duaer-spec 0.25.0 → 0.26.0

package/README.md

@@ -10,10 +10,11 @@ into a checkable Brief (goal / boundary / acceptance), implements only that
 scope, and only claims done when acceptance is met.
 
 The product pitch is **norms → satisfactory delivery** — not teaching humans
-to operate Spec phases or slash commands.
+to operate Spec phases, slash commands, or a professional delivery process.
 
-You should **not** have to operate the workflow. After a one-time install, you
-talk; the employee runs the process.
+You do **not** need to understand that process. After a one-time install, talk
+in **Cursor** or **Claude Code** in plain language; the employee runs the
+workflow.
 
 ## One-time install
 
@@ -39,9 +40,12 @@ The CLI also prints a one-line hint when a newer npm release exists
 npx duaer-spec update
 ```
 
-Then use **Cursor**, **Claude Code**, **Codex**, **Copilot**, **Windsurf**, **Cline**, **Continue**, **Gemini**, or **Aider** — describe what you want.
+Then use **Cursor** or **Claude Code** — describe what you want in plain
+language. You do not need to learn Spec phases or slash commands.
 
-Install writes host adapters for each (see [`ADOPT.md`](ADOPT.md)).
+`init` / `update` also write optional adapters for other editors (see
+[`ADOPT.md`](ADOPT.md)). FDE console dispatch launches **Cursor Agent** or
+**Claude Code** only.
 
 Chinese guide: [`README.zh-CN.md`](README.zh-CN.md)
 
@@ -54,20 +58,20 @@ After install, these are the capabilities you actually use:
 - **You talk; it runs the process.** No slash commands, no “you must run Spec phases.”
 - **Clarify before coding.** It turns your ask into: what to build, what not to build, how we know it’s done.
 - **Done means accepted.** It only claims finished when acceptance checks out — not when code merely exists.
-- **Works in many hosts.** Cursor, Claude Code, Codex, Copilot, Windsurf, Cline, Continue, Gemini, Aider.
+- **Supported in Cursor and Claude Code.** Install once, then speak there — no professional Spec workflow required. (Optional adapters for other editors: [`ADOPT.md`](ADOPT.md).)
 
-### In the browser: Duaer-spec FDE (live desk)
+### In the browser: Duaer-spec FDE (live console)
 
-A local web desk (`duaer live`) for “chat → confirm → dispatch → watch → open result”:
+A local web console (`duaer live`) for “chat → confirm → dispatch → watch → open result”:
 
 - **Projects.** Create and switch projects; each keeps its own chat and progress.
-- **Chat until the ask is clear.** Left chat; center confirm cards (goal / out of scope / checkable acceptance). Vague acceptance cannot be confirmed.
+- **Chat until the ask is clear.** Left chat; center confirm cards (goal / out of scope / checkable acceptance). Vague acceptance cannot be confirmed. Beside **Send**, tap **Record** to capture speech; with a **speech recognition (STT) model** in Settings (OpenAI-compatible `/audio/transcriptions`, e.g. whisper-1), the clip is decoded into the composer—edit, then send.
 - **Split big work into modules.** Confirm module by module; nothing starts until you kick off after the ones you care about are confirmed.
-- **Architecture diagram.** After requirements, review the system diagram; click it for fullscreen; kickoff waits until architecture is confirmed.
+- **Architecture diagram.** After requirements, review the system diagram; click it for fullscreen; kickoff waits until architecture is confirmed. If chat claims the diagram is ready but the middle panel is empty, the same bubble offers Regenerate — it does not leave a false “ready” line.
 - **Dispatch digital employees to code.** Pick a product folder and Cursor Agent or Claude Code; work happens in an isolated worktree, not directly on your long-lived branches.
-- **Visible task progress.** Work is a checklist; each finished item is checked off and the desk refreshes; dependent tasks release in waves so agents don’t pile on blocked work.
+- **Visible task progress.** Work is a checklist; each finished item is checked off and the console refreshes; dependent tasks release in waves so agents don’t pile on blocked work.
 - **Employee roles.** Top-bar **digital employee** directory (Implementer / Functional regression / Deployer). Optional parallel workers on the same CLI. See the next section.
-- **Locales.** Desk UI: zh-CN, zh-TW, en, ja, ko, es, pt-BR, fr, de, ru, vi.
+- **Locales.** Console UI: zh-CN, zh-TW, en, ja, ko, es, pt-BR, fr, de, ru, vi.
 - **Deliverables and results.** Open a deliverables page (requirements timeline,
   architecture, tasks, delivery); when done, open the page or the project folder.
 - **Revise when it’s not right.** Continue in left chat → re-confirm → relaunch on
@@ -78,8 +82,8 @@ A local web desk (`duaer live`) for “chat → confirm → dispatch → watch
   facts autofill; defects join the same project timeline as revisions.
 - **Quick nav.** Chat-column edge control jumps to chat bottom · confirm card ·
   result · progress.
-- **Dispatch center.** Own page for the task graph with desk top nav; long chains
-  wrap; nodes show live status.
+- **Dispatch center.** Own page for the task graph and each project’s system
+  architecture, with console top nav; long chains wrap; nodes show live status.
 
 ### Digital employees: three roles
 
@@ -93,19 +97,19 @@ Open **Employees** in the top bar. Runtime is still Cursor Agent or Claude Code;
 
 With multiple workers, the **last lane** owns functional regression (and deploy tasks); implement work stays on earlier lanes. With one worker, the same CLI session runs roles in task order.
 
-Before a job can stay accepted, the desk runs the commands in the product `.duaer/memory/verify.json`. A non-zero exit, or a missing contract, reopens delivery. Docs-only work must set `{ "waiver": "docs-only" }` explicitly.
+Before a job can stay accepted, the console runs the commands in the product `.duaer/memory/verify.json`. A non-zero exit, or a missing contract, reopens delivery. Docs-only work must set `{ "waiver": "docs-only" }` explicitly.
 
 ### Deploy: supported hosts
 
-Before kickoff, pick a **planned host** on the desk. Coding follows that platform; the **Deployer** employee publishes and writes the public URL into delivery preview.
+Before kickoff, pick a **planned host** on the console. Coding follows that platform; the **Deployer** employee publishes and writes the public URL into delivery preview.
 
 | Option | Good for | You prepare |
 |---|---|---|
 | **No deploy yet** | Local delivery only | Nothing |
 | **GitHub Pages** | Static / docs sites; default when hosting is implied but no host was chosen | GitHub repo + `gh` auth |
-| **Cloudflare** | Pages for static; Workers for edge APIs | API Token + Account ID in desk Settings |
-| **Alibaba Cloud** | OSS (+ CDN) for static; Function Compute and similar for APIs | AccessKey in desk Settings |
-| **AWS** | S3 + CloudFront for static; Amplify / Lambda and similar when needed | Access keys in desk Settings |
+| **Cloudflare** | Pages for static; Workers for edge APIs | API Token + Account ID in console Settings |
+| **Alibaba Cloud** | OSS (+ CDN) for static; Function Compute and similar for APIs | AccessKey in console Settings |
+| **AWS** | S3 + CloudFront for static; Amplify / Lambda and similar when needed | Access keys in console Settings |
 
 Detail: [`docs/agent/deploy-targets.md`](docs/agent/deploy-targets.md) · GitHub default path: [`docs/agent/deploy-github.md`](docs/agent/deploy-github.md)
 
@@ -126,13 +130,13 @@ The sections below are **how to run it**; this section is **what it can do**.
 Step skills and CLI checks exist for agents and power users — not as the
 everyday human UI.
 
-## Duaer-spec FDE (live desk)
+## Duaer-spec FDE (live console)
 
 Isolated from product repos. Config and Briefs live under `~/.duaer/live/`.
 **Duaer-spec FDE** = Field Development Environment.
 
 ```bash
-# Start anytime (opens the desk even if the model is not configured yet)
+# Start anytime (opens the console even if the model is not configured yet)
 duaer live
 
 # In the page Settings, or via CLI:
@@ -142,21 +146,21 @@ duaer live config --provider deepseek --api-key sk-...
 duaer live repo add
 ```
 
-`duaer live` opens the browser to the desk URL (default `http://127.0.0.1:8787`).
-If the model is missing, use **Settings** in the page. The desk is **full-width**
+`duaer live` opens the browser to the console URL (default `http://127.0.0.1:8787`).
+If the model is missing, use **Settings** in the page. The console is **full-width**
 with three columns: **chat** | **modules / confirm / architecture / kickoff /
 revise** | **task progress**. Detail:
 [`docs/agent/live-desk.md`](docs/agent/live-desk.md).
 
 **Verify (maintainers / agents):** after Duaer-spec FDE UI or validate-gate changes,
-run `npm run test:live` (L3 smoke: desk shell markers + `/api/validate` with a
+run `npm run test:live` (L3 smoke: console shell markers + `/api/validate` with a
 mock LLM — no paid API).
 
 **Flow**
 
 1. Open or create a **project**, then chat. The **Projects** drawer shows each
    project’s **delivery status**; open **Deliverables** when the session has
-   content. Messy multi-topic talk is fine: the desk evolves **`modules[]`**
+   content. Messy multi-topic talk is fine: the console evolves **`modules[]`**
    (tabs). Chat bubbles render **Markdown** (`**bold**`, code, links).  
 2. Per module: fill the confirm card (goal / out-of-scope / **checkable**
    acceptance / assumptions). Vague acceptance fails the gate; validate must
@@ -190,7 +194,7 @@ mock LLM — no paid API).
    → relaunch on the same worktree (preempt leftover Agents before enqueue).
    Progress tracks this revision’s `R{n}-*` tasks. Failed launch rolls back the
    Brief Revision; status shows Terminal busy/queue per lane.  
-9. If the job needs hosting / a public URL, pick a **planned host** on the desk
+9. If the job needs hosting / a public URL, pick a **planned host** on the console
    (Cloudflare / Alibaba Cloud / AWS / GitHub Pages) so code matches that
    platform ([`docs/agent/deploy-targets.md`](docs/agent/deploy-targets.md)).
    Default when unspecified: **GitHub CLI (`gh`) + Actions**

package/README.zh-CN.md

@@ -6,11 +6,11 @@
 
 **Duaer** 把 AI 编程助手变成能**满意交付**的**数字员工**：你用自然语言说意图，它先把需求规范成可检查的 Brief（目标 / 边界 / 验收），再按范围实现，对照验收通过才算完成。
 
-主打的是 **规范 → 满意交付**——不是让人学会操作 Spec 阶段或 slash 命令。
+主打的是 **规范 → 满意交付**——不是让人学会 Spec 阶段、slash 命令，或一整套专业交付流程。
 
-你**不用**操作流程。装一次之后，在 **Cursor** / **Claude Code** / **Codex** / **Copilot** / **Windsurf** / **Cline** / **Continue** / **Gemini** / **Aider** 里说话即可。
+你**不必懂**那些流程。装一次之后，在 **Cursor** 或 **Claude Code** 里用平常话交代即可；流程由数字员工自己跑。
 
-`init` / `update` 会写入各宿主适配文件（详见 [`ADOPT.md`](ADOPT.md)）。
+`init` / `update` 还会给其它编辑器写可选适配（见 [`ADOPT.md`](ADOPT.md)）；控制台派工只启动 **Cursor Agent** 或 **Claude Code**。
 
 ## 我们有什么
 
@@ -21,29 +21,29 @@
 - **你平常怎么说，就怎么交代。** 不用记 slash 命令、不用自己点「写 Spec / 拆任务」。
 - **先谈清楚再动手。** 它会把你的想法整理成：要做什么、不做什么、怎样算做完。
 - **做完才算完。** 对照验收通过才会说「交付好了」；半成品不会假装完成。
-- **多宿主可用。** Cursor、Claude Code、Codex、Copilot、Windsurf、Cline、Continue、Gemini、Aider 都行。
+- **支持 Cursor 与 Claude Code。** 装一次后直接在里面说话，不必学专业交付流程。（其它编辑器可选适配见 [`ADOPT.md`](ADOPT.md)。）
 
-### 在浏览器里：Duaer-spec FDE（现场台面）
+### 在浏览器里：Duaer-spec FDE（现场控制台）
 
-本机开一个网页台面（`duaer live`），专门管「从聊需求到派工、看进度、看结果」：
+本机开一个网页控制台（`duaer live`），专门管「从聊需求到派工、看进度、看结果」：
 
 - **项目管理。** 建项目、切换项目；每个项目自己的对话和进度，互不搅在一起。
-- **聊着把需求聊清楚。** 左边聊天；中间把目标、边界、验收写成确认卡；空话验收过不了，得能检查才给确认。
+- **聊着把需求聊清楚。** 左边聊天；中间把目标、边界、验收写成确认卡；空话验收过不了，得能检查才给确认。发送旁可点 **录音**；在设置里配置 **语音识别模型**（兼容 OpenAI `/audio/transcriptions`，如 whisper-1）后，录完会解码成文字填入输入框，改完再发送。
 - **大需求可拆成多个模块。** 一块一块确认，确认完一块不会立刻开工，全部谈妥再一起派。
-- **系统架构图。** 确认需求后一起看架构；点图可全屏查看；架构确认后才能派工。
+- **系统架构图。** 确认需求后一起看架构；点图可全屏查看；架构确认后才能派工。若对话称已生成但中间栏没有图，同一条气泡会改成提示并提供「重新生成架构图」，不会留下假的「已生成」。
 - **一键派数字员工写代码。** 选产品目录、选 Cursor Agent 或 Claude Code，点开工；它在隔离目录里改代码，不直接乱动你的主分支。
-- **任务进度看得见。** 任务拆成一条条可勾选项；做完勾一项，台面实时刷新；有依赖的任务会排队放行，不会一窝蜂乱做。
+- **任务进度看得见。** 任务拆成一条条可勾选项；做完勾一项，控制台实时刷新；有依赖的任务会排队放行，不会一窝蜂乱做。
 - **数字员工分工。** 顶部有「数字员工」目录（实现 / 功能回归 / 部署）；可开多个并行（同一套 CLI）。详见下一节。
-- **中 / 繁中 / 英 / 日 / 韩 / 西 / 葡 / 法 / 德 / 俄 / 越界面。** 台面文案可切换语种。
+- **中 / 繁中 / 英 / 日 / 韩 / 西 / 葡 / 法 / 德 / 俄 / 越界面。** 控制台文案可切换语种。
 - **看交付物和结果。** 可打开交付物页（需求时间线、架构、任务、交付记录）；做完了可打开页面或项目文件夹。
 - **不满意就继续改。** 左侧继续聊改进 → 再确认 → 同一次任务上续派，不用从头建仓。部署后结果栏仍可「再改一版」和「修 bug」。
 - **修 bug。** 走缺陷卡（不拆多模块），默认跳过架构，短任务池，开 `fix/` 分支（可选从 `main` 出紧急修复）。交付过的地址/启动信息会自动带入；缺陷和迭代排在同一条项目时间线上。
 - **快速导航。** 对话栏右缘「导航」可跳到：对话底部 · 确认卡 · 结果 · 进度。
-- **调度中心。** 独立页看派工图，顶栏与台面一致；长任务链会换行，节点显示实时状态。
+- **调度中心。** 独立页看派工图与各项目的系统架构，顶栏与控制台一致；长任务链会换行，节点显示实时状态。
 
 ### 数字员工：三种角色
 
-台面顶部点「数字员工」可看目录。运行时仍是 Cursor Agent 或 Claude Code；差别在**任务角色**——谁做实现、谁做回归、谁做部署。
+控制台顶部点「数字员工」可看目录。运行时仍是 Cursor Agent 或 Claude Code；差别在**任务角色**——谁做实现、谁做回归、谁做部署。
 
 | 角色 | 干什么 | 不干什么 |
 |---|---|---|
@@ -53,19 +53,19 @@
 
 多人并行时：默认末位泳道承担**功能回归**（以及部署任务）；实现任务在前面的泳道。一个人时：同一 CLI 会话按任务角色依次做完。
 
-台面在放行前会自己跑产品仓 `.duaer/memory/verify.json` 里的命令。退出码不是 0，或没有这份契约，就不会保持「已交付」。纯文档任务要显式写 `{ "waiver": "docs-only" }`。
+控制台在放行前会自己跑产品仓 `.duaer/memory/verify.json` 里的命令。退出码不是 0，或没有这份契约，就不会保持「已交付」。纯文档任务要显式写 `{ "waiver": "docs-only" }`。
 
 ### 部署：支持的几种方式
 
-开工前在台面选 **计划托管平台**。选好后，数字员工按该平台约束写代码，并由**部署员工**负责上线（把公网地址写进交付预览）。
+开工前在控制台选 **计划托管平台**。选好后，数字员工按该平台约束写代码，并由**部署员工**负责上线（把公网地址写进交付预览）。
 
 | 方式 | 适合做什么 | 你要准备什么 |
 |---|---|---|
 | **暂不部署** | 只在本地做完、先不发公网 | 无 |
 | **GitHub Pages** | 静态站 / 文档站；未选平台但文案要上线时的默认 | GitHub 仓库与 `gh` 登录 |
-| **Cloudflare** | 静态站用 Pages；接口 / 边缘逻辑用 Workers | 在台面设置里保存 API Token 与 Account ID |
-| **阿里云** | 静态多用 OSS（+ CDN）；接口可用函数计算等 | 在台面设置里保存 AccessKey |
-| **AWS** | 静态多用 S3 + CloudFront；也可用 Amplify / Lambda 等 | 在台面设置里保存 Access Key |
+| **Cloudflare** | 静态站用 Pages；接口 / 边缘逻辑用 Workers | 在控制台设置里保存 API Token 与 Account ID |
+| **阿里云** | 静态多用 OSS（+ CDN）；接口可用函数计算等 | 在控制台设置里保存 AccessKey |
+| **AWS** | 静态多用 S3 + CloudFront；也可用 Amplify / Lambda 等 | 在控制台设置里保存 Access Key |
 
 细则：[`docs/agent/deploy-targets.md`](docs/agent/deploy-targets.md) · GitHub 默认路径：[`docs/agent/deploy-github.md`](docs/agent/deploy-github.md)
 
@@ -120,7 +120,7 @@ npx duaer-spec update
 **Duaer-spec FDE** = Field Development Environment（现场开发）。
 
 ```bash
-# 启动（任意目录都行，不会写入当前项目；未配模型也会打开台面）
+# 启动（任意目录都行，不会写入当前项目；未配模型也会打开控制台）
 duaer live
 
 # 在打开的页面「设置」里填模型，或用 CLI：
@@ -134,17 +134,17 @@ duaer live config --base-url https://api.openai.com/v1 --api-key sk-... --model
 ```
 
 启动后会自动打开浏览器（默认 `http://127.0.0.1:8787`）。未配置模型时请在页面
-**设置**里填写并保存。台面**全屏宽**，三列：
+**设置**里填写并保存。控制台**全屏宽**，三列：
 **对话** | **模块 / 确认 / 架构 / 开工 / 改进** | **任务进度**。细则见
 [`docs/agent/live-desk.md`](docs/agent/live-desk.md)。
 
 **验收（维护者 / 数字员工）：** 改 Duaer-spec FDE UI 或确认校验门禁后，跑
-`npm run test:live`（L3 冒烟：台面结构标记 + `/api/validate`，内置 mock 模型，
+`npm run test:live`（L3 冒烟：控制台结构标记 + `/api/validate`，内置 mock 模型，
 不产生付费 API 调用）。
 
 **流程**
 
-1. 打开或新建**项目**，再聊天。右上角「项目」列表会显示每个项目的**交付状态**；有内容时可直接打开**交付物**。话题可以乱跳：台面会演进 **`modules[]`**（模块页签）。气泡支持 **Markdown**（加粗、代码、链接）。  
+1. 打开或新建**项目**，再聊天。右上角「项目」列表会显示每个项目的**交付状态**；有内容时可直接打开**交付物**。话题可以乱跳：控制台会演进 **`modules[]`**（模块页签）。气泡支持 **Markdown**（加粗、代码、链接）。  
 2. **按模块**填确认卡（目标 / 不做 / **可检查验收** / 假设）。空泛验收过不了门禁；须校验通过才能点确认。确认只**锁住该模块卡**，**不会**立刻写 Brief 或开工。模块之间可随时切换。  
 3. 关心的模块都确认完 → 审 / 确认**架构**（Archify）→ **开工（kickoff）**。开工才写 Brief，建带 **`dependsOn` 的任务池**，并分派。默认 **1** 个数字员工；可选 **N** 个同 CLI 并行。共享任务给 1 号；模块任务轮询。每个 worker 独占一条 **Terminal 队列泳道**（`live-terminal` / `live-terminal/w2`…）。  
 4. 浏览 / 扫描 / 点选产品仓，或粘贴路径 / 项目名。路径不存在会自动创建；可先设**产品父目录**，再填短名在其下新建。非 git 目录会 `git init -b develop`；没有 develop/main/master 会建本地 `develop`。  
@@ -155,7 +155,7 @@ duaer live config --base-url https://api.openai.com/v1 --api-key sk-... --model
 6. 编辑「启动命令」（须以 `Duaer` 开头）→ 开工建 `.worktree/feat-*`，并为每条 worker 泳道打开 **Terminal**。  
 7. 右侧进度列先提示**下一步**（阶段条），并轮询 `tasks.md`。进度上方 **查看交付物** 会打开生成的 HTML 页（需求文档时间线、确认书、架构、任务池、交付等）。交付时数字员工先更新产品仓 **README**，再把 `delivery.json` 标为 `accepted`。然后出现 **结果**（有页面时 **查看结果**；否则 **打开项目目录**）。  
 8. 成品不满意：点 **继续改进（左侧对话）** → 确认改进卡（同样须校验）→ 再确认架构（保留或重画）→ 同 worktree 续派（先清残留 Agent 再入队）。进度跟本轮 `R{n}-*`。启动失败会回滚本轮 Brief Revision；状态行按泳道显示 Terminal 忙闲/队列。  
-9. 若需要上线 / 公网：在台面选 **计划托管平台**（Cloudflare / 阿里云 / AWS / GitHub Pages），说明见 [`docs/agent/deploy-targets.md`](docs/agent/deploy-targets.md)。未指定时默认 **GitHub CLI（`gh`）+ Actions**（[`docs/agent/deploy-github.md`](docs/agent/deploy-github.md)）
+9. 若需要上线 / 公网：在控制台选 **计划托管平台**（Cloudflare / 阿里云 / AWS / GitHub Pages），说明见 [`docs/agent/deploy-targets.md`](docs/agent/deploy-targets.md)。未指定时默认 **GitHub CLI（`gh`）+ Actions**（[`docs/agent/deploy-github.md`](docs/agent/deploy-github.md)）
 
 ### 文档语言
 

package/bin/duaer-live.mjs

@@ -354,6 +354,9 @@ function emptyLiveConfig() {
     baseUrl: "",
     apiKey: "",
     model: "",
+    sttBaseUrl: "",
+    sttApiKey: "",
+    sttModel: "",
     preferredAgentId: "",
     projectsRoot: "",
     activeProjectPath: "",
@@ -383,6 +386,15 @@ function readConfig() {
       baseUrl: String(raw.baseUrl || process.env.DUAER_LIVE_BASE_URL || "").trim(),
       apiKey: String(raw.apiKey || process.env.DUAER_LIVE_API_KEY || "").trim(),
       model: String(raw.model || process.env.DUAER_LIVE_MODEL || "").trim(),
+      sttBaseUrl: String(
+        raw.sttBaseUrl || process.env.DUAER_LIVE_STT_BASE_URL || "",
+      ).trim(),
+      sttApiKey: String(
+        raw.sttApiKey || process.env.DUAER_LIVE_STT_API_KEY || "",
+      ).trim(),
+      sttModel: String(
+        raw.sttModel || process.env.DUAER_LIVE_STT_MODEL || "",
+      ).trim(),
       preferredAgentId: String(raw.preferredAgentId || "").trim(),
       projectsRoot: String(raw.projectsRoot || "").trim(),
       activeProjectPath: String(raw.activeProjectPath || "").trim(),
@@ -427,6 +439,9 @@ function writeConfig(partial) {
     baseUrl: pick("baseUrl"),
     apiKey: pick("apiKey"),
     model: pick("model"),
+    sttBaseUrl: pick("sttBaseUrl"),
+    sttApiKey: pick("sttApiKey"),
+    sttModel: pick("sttModel"),
     preferredAgentId: pick("preferredAgentId"),
     projectsRoot:
       partial.projectsRoot !== undefined
@@ -452,6 +467,21 @@ function configReady(cfg = readConfig()) {
   return Boolean(cfg.baseUrl && cfg.apiKey && cfg.model);
 }
 
+function sttConfigReady(cfg = readConfig()) {
+  return Boolean(cfg.sttBaseUrl && cfg.sttApiKey && cfg.sttModel);
+}
+
+/** OpenAI-compatible audio transcriptions URL from a Base URL. */
+function sttTranscriptionsUrl(baseUrl) {
+  const base = String(baseUrl || "")
+    .trim()
+    .replace(/\/$/, "");
+  if (!base) return "";
+  if (/\/audio\/transcriptions$/i.test(base)) return base;
+  if (/\/v1$/i.test(base)) return `${base}/audio/transcriptions`;
+  return `${base}/v1/audio/transcriptions`;
+}
+
 /** Both AccessKey ID and Secret present → 阿里云 may appear in deploy UI. */
 function hasAliyunCredentials(cfg = readConfig()) {
   return Boolean(
@@ -560,6 +590,10 @@ function publicConfig(cfg = readConfig()) {
     baseUrl: cfg.baseUrl || "",
     model: cfg.model || "",
     hasApiKey: Boolean(cfg.apiKey),
+    sttReady: sttConfigReady(cfg),
+    sttBaseUrl: cfg.sttBaseUrl || "",
+    sttModel: cfg.sttModel || "",
+    hasSttApiKey: Boolean(cfg.sttApiKey),
     hasAliyunCredentials: hasAliyunCredentials(cfg),
     hasCloudflareCredentials: hasCloudflareCredentials(cfg),
     hasAwsCredentials: hasAwsCredentials(cfg),
@@ -718,7 +752,7 @@ const BUG_CHAT_PROMPT = `你是「Duaer-spec FDE」缺陷助手。用户要修 b
    - outOfScope = 本次不改什么
    - acceptance = 怎么算修好（复现关闭 + 可核对结果/命令）
    - assumptions = 环境 / 疑似原因 / 是否线上紧急（优先写入系统给的「项目交付上下文」）
-4. **禁止向用户追问**网页/成品 URL、服务启动命令或脚本、机器/端口/依赖等运行环境、疑似原因、是否线上紧急——这些由台面从项目交付自动补齐。疑似原因未知时写「待复现定位」；线上紧急默认「否」（从 develop 出 fix/），仅当用户明确说线上/生产/紧急才标「是」。
+4. **禁止向用户追问**网页/成品 URL、服务启动命令或脚本、机器/端口/依赖等运行环境、疑似原因、是否线上紧急——这些由控制台从项目交付自动补齐。疑似原因未知时写「待复现定位」；线上紧急默认「否」（从 develop 出 fix/），仅当用户明确说线上/生产/紧急才标「是」。
 5. ready=true 表示缺陷卡可确认。不要催派工。不要写代码。不要假设用户没给的仓库路径（上下文里有则直接用）。
 6. 只要问题是让用户做选择，必须在 options 填 2～5 个短选项（≤20字）。
 7. 输出格式（严格）：
@@ -795,7 +829,7 @@ const BUG_ACCEPT_PROMPT = `你是「Duaer-spec FDE」缺陷验收官。用户即
 1. goal 是否写清现象与复现（可执行，不要堆无关功能）
 2. acceptance 是否可客观检查：复现关闭 + 打开/看到/命令通过等；禁止仅「修好了/更好用」
 3. outOfScope 是否划清本次不改什么（可简短）
-4. assumptions：环境/URL/启动/紧急/疑似原因由台面交付上下文补齐——**禁止**因缺少网页地址、启动命令、运行环境、疑似原因、是否线上紧急而 failed
+4. assumptions：环境/URL/启动/紧急/疑似原因由控制台交付上下文补齐——**禁止**因缺少网页地址、启动命令、运行环境、疑似原因、是否线上紧急而 failed
 
 规则：
 - 若小改即可通过：修订四块，passed=true；assumptions 空缺时写入合理默认（疑似原因=待复现定位；线上紧急=否）
@@ -823,6 +857,69 @@ function llmTimeoutError(ms) {
   );
 }
 
+/**
+ * POST audio to OpenAI-compatible /audio/transcriptions.
+ * @param {object} cfg
+ * @param {{ buffer: Buffer, mimeType?: string, filename?: string, language?: string }} audio
+ */
+async function callSttTranscribe(cfg, audio) {
+  const url = sttTranscriptionsUrl(cfg.sttBaseUrl);
+  if (!url) throw new Error("STT baseUrl missing");
+  const mime = String(audio.mimeType || "audio/webm").trim() || "audio/webm";
+  const filename =
+    String(audio.filename || "").trim() ||
+    (mime.includes("mp4") || mime.includes("m4a")
+      ? "voice.m4a"
+      : mime.includes("mpeg") || mime.includes("mp3")
+        ? "voice.mp3"
+        : mime.includes("wav")
+          ? "voice.wav"
+          : mime.includes("ogg")
+            ? "voice.ogg"
+            : "voice.webm");
+  const form = new FormData();
+  form.append("file", new Blob([audio.buffer], { type: mime }), filename);
+  form.append("model", String(cfg.sttModel).trim());
+  const lang = String(audio.language || "").trim();
+  if (lang) form.append("language", lang);
+  const ac = new AbortController();
+  const timer = setTimeout(() => ac.abort(), LLM_TIMEOUT_MS);
+  try {
+    const res = await fetch(url, {
+      method: "POST",
+      headers: {
+        authorization: `Bearer ${cfg.sttApiKey}`,
+      },
+      body: form,
+      signal: ac.signal,
+    });
+    const text = await res.text();
+    let data = null;
+    try {
+      data = text ? JSON.parse(text) : null;
+    } catch {
+      data = null;
+    }
+    if (!res.ok) {
+      const msg =
+        (data && (data.error?.message || data.error || data.message)) ||
+        text.slice(0, 240) ||
+        `HTTP ${res.status}`;
+      throw new Error(String(msg));
+    }
+    const transcript = String(
+      data?.text || data?.transcript || data?.result || "",
+    ).trim();
+    if (!transcript) throw new Error("empty transcript");
+    return transcript;
+  } catch (err) {
+    if (err?.name === "AbortError") throw llmTimeoutError(LLM_TIMEOUT_MS);
+    throw err;
+  } finally {
+    clearTimeout(timer);
+  }
+}
+
 async function callChatModel(cfg, messages, systemPrompt = SYSTEM_PROMPT) {
   const base = cfg.baseUrl.replace(/\/$/, "");
   const url = `${base}/chat/completions`;
@@ -3817,7 +3914,7 @@ Brief: ${featureDir}
 4. 每完成 tasks.md 中的一步，立刻把该行改成 - [x]（Duaer-spec FDE 靠此显示细粒度进度与编排放行）
 4a. 只做当前编排波次已放行的任务；未满足 dependsOn / 未放行的任务不要开工；本波勾完后：若还有后续波次则输出「本波完成，退出等编排器」并立刻结束本次 CLI 会话（不要挂起；编排器用 --continue 放行）；若已无后续波次则继续完成 stamp accepted
 4b. 拆任务（强制）：必须拆到原子任务且每条可独立验证；每个勾选项只覆盖一个验收点；进度只能用 tasks.md 的 - [ ]/- [x] 监控（立刻勾选，不要攒到最后）；不要把多项验收揉进同一条；不要人为限制条数。若仍偏粗，先按 Acceptance 扩成「一条验收一勾选」（仍用 T00x），保存后再做
-5. 对照 Acceptance 全部满足后，才 stamp ${path.join(featureDir, "delivery.json")} 为 accepted——tasks.md 全部勾完还不够，必须 stamp；禁止停在 Job not accepted yet。台面会按产品仓 .duaer/memory/verify.json 自己跑 commands；退出码非 0 或缺少契约时会把 accepted 打回 open。不要把已有 commands 改成 waiver 来跳过
+5. 对照 Acceptance 全部满足后，才 stamp ${path.join(featureDir, "delivery.json")} 为 accepted——tasks.md 全部勾完还不够，必须 stamp；禁止停在 Job not accepted yet。控制台会按产品仓 .duaer/memory/verify.json 自己跑 commands；退出码非 0 或缺少契约时会把 accepted 打回 open。不要把已有 commands 改成 waiver 来跳过
 5b. 交付前必须更新产品仓 README（说明文档）：与本次交付一致——做什么、模块/验收要点、如何运行或打开；需求变了就改 README，不要只改代码。英文 README 不得出现中文；若项目是中文说明则用 README.zh-CN.md（或项目既有约定），可夹英文术语
 6. 必须在 delivery.json 写入 preview.url（满意交付的必填证据）：必须是可打开的成品入口——HTTP 服务用 http://localhost:…；静态页用 index.html 等 HTML。禁止把 docs/**/*.md 等说明文档当作 preview.url——不要因「没有页面」而省略
 6b. 若交付是 HTTP 服务：验收前必须先把服务跑起来（如 npm start），确认能打开 preview.url 后再 stamp accepted；不要只写地址却不启动
@@ -5884,7 +5981,9 @@ async function handleApi(req, res) {
         "awsRegion",
         "clearAwsCredentials",
       ];
+      const sttKeys = ["sttBaseUrl", "sttApiKey", "sttModel", "clearStt"];
       const noHostCreds = hostCredKeys.every((k) => body[k] === undefined);
+      const noStt = sttKeys.every((k) => body[k] === undefined);
       const onlyProjectsRoot =
         body.projectsRoot !== undefined &&
         body.baseUrl === undefined &&
@@ -5892,7 +5991,8 @@ async function handleApi(req, res) {
         body.model === undefined &&
         body.preferredAgentId === undefined &&
         body.activeProjectPath === undefined &&
-        noHostCreds;
+        noHostCreds &&
+        noStt;
       const onlyActiveProject =
         body.activeProjectPath !== undefined &&
         body.baseUrl === undefined &&
@@ -5900,7 +6000,8 @@ async function handleApi(req, res) {
         body.model === undefined &&
         body.preferredAgentId === undefined &&
         body.projectsRoot === undefined &&
-        noHostCreds;
+        noHostCreds &&
+        noStt;
       const onlyHostCreds =
         !noHostCreds &&
         body.baseUrl === undefined &&
@@ -5908,7 +6009,17 @@ async function handleApi(req, res) {
         body.model === undefined &&
         body.preferredAgentId === undefined &&
         body.projectsRoot === undefined &&
-        body.activeProjectPath === undefined;
+        body.activeProjectPath === undefined &&
+        noStt;
+      const onlyStt =
+        !noStt &&
+        body.baseUrl === undefined &&
+        body.apiKey === undefined &&
+        body.model === undefined &&
+        body.preferredAgentId === undefined &&
+        body.projectsRoot === undefined &&
+        body.activeProjectPath === undefined &&
+        noHostCreds;
       let projectsRoot = body.projectsRoot;
       if (projectsRoot !== undefined) {
         const raw = String(projectsRoot || "").trim();
@@ -5966,6 +6077,23 @@ async function handleApi(req, res) {
           hostPartial.awsRegion = body.awsRegion;
         }
       }
+      const sttPartial = {};
+      if (body.clearStt) {
+        sttPartial.sttBaseUrl = "";
+        sttPartial.sttApiKey = "";
+        sttPartial.sttModel = "";
+      } else {
+        if (body.sttBaseUrl !== undefined) {
+          sttPartial.sttBaseUrl = body.sttBaseUrl;
+        }
+        if (body.sttModel !== undefined) {
+          sttPartial.sttModel = body.sttModel;
+        }
+        if (body.sttApiKey !== undefined) {
+          const sttKey = String(body.sttApiKey || "").trim();
+          if (sttKey) sttPartial.sttApiKey = sttKey;
+        }
+      }
       const next = writeConfig({
         baseUrl: body.baseUrl,
         apiKey: body.apiKey,
@@ -5974,8 +6102,9 @@ async function handleApi(req, res) {
         projectsRoot,
         activeProjectPath,
         ...hostPartial,
+        ...sttPartial,
       });
-      if (onlyProjectsRoot || onlyActiveProject || onlyHostCreds) {
+      if (onlyProjectsRoot || onlyActiveProject || onlyHostCreds || onlyStt) {
         send(res, 200, publicConfig(next));
         return;
       }
@@ -5995,6 +6124,55 @@ async function handleApi(req, res) {
     return;
   }
 
+  if (req.method === "POST" && url.pathname === "/api/transcribe") {
+    const cfg = readConfig();
+    if (!sttConfigReady(cfg)) {
+      send(res, 400, {
+        error: "请先在设置中配置语音识别模型（STT Base URL / API Key / Model）",
+        code: "NEED_STT",
+        ...publicConfig(cfg),
+      });
+      return;
+    }
+    try {
+      const body = await readJson(req);
+      const b64 = String(body.audioBase64 || body.audio || "").trim();
+      if (!b64) {
+        send(res, 400, { error: "audioBase64 required" });
+        return;
+      }
+      const raw = b64.replace(/^data:[^;]+;base64,/, "");
+      let buffer;
+      try {
+        buffer = Buffer.from(raw, "base64");
+      } catch {
+        send(res, 400, { error: "invalid audioBase64" });
+        return;
+      }
+      if (!buffer.length) {
+        send(res, 400, { error: "empty audio" });
+        return;
+      }
+      if (buffer.length > 12 * 1024 * 1024) {
+        send(res, 400, { error: "audio too large (max 12MB)" });
+        return;
+      }
+      const text = await callSttTranscribe(cfg, {
+        buffer,
+        mimeType: String(body.mimeType || body.type || "").trim(),
+        filename: String(body.filename || "").trim(),
+        language: String(body.language || "").trim(),
+      });
+      send(res, 200, { text, model: cfg.sttModel });
+    } catch (err) {
+      send(res, 502, {
+        error: err instanceof Error ? err.message : "transcribe failed",
+        code: "STT_FAILED",
+      });
+    }
+    return;
+  }
+
   if (req.method === "POST" && url.pathname === "/api/chat") {
     const cfg = readConfig();
     if (!configReady(cfg)) {
@@ -7358,7 +7536,7 @@ function serve(port) {
     console.log(`Brief 写入 ${jobsRoot()}  （不会写入你当前业务仓库）`);
     if (!cfg.ready) {
       console.log(
-        `模型尚未配置 — 已打开台面，请在页面「设置」里填写模型后保存。`,
+        `模型尚未配置 — 已打开控制台，请在页面「设置」里填写模型后保存。`,
       );
     } else {
       console.log(`模型      ${cfg.model} @ ${cfg.baseUrl}`);

package/bin/live-bug-context.mjs

@@ -127,7 +127,7 @@ export function collectBugProjectContext(input = {}) {
  */
 export function formatBugProjectContextBlock(ctx) {
   const c = ctx && typeof ctx === "object" ? ctx : {};
-  const lines = ["【项目交付上下文——已从台面/交付得知，禁止再向用户追问这些项】"];
+  const lines = ["【项目交付上下文——已从控制台/交付得知，禁止再向用户追问这些项】"];
   if (c.projectPath) lines.push(`- 仓库路径: ${c.projectPath}`);
   if (c.previewUrl) lines.push(`- 成品/预览地址: ${c.previewUrl}`);
   else if (c.inferredPreviewUrl) {

package/bin/live-deliverables.mjs

@@ -419,7 +419,7 @@ export function buildDeliverablesModel(session, opts = {}) {
             ? pickLang(lang, {
                 en: "Architecture confirmed on the desk.",
                 ja: "デスクでアーキテクチャを確認済みです。",
-                zh: "已在台面确认架构。",
+                zh: "已在控制台确认架构。",
               })
             : "",
         },

package/bin/live-verify-gate.mjs

@@ -360,7 +360,7 @@ Brief: ${featureDir || ""}
 产品仓需要 \`.duaer/memory/verify.json\`。
 - 有可跑测试时写入 { "commands": ["<真实命令>"] }（例如 npm test）
 - 只有纯文档、没有可跑命令时才允许 { "waiver": "docs-only" }
-不要 stamp delivery.json 为 accepted。台面会自己复跑；失败会把 accepted 打回 open。
+不要 stamp delivery.json 为 accepted。控制台会自己复跑；失败会把 accepted 打回 open。
 `;
   }
   return `Duaer
@@ -373,6 +373,6 @@ Brief: ${featureDir || ""}
 输出尾部:
 ${nudge.tail || "(无)"}
 
-修好并使上述命令退出码为 0。不要把已有 commands 改成 waiver 来跳过。不要 stamp accepted；台面会复跑，通过后才允许 accepted。
+修好并使上述命令退出码为 0。不要把已有 commands 改成 waiver 来跳过。不要 stamp accepted；控制台会复跑，通过后才允许 accepted。
 `;
 }

package/docs/agent/e2e-test-plan.md

@@ -164,6 +164,11 @@ the table for their app.
 | E2E-192 | Project timeline merges initial/revise/bug by time; chat quick-nav jumps to 4 anchors | `npm test` + `npm run test:live` + manual |
 | E2E-193 | Chat quick-nav stays visible inside chat column (not clipped/covered) | `npm run test:live` + manual |
 | E2E-194 | After deploy, result bar still offers 再改一版 and 修 bug | `npm run test:live` + manual |
+| E2E-195 | Architecture claim-ready without IR rewrites the same chat bubble (no leftover false “diagram ready”) + regenerate | `npm run test:live` + manual |
+| E2E-196 | Dispatch center per-project view switch: Dispatch graph | Architecture (saved architecture.url) | `npm test` + manual |
+| E2E-197 | Composer Record + Send in compact `.composer-actions`; with STT configured, stop → `/api/transcribe` fills `#input`; without STT, Record opens Settings speech-model block | `npm run test:live` + manual |
+| E2E-198 | Settings speech recognition model block saves `sttBaseUrl` / `sttApiKey` / `sttModel`; publicConfig exposes `sttReady` | `npm test` + `npm run test:live` |
+| E2E-199 | Record click with no STT shows bot guide and opens Settings → speech model | `npm run test:live` + manual |
 
 | E2E-156 | Accepted chat bubble preview URL is a clickable `<a>` (http(s) or /api/…) | `npm test` + `npm run test:live` |
 | E2E-157 | `http://host/api/result/…` chat links render as one clean `<a>` (no nested HTML garble) | `npm test` |
@@ -393,4 +398,9 @@ the table for their app.
 | E2E-192 | `247-timeline-quicknav` | timeline sort + chat quick nav |
 | E2E-193 | `248-quicknav-visible` | quick-nav not clipped by overflow |
 | E2E-194 | `249-result-cta-after-deploy` | result CTAs remain after deploy |
+| E2E-195 | `251-arch-claim-no-diagram` | rewrite claim-ready bubble when IR missing |
+| E2E-196 | `255-dispatch-arch-view` | dispatch center per-project architecture view |
+| E2E-197 | `256-chat-voice` / `257-composer-voice-compact` | composer voice / Record beside Send |
+| E2E-198 | `258-chat-stt` | STT model settings + /api/transcribe |
+| E2E-199 | `259-voice-guide-stt` | Record without STT opens Settings |