创作 · 美学纲领与交互范式",
+      outputTags: ["设计.美学纲领与交互范式"],
+    });
+    expect(text).toContain("待改底稿");
+    expect(text).toContain("用户.worker答复");
+    expect(text).toContain("偏冷");
+    expect(text).toContain("再冷一点");
+  });
+
  it("falls back to JSON inputs without segments", () => {
    const bb = new Blackboard();
    const text = assembleWorkerContext({
--- a/tests/context-fragment.test.ts
+++ b/tests/context-fragment.test.ts
@@ -4,6 +4,8 @@ import {
  projectFragmentContent,
  contextFragmentToView,
  extractFragmentAskSidecar,
+  isUsableContextFragment,
+  expectsContextFragmentTag,
 } from "../src/skills/context-fragment.js";
 import {
  parseContextOrder,
@@ -74,6 +76,26 @@ describe("context-fragment.v1", () => {
      "其它",
    ]);
  });
+
+  it("isUsableContextFragment requires non-empty 正文", () => {
+    expect(
+      isUsableContextFragment({
+        schema: "context-fragment.v1",
+        技能: "美学纲领与交互范式",
+        brief: "孤立免疫",
+        正文: { 美学纲领: { 体验内核: "特权" } },
+      }),
+    ).toBe(true);
+    expect(
+      isUsableContextFragment({
+        schema: "context-fragment.v1",
+        技能: "美学纲领与交互范式",
+        brief: "孤立免疫",
+      }),
+    ).toBe(false);
+    expect(expectsContextFragmentTag("设计.美学纲领与交互范式")).toBe(true);
+    expect(expectsContextFragmentTag("设计.创作流程")).toBe(false);
+  });
 });

 describe("context-order.v1", () => {
--- a/tests/creation-flow.test.ts
+++ b/tests/creation-flow.test.ts
@@ -6,6 +6,7 @@ import {
  artifactTagForStep,
  extractModuleOpening,
  formatCreationFlowForUser,
+  formatFlowProgressForAgent,
  formatModuleCatalogForAgent,
  formatRecipeCatalogForAgent,
  formatSelectedRecipeForAgent,
@@ -14,16 +15,23 @@ import {
  loadModuleCatalog,
  loadModulePrompt,
  loadRecipeCatalog,
+  mergeCreationFlowPreservingAccepted,
  needsFlowExpansion,
  nextPendingStep,
  parseCreationFlow,
+  pickRecordedStepId,
  parseModuleCatalog,
  parseModuleOpeningState,
  parseRecipeCatalog,
  parseRecipeYaml,
  parseSelectedRecipeRef,
  resolveDesignStepBinding,
+  shouldSkipModuleOpening,
+  isProgressPointerTag,
+  stringifyCreationFlow,
+  creationFlowFromRecipeSeed,
  validateCreationFlow,
+  validateStepReadyToRun,
 } from "../src/skills/creation-flow.js";
 import { loadWorkerSkillWithContext } from "../src/skills/loader.js";

@@ -67,6 +75,79 @@ describe("creation-flow", () => {
    });
  });

+  it("keeps accepted steps intact when the planner rewrites the flow", () => {
+    const prev = `{
+      "status": "open",
+      "steps": [
+        { "id": "美学纲领与交互范式", "name": "美学纲领与交互范式", "depends_on": [] }
+      ]
+    }`;
+    const rewritten = `{
+      "status": "open",
+      "steps": [
+        { "id": "美学纲领与交互范式", "name": "美学纲领与交互范式", "depends_on": ["实现机制"] },
+        { "id": "实现机制", "name": "实现机制", "depends_on": [] }
+      ]
+    }`;
+    const merged = mergeCreationFlowPreservingAccepted({
+      prevRaw: prev,
+      nextRaw: rewritten,
+      acceptedStepIds: ["美学纲领与交互范式"],
+    });
+    expect(merged.restored).toEqual(["美学纲领与交互范式"]);
+    const flow = parseCreationFlow(merged.raw)!;
+    expect(flow.steps[0]?.depends_on).toEqual([]);
+    expect(flow.steps.map((s) => s.id)).toEqual([
+      "美学纲领与交互范式",
+      "实现机制",
+    ]);
+  });
+
+  it("re-inserts an accepted step the planner dropped", () => {
+    const prev = `{
+      "status": "open",
+      "steps": [
+        { "id": "美学纲领与交互范式", "name": "美学纲领与交互范式", "depends_on": [] }
+      ]
+    }`;
+    const rewritten = `{
+      "status": "open",
+      "steps": [{ "id": "实现机制", "name": "实现机制", "depends_on": [] }]
+    }`;
+    const merged = mergeCreationFlowPreservingAccepted({
+      prevRaw: prev,
+      nextRaw: rewritten,
+      acceptedStepIds: ["美学纲领与交互范式"],
+    });
+    expect(parseCreationFlow(merged.raw)!.steps.map((s) => s.id)).toEqual([
+      "美学纲领与交互范式",
+      "实现机制",
+    ]);
+  });
+
+  it("leaves an untouched rewrite alone", () => {
+    const prev = `{
+      "status": "open",
+      "steps": [
+        { "id": "美学纲领与交互范式", "name": "美学纲领与交互范式", "depends_on": [] }
+      ]
+    }`;
+    const rewritten = `{
+      "status": "open",
+      "steps": [
+        { "id": "美学纲领与交互范式", "name": "美学纲领与交互范式", "depends_on": [] },
+        { "id": "实现机制", "name": "实现机制", "depends_on": ["美学纲领与交互范式"] }
+      ]
+    }`;
+    const merged = mergeCreationFlowPreservingAccepted({
+      prevRaw: prev,
+      nextRaw: rewritten,
+      acceptedStepIds: ["美学纲领与交互范式"],
+    });
+    expect(merged.restored).toEqual([]);
+    expect(merged.raw).toBe(rewritten);
+  });
+
  it("allows duplicate capability names with distinct ids", () => {
    const flow = parseCreationFlow(`{
      "status": "open",
@@ -171,6 +252,29 @@ describe("creation-flow", () => {
    });
    expect(view.steps[2]?.occurrence).toBe(2);
    expect(view.steps[2]?.repeatable).toBe(true);
+    expect(view.steps[0]?.runState).toBe("current");
+    expect(view.steps[1]?.runState).toBe("pending");
+    expect(view.steps[2]?.runState).toBe("pending");
+  });
+
+  it("marks accepted steps done and the next pending current", () => {
+    const flow = parseCreationFlow(`{
+      "brief": "网恋回合",
+      "status": "open",
+      "steps": [
+        { "name": "美学纲领与交互范式", "depends_on": [] },
+        { "name": "生成规则", "depends_on": ["美学纲领与交互范式"] },
+        { "id": "生成规则#2", "name": "生成规则", "depends_on": ["生成规则"] }
+      ]
+    }`)!;
+    const view = formatCreationFlowForUser(flow, sampleCatalog, [
+      "美学纲领与交互范式",
+    ]);
+    expect(view.steps.map((s) => s.runState)).toEqual([
+      "done",
+      "current",
+      "pending",
+    ]);
  });

  it("loads world-simulator catalog and formats for agent", async () => {
@@ -273,6 +377,18 @@ steps:
    expect(formatted).toContain("用户已选配方");
    expect(formatted).toContain("核心思路");
    expect(formatted).toContain("增量");
+
+    const seeded = creationFlowFromRecipeSeed(detail);
+    expect(seeded?.steps[0]?.name).toBe("美学纲领与交互范式");
+    expect(seeded?.status).toBe("open");
+    const raw = stringifyCreationFlow(seeded!);
+    expect(parseCreationFlow(raw)?.steps[0]?.id).toBe("美学纲领与交互范式");
+    expect(creationFlowFromRecipeSeed({ seed: null })).toBeNull();
+    expect(
+      creationFlowFromRecipeSeed({
+        seed: { version: 1, status: "open", steps: [] },
+      }),
+    ).toBeNull();
  });

  it("falls back to legacy hint when methodology absent", () => {
@@ -326,9 +442,50 @@ recipes:
    expect(selected.promptBody).toContain("核心思路");
    expect(selected.promptBody).toContain("何时用");
    expect(selected.promptBody).toContain("增量");
+    expect(selected.promptBody).toContain("【流程进度】");
    expect(selected.promptBody).not.toContain("【导演】用户尚未手动选择");
  });

+  it("tells design-flow which steps are done and not to reschedule them", async () => {
+    const flowRaw = JSON.stringify({
+      steps: [{ name: "美学纲领与交互范式", depends_on: [] }],
+    });
+    const loaded = await loadWorkerSkillWithContext(
+      "world-simulator",
+      "design-flow",
+      undefined,
+      {
+        selectedRecipeRef: "world-simulator",
+        flowRaw,
+        acceptedStepNames: ["美学纲领与交互范式"],
+        filledArtifactTags: ["设计.美学纲领与交互范式"],
+      },
+    );
+    expect(loaded.promptBody).toContain("【流程进度】");
+    expect(loaded.promptBody).toContain("已完成：");
+    expect(loaded.promptBody).toContain("美学纲领与交互范式");
+    expect(loaded.promptBody).toContain("已验收");
+    expect(loaded.promptBody).toContain("禁止再排一次执行");
+    expect(loaded.promptBody).toContain("生成规则");
+  });
+
+  it("formatFlowProgressForAgent keeps seed steps as draft not new work", () => {
+    const flow = parseCreationFlow(`{
+      "steps": [
+        { "name": "美学纲领与交互范式", "depends_on": [] }
+      ]
+    }`)!;
+    const text = formatFlowProgressForAgent({
+      flow,
+      acceptedStepIds: [],
+      catalog: sampleCatalog,
+    });
+    expect(text).toContain("草案已有、尚未验收");
+    expect(text).toContain("美学纲领与交互范式");
+    expect(text).toContain("不要当作新规划再写一遍");
+    expect(text).toContain("可反复追加：生成规则、具体实例");
+  });
+
  it("injects aesthetics-interaction module prompt into design-step", async () => {
    const flowRaw = JSON.stringify({
      steps: [{ name: "美学纲领与交互范式", depends_on: [] }],
@@ -344,6 +501,8 @@ recipes:
      },
    );
    expect(loaded.worker.outputTags).toContain("设计.美学纲领与交互范式");
+    expect(loaded.worker.outputTags).not.toContain("创作.当前步骤");
+    expect(loaded.worker.inputTags).toContain("创作.当前步骤");
    expect(loaded.promptBody).toContain("美学纲领与交互范式");
    expect(loaded.promptBody).toContain("体验契约");
    expect(loaded.promptBody).toContain("程序开场");
@@ -386,6 +545,35 @@ recipes:
      美学纲领与交互范式: "shown",
    });

+    expect(
+      shouldSkipModuleOpening({
+        demand: "丧尸世界只有我不会被感染",
+        dependsOn: [],
+      }),
+    ).toBe(true);
+    expect(
+      shouldSkipModuleOpening({
+        demand: "丧尸世界只有我不会被感染",
+        dependsOn: ["美学纲领与交互范式"],
+      }),
+    ).toBe(false);
+    expect(
+      shouldSkipModuleOpening({
+        demand: "",
+        dependsOn: [],
+      }),
+    ).toBe(false);
+    expect(
+      shouldSkipModuleOpening({
+        demand: "已有需求",
+        dependsOn: [],
+        phase: "shown",
+      }),
+    ).toBe(false);
+    expect(isProgressPointerTag("创作.当前步骤")).toBe(true);
+    expect(isProgressPointerTag("设计.美学纲领与交互范式")).toBe(false);
+    expect(isProgressPointerTag("设计.创作流程")).toBe(false);
+
    const binding = await resolveDesignStepBinding({
      skillPackRoot: "dialogue/world-simulator",
      flowRaw: JSON.stringify({
@@ -428,9 +616,11 @@ recipes:
    const missing = parseCreationFlow(`{
      "steps": [{ "name": "生成规则", "depends_on": [] }]
    }`)!;
-    const bad = validateCreationFlow(missing, catalog);
-    expect(bad.ok).toBe(false);
-    expect(bad.errors.some((e) => e.includes("target"))).toBe(true);
+    // 编排期允许空壳；必填 params 在确认开干时钉齐
+    expect(validateCreationFlow(missing, catalog).ok).toBe(true);
+    const stepReady = validateStepReadyToRun(missing.steps[0]!, catalog);
+    expect(stepReady.ok).toBe(false);
+    expect(stepReady.errors.some((e) => e.includes("target"))).toBe(true);

    const block = formatModuleCatalogForAgent(catalog!);
    expect(block).toContain("编排参数");
@@ -483,6 +673,43 @@ recipes:
    ).toBeNull();
  });

+  it("pickRecordedStepId ignores LLM junk on 创作.当前步骤", () => {
+    const flow = parseCreationFlow(`{
+      "status": "open",
+      "steps": [
+        { "name": "美学纲领与交互范式", "depends_on": [] },
+        { "name": "生成规则", "depends_on": ["美学纲领与交互范式"] }
+      ]
+    }`)!;
+    expect(
+      pickRecordedStepId({
+        flow,
+        alreadyAccepted: [],
+        pinnedUnitId: "美学纲领与交互范式",
+        writtenCurrentStep: JSON.stringify({
+          schema: "context-fragment.v1",
+          正文: { 交互范式: {}, 美学纲领: {} },
+        }),
+      }),
+    ).toBe("美学纲领与交互范式");
+    expect(
+      pickRecordedStepId({
+        flow,
+        alreadyAccepted: [],
+        pinnedUnitId: "flow",
+        writtenCurrentStep: "交互范式",
+      }),
+    ).toBe("美学纲领与交互范式");
+    expect(
+      pickRecordedStepId({
+        flow,
+        alreadyAccepted: ["美学纲领与交互范式"],
+        pinnedUnitId: "",
+        writtenCurrentStep: "",
+      }),
+    ).toBe("生成规则");
+  });
+
  it("open flow needs expansion after listed steps accepted", () => {
    const flow = parseCreationFlow(`{
      "status": "open",
--- a/tests/display-labels.test.ts
+++ b/tests/display-labels.test.ts
@@ -3,6 +3,7 @@ import {
  displayStageLabel,
  displayWorkerLabel,
  formatWorkerDisplayTitle,
+  reviewComposerCopy,
 } from "../src/server/display-labels.js";

 describe("display-labels", () => {
@@ -27,4 +28,22 @@ describe("display-labels", () => {
    );
    expect(displayWorkerLabel("worker:narrator")).toBe("执行单元 · 叙事转述");
  });
+
+  it("maps review composer copy by worker", () => {
+    const flow = reviewComposerCopy("design-flow");
+    expect(flow.acceptLabel).toBe("确认编排");
+    expect(flow.submitLabel).toBe("按意见改编排");
+    expect(flow.kicker).toBe("核对编排");
+    expect(flow.tone).toBe("plan");
+
+    const step = reviewComposerCopy("design-step");
+    expect(step.acceptLabel).toBe("接受");
+    expect(step.submitLabel).toBe("按意见修改");
+    expect(step.kicker).toBe("待验收");
+    expect(step.tone).toBe("accept");
+
+    const opening = reviewComposerCopy("opening-generator");
+    expect(opening.acceptLabel).toBe("选定此开场");
+    expect(opening.kicker).toBe("待选定");
+  });
 });
--- a/tests/phase-machine.test.ts
+++ b/tests/phase-machine.test.ts
@@ -83,7 +83,7 @@ describe("phase machine", () => {
      outputTags: ["设计.worker集.草稿"],
    });
    session = { ...session, artifacts: [artifact] };
-    session = applyEvent(session, {
+    const completed = applyEvent(session, {
      type: "worker_completed",
      payload: {
        artifactId: artifact.id,
@@ -95,12 +95,22 @@ describe("phase machine", () => {
          },
        ],
      },
-    }).session;
+    });
+    session = completed.session;

    expect(session.waitingReason?.kind).toBe("review_artifact");
    if (session.waitingReason?.kind !== "review_artifact") return;
    expect(session.waitingReason.questions?.length).toBe(1);
    expect(session.waitingReason.questions?.[0]?.required).toBe(false);
+    // 追问只挂 waitingReason，不另发「可选追问」消息（与产物同面展示）
+    expect(
+      completed.effects.some(
+        (e) =>
+          e.type === "emit_message" &&
+          typeof e.message === "string" &&
+          /可选追问/.test(e.message),
+      ),
+    ).toBe(false);

    // 不答追问，直接接受产物
    session = applyEvent(session, {
@@ -159,4 +169,152 @@ describe("phase machine", () => {
    }).session;
    expect(session.slots.designInstanceReady).toBeUndefined();
  });
+
+  it("rejecting artifact with feedback immediately reruns the same worker", () => {
+    let session = createSession("default");
+    const artifact = createArtifact({
+      workerId: "design-step",
+      stepId: "step-aesthetics",
+      outputTags: ["设计.美学纲领"],
+    });
+    session = {
+      ...session,
+      artifacts: [artifact],
+      pendingArtifactId: artifact.id,
+      currentStepId: "step-aesthetics",
+      phase: "waiting_user",
+      waitingReason: { kind: "review_artifact", artifactId: artifact.id },
+      slots: {
+        "用户.worker答复": "【追问作答】\n问：偏好？\n答：偏冷",
+      },
+    };
+
+    const result = applyEvent(session, {
+      type: "user_rejected_artifact",
+      payload: { artifactId: artifact.id, reason: "语气再冷一点" },
+    });
+
+    expect(result.session.phase).toBe("running");
+    expect(result.session.waitingReason).toBeUndefined();
+    expect(result.session.slots["用户.修订说明"]).toBe("语气再冷一点");
+    // 追问答复不得被修订说明冲掉
+    expect(result.session.slots["用户.worker答复"]).toContain("偏冷");
+    expect(result.session.artifacts[0]?.status).toBe("revision_requested");
+    // 只有这一份底稿可在重跑写不出新版时退回
+    expect(result.session.slots.revisionTargetArtifactId).toBe(artifact.id);
+    expect(result.session.currentWorkerId).toBe("design-step");
+    expect(result.session.currentStepId).toBe("step-aesthetics");
+    expect(result.effects).toEqual([
+      { type: "run_worker", workerId: "design-step" },
+    ]);
+  });
+
+  it("failed revision rerun falls back to reviewing the previous artifact", () => {
+    let session = createSession("default");
+    const artifact = {
+      ...createArtifact({
+        workerId: "design-step",
+        stepId: "step-aesthetics",
+        outputTags: ["设计.美学纲领"],
+      }),
+      status: "revision_requested" as const,
+    };
+    session = {
+      ...session,
+      artifacts: [artifact],
+      currentWorkerId: "design-step",
+      currentStepId: "step-aesthetics",
+      phase: "running",
+      slots: { revisionTargetArtifactId: artifact.id },
+    };
+
+    const result = applyEvent(session, {
+      type: "worker_revision_produced_nothing",
+      payload: {
+        artifactId: artifact.id,
+        questions: ["这一步还没写出可验收的产物。请再补一点你最在意的体验。"],
+      },
+    });
+
+    expect(result.session.phase).toBe("waiting_user");
+    expect(result.session.waitingReason?.kind).toBe("review_artifact");
+    expect(result.session.pendingArtifactId).toBe(artifact.id);
+    expect(result.session.artifacts[0]?.status).toBe("under_review");
+    // 重试题降级为可选追问，用户仍可直接接受上一版
+    const reason = result.session.waitingReason;
+    expect(reason?.kind === "review_artifact" && reason.questions?.[0]?.required).toBe(
+      false,
+    );
+    // 退回后不再是重跑目标，避免下一步无产物时又被挂回来
+    expect(result.session.slots.revisionTargetArtifactId).toBeUndefined();
+  });
+
+  it("rejecting artifact without feedback waits for revision instruction", () => {
+    let session = createSession("default");
+    const artifact = createArtifact({
+      workerId: "design-step",
+      outputTags: ["设计.美学纲领"],
+    });
+    session = {
+      ...session,
+      artifacts: [artifact],
+      pendingArtifactId: artifact.id,
+      phase: "waiting_user",
+      waitingReason: { kind: "review_artifact", artifactId: artifact.id },
+    };
+
+    const rejected = applyEvent(session, {
+      type: "user_rejected_artifact",
+      payload: { artifactId: artifact.id },
+    });
+    expect(rejected.session.waitingReason?.kind).toBe("revision");
+    expect(rejected.session.artifacts[0]?.status).toBe("rejected");
+
+    const resumed = applyEvent(rejected.session, {
+      type: "user_submitted_input",
+      payload: { text: "加一点冲突感" },
+    });
+    expect(resumed.session.phase).toBe("running");
+    expect(resumed.session.slots["用户.修订说明"]).toBe("加一点冲突感");
+    expect(resumed.effects).toEqual([
+      { type: "run_worker", workerId: "design-step" },
+    ]);
+  });
+
+  it("accepting design-step enters next_intent instead of invoking main agent", () => {
+    let session = createSession("default");
+    const artifact = createArtifact({
+      workerId: "design-step",
+      stepId: "美学纲领与交互范式",
+      outputTags: ["设计.美学纲领与交互范式"],
+    });
+    session = {
+      ...session,
+      artifacts: [artifact],
+      pendingArtifactId: artifact.id,
+      phase: "waiting_user",
+      waitingReason: { kind: "review_artifact", artifactId: artifact.id },
+    };
+
+    const accepted = applyEvent(session, {
+      type: "user_accepted_artifact",
+      payload: { artifactId: artifact.id },
+    });
+    expect(accepted.session.waitingReason?.kind).toBe("next_intent");
+    expect(accepted.effects.some((e) => e.type === "invoke_main_agent")).toBe(
+      false,
+    );
+
+    const continued = applyEvent(accepted.session, {
+      type: "user_submitted_input",
+      payload: { text: "接下来写怪物生成规则" },
+    });
+    expect(continued.session.phase).toBe("running");
+    expect(continued.session.slots["用户.下一步意向"]).toBe(
+      "接下来写怪物生成规则",
+    );
+    expect(continued.effects).toEqual([
+      { type: "propose_next_creation_step" },
+    ]);
+  });
 });
--- a/tests/question-protocol.test.ts
+++ b/tests/question-protocol.test.ts
@@ -1,8 +1,14 @@
 import { describe, expect, it } from "vitest";
 import {
+  appendAskedQuestions,
+  dropAlreadyAskedQuestions,
  formatQuestionAnswersForAi,
  formatQuestionAnswersForDisplay,
+  isModuleOpeningQuestions,
+  MODULE_OPENING_QUESTION_ID,
+  mergeQuestionsPreferFragment,
  normalizeQuestions,
+  parseAskedQuestions,
 } from "../src/skills/question-protocol.js";

 describe("normalizeQuestions", () => {
@@ -28,6 +34,131 @@ describe("normalizeQuestions", () => {
  });
 });

+describe("isModuleOpeningQuestions", () => {
+  it("recognizes the single module-opening question", () => {
+    const qs = normalizeQuestions([
+      {
+        id: MODULE_OPENING_QUESTION_ID,
+        prompt: "本轮生成对象已钉死。请直接谈…",
+        allowOther: true,
+        required: true,
+      },
+    ]);
+    expect(isModuleOpeningQuestions(qs)).toBe(true);
+  });
+
+  it("rejects ordinary worker questions", () => {
+    const qs = normalizeQuestions([
+      { id: "q1", prompt: "要不要池？" },
+    ]);
+    expect(isModuleOpeningQuestions(qs)).toBe(false);
+    expect(isModuleOpeningQuestions([])).toBe(false);
+    expect(isModuleOpeningQuestions(null)).toBe(false);
+  });
+});
+describe("mergeQuestionsPreferFragment", () => {
+  it("keeps the 示例 version when askUser repeats the same options", () => {
+    const ask = normalizeQuestions([
+      {
+        prompt: "为贴近你要的质感：你最想反复感受到的是哪一种？",
+        options: [
+          "隐于都市的日常从容：所有人都只是凡人，而你心里清楚世界是你的",
+          "身份揭示的瞬间翻转：前一刻还被人平视甚至冒犯，后一刻万人噤声",
+        ],
+      },
+      {
+        prompt: "在你亮明身份之前，你希望普通世界如何对待你？",
+        options: [
+          "基本礼貌但无特别敬畏，偶尔会有微小冒犯",
+          "因你气质不凡而隐隐客气，但没人知道原因",
+        ],
+      },
+    ]);
+    const frag = normalizeQuestions([
+      {
+        prompt:
+          "你最想反复感受到的是哪一种？\n示例：比如：走在现代街头，有人撞了你一下。",
+        options: [
+          "隐于都市的日常从容：所有人都只是凡人，而你心里清楚世界是你的",
+          "身份揭示的瞬间翻转：前一刻还被人平视甚至冒犯，后一刻万人噤声",
+        ],
+      },
+      {
+        prompt:
+          "在你亮明身份之前，你希望普通世界如何对待你？\n示例：这样我才好确定叙事里普通人对你的默认距离。",
+        options: [
+          "基本礼貌但无特别敬畏，偶尔会有微小冒犯",
+          "因你气质不凡而隐隐客气，但没人知道原因",
+        ],
+      },
+    ]);
+    const merged = mergeQuestionsPreferFragment(ask, frag);
+    expect(merged).toHaveLength(2);
+    expect(merged[0]?.prompt).toContain("示例：");
+    expect(merged[1]?.prompt).toContain("示例：");
+  });
+
+  it("keeps askUser-only questions that are not in the fragment", () => {
+    const ask = normalizeQuestions([
+      { prompt: "人称？", options: ["第二人称", "第一人称"] },
+    ]);
+    const frag = normalizeQuestions([
+      {
+        prompt: "暴力尺度？\n示例：锈迹即可",
+        options: ["少而锋利", "常态压抑"],
+      },
+    ]);
+    const merged = mergeQuestionsPreferFragment(ask, frag);
+    expect(merged).toHaveLength(2);
+    expect(merged[0]?.prompt).toContain("暴力尺度");
+    expect(merged[1]?.prompt).toBe("人称？");
+  });
+});
+
+describe("已问追问去重", () => {
+  const aesthetics = normalizeQuestions([
+    {
+      id: "focus",
+      prompt: "「永远有 1 万元余额」这个设定，你希望体验更偏向哪种？",
+      options: [
+        "用这 1 万元在城市里精打细算地活下去",
+        "靠这 1 万元撬动关系、机会与影响力",
+      ],
+    },
+  ]);
+
+  it("编排层重抛上一步问过的题时整题丢掉", () => {
+    const history = appendAskedQuestions([], aesthetics);
+    const replanned = normalizeQuestions([
+      {
+        id: "focus",
+        prompt: "「永远有 1 万元余额」这个设定，你希望体验更偏向哪种？",
+        options: [
+          "用这 1 万元在城市里精打细算地活下去",
+          "靠这 1 万元撬动关系、机会与影响力",
+        ],
+      },
+    ]);
+    expect(dropAlreadyAskedQuestions(replanned, history)).toHaveLength(0);
+  });
+
+  it("新问题照常保留", () => {
+    const history = appendAskedQuestions([], aesthetics);
+    const fresh = normalizeQuestions([
+      { prompt: "下一步先排实现机制还是叙事指南？", options: ["实现机制"] },
+    ]);
+    expect(dropAlreadyAskedQuestions(fresh, history)).toHaveLength(1);
+  });
+
+  it("留痕可经 slots 往返，且同题不重复堆积", () => {
+    const once = appendAskedQuestions([], aesthetics);
+    const twice = appendAskedQuestions(once, aesthetics);
+    expect(twice).toHaveLength(1);
+    const roundTripped = parseAskedQuestions(JSON.stringify(twice));
+    expect(dropAlreadyAskedQuestions(aesthetics, roundTripped)).toHaveLength(0);
+  });
+});
+
 describe("formatQuestionAnswers", () => {
  it("AI payload keeps Q+A; display omits prompts", () => {
    const questions = normalizeQuestions([
--- a/tests/tool-loop.test.ts
+++ b/tests/tool-loop.test.ts
@@ -5,6 +5,7 @@ import {
  MockLlmProvider,
 } from "../src/llm/client.js";
 import { runMainAgentToolLoop } from "../src/main-agent/tool-loop.js";
+import { MAIN_AGENT_TOOL_DEFINITIONS } from "../src/main-agent/tools.js";
 import { toolCallToDecision } from "../src/runtime/tool-registry.js";
 import { createSession } from "../src/runtime/phase-machine.js";

@@ -22,6 +23,12 @@ describe("main agent tool loop", () => {
    listArtifacts: () => [],
  };

+  it("exposes only DAG routing tools", () => {
+    expect(
+      MAIN_AGENT_TOOL_DEFINITIONS.map((tool) => tool.function.name),
+    ).toEqual(["read_blackboard", "run_worker"]);
+  });
+
  it("runs loop tools then returns terminal decision", async () => {
    const llm = new MockLlmProvider([
      createMockToolCall("read_blackboard", { tags: ["book.brief"] }),
--- a/tests/worker-set-sanitize.test.ts
+++ b/tests/worker-set-sanitize.test.ts
@@ -119,4 +119,114 @@ describe("sanitizeWorkerSetOutputs", () => {
    expect(result.askAssessment).toContain("美学纲领 6/10");
    expect(result.askAssessment).toContain("还需确认");
  });
+
+  it("drops askUser copies of fragment 追问 and keeps the 示例 prompt", () => {
+    const result = parseWorkerResponseForTest(
+      JSON.stringify({
+        outputs: {
+          "设计.美学纲领与交互范式": {
+            schema: "context-fragment.v1",
+            brief: "孤立免疫",
+            正文: { 美学纲领: { 体验内核: "特权与惊惶" } },
+            追问: {
+              导语: "还需确认：",
+              题目: [
+                {
+                  问: "你最想反复感受到的是哪一种？",
+                  建议选项: ["日常从容", "揭示翻转"],
+                  示例: "袖口唐纹",
+                },
+              ],
+            },
+          },
+        },
+        summary: "美学纲领 · 孤立免疫",
+        askUser: [
+          {
+            prompt: "为贴近你要的质感：你最想反复感受到的是哪一种？",
+            options: ["日常从容", "揭示翻转"],
+          },
+        ],
+      }),
+      ["设计.美学纲领与交互范式"],
+    );
+    expect(result.askUser).toHaveLength(1);
+    expect(result.askUser?.[0]?.prompt).toContain("示例：袖口唐纹");
+    expect(result.askUser?.[0]?.prompt).not.toContain("为贴近你要的质感");
+  });
+
+  it("recovers a root-level context-fragment into the artifact tag", () => {
+    const frag = {
+      schema: "context-fragment.v1",
+      技能: "美学纲领与交互范式",
+      brief: "孤立免疫",
+      正文: {
+        设定逻辑: { 变造: { 原型: "丧尸", 变点: "唯我免疫" } },
+        交互范式: { 代入: "完全" },
+        美学纲领: { 体验内核: "人人录我" },
+      },
+    };
+    const result = parseWorkerResponseForTest(JSON.stringify(frag), [
+      "设计.美学纲领与交互范式",
+    ]);
+    expect(result.outputs["设计.美学纲领与交互范式"]).toContain("人人录我");
+    expect(result.askUser).toBeUndefined();
+  });
+
+  it("moves a fragment written to 创作.当前步骤 onto the artifact tag", () => {
+    const frag = {
+      schema: "context-fragment.v1",
+      技能: "美学纲领与交互范式",
+      brief: "孤立免疫",
+      正文: { 美学纲领: { 体验内核: "特权与惊惶" } },
+    };
+    const result = parseWorkerResponseForTest(
+      JSON.stringify({
+        outputs: { "创作.当前步骤": frag },
+        summary: "误写入指针",
+      }),
+      ["设计.美学纲领与交互范式", "创作.当前步骤"],
+    );
+    expect(result.outputs["创作.当前步骤"]).toBeUndefined();
+    expect(result.outputs["设计.美学纲领与交互范式"]).toContain("特权与惊惶");
+  });
+
+  it("does not accept a fragment without 正文; keeps 追问 for the LLM loop", () => {
+    const result = parseWorkerResponseForTest(
+      JSON.stringify({
+        outputs: {
+          "设计.美学纲领与交互范式": {
+            schema: "context-fragment.v1",
+            技能: "美学纲领与交互范式",
+            brief: "在唯一免疫的末日联接中体验绝望掌控感",
+            mount: "world-simulator",
+            追问: {
+              导语: "还差站位",
+              题目: [{ 问: "你代入吗？", 建议选项: ["完全代入", "旁观"] }],
+            },
+          },
+        },
+        summary: "半残",
+      }),
+      ["设计.美学纲领与交互范式"],
+    );
+    expect(result.outputs["设计.美学纲领与交互范式"]).toBeUndefined();
+    expect(result.askUser?.length).toBeGreaterThan(0);
+    expect(result.askUser?.[0]?.prompt).toContain("你代入吗");
+  });
+
+  it("asks to continue the loop when fragment JSON is unusable", () => {
+    const result = parseWorkerResponseForTest(
+      JSON.stringify({
+        outputs: {
+          "设计.美学纲领与交互范式":
+            '{"schema":"context-fragment.v1","brief":"孤立免疫","mount":"world-simulator"',
+        },
+        summary: "截断",
+      }),
+      ["设计.美学纲领与交互范式"],
+    );
+    expect(result.outputs["设计.美学纲领与交互范式"]).toBeUndefined();
+    expect(result.askUser?.[0]?.prompt).toContain("还没写出可验收的产物");
+  });
 });
--- a/web/agent-ui.js
+++ b/web/agent-ui.js
--- a/web/app.js
+++ b/web/app.js
--- a/web/display-labels.js
+++ b/web/display-labels.js
@@ -93,3 +93,64 @@ export function formatWorkerDisplayTitle(workerId, action = null) {
  if (action === "running") return `${base} · 执行中`;
  return base;
 }
+
+/**
+ * `review_artifact` 底栏 / 工作面文案。流程编排核对工作流计划，不是「验收通过」。
+ * 须与 src/server/display-labels.ts `reviewComposerCopy` 保持一致。
+ */
+export function reviewComposerCopy(workerId, opts = {}) {
+  const hasQuestions = Boolean(opts.hasQuestions);
+  const id = String(workerId ?? "").trim();
+
+  if (id === "design-flow") {
+    return {
+      chip: "编排",
+      kicker: "核对编排",
+      taskLabel: "编排",
+      taskTitle: "核对工作流计划",
+      hint: hasQuestions
+        ? "同意就按这个排；要改就写意见，在现有计划上改。下方追问可选答。"
+        : "同意就按这个排；要改就写意见，在现有计划上改。",
+      submitLabel: "按意见改编排",
+      acceptLabel: "确认编排",
+      placeholder: "改编排意见…",
+      emptyEnterHint: "空 Enter＝确认编排",
+      tone: "plan",
+      kind: "flow",
+    };
+  }
+
+  if (id === "opening-generator") {
+    return {
+      chip: "开场",
+      kicker: "待选定",
+      taskLabel: "开场",
+      taskTitle: "选定开场白",
+      hint: hasQuestions
+        ? "同意就选定此开场；要改就写意见。下方追问可选答。"
+        : "同意就选定此开场；要改就写意见。",
+      submitLabel: "按意见修改",
+      acceptLabel: "选定此开场",
+      placeholder: "修改意见…",
+      emptyEnterHint: "空 Enter＝选定此开场",
+      tone: "plan",
+      kind: "opening",
+    };
+  }
+
+  return {
+    chip: "验收",
+    kicker: "待验收",
+    taskLabel: "验收",
+    taskTitle: "验收产物",
+    hint: hasQuestions
+      ? "同意就接受；要改就写意见，在现有产物上改。下方追问可选答。"
+      : "同意就接受；要改就写意见，在现有产物上改。",
+    submitLabel: "按意见修改",
+    acceptLabel: "接受",
+    placeholder: "修改意见…",
+    emptyEnterHint: "空 Enter＝接受",
+    tone: "accept",
+    kind: "accept",
+  };
+}
--- a/web/index.html
+++ b/web/index.html
@@ -35,7 +35,7 @@

        <div class="rail-panels">
          <div class="rail-tab-panel active" id="rail-panel-books" role="tabpanel" data-rail-panel="books">
-            <nav class="path-bar" id="book-path-bar" aria-label="作品路径"></nav>
+            <nav class="path-bar" id="book-path-bar" aria-label="面包屑" hidden></nav>
            <div class="book-list" id="book-list"></div>
            <div id="book-box-overlay" class="book-box-overlay" hidden aria-hidden="true"></div>
          </div>
@@ -60,14 +60,19 @@
            </div>
          </div>
          <div class="main-head-right">
-            <a class="link" href="/modules.html">技能</a>
-            <a class="link" href="/shells.html">呈现壳</a>
-            <a class="link" href="/stats.html">Token</a>
-            <a class="link" href="/settings.html">设置</a>
            <button type="button" class="btn-sm" id="btn-save-instance" hidden>保存定稿</button>
            <button type="button" class="btn-sm" id="btn-save-play" hidden>保存游玩</button>
            <button type="button" class="btn-sm" id="btn-export" disabled>导出</button>
            <span class="status" id="status-pill"><span class="status-dot" id="status-dot"></span><span id="status-text">—</span></span>
+            <details class="more-menu">
+              <summary class="btn-sm more-menu-sum" title="更多页面">更多</summary>
+              <div class="more-menu-panel" role="menu">
+                <a class="more-menu-item" href="/modules.html" role="menuitem">技能</a>
+                <a class="more-menu-item" href="/shells.html" role="menuitem">呈现壳</a>
+                <a class="more-menu-item" href="/stats.html" role="menuitem">Token</a>
+                <a class="more-menu-item" href="/settings.html" role="menuitem">设置</a>
+              </div>
+            </details>
          </div>
        </header>

@@ -77,8 +82,28 @@
            <div id="message-feed" class="message-feed">
              <p class="empty">点击左侧 + 新建作品</p>
            </div>
+            <div id="workspace-stage" class="workspace-stage" hidden aria-label="验收工作面"></div>
            <div id="questions-card-host" class="questions-card-host" hidden></div>
          </section>
+          <aside id="coord-rail" class="coord-rail" hidden aria-label="此前对话">
+            <button
+              type="button"
+              class="coord-rail-toggle"
+              id="btn-coord-toggle"
+              aria-expanded="false"
+              title="展开 / 收起此前对话"
+            >
+              <span class="coord-rail-toggle-label">对话</span>
+              <span id="coord-drawer-count" class="coord-drawer-count"></span>
+            </button>
+            <div class="coord-rail-panel">
+              <header class="coord-rail-head">
+                <span class="coord-rail-title">此前对话</span>
+                <button type="button" class="btn-sm coord-rail-collapse" id="btn-coord-collapse" title="收起" aria-label="收起此前对话">›</button>
+              </header>
+              <div id="coord-drawer-body" class="coord-drawer-body"></div>
+            </div>
+          </aside>
        </div>

        <footer class="composer" id="composer"></footer>
@@ -119,12 +144,13 @@
      <button type="button" class="context-menu-item" id="msg-menu-context" data-msg-menu-action="view-context" hidden>查看上下文</button>
      <button type="button" class="context-menu-item" id="msg-menu-edit" data-msg-menu-action="edit">编辑</button>
      <button type="button" class="context-menu-item" id="msg-menu-copy" data-msg-menu-action="copy">复制</button>
-      <button type="button" class="context-menu-item context-menu-danger" id="msg-menu-delete" data-msg-menu-action="delete">删除</button>
+      <button type="button" class="context-menu-item context-menu-separator-before" id="msg-menu-rollback" data-msg-menu-action="rollback" hidden>回退到这里</button>
+      <button type="button" class="context-menu-item context-menu-danger" id="msg-menu-delete" data-msg-menu-action="delete">从这里重开</button>
    </menu>

-    <dialog id="dialog-context-trace" class="dialog dialog-wide">
+    <dialog id="dialog-context-trace" class="dialog dialog-wide" aria-labelledby="context-trace-title">
      <form method="dialog" class="context-trace-form">
-        <h2>请求上下文</h2>
+        <h2 id="context-trace-title">上下文调用日志</h2>
        <p class="dialog-desc" id="context-trace-meta"></p>
        <pre id="context-trace-body" class="context-trace-body"></pre>
        <div class="dialog-actions">
--- a/web/questions-ui.js
+++ b/web/questions-ui.js
@@ -2,10 +2,15 @@
 * 结构化询问卡：点字母选中，点文案编辑；左右分页。
 * 询问是对同一次发送的可选增强：有选中则问+答拼接；未选中则不带问。
 * 点发送即答复 → 收起本轮卡（卡上无独立发送钮）。
- * 产物验收挂载题：底栏有字优先按「改产物」发送；无字时才提交所选答案。
+ * 产物验收挂载题：选项 + 底栏意见一并随「按意见修改」写回产物；空内容点该钮不等于接受。
 * 分页 / 跳过为悬浮控件，不占独立 header/footer 行。
+ *
+ * 能力「默认问题」(id=module-opening) 不是追问：对齐美学纲领开局，走说话面 + openingGuide。
 */

+/** 与 phase-runtime maybeAskModuleOpening 写入的 id 对齐 */
+export const MODULE_OPENING_QUESTION_ID = "module-opening";
+
 function esc(s) {
  return String(s ?? "")
    .replace(/&/g, "&amp;")
@@ -14,6 +19,22 @@ function esc(s) {
    .replace(/"/g, "&quot;");
 }

+/** 当前等待是否为能力默认问题（应呈现为开场引导，而非答题卡） */
+export function isModuleOpeningWaiting(view) {
+  const wr = view?.waitingReason;
+  if (wr?.kind !== "worker_questions" || !Array.isArray(wr.questions)) return false;
+  const qs = normalizeQuestions(wr.questions);
+  return qs.length === 1 && qs[0].id === MODULE_OPENING_QUESTION_ID;
+}
+
+/** 取出默认问题正文（供 openingGuide / 说话面展示） */
+export function getModuleOpeningPrompt(view) {
+  if (!isModuleOpeningWaiting(view)) return null;
+  const qs = normalizeQuestions(view.waitingReason.questions);
+  const prompt = qs[0]?.prompt?.trim();
+  return prompt || null;
+}
+
 function normalizeQuestions(raw) {
  if (!Array.isArray(raw)) return [];
  return raw
@@ -64,6 +85,8 @@ export function getActiveQuestions(view) {
  // 默认每页 1 题，左右切换；勿一次堆多题
  const pageSize = Math.max(1, Number(wr.pageSize) || 1);
  if (wr.kind === "worker_questions" && wr.questions?.length) {
+    // 能力默认问题：不进答题卡（与美学纲领 openingGuide 同形）
+    if (isModuleOpeningWaiting(view)) return null;
    return {
      questions: normalizeQuestions(wr.questions),
      workerId: wr.workerId,
@@ -114,6 +137,18 @@ export function getActiveQuestions(view) {
  return null;
 }

+/**
+ * 本轮题是否已答复收起（卡不会再画出来）。
+ * 此时若仍把主区排成答题面，用户只会看到一片空壳。
+ */
+export function isQuestionCardDismissed(view, host) {
+  const el = host ?? document.getElementById("questions-card-host");
+  const active = getActiveQuestions(view);
+  if (!el?._qDismissed || !active?.questions?.length) return false;
+  const key = JSON.stringify(active.questions.map((q) => q.id));
+  return el._qDismissed.sessionId === view?.id && el._qDismissed.key === key;
+}
+
 function persistDraftsFromDom(host, state) {
  host.querySelectorAll("[data-draft]").forEach((el) => {
    const key = el.getAttribute("data-draft");
@@ -224,7 +259,9 @@ export function clearQuestionCardState(host, view) {
  host._qState = null;
  host.hidden = true;
  host.innerHTML = "";
-  host.classList.remove("is-open");
+  host.classList.remove("is-open", "is-expanded", "has-pager", "has-skip");
+  host.removeAttribute("role");
+  host.removeAttribute("aria-label");
 }

 /**
@@ -238,7 +275,9 @@ export function renderQuestionsCard(host, view, _handlers = {}) {
  if (!active?.questions?.length) {
    host.hidden = true;
    host.innerHTML = "";
-    host.classList.remove("is-open");
+    host.classList.remove("is-open", "is-expanded", "has-pager", "has-skip");
+    host.removeAttribute("role");
+    host.removeAttribute("aria-label");
    host._qDismissed = null;
    return false;
  }
@@ -251,7 +290,9 @@ export function renderQuestionsCard(host, view, _handlers = {}) {
  ) {
    host.hidden = true;
    host.innerHTML = "";
-    host.classList.remove("is-open");
+    host.classList.remove("is-open", "is-expanded", "has-pager", "has-skip");
+    host.removeAttribute("role");
+    host.removeAttribute("aria-label");
    return false;
  }

@@ -264,11 +305,14 @@ export function renderQuestionsCard(host, view, _handlers = {}) {
        sessionId: view.id,
        key: qKey,
        page: 0,
+        expanded: false,
        // questionId -> { optionId?, text, otherText? }
        answers: {},
        drafts: {}, // option key -> edited label
      };
+  if (typeof state.expanded !== "boolean") state.expanded = false;
  host._qState = state;
+  host.classList.toggle("is-expanded", state.expanded);

  const { questions, pageSize, assessment, optional, skipLabel, dismissOnAccept } =
    active;
@@ -303,20 +347,24 @@ export function renderQuestionsCard(host, view, _handlers = {}) {
    ? `<button type="button" class="qcard-skip" data-qskip title="Esc">${esc(skipLabel || "跳过")}</button>`
    : "";

-  const chromeClass = [
-    "qcard",
+  const chromeMods = [
    pageCount > 1 ? "has-pager" : "",
    optional ? "has-skip" : "",
-  ]
-    .filter(Boolean)
-    .join(" ");
+  ].filter(Boolean);
+
+  host.classList.remove("has-pager", "has-skip");
+  for (const c of chromeMods) host.classList.add(c);
+  host.setAttribute("role", "dialog");
+  host.setAttribute("aria-label", "询问");

  host.innerHTML = `
-    <div class="${chromeClass}" role="dialog" aria-label="询问">
      <div class="qcard-float">
        ${pagerHtml}
        ${skipHtml}
      </div>
+      <button type="button" class="qcard-top" data-qtop title="${state.expanded ? "点击收起" : "点击展开"}" aria-label="${state.expanded ? "收起询问卡" : "展开询问卡"}">
+        <span class="qcard-top-grip" aria-hidden="true"></span>
+      </button>
      <div class="qcard-body">
        ${dismissNote}
        ${assessmentHtml}
@@ -364,8 +412,43 @@ export function renderQuestionsCard(host, view, _handlers = {}) {
            </section>`;
          })
          .join("")}
-      </div>
-    </div>`;
+      </div>`;
+
+  const setExpanded = (next) => {
+    state.expanded = next;
+    host.classList.toggle("is-expanded", next);
+    const top = host.querySelector("[data-qtop]");
+    if (top) {
+      top.title = next ? "点击收起" : "点击展开";
+      top.setAttribute("aria-label", next ? "收起询问卡" : "展开询问卡");
+    }
+  };
+
+  host.querySelector("[data-qtop]")?.addEventListener("click", (e) => {
+    e.preventDefault();
+    e.stopPropagation();
+    setExpanded(!state.expanded);
+  });
+
+  host.querySelector(".qcard-body")?.addEventListener("click", (e) => {
+    const t = e.target;
+    if (!(t instanceof Element)) return;
+    if (
+      t.closest(
+        ".qcard-letter, .qcard-other-input, [contenteditable=true], .qcard-nav, .qcard-skip, .qcard-pager",
+      )
+    ) {
+      return;
+    }
+    // 展开后点题干 / 评估区收起；收起时点内容区展开
+    if (state.expanded) {
+      if (t.closest(".qcard-prompt, .qcard-assessment, .qcard-optional-note")) {
+        setExpanded(false);
+      }
+      return;
+    }
+    setExpanded(true);
+  });

  host.querySelector("[data-qskip]")?.addEventListener("click", () => {
    _handlers.onSkipQuestions?.();
@@ -401,17 +484,37 @@ export function renderQuestionsCard(host, view, _handlers = {}) {
        text = (labelEl?.textContent ?? state.drafts[draftKey] ?? "").trim();
        state.answers[qid] = { optionId: oid, text };
      }
-      renderQuestionsCard(host, view, _handlers);
+      // 只切换选中态，避免整卡 innerHTML 重绘把 .qcard-body 滚回顶部
+      section.querySelectorAll(".qcard-opt").forEach((li) => {
+        li.classList.toggle("is-selected", li.getAttribute("data-oid") === oid);
+      });
+      if (!state.expanded) setExpanded(true);
+      _handlers.onAnswersChange?.();
    });
  });

+  host.querySelectorAll("[data-other]").forEach((el) => {
+    el.addEventListener("keydown", (e) => {
+      if (e.key !== "Enter" || e.shiftKey || e.isComposing) return;
+      e.preventDefault();
+      persistDraftsFromDom(host, state);
+      el.blur();
+      _handlers.onCardEnter?.();
+    });
+    el.addEventListener("input", () => _handlers.onAnswersChange?.());
+  });
+
  host.querySelectorAll("[data-draft]").forEach((el) => {
    el.addEventListener("keydown", (e) => {
-      if (e.key === "Enter") {
-        e.preventDefault();
-        el.blur();
-      }
+      if (e.key !== "Enter" || e.shiftKey || e.isComposing) return;
+      e.preventDefault();
+      persistDraftsFromDom(host, state);
+      const key = el.getAttribute("data-draft");
+      if (key) state.drafts[key] = el.textContent ?? "";
+      el.blur();
+      _handlers.onCardEnter?.();
    });
+    el.addEventListener("input", () => _handlers.onAnswersChange?.());
    el.addEventListener("blur", () => {
      const key = el.getAttribute("data-draft");
      if (key) state.drafts[key] = el.textContent ?? "";
--- a/web/shells.css
+++ b/web/shells.css
@@ -1,62 +1,124 @@
-/* 呈现壳预览页
- * 注意：本页也引入 styles.css（present 样式），会给 body 设 overflow:hidden；
- * 必须在此覆盖，否则侧栏后几项点不到。
- */
+/* 呈现壳预览页 — 主工作面 = 壳画布；侧栏只负责换壳 */
+
 html,
 body.shells-page {
  height: auto;
  min-height: 100%;
  overflow: auto;
+  background: var(--st-bg, #f4f4f5);
 }

 .shells-page .st-shell.shells-shell {
-  max-width: 1180px;
+  max-width: 1280px;
  align-items: start;
+  grid-template-columns: 200px minmax(0, 1fr);
 }

 .shells-page .st-rail {
-  --st-rail-w: 220px;
-  width: var(--st-rail-w);
+  width: 200px;
  position: sticky;
  top: var(--st-top-h, 48px);
  max-height: calc(100dvh - var(--st-top-h, 48px));
  overflow-y: auto;
  min-height: 0;
  align-self: start;
+  gap: 2px;
 }

-.shells-tone-label {
-  margin-top: 1rem !important;
-}
-
-.shells-density {
-  display: flex;
-  flex-wrap: wrap;
-  gap: 6px 10px;
-  align-items: baseline;
-  margin: 0.5rem 0 0.35rem;
-  padding: 8px 10px;
-  border-radius: 8px;
-  background: color-mix(in srgb, var(--accent-soft, #e8f0fe) 70%, transparent);
-  font-size: 13px;
-  line-height: 1.45;
-}
-
-.shells-density span {
+.st-rail-hint {
+  margin: 0 8px 10px;
  font-size: 11px;
-  font-weight: 700;
-  letter-spacing: 0.04em;
-  color: var(--muted, #666);
-  text-transform: uppercase;
+  line-height: 1.45;
+  color: var(--st-faint, #a1a1aa);
 }

 .shells-list {
  display: flex;
  flex-direction: column;
-  gap: 4px;
+  gap: 2px;
  flex: 0 0 auto;
 }

+.shells-list .st-rail-item {
+  display: flex;
+  flex-direction: column;
+  align-items: flex-start;
+  gap: 2px;
+  padding: 8px 10px;
+  line-height: 1.25;
+}
+
+.shells-list .shell-item-name {
+  font-size: 13px;
+  font-weight: 600;
+  color: inherit;
+}
+
+.shells-list .shell-item-task {
+  font-size: 11px;
+  font-weight: 400;
+  color: var(--st-faint, #a1a1aa);
+}
+
+.shells-list .st-rail-item.active .shell-item-task {
+  color: color-mix(in srgb, var(--st-accent, #2563eb) 65%, #71717a);
+}
+
+.shells-main {
+  min-height: calc(100dvh - var(--st-top-h, 48px));
+  display: flex;
+  flex-direction: column;
+}
+
+.shells-toolbar {
+  display: flex;
+  flex-wrap: wrap;
+  align-items: flex-end;
+  gap: 12px 20px;
+  padding: 14px 24px 12px;
+  border-bottom: 1px solid var(--st-border, #e4e4e7);
+  background: var(--st-surface, #fff);
+  position: sticky;
+  top: var(--st-top-h, 48px);
+  z-index: 3;
+}
+
+.shells-toolbar-block {
+  display: flex;
+  flex-direction: column;
+  gap: 6px;
+  min-width: 0;
+}
+
+.shells-toolbar-label {
+  font-size: 11px;
+  font-weight: 600;
+  letter-spacing: 0.04em;
+  text-transform: uppercase;
+  color: var(--st-faint, #a1a1aa);
+}
+
+.shells-toolbar-meta {
+  margin-left: auto;
+  align-items: flex-end;
+  text-align: right;
+  max-width: 280px;
+}
+
+.shells-toolbar-meta .shells-meta-title {
+  margin: 0;
+  font-size: 14px;
+  font-weight: 650;
+  color: var(--st-text, #18181b);
+}
+
+.shells-toolbar-meta .shells-meta-sub {
+  margin: 2px 0 0;
+  font-size: 12px;
+  color: var(--st-muted, #71717a);
+  line-height: 1.4;
+}
+
 .shells-tones {
  display: flex;
  flex-wrap: wrap;
@@ -66,65 +128,102 @@ body.shells-page {
 .shells-tone-chip {
  font: inherit;
  font-size: 12px;
-  padding: 4px 10px;
-  border-radius: 999px;
-  border: 1px solid var(--border, #ddd);
-  background: var(--surface, #fff);
+  padding: 5px 10px;
+  border-radius: 6px;
+  border: 1px solid var(--st-border, #e4e4e7);
+  background: var(--st-surface, #fff);
  cursor: pointer;
-  color: inherit;
+  color: var(--st-muted, #71717a);
+}
+
+.shells-tone-chip:hover {
+  background: #f4f4f5;
+  color: var(--st-text, #18181b);
+}
+
+.shells-tone-chip:focus-visible {
+  outline: 2px solid var(--st-accent, #2563eb);
+  outline-offset: 2px;
 }

 .shells-tone-chip.active {
  border-color: transparent;
-  background: var(--accent-soft, #e8f0fe);
+  background: var(--st-accent-soft, #dbeafe);
+  color: var(--st-accent, #2563eb);
  font-weight: 600;
 }

-.shells-body {
-  display: flex;
-  flex-direction: column;
-  gap: 1.25rem;
-}
-
-.shells-meta p {
-  margin: 0 0 0.35rem;
-  line-height: 1.5;
-}
-
-.shells-preview-label,
-.shells-tweaks h2,
-.shells-help h2 {
-  margin: 0 0 0.5rem;
-  font-size: 13px;
-  font-weight: 700;
-  color: var(--muted, #666);
+/* 主工作面：尽量大的阅读画布 */
+.shells-preview-wrap {
+  flex: 1 1 auto;
+  padding: 20px 24px 28px;
+  background: #ebebed;
 }

 .shells-preview {
-  min-height: 280px;
-  padding: 16px;
-  border-radius: 12px;
-  border: 1px dashed var(--border, #ccc);
-  background:
-    radial-gradient(ellipse at 20% 0%, rgba(15, 118, 110, 0.06), transparent 50%),
-    var(--bg, #f6f6f4);
+  min-height: min(70vh, 720px);
+  padding: 20px 16px 28px;
+  border-radius: 10px;
+  border: 1px solid var(--st-border, #e4e4e7);
+  background: #f4f4f5;
+  overflow: visible;
 }

 .shells-preview .present-shell {
-  max-width: 640px;
+  max-width: 42rem;
  margin: 0 auto;
-  box-shadow: 0 10px 28px rgba(28, 25, 23, 0.07);
+  box-shadow: none;
 }

 .shells-preview[data-shell="turn_panel"] .present-shell,
 .shells-preview[data-shell="split_board"] .present-shell,
 .shells-preview[data-shell="chapter_reader"] .present-shell {
-  max-width: 780px;
+  max-width: 52rem;
 }

-/* 预览区不裁切长正文 */
-.shells-preview {
-  overflow: visible;
+.shells-preview .present-shell-badge {
+  display: none;
+}
+
+.shells-details {
+  margin: 0 24px 28px;
+  border: 1px solid var(--st-border, #e4e4e7);
+  border-radius: 8px;
+  background: var(--st-surface, #fff);
+}
+
+.shells-details > summary {
+  cursor: pointer;
+  padding: 12px 14px;
+  font-size: 13px;
+  font-weight: 600;
+  color: var(--st-muted, #71717a);
+  list-style: none;
+}
+
+.shells-details > summary::-webkit-details-marker {
+  display: none;
+}
+
+.shells-details > summary::before {
+  content: "▸ ";
+  color: var(--st-faint, #a1a1aa);
+}
+
+.shells-details[open] > summary::before {
+  content: "▾ ";
+}
+
+.shells-details[open] > summary {
+  border-bottom: 1px solid var(--st-border, #e4e4e7);
+  color: var(--st-text, #18181b);
+}
+
+.shells-details-body {
+  padding: 14px 16px 16px;
+  display: flex;
+  flex-direction: column;
+  gap: 14px;
 }

 .shells-tweak-grid {
@@ -144,30 +243,68 @@ body.shells-page {
  font: inherit;
  padding: 6px 8px;
  border-radius: 6px;
-  border: 1px solid var(--border, #ddd);
+  border: 1px solid var(--st-border, #e4e4e7);
+  max-width: 320px;
 }

-.shells-help ul {
+.shells-help-list {
  margin: 0;
-  padding-left: 1.2rem;
+  padding-left: 1.15rem;
  line-height: 1.55;
  font-size: 13px;
-}
-
-.shells-help .muted {
-  font-size: 12px;
-  color: var(--muted, #888);
+  color: var(--st-muted, #71717a);
 }

@media (max-width: 800px) {
+  .shells-page .st-shell.shells-shell {
+    grid-template-columns: 1fr;
+  }
+
  .shells-page .st-rail {
    position: static;
    max-height: none;
    width: auto;
    overflow: visible;
+    border-right: none;
+    border-bottom: 1px solid var(--st-border, #e4e4e7);
+  }
+
+  .shells-list {
+    flex-direction: row;
+    flex-wrap: wrap;
+  }
+
+  .shells-list .st-rail-item {
+    width: auto;
+    max-width: 46%;
+  }
+
+  .st-rail-hint,
+  .st-rail-note {
+    display: none;
+  }
+
+  .shells-toolbar {
+    position: static;
+  }
+
+  .shells-toolbar-meta {
+    margin-left: 0;
+    text-align: left;
+    align-items: flex-start;
+    max-width: none;
+    width: 100%;
+  }
+
+  .shells-preview-wrap {
+    padding: 12px;
  }

  .shells-preview .present-shell {
    max-width: 100%;
  }
-}
+
+  .shells-details {
+    margin: 0 12px 20px;
+  }
+}
--- a/web/shells.html
+++ b/web/shells.html
@@ -28,44 +28,42 @@

    <div class="st-shell shells-shell">
      <aside class="st-rail" aria-label="壳列表">
-        <p class="st-rail-label">呈现壳</p>
+        <p class="st-rail-label">选壳</p>
+        <p class="st-rail-hint">按「要同时看见什么」选，不是按好看选。</p>
        <nav class="shells-list" id="shell-list" role="tablist"></nav>
-        <p class="st-rail-label shells-tone-label">装饰语气</p>
-        <div class="shells-tones" id="tone-list" role="group" aria-label="tone_chrome"></div>
-        <p class="st-rail-label shells-tone-label">正文长度</p>
-        <div class="shells-tones" id="length-list" role="group" aria-label="body length"></div>
        <p class="st-rail-note">
-          正文按 600~3000 字设计：随字数撑高、页面滚动；侧栏/选择坞钉住。创作只改内容绑定。
+          主任务永远是读正文（常 600~3000 字）。侧栏只换壳；长度/语气在预览顶栏。
        </p>
      </aside>

-      <main class="st-main">
-        <header class="st-main-head">
-          <div>
-            <h1 id="panel-title">呈现壳预览</h1>
-            <p class="st-sub" id="panel-subtitle">用示例数据看布局与可调项</p>
+      <main class="st-main shells-main">
+        <div class="shells-toolbar" role="toolbar" aria-label="预览条件">
+          <div class="shells-toolbar-block">
+            <span class="shells-toolbar-label" id="length-label">正文长度</span>
+            <div class="shells-tones" id="length-list" role="group" aria-labelledby="length-label"></div>
          </div>
-        </header>
-        <div class="st-body shells-body">
-          <section class="shells-meta" id="shell-meta"></section>
-          <section class="shells-preview-wrap">
-            <h2 class="shells-preview-label">预览</h2>
-            <div id="shell-preview" class="shells-preview"></div>
-          </section>
-          <section class="shells-tweaks">
-            <h2>本页可调（对应创作「微调轴」）</h2>
-            <div class="shells-tweak-grid" id="tweak-controls"></div>
-          </section>
-          <section class="shells-help">
-            <h2>设计意图</h2>
-            <ul>
-              <li>正文默认装得下单轮 <strong>600~3000 字</strong>：区域随内容长高，由页面/消息区滚动，不塞进固定小框。</li>
-              <li>双栏壳：侧栏 sticky；选择坞：选项区 sticky 贴底——长文读完控件还在。</li>
-              <li>壳差在「谁占主柱」；可改的是字段/显示名/行动开关/tone，不是重画布局。</li>
-            </ul>
-            <p class="muted">说明文档：<code>docs/play-presentation-shells.md</code></p>
-          </section>
+          <div class="shells-toolbar-block">
+            <span class="shells-toolbar-label" id="tone-label">语气</span>
+            <div class="shells-tones" id="tone-list" role="group" aria-labelledby="tone-label"></div>
+          </div>
+          <div class="shells-toolbar-block shells-toolbar-meta" id="shell-meta"></div>
        </div>
+
+        <section class="shells-preview-wrap" aria-label="壳预览主工作面">
+          <div id="shell-preview" class="shells-preview"></div>
+        </section>
+
+        <details class="shells-details">
+          <summary>微调轴与说明</summary>
+          <div class="shells-details-body">
+            <div class="shells-tweak-grid" id="tweak-controls"></div>
+            <ul class="shells-help-list">
+              <li>主工作面 = 上面的壳。读长文时正文撑高，页面滚动；侧栏 / 选择坞钉住。</li>
+              <li>壳差在「必须同时看见什么」：只正文 / 正文+状态 / 正文+线索 / 正文+选项…</li>
+              <li>创作只改字段绑定与显示名，不发明新布局。文档：<code>docs/play-presentation-shells.md</code></li>
+            </ul>
+          </div>
+        </details>
      </main>
    </div>

--- a/web/shells.js
+++ b/web/shells.js
@@ -1,71 +1,64 @@
 /**
- * 呈现壳预览页：切换壳 / 装饰语气 / 正文长度，用示例数据渲染 present-shells。
- * 正文设计目标：单轮常见 600~3000 字 —— 随字数撑高，外层滚动；侧栏/选项坞钉住。
+ * 呈现壳预览：主工作面 = 壳画布。
+ * 按「要同时看见什么」选壳；正文按 600~3000 字撑高滚动。
 */
 import {
  PRESENT_SHELL_IDS,
  renderPresentShellHtml,
 } from "./present-shells.js";

+/** task = 同时可见关系（选壳依据）；name = 短名 */
 const SHELL_INFO = {
  prose: {
    name: "纯散文",
-    blurb: "整块都是正文，无 HUD。长文直接往下滚。",
-    density: "正文随字数伸展；无侧栏争面积",
-    scenes: "轻对话、纯叙事、不要状态栏",
+    task: "只要正文",
+    blurb: "无 HUD。读完即止。",
  },
  chat_monitor: {
-    name: "对话 + 监控",
-    blurb: "顶栏芯片固定矮；主聊/场面再长也只往下长。",
-    density: "监控一行；正文撑高滚动",
-    scenes: "网恋 / 日常扮演 / 轻 AIRP",
+    name: "对话+监控",
+    task: "正文 + 扫状态",
+    blurb: "顶栏芯片矮；主聊/场面往下长。",
  },
  spotlight: {
    name: "场面主视",
-    blurb: "芯片一行 + 大正文柱。适合一幕写满。",
-    density: "正文主导并伸展；监控不占高",
-    scenes: "电影感 RP、一幕一景、少机制",
+    task: "大正文 + 弱状态",
+    blurb: "无聊天气泡框；场面即画布。",
  },
  turn_panel: {
    name: "回合面板",
-    blurb: "场面叙述再长，右侧交互物钉在视口内对照。",
-    density: "正文撑高；侧栏 sticky；行动在下",
-    scenes: "资源、检定、场景交互、跑团感",
+    task: "场面 + 物 + 行动",
+    blurb: "对照交互物；行动在下。",
  },
  split_board: {
    name: "双栏看板",
-    blurb: "左栏长文滚动，右栏线索钉住——对照读。",
-    density: "左正文伸展；右侧栏 sticky",
-    scenes: "调查、推理、多线索并行",
+    task: "正文 ‖ 线索",
+    blurb: "左右对照；侧栏钉住。",
  },
  choice_dock: {
    name: "选择坞",
-    blurb: "先读完长局面；选项坞 sticky 贴视口底，不用翻回顶。",
-    density: "正文撑高；底部坞钉住",
-    scenes: "选项驱动、AVG、分支关口",
+    task: "局面 + 选项",
+    blurb: "读完再选；坞贴视口底。",
  },
  chapter_reader: {
    name: "章节阅读",
-    blurb: "章题 + 长阅读柱；窄进度侧栏 sticky。",
-    density: "正文伸展为主；侧栏窄且钉住",
-    scenes: "长文 / 爽文 / 先纲后章",
+    task: "章正文 + 进度",
+    blurb: "阅读柱为主；进度窄栏钉住。",
  },
 };

 const TONES = [
  { id: "default", label: "默认" },
-  { id: "messenger", label: "讯息感" },
-  { id: "book", label: "书页感" },
-  { id: "terminal", label: "终端感" },
+  { id: "messenger", label: "讯息" },
+  { id: "book", label: "书页" },
+  { id: "terminal", label: "终端" },
 ];

 const LENGTHS = [
-  { id: "short", label: "短（~120）" },
-  { id: "mid", label: "中（~800）" },
-  { id: "long", label: "长（~2500）" },
+  { id: "short", label: "短" },
+  { id: "mid", label: "中 ·800" },
+  { id: "long", label: "长 ·2500" },
 ];

-/** 中等长度 ≈800 字 */
 const BODY_MID = `雨还在下。你把伞往她那边偏了偏，她没说话，只是把购物袋往怀里收紧了一点。水从伞骨边缘连成线，砸在柏油路上，溅起细碎的白点。

 路口的灯跳成绿色。你们一起迈步——这一次，谁也没有先松开。对面便利店的玻璃上映着两个人并肩的轮廓，忽明忽暗，像谁随手画的草稿。
@@ -82,7 +75,6 @@ const BODY_MID = `雨还在下。你把伞往她那边偏了偏，她没说话

 门开了一条缝。暖光漏出来，照在你们脚边的水渍上。她没有立刻进去，只是抬眼看你，像在等一句还没说出口的话。你把伞收拢，水仍顺着伞尖滴到地毯边缘。谁也没有提「进来坐坐」，但空气里已经有了那句话的形状。`;

-/** 长文 ≈2500 字（重复段落拼成，预览用） */
 const BODY_LONG = `${BODY_MID}

 天桥下的风从另一头灌过来，把塑料袋吹得啪啪响。你想起上一次走过这里，还是夏天，蝉声吵得人没法好好说话；如今只剩雨，以及偶尔驶过的车灯，把积水切成一条条亮痕。
@@ -115,9 +107,7 @@ const BODY_LONG = `${BODY_MID}

 「有的不用还。」她看着河对岸的塔吊，「有的还了，反而像没发生过。」

-塔吊缓缓旋转，吊臂在灰蓝天空里划出钝角。你忽然明白，呈现壳要解决的不是「字大一点」，而是这一长段话落在界面里时，读者还愿不愿意往下看——状态栏不能抢戏，选项不能在中途消失，侧栏得钉住好让人对照线索。
-
-你们在桥上站了一会儿。风把伞套吹得鼓起，又瘪下去。她把热饮塞进你手里：「暖的。你昨晚肯定没睡好。」你没有否认。杯壁的温度从掌心爬上手腕，像把雨夜重新烘了一遍，只留下能说出口的那一层。
+塔吊缓缓旋转，吊臂在灰蓝天空里划出钝角。你们在桥上站了一会儿。风把伞套吹得鼓起，又瘪下去。她把热饮塞进你手里：「暖的。你昨晚肯定没睡好。」你没有否认。杯壁的温度从掌心爬上手腕，像把雨夜重新烘了一遍，只留下能说出口的那一层。

 分开前，她抬手点了点伞：「这个，先放你那儿。等下一场雨。」

@@ -231,7 +221,7 @@ function buildTweaks() {
    show_suggested_actions: state.showActions,
    block_labels: { ...state.blockLabels },
    empty_states: {
-      aside: "（侧栏空：可在创作里改空态文案）",
+      aside: "（侧栏空）",
      footer: "（文末空）",
    },
  };
@@ -252,27 +242,25 @@ function buildPacket() {
 function render() {
  const info = SHELL_INFO[state.shell] || {
    name: state.shell,
+    task: "",
    blurb: "",
-    scenes: "",
-    density: "",
  };
  const packet = buildPacket();
-  document.getElementById("panel-title").textContent = info.name;
-  document.getElementById("panel-subtitle").textContent = `${info.scenes} · 正文约 ${packet._chars} 字（不计空白）`;

  document.getElementById("shell-meta").innerHTML = `
-    <p><strong>${esc(info.name)}</strong>（<code>${esc(state.shell)}</code>）</p>
-    <p>${esc(info.blurb)}</p>
-    <p class="shells-density"><span>长文策略</span>${esc(info.density)} · 当前试读 <strong>${packet._chars}</strong> 字</p>
-    <p class="muted">适用：${esc(info.scenes)}。请用上方「正文长度」切到长文，确认侧栏/选项是否仍可用。</p>
+    <p class="shells-meta-title">${esc(info.name)}</p>
+    <p class="shells-meta-sub">${esc(info.task)} · 试读 ${packet._chars} 字<br>${esc(info.blurb)}</p>
  `;

  const preview = document.getElementById("shell-preview");
  preview.dataset.shell = state.shell;
-  preview.innerHTML = renderPresentShellHtml(packet, esc, { tweaks: buildTweaks() });
+  preview.innerHTML = renderPresentShellHtml(packet, esc, {
+    tweaks: buildTweaks(),
+  });

  document.querySelectorAll("[data-shell]").forEach((btn) => {
    btn.classList.toggle("active", btn.getAttribute("data-shell") === state.shell);
+    btn.setAttribute("aria-selected", btn.getAttribute("data-shell") === state.shell ? "true" : "false");
  });
  document.querySelectorAll("[data-tone]").forEach((btn) => {
    btn.classList.toggle("active", btn.getAttribute("data-tone") === state.tone);
@@ -289,8 +277,9 @@ function mount() {
  const list = document.getElementById("shell-list");
  list.innerHTML = PRESENT_SHELL_IDS.map((id) => {
    const info = SHELL_INFO[id];
-    return `<button type="button" class="st-rail-item" data-shell="${esc(id)}" role="tab">
-      ${esc(info?.name || id)}
+    return `<button type="button" class="st-rail-item" data-shell="${esc(id)}" role="tab" aria-selected="false">
+      <span class="shell-item-name">${esc(info?.name || id)}</span>
+      <span class="shell-item-task">${esc(info?.task || "")}</span>
    </button>`;
  }).join("");

@@ -307,15 +296,15 @@ function mount() {
  document.getElementById("tweak-controls").innerHTML = `
    <label class="field">
      <input type="checkbox" id="tweak-show-actions" />
-      显示「建议行动」（对应微调轴 show_suggested_actions）
+      显示建议行动（show_suggested_actions）
    </label>
    <label class="field">
-      正文区显示名（block_labels.body）
-      <input type="text" id="tweak-label-body" placeholder="例如：短信记录 / 场面 / 本章" />
+      正文显示名（block_labels.body）
+      <input type="text" id="tweak-label-body" placeholder="短信记录 / 场面 / 本章" />
    </label>
    <label class="field">
-      监控区显示名（block_labels.monitor）
-      <input type="text" id="tweak-label-monitor" placeholder="例如：状态 / 今日" />
+      监控显示名（block_labels.monitor）
+      <input type="text" id="tweak-label-monitor" placeholder="状态 / 今日" />
    </label>
  `;

--- a/web/styles.css
+++ b/web/styles.css