@ww-ai-lab/openclaw-office 2026.4.13 → 2026.5.10-beta.1

package/README.en.md

@@ -43,6 +43,7 @@ Full system management interface with dedicated pages:
 | **Agents**    | Agent list/create/delete, detail tabs (Overview, Channels, Cron, Skills, Tools, Files)                         |
 | **Channels**  | Channel cards, configuration dialogs, stats, WhatsApp QR binding                                               |
 | **Skills**    | Skill marketplace, install options, skill detail dialogs                                                       |
+| **Skill Workbench** ✨ | **Skills development platform** — nested routes (list / create / detail), AI-assisted creation and editing via side chat, automated FLOWCHART.md generation with pure-Markdown multi-chart preview |
 | **Cron**      | Scheduled task management and statistics                                                                       |
 | **Settings**  | Provider management (add/edit/model editor), appearance, Gateway, developer, advanced, about, update            |
 
@@ -60,6 +61,20 @@ Full system management interface with dedicated pages:
 
 ---
 
+## New: Skill Workbench (Skills Development Platform)
+
+Starting from `2026.5.10-beta.1`, the console ships a new **Skill Workbench** — a full AI-assisted development environment for Skills:
+
+- **Nested three-page routing**: `/skill-workbench` (list), `/skill-workbench/create` (wizard), `/skill-workbench/:slug` (detail)
+- **Chat-side-panel driven development**: create and modify Skills through a conversation; file changes are flushed to disk in real time
+- **Automated FLOWCHART.md generation**: a built-in guard skill (`skill-workbench-mermaid-guard`) produces compliant, colored Mermaid flowcharts (single- or multi-chart mode) on demand
+- **Pure-Markdown multi-chart preview**: the flowchart panel renders multiple Mermaid fenced blocks directly through the Markdown pipeline
+- **Zero-setup default skill install**: on first entry, the embedded server copies bundled default skills from the npm package into `~/.openclaw/workspace/skills/` — no manual install required
+
+See the full guide in [SKILL-WORKBENCH.md](./SKILL-WORKBENCH.md).
+
+---
+
 ## Tech Stack
 
 | Layer            | Technology                                      |

package/README.md

@@ -43,6 +43,7 @@
 | **Agents**    | Agent 列表/创建/删除，详情多 Tab（Overview/Channels/Cron/Skills/Tools/Files）                                        |
 | **Channels**  | 渠道卡片、配置对话框、统计、WhatsApp QR 绑定流程                                                                     |
 | **Skills**    | 技能市场、安装选项、技能详情                                                                                         |
+| **Skill Workbench** ✨ | **Skills 开发平台** — 三页面（列表 / 创建 / 详情）嵌套路由，聊天侧边栏驱动 AI 协作创建和修改 Skill，FLOWCHART.md 自动生成并提供纯 Markdown 多图预览 |
 | **Cron**      | 定时任务管理和统计                                                                                                   |
 | **Settings**  | Provider 管理（添加/编辑/模型编辑器）、外观/Gateway/开发者/高级/关于/更新                                            |
 
@@ -60,6 +61,20 @@
 
 ---
 
+## 新特性：Skill Workbench（Skills 开发平台）
+
+从 `2026.5.10-beta.1` 起，控制台新增了 **Skill Workbench**（Skills 开发平台），把「创建 Skill / 修改 Skill / 生成流程图」整合成一个完整的 AI 协作开发环境：
+
+- **三页面嵌套路由**：`/skill-workbench` 列表页、`/skill-workbench/create` 创建向导、`/skill-workbench/:slug` 详情页
+- **聊天侧边栏驱动开发**：在侧边栏和 AI 对话即可创建、修改 Skill，文件变动实时同步磁盘
+- **FLOWCHART.md 自动生成**：内置 [`skill-workbench-mermaid-guard`](./SKILL-WORKBENCH.md) 守卫技能，一键生成符合规范的彩色 Mermaid 流程图，支持单图与多图模式
+- **纯 Markdown 多图预览**：流程图预览完全走 Markdown 渲染链路，一次渲染多个 Mermaid 代码块
+- **默认技能自动安装**：首次进入工作台时，嵌入式服务端会自动把 npm 包内置的默认 Skill 拷贝到 `~/.openclaw/workspace/skills/`，无需手工安装
+
+详细使用指南见 [SKILL-WORKBENCH.md](./SKILL-WORKBENCH.md)。
+
+---
+
 ## 技术栈
 
 | 层       | 技术                                        |

package/SKILL-WORKBENCH.md

@@ -0,0 +1,173 @@
+# Skill Workbench — Skills 开发平台使用指南
+
+> 一个面向 OpenClaw Skills 的可视化开发平台，集成在 OpenClaw Office 控制台中。
+> 发布版本：`2026.5.10-beta.1` 起正式提供。
+
+---
+
+## 1. 定位
+
+**Skill Workbench** 把"Skill 开发"从写 Markdown 的纯文本工作流，升级为一个**聊天驱动 + 可视预览 + 一键生成流程图**的开发平台：
+
+- 所有产物都是普通的 Markdown / YAML 文件，落盘在 `~/.openclaw/workspace/skills/<skill-slug>/` 下
+- 聊天侧边栏直接挂载到 OpenClaw Gateway，借助 AI 与本地 Skill 文件工具完成增删改
+- 详情页提供 `SKILL.md`、`FLOWCHART.md` 及其它自定义文件的切换浏览与原地编辑
+- 流程图走纯 Markdown 渲染链路，天然支持**一份 FLOWCHART.md 内多个 Mermaid 代码块**
+
+可以把它理解为 "OpenClaw Skills 版的 Notion + VS Code"：写作在前，文件在后。
+
+---
+
+## 2. 入口与路由
+
+工作台采用**嵌套路由三页面**结构：
+
+| 路径                                | 页面             | 作用                                                           |
+| ----------------------------------- | ---------------- | -------------------------------------------------------------- |
+| `/skill-workbench`                  | 列表首页         | 展示本地已安装的 Skills，提供创建入口和搜索                    |
+| `/skill-workbench/create`           | 创建向导         | 与 AI 对话，基于需求描述生成全新的 Skill 骨架与 `SKILL.md`     |
+| `/skill-workbench/:slug`            | 详情页           | 浏览 / 编辑某个 Skill 的全部文件；默认展示 `FLOWCHART.md` 预览 |
+
+从控制台左侧导航即可进入；子路由之间通过 `AppShell` 内部的 `Outlet` 切换，不会打断底部 Chat Dock 的会话状态。
+
+---
+
+## 3. 创建一个新 Skill
+
+1. 进入 `/skill-workbench`，点击 **"新建 Skill"**。
+2. 在创建向导页的侧边栏与 AI 对话描述你的需求（例如"帮我生成一个能把长文章总结成三点摘要的 Skill"）。
+3. AI 会基于内置 [`skill-workbench-creator`] 默认技能，产出：
+   - 规范的 `SKILL.md`（带 YAML frontmatter：`name`、`description`）
+   - 初版工作流程描述
+4. 保存后 Skill 立即落盘到 `~/.openclaw/workspace/skills/<slug>/`，并在列表页可见。
+5. 首次保存后通常**紧接着让 AI 生成 `FLOWCHART.md`**（见第 5 节）。
+
+---
+
+## 4. 编辑已有 Skill
+
+在 `/skill-workbench/:slug` 详情页：
+
+- **左侧文件树** — 列出该 Skill 目录下所有文件，点击即可浏览；`FLOWCHART.md` 始终排在最上方
+- **中间主视图** — 根据选中项切换：
+  - 选中 `FLOWCHART.md` → 进入纯 Markdown 多图预览
+  - 选中其它文件 → 进入 `SkillFileViewer` 的原地编辑器
+- **右侧聊天栏（可折叠）** — 点击 "修改此 Skill" 开启专用会话：
+  - 会话 key 会带时间戳，保证与历史对话隔离
+  - 会话启动时会注入当前 Skill 目录与默认守卫技能作为 system 上下文
+  - 用自然语言描述修改点，AI 会直接用 `read` / `write` / `edit` 工具改盘
+
+所有变更都是**即时落盘**的——刷新页面即可看到最新内容；不需要保存按钮。
+
+---
+
+## 5. 流程图：FLOWCHART.md 一键生成
+
+### 交互
+
+在详情页选中 **FLOWCHART.md**：
+
+- 若该 Skill 尚未有 `FLOWCHART.md`，预览区会显示 **"一键生成流程图"** 按钮
+- 点击后工作台会：
+  1. 自动发送符合 `skill-workbench-mermaid-guard` 规范的生成指令
+  2. 流式写出 Mermaid 源
+  3. 把完整的 `FLOWCHART.md` 写回到磁盘
+  4. 预览区自动刷新为最终渲染结果
+
+### 纯 Markdown 多图预览
+
+预览面板不再做特殊裁剪：**整份 `FLOWCHART.md` 作为 Markdown 文档渲染**，内部的每一个 ```mermaid 代码块都会各自被 `MermaidPreview` 渲染成独立 SVG。这意味着：
+
+- 一个 Skill 可以拥有一张总览图 + 若干分阶段子图
+- 每个代码块前可以加 `##` 二级标题，互相之间可以穿插文字说明
+- 预览体验与 GitHub / VS Code 等 Markdown 预览保持一致
+
+### 颜色 / 结构规范
+
+流程图生成严格遵循内置守卫技能 `skill-workbench-mermaid-guard`：
+
+| 节点类型       | classDef    | 颜色含义 | 使用场景                                          |
+| -------------- | ----------- | -------- | ------------------------------------------------- |
+| 开始节点       | `startNode` | 绿色     | 流程入口                                          |
+| 成功结束       | `endOk`     | 蓝色     | 正常/成功出口                                     |
+| 失败/错误结束  | `endErr`    | 红色     | 错误/拒绝出口                                     |
+| 决策节点       | `decision`  | 琥珀色   | `{...}` 菱形判断                                  |
+| 普通步骤       | `process`   | 浅蓝     | 执行 / 操作节点                                   |
+| 阶段节点       | `phase`     | 紫色     | 总览图中代表一个阶段（内部可能展开为一张子流程图）|
+
+---
+
+## 6. 默认技能的自动安装
+
+工作台依赖两个"默认技能"注入到 AI 的 system 上下文：
+
+| Skill slug                           | 作用                                                  |
+| ------------------------------------ | ----------------------------------------------------- |
+| `skill-workbench-creator`            | 引导 AI 生成规范的 `SKILL.md` 骨架                    |
+| `skill-workbench-mermaid-guard` ✨   | 保障 `FLOWCHART.md` 输出的格式 / 颜色 / 规范           |
+
+从 `2026.5.10-beta.1` 起：
+
+- `skill-workbench-mermaid-guard` 的完整源码随 npm 包一起发布（放在 `bin/skills/` 目录下）
+- 嵌入式 Node 服务端新增 `POST /api/workspace-skills/ensure-defaults` 端点
+- 前端在每次打开 / 修改工作台（调用 `enterWorkbench()`）之前，都会先调一次该端点
+- 如果 `~/.openclaw/workspace/skills/<slug>/SKILL.md` 不存在，服务端会**自动递归拷贝**内置版本到用户工作区
+- 已存在则不做任何事（幂等）；整个过程对用户无感
+
+如果你手动删掉了 `~/.openclaw/workspace/skills/skill-workbench-mermaid-guard/`，下一次进入工作台会被重新生成。
+
+---
+
+## 7. 与 OpenClaw Gateway 的协作关系
+
+工作台的数据流如下：
+
+```
+Browser ── /api/workspace-skills/*  ──►  Embedded Node server (bin/openclaw-office.js)
+    │                                           │
+    │                                           └─► Local FS: ~/.openclaw/workspace/skills/
+    │
+    └── WebSocket /gateway-ws ─────────►  OpenClaw Gateway (chat & tool calls)
+```
+
+- **文件 CRUD** 直接走嵌入式服务端，读写 `~/.openclaw/workspace/skills/`；完全不经 Gateway
+- **聊天 / 工具调用** 走 Gateway WebSocket；AI 使用 Gateway 提供的 `read` / `write` / `edit` 等工具直接改本地文件
+- **Git 提交**（可选）由嵌入式服务端 `commit` 端点完成，默认提交消息为 `chore(skill): update <slug>/<file> via OpenClaw Office`
+
+---
+
+## 8. 常见问题
+
+**Q: 流程图生成一直卡在 "生成中…"？**
+A: 检查 Gateway 连接状态（顶部应为 Connected），并确认所选 Agent 有可用的模型 Provider。生成完成的标志是 `FLOWCHART.md` 被写入磁盘；若 AI 只是回复了解释文字但没有真正写文件，请再次点击 "一键生成"，守卫技能会强制直接调用 write 工具。
+
+**Q: 自动安装默认技能失败怎么办？**
+A: 进入工作台时浏览器控制台会打印 `[skill-workbench] ensure default skills failed:` 警告。失败不会阻塞工作台本身；你可以手动从本项目的 `bin/skills/skill-workbench-mermaid-guard/` 拷贝到 `~/.openclaw/workspace/skills/` 作为 fallback。
+
+**Q: 我想用我自己的守卫技能替换默认的，行吗？**
+A: 可以。只要在 `~/.openclaw/workspace/skills/skill-workbench-mermaid-guard/SKILL.md` 存在自定义实现，`ensure-defaults` 就不会覆盖——它只在缺失时安装。
+
+**Q: 文件跟不上磁盘变化？**
+A: 详情页的文件树与 FLOWCHART 预览会在每一轮聊天流结束时自动刷新；若仍有偏差，切换选中的文件或返回列表页再进入即可强制重新拉取。
+
+---
+
+## 9. 相关文件
+
+- 前端入口：[src/components/pages/SkillWorkbenchPage.tsx](./src/components/pages/)、`SkillWorkbenchCreatePage.tsx`、`SkillWorkbenchDetailPage.tsx`
+- 工作台状态：[src/store/console-stores/skill-workbench-store.ts](./src/store/console-stores/skill-workbench-store.ts)
+- 嵌入式 API：[bin/openclaw-office.js](./bin/openclaw-office.js) 中的 `handleWorkspaceSkillsApi`
+- 内置默认技能：[bin/skills/skill-workbench-mermaid-guard/](./bin/skills/skill-workbench-mermaid-guard/)
+- 前端客户端：[src/gateway/workspace-skills-client.ts](./src/gateway/workspace-skills-client.ts)
+
+---
+
+## 10. 版本记录
+
+- **2026.5.10-beta.1** — 首次对外发布 Skill Workbench
+  - 三页面嵌套路由
+  - FLOWCHART.md 纯 Markdown 多图预览
+  - 内置 `skill-workbench-mermaid-guard` 并提供 `ensure-defaults` 自动安装
+  - 创建模式首轮 session & skill 注入修复
+
+[`skill-workbench-creator`]: https://github.com/openclaw/openclaw

package/bin/openclaw-office.js

@@ -3,10 +3,11 @@
 import { createServer, request as httpRequest } from "node:http";
 import { request as httpsRequest } from "node:https";
 import { readFileSync, existsSync, mkdirSync } from "node:fs";
-import { readFile, writeFile, access, readdir, unlink, mkdir } from "node:fs/promises";
-import { resolve, join, extname } from "node:path";
+import { readFile, writeFile, access, readdir, unlink, mkdir, stat } from "node:fs/promises";
+import { resolve, join, extname, dirname, relative } from "node:path";
 import { fileURLToPath } from "node:url";
 import { networkInterfaces, homedir } from "node:os";
+import { execFile } from "node:child_process";
 
 const __dirname = fileURLToPath(new URL(".", import.meta.url));
 const distDir = resolve(__dirname, "..", "dist");
@@ -159,6 +160,9 @@ const runtimeConfig = JSON.stringify({
 });
 const configScript = `<script>window.__OPENCLAW_CONFIG__=${runtimeConfig};</script>`;
 const gatewayWsPrefixes = new Set(["/gateway-ws", "/api/gateway/ws"]);
+const WORKSPACE_SKILLS_DIR = join(homedir(), ".openclaw", "workspace", "skills");
+const BUNDLED_SKILLS_DIR = resolve(__dirname, "skills");
+const DEFAULT_BUNDLED_SKILL_SLUGS = ["skill-workbench-mermaid-guard"];
 
 async function tryReadFile(filePath) {
   try {
@@ -379,6 +383,161 @@ async function safeReaddir(dir) {
   }
 }
 
+function ensureInside(baseDir, targetPath) {
+  const normalizedBase = `${resolve(baseDir)}/`;
+  const normalizedTarget = resolve(targetPath);
+  if (normalizedTarget !== resolve(baseDir) && !normalizedTarget.startsWith(normalizedBase)) {
+    throw new Error("Invalid workspace skill path");
+  }
+  return normalizedTarget;
+}
+
+async function listWorkspaceSkillFiles(skillSlug) {
+  const skillDir = ensureInside(WORKSPACE_SKILLS_DIR, join(WORKSPACE_SKILLS_DIR, skillSlug));
+  const entries = await readdir(skillDir, { withFileTypes: true });
+  const files = [];
+
+  for (const entry of entries) {
+    const fullPath = join(skillDir, entry.name);
+    if (entry.isDirectory()) {
+      const childFiles = await listWorkspaceSkillFiles(join(skillSlug, entry.name));
+      files.push(...childFiles.map((file) => ({
+        ...file,
+        name: `${entry.name}/${file.name}`,
+      })));
+      continue;
+    }
+
+    const info = await stat(fullPath);
+    files.push({
+      name: entry.name,
+      size: info.size,
+      modifiedAt: info.mtimeMs,
+    });
+  }
+
+  return files.sort((left, right) => left.name.localeCompare(right.name));
+}
+
+async function readWorkspaceSkillFile(skillSlug, name) {
+  const skillDir = ensureInside(WORKSPACE_SKILLS_DIR, join(WORKSPACE_SKILLS_DIR, skillSlug));
+  const filePath = ensureInside(skillDir, join(skillDir, name));
+  const [content, info] = await Promise.all([
+    readFile(filePath, "utf-8"),
+    stat(filePath),
+  ]);
+
+  return {
+    name,
+    content,
+    size: info.size,
+    modifiedAt: info.mtime.toISOString(),
+  };
+}
+
+async function writeWorkspaceSkillFile(skillSlug, name, content) {
+  const skillDir = ensureInside(WORKSPACE_SKILLS_DIR, join(WORKSPACE_SKILLS_DIR, skillSlug));
+  const filePath = ensureInside(skillDir, join(skillDir, name));
+  await mkdir(dirname(filePath), { recursive: true });
+  await writeFile(filePath, content, "utf-8");
+  const info = await stat(filePath);
+  return { name, size: info.size, modifiedAt: info.mtime.toISOString() };
+}
+
+async function findGitRoot(startDir) {
+  let current = resolve(startDir);
+  for (let i = 0; i < 10; i++) {
+    try {
+      await access(join(current, ".git"));
+      return current;
+    } catch {
+      /* walk up */
+    }
+    const parent = dirname(current);
+    if (parent === current) return null;
+    current = parent;
+  }
+  return null;
+}
+
+function runGit(repoRoot, args) {
+  return new Promise((resolveRun) => {
+    execFile(
+      "git",
+      ["-C", repoRoot, ...args],
+      { timeout: 5000, encoding: "utf-8" },
+      (err, stdout, stderr) => {
+        const code = err && typeof err.code === "number" ? err.code : err ? 1 : 0;
+        resolveRun({ stdout: String(stdout ?? ""), stderr: String(stderr ?? ""), code });
+      },
+    );
+  });
+}
+
+async function copyDirRecursive(srcDir, destDir) {
+  await mkdir(destDir, { recursive: true });
+  const entries = await readdir(srcDir, { withFileTypes: true });
+  for (const entry of entries) {
+    const srcPath = join(srcDir, entry.name);
+    const destPath = join(destDir, entry.name);
+    if (entry.isDirectory()) {
+      await copyDirRecursive(srcPath, destPath);
+    } else if (entry.isFile()) {
+      const data = await readFile(srcPath);
+      await writeFile(destPath, data);
+    }
+  }
+}
+
+async function ensureBundledSkillInstalled(skillSlug) {
+  const targetDir = ensureInside(WORKSPACE_SKILLS_DIR, join(WORKSPACE_SKILLS_DIR, skillSlug));
+  const targetSkillMd = join(targetDir, "SKILL.md");
+  try {
+    await access(targetSkillMd);
+    return { slug: skillSlug, status: "present" };
+  } catch {
+    // Not installed yet — attempt to install from bundled source.
+  }
+
+  const bundledDir = resolve(BUNDLED_SKILLS_DIR, skillSlug);
+  const bundledSkillMd = join(bundledDir, "SKILL.md");
+  try {
+    await access(bundledSkillMd);
+  } catch {
+    return { slug: skillSlug, status: "missing_bundle" };
+  }
+
+  try {
+    await copyDirRecursive(bundledDir, targetDir);
+    return { slug: skillSlug, status: "installed" };
+  } catch (err) {
+    return { slug: skillSlug, status: "error", error: String(err?.message ?? err) };
+  }
+}
+
+async function commitWorkspaceSkillFile(skillSlug, name, message) {
+  const skillDir = ensureInside(WORKSPACE_SKILLS_DIR, join(WORKSPACE_SKILLS_DIR, skillSlug));
+  const filePath = ensureInside(skillDir, join(skillDir, name));
+  const repoRoot = await findGitRoot(dirname(filePath));
+  if (!repoRoot) return { committed: false, reason: "not_a_git_repo" };
+
+  const rel = relative(repoRoot, filePath);
+  const add = await runGit(repoRoot, ["add", "--", rel]);
+  if (add.code !== 0) {
+    return { committed: false, reason: "failure", error: add.stderr || add.stdout };
+  }
+  const commit = await runGit(repoRoot, ["commit", "--only", "--", rel, "-m", message]);
+  if (commit.code !== 0) {
+    const out = `${commit.stdout}\n${commit.stderr}`.toLowerCase();
+    if (out.includes("nothing to commit") || out.includes("no changes added")) {
+      return { committed: false, reason: "nothing_to_commit" };
+    }
+    return { committed: false, reason: "failure", error: commit.stderr || commit.stdout };
+  }
+  const rev = await runGit(repoRoot, ["rev-parse", "--short", "HEAD"]);
+  return { committed: true, commit: rev.stdout.trim() };
+}
+
 async function readSessionMessages(sessionDir) {
   const files = await safeReaddir(sessionDir);
   const dayFiles = files.filter(isDayFile).sort();
@@ -568,6 +727,104 @@ async function handleChatCacheApi(req, res, pathname) {
   return false;
 }
 
+async function handleWorkspaceSkillsApi(req, res, pathname) {
+  if (req.method === "OPTIONS") {
+    res.writeHead(204, {
+      "Access-Control-Allow-Origin": "*",
+      "Access-Control-Allow-Methods": "GET, POST, OPTIONS",
+      "Access-Control-Allow-Headers": "Content-Type",
+    });
+    res.end();
+    return true;
+  }
+
+  const url = new URL(req.url || "/", `http://${req.headers.host}`);
+
+  if (pathname === "/api/workspace-skills/list" && req.method === "GET") {
+    const skillSlug = url.searchParams.get("slug");
+    if (!skillSlug) {
+      sendJson(res, 400, { error: "Missing slug" });
+      return true;
+    }
+
+    const files = await listWorkspaceSkillFiles(skillSlug);
+    sendJson(res, 200, {
+      agentId: skillSlug,
+      workspace: "~/.openclaw/workspace/skills",
+      files,
+    });
+    return true;
+  }
+
+  if (pathname === "/api/workspace-skills/file" && req.method === "GET") {
+    const skillSlug = url.searchParams.get("slug");
+    const name = url.searchParams.get("name");
+    if (!skillSlug || !name) {
+      sendJson(res, 400, { error: "Missing slug or name" });
+      return true;
+    }
+
+    const file = await readWorkspaceSkillFile(skillSlug, name);
+    sendJson(res, 200, {
+      agentId: skillSlug,
+      workspace: "~/.openclaw/workspace/skills",
+      file,
+    });
+    return true;
+  }
+
+  if (pathname === "/api/workspace-skills/save" && req.method === "POST") {
+    const body = await readRequestBody(req);
+    const skillSlug = typeof body.slug === "string" ? body.slug : "";
+    const name = typeof body.name === "string" ? body.name : "";
+    const content = typeof body.content === "string" ? body.content : null;
+    if (!skillSlug || !name || content === null) {
+      sendJson(res, 400, { error: "Missing slug, name or content" });
+      return true;
+    }
+    const info = await writeWorkspaceSkillFile(skillSlug, name, content);
+    sendJson(res, 200, { agentId: skillSlug, file: info });
+    return true;
+  }
+
+  if (pathname === "/api/workspace-skills/commit" && req.method === "POST") {
+    const body = await readRequestBody(req);
+    const skillSlug = typeof body.slug === "string" ? body.slug : "";
+    const name = typeof body.name === "string" ? body.name : "";
+    const message =
+      typeof body.message === "string" && body.message.trim().length > 0
+        ? body.message
+        : `chore(skill): update ${skillSlug}/${name} via OpenClaw Office`;
+    if (!skillSlug || !name) {
+      sendJson(res, 400, { error: "Missing slug or name" });
+      return true;
+    }
+    const result = await commitWorkspaceSkillFile(skillSlug, name, message);
+    sendJson(res, 200, result);
+    return true;
+  }
+
+  if (pathname === "/api/workspace-skills/ensure-defaults" && req.method === "POST") {
+    const body = await readRequestBody(req);
+    const rawSlugs = Array.isArray(body.slugs) ? body.slugs : null;
+    const slugs = (rawSlugs && rawSlugs.length > 0 ? rawSlugs : DEFAULT_BUNDLED_SKILL_SLUGS)
+      .filter((s) => typeof s === "string" && s.length > 0 && !s.includes("/") && !s.includes(".."));
+    await mkdir(WORKSPACE_SKILLS_DIR, { recursive: true });
+    const results = [];
+    for (const slug of slugs) {
+      try {
+        results.push(await ensureBundledSkillInstalled(slug));
+      } catch (err) {
+        results.push({ slug, status: "error", error: String(err?.message ?? err) });
+      }
+    }
+    sendJson(res, 200, { results });
+    return true;
+  }
+
+  return false;
+}
+
 const server = createServer(async (req, res) => {
   const url = new URL(req.url || "/", `http://${req.headers.host}`);
   const pathname = decodeURIComponent(url.pathname);
@@ -583,6 +840,16 @@ const server = createServer(async (req, res) => {
     }
   }
 
+  if (pathname.startsWith("/api/workspace-skills/")) {
+    try {
+      const handled = await handleWorkspaceSkillsApi(req, res, pathname);
+      if (handled) return;
+    } catch (err) {
+      sendJson(res, 500, { error: String(err) });
+      return;
+    }
+  }
+
   // Serve injected index.html for root and SPA routes
   if (pathname === "/" || pathname === "/index.html") {
     const html = await getIndexHtml();