docs: add annotated screenshots with arrows (README + guides)

Check off completed v0.6.x items and add new planned features:
agent inline editing, gateway restart button, settings tab, bot connectivity test.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

DEVLOG.md

@@ -1,7 +1,239 @@
 # OpenClaw Manager — 开发日志
 
-> 最后更新：2026-02-25
+> 最后更新：2026-02-27
-> 当前版本：v0.5.2
+> 当前版本：v0.7.1
+
+---
+
+## v0.7.1 更新日志（2026-02-27）
+
+### New Features
+
+**Dashboard Guidance Block (Telegram-first usage + safety)**
+- Added a concise guidance card under Dashboard explaining OCM's intended Telegram workflow
+- Clarifies prerequisites: users should already have basic OpenClaw CLI experience
+- Clarifies purpose: OCM mainly helps visualize and update `openclaw.json` for easier main-agent/sub-agent management and multi-tree structures
+- Added explicit Telegram security checklist:
+  - BotFather `Allow Groups = ON`
+  - BotFather `Group Privacy = OFF`
+  - Keep each group private (only you + agent/sub-agents), do not invite others
+
+### Improvements
+
+**Model Dropdown Source — only `openclaw models list`**
+- Model dropdown options are now sourced from real CLI output (`openclaw models list`) instead of built-in/static lists or config-only additions
+- Applies to all relevant selectors: agent/sub-agent creation forms, agent inline model switcher, primary model selector, and fallback picker
+- Added warning message when model list loading/parsing fails
+
+### Technical Notes
+
+- Added model ID parser for CLI output and short-term cache for model list fetches
+- `GET /api/models` now returns:
+  - `knownModels` from CLI
+  - `modelListError` for UI warning display
+- Startup script banners updated to `v0.7.1` (`start.sh` / `start.bat`)
+
+---
+
+## v0.7.0 更新日志（2026-02-27）
+
+### New Features
+
+**Dashboard Redesign — Circular Gauge System Health**
+- CPU%, RAM%, DISK% displayed as SVG circular gauges with colour-coded arcs (green < 70%, amber < 90%, red ≥ 90%)
+- System info card: hostname, OS, Node.js, CPU model, cores, uptime, load average (1/5/15 min)
+- Agent overview: separate main agent count and sub-agent count (plus total)
+- Auto-refresh toggle: when enabled, dashboard polls `/api/dashboard` every 10 seconds
+- Backend now measures CPU usage via `os.cpus()` delta sampling (200ms interval) and returns `loadAvg` from `os.loadavg()`
+
+**Sub-Agent Creation Flow — Step-by-Step Guide Rewrite**
+- Expanded from 3 steps to 5 detailed steps covering the full BotFather → Telegram Group → OCM workflow
+- Step 1: Create new Bot via `/newbot` in BotFather
+- Step 2: Disable Group Privacy via `/mybots` → Bot Settings → Group Privacy → Turn off
+- Step 3: Create Telegram group, add Bot (with security warning)
+- Step 4: Get Group ID from gateway logs
+- Step 5: Fill in the form
+- New "Your Telegram User ID" input field — auto-writes to `channels.telegram.allowFrom` whitelist on creation
+- Security warning banner: "Do NOT add other people to this group" with explanation about API cost risks
+
+**Setup Page — English Localization**
+- All Chinese text in the initial directory selection page translated to English
+- Includes: title, description, labels, placeholders, error messages, button text
+
+### Bug Fixes
+
+**Agent tree: sub-agents not grouped under main (old config format)**
+- **Symptom**: On machines with openclaw.json created before OCM v0.6 multi-bot support, all sub-agents displayed as independent cards instead of nested under main
+- **Root cause**: Old-format bindings lack `accountId` field in sub-agent entries. The tree builder relied on `parentAccountId` to match sub-agents to roots, but old bindings only had `channel` + `peer` — no `accountId`. So `parentAccountId` was null and no parent match was found
+- **Fix**: Fallback inference — if a sub-agent has a peer binding but no `accountId`, automatically infer `parentAccountId` from main's binding (`accountId`) or fall back to `'default'`
+
+### Technical Notes
+
+- **CPU usage measurement:** Two `os.cpus()` snapshots 200ms apart, calculating idle-to-total ratio across all cores
+- **Gauge rendering:** Pure SVG arcs with `stroke-dasharray` animation, no external libraries. 270° arc (gap at bottom), colour transitions via `gaugeColor()` function
+- **allowFrom auto-config:** `POST /api/agents` now accepts optional `telegramUserId` field (numeric string), appends to `channels.telegram.allowFrom[]` if not already present
+- **Old-config compatibility:** Sub-agent bindings without `accountId` are now auto-assigned to the main agent's bot, maintaining backward compatibility with configs created before multi-bot support
+
+---
+
+## v0.6.6 更新日志（2026-02-27）
+
+### Bug Fixes
+
+**Agent tree: main agent not detected as root**
+- **Symptom**: Sub-agents displayed as independent agents instead of under their parent. All agents appeared as orphan roots.
+- **Root cause**: `main` agent in OpenClaw is the default catch-all and typically has NO explicit binding in `openclaw.json`. The `hasOwnBot` detection required an explicit binding with `accountId && !peer`, so `main` was classified as a non-root. Without `main` being a root, sub-agents' `parentAccountId` had no matching root to link to.
+- **Fix**:
+  - `main` is now always treated as a root agent regardless of binding existence
+  - `main`'s `accountId` is inferred from `channels.telegram.accounts`: first unclaimed account (not explicitly bound to another agent), or first account as fallback
+  - Sub-agents with `parentAccountId` matching `main`'s inferred `accountId` now correctly appear under `main` in the tree
+
+---
+
+## v0.6.5 更新日志（2026-02-26）
+
+### New Features
+
+**Dashboard Tab (🏠 Dashboard)**
+- New default landing tab showing system overview at a glance
+- **System card**: hostname, OS, Node.js version, CPU model/cores, uptime, memory usage with progress bar
+- **Gateway card**: process status indicator (running/stopped/unknown) with coloured dot, port, PID, HTTP ping reachability
+- **Agents card**: total agent count, last session activity timestamp (Brisbane time), OCM version, server time
+- **Storage card**: OpenClaw dir size, disk usage with progress bar, free space
+- New `GET /api/dashboard` endpoint aggregates all info (ps grep + curl ping + fs stat)
+- Lazy-loaded on tab switch; auto-loaded on first page load
+
+**Cache-Control + Version-Based Cache Busting**
+- All HTTP responses now include `Cache-Control: no-store, no-cache, must-revalidate`
+- HTML responses include `ETag: "ocm-<version>"` for version-based cache validation
+- All responses include `X-OCM-Version` header
+- Browser `api()` function checks `X-OCM-Version` against client version; shows toast notification when server has been updated, prompting user to refresh
+- Prevents stale frontend from calling deleted/changed API endpoints after update
+
+### Cleanup
+
+**Old `/api/agents/main` endpoint fully removed** (confirmed via code search — no residual references)
+
+### Technical Notes
+
+- Dashboard gateway detection: `ps aux | grep openclaw.*gateway` for process status, `curl --max-time 2 http://127.0.0.1:<port>` for HTTP ping
+- Disk usage: `df -k` for filesystem stats, `du -sk` for OpenClaw dir size
+- Agent last activity: scans `~/.openclaw/agents/*/sessions/*.jsonl` mtime
+- Version cache busting: `OCM_CLIENT_VERSION` is injected into browser script via template literal `${APP_VERSION}`, compared against `X-OCM-Version` response header on every API call
+
+---
+
+## v0.6.0 更新日志（2026-02-26）
+
+### 重大 Bug 修复
+
+**Add Agent 覆盖已有 Bot Token（数据破坏性 Bug）**
+- **症状**：通过 Add Agent 表单添加新 agent 时，直接覆盖 `channels.telegram.botToken`，导致所有已有 agent（包括 sub-agent）全部失效
+- **根本原因**：`POST /api/agents/main` 端点无条件覆盖 `channels.telegram.botToken` 字段，没有保护已有配置
+- **修复**：
+  - 彻底删除 `POST /api/agents/main` 端点
+  - 新建 `POST /api/agents/bot` 端点，使用 OpenClaw 的 `channels.telegram.accounts` 多 bot 结构
+  - 每个新 agent 获得独立的 `accountId` 和 `botToken`，绝不覆盖已有 token
+  - 自动迁移：首次添加新 bot 时，自动将旧格式（顶层 `botToken`）迁移到 `accounts.default`
+- **数据恢复**：程序在修改前自动创建 `openclaw.json.create.*` 备份，可通过 `cp` 恢复
+
+**浏览器 Popover API 命名冲突**
+- **症状**：点击 "＋ Add Agent" / "＋ Add Sub-Agent" 按钮报错 `NotSupportedError: Failed to execute 'togglePopover' on 'HTMLElement'`
+- **根本原因**：自定义函数 `togglePopover()` 与浏览器原生 Popover API 的 `HTMLElement.togglePopover()` 方法冲突
+- **修复**：函数重命名为 `showConfigPop()`
+
+### 架构变更
+
+**多 Bot 支持（Multi-Account）**
+- 支持 OpenClaw 的 `channels.telegram.accounts` 结构，每个主 agent 可绑定独立的 Telegram bot
+- 配置格式：
+  ```json
+  {
+    "channels": { "telegram": { "accounts": {
+      "default": { "botToken": "TOKEN_A" },
+      "research": { "botToken": "TOKEN_B" }
+    }}}
+  }
+  ```
+- `GET /api/agents` 返回 `hasOwnBot` 字段，标识 agent 是否拥有独立 bot
+- Sub-Agent 表单新增 "Parent Agent" 下拉，选择共享哪个 bot（不再硬编码 "main"）
+
+**去除 Landing Page，直接进入主程序**
+- 移除模式选择首页（Sub-agent / Multi-agent 二选一）
+- 启动后直接进入 Agent 管理页面
+- 移除 landing page 相关 HTML、CSS、JS、i18n keys
+
+**Agent 页面重设计（popover 配置窗口）**
+- 不再使用左右分屏布局
+- "＋ Add Agent" / "＋ Add Sub-Agent" 按钮居中显示在 agent 树上方
+- 点击按钮弹出浮动配置窗口（popover），包含引导步骤和表单
+- Agent 树宽度限制 720px 居中，多个 agent 树纵向排列
+
+### 功能改进
+
+**Stats 重写 — 从 session JSONL 文件解析真实用量**
+- 不再从 `gateway.log` 解析（之前一直是 0 数据）
+- 改为扫描 `~/.openclaw/agents/*/sessions/*.jsonl`
+- 解析 `type: "message"` + `role: "assistant"` 的 `message.usage` 字段
+- 新增维度：By Agent（每个 agent 的用量）、Cache Read tokens
+- 测试验证：成功解析出 990 条请求、6 个 agent、10 个模型的真实数据
+
+**Model 下拉 — 只显示已注册模型**
+- 不再使用硬编码的 KNOWN_MODELS 列表
+- 改为从 `openclaw.json` 的 `agents.defaults.models` 读取实际注册的模型
+
+**Agent 树事件委托**
+- `renderAgents()` 中的按钮不再使用 `onclick="func('escaped-string')"` 内联写法
+- 改用 `data-action` / `data-id` 属性 + 事件委托（`agentTreeAction`），避免 template literal 转义问题
+
+**响应式布局**
+- `main` 容器改为 `max-width:100%`，适配不同屏幕宽度
+- 新增 `@media (max-width: 600px)` 断点：侧边导航折叠、按钮纵向排列
+
+**备份时间戳改为 Brisbane 时区**
+- `brisbaneTimestamp()` 函数，所有备份文件名使用 `Australia/Brisbane` 时区
+- 重要：系统所有时间显示统一按 Brisbane 处理
+
+**启动脚本全面重写**
+- `start.sh` / `start.bat` 全部英文
+- 端口冲突时自动 kill 旧进程，而非报错退出
+- 支持 `--host` 参数
+
+**README 重写**
+- 移除所有空的 screenshot 占位符（含敏感 ID 的截图已删除）
+- 更新功能描述匹配 v0.6
+- 精简结构，保留中英文双语
+
+### 新功能 i18n 策略
+
+- v0.6 新增的所有 UI 文案仅提供英文
+- 中文翻译推迟到 v1.0 正式版
+
+### 技术备忘
+
+- **Template literal 转义规则**：MAIN_HTML_SCRIPT 是反引号模板字符串
+  - `\n` → 真实换行（浏览器 JS 字符串跨行 → SyntaxError），必须写 `\\n`
+  - `\'` → `'`（无法用于 onclick 里的引号转义），改用 data 属性 + 事件委托
+  - `\`` → `` ` ``（嵌套模板字符串在 evaluated output 中正常工作）
+- **`assertBrowserScriptSyntax()`** 在启动时检查 MAIN_HTML_SCRIPT 的 evaluated 值
+- **OpenClaw config 多 bot 格式**：`channels.telegram.accounts.<id>.botToken`，binding 用 `accountId` 路由
+
+### 下一步（v0.7+）
+
+- [x] Dashboard 首页 tab（系统信息 + OpenClaw health 状态）— done in v0.6.5
+- [x] HTTP 响应加 `Cache-Control: no-store` + version-based cache busting — done in v0.6.5
+- [x] 彻底删除旧 `/api/agents/main` 端点残留（确认已清除）— confirmed in v0.6.5
+- [x] Agent tree: independent roots per bot + side-by-side layout — done in v0.6.5
+- [x] Agent tree: expand/collapse toggle for sub-agents — done in v0.6.5
+- [x] Add Agent: creates `agents/<id>/sessions/` runtime dirs + SOUL.md personality — done in v0.6.5
+- [x] Agent tree: main agent always root with inferred accountId — done in v0.6.6
+- [ ] Dashboard: auto-refresh every 30s when tab is active
+- [ ] Dashboard: OpenClaw version display (from `openclaw --version`)
+- [ ] Agent page: edit agent name, purpose, personality inline
+- [ ] Agent page: restart gateway button with status feedback
+- [ ] Settings tab: edit openclaw.json key fields via form (model, workspace defaults)
+- [ ] Telegram bot connectivity test button (ping bot API from server)
+- [ ] DEVLOG.md 中文 → 逐步迁移为英文
 
 ---
 

OpenClaw Manager.app/Contents/Info.plist

@@ -9,9 +9,9 @@
     <key>CFBundleIdentifier</key>
     <string>com.openclaw.manager</string>
     <key>CFBundleVersion</key>
-    <string>0.5.0</string>
+    <string>0.7.1</string>
     <key>CFBundleShortVersionString</key>
-    <string>0.5.0</string>
+    <string>0.7.1</string>
     <key>CFBundlePackageType</key>
     <string>APPL</string>
     <key>CFBundleIconFile</key>

README.md

@@ -1,116 +1,101 @@
 # OpenClaw Manager (OCM)
 
-> A zero-dependency, single-file local dashboard for [OpenClaw](https://github.com/anthropics/openclaw) — manage sub-agents, run commands in a built-in CLI, and stop hand-editing JSON.
+A zero-dependency, single-file web dashboard for [OpenClaw](https://github.com/anthropics/openclaw). Manage agents, monitor token usage, and run commands — all from your browser, no `npm install` required.
 
-[中文说明](#中文说明)
+[中文说明](#中文说明) · [中文使用说明（含截图）](docs/USAGE_GUIDE.zh-CN.md) · [English Guide (with screenshots)](docs/USAGE_GUIDE.en.md)
 
 ---
 
-## Quick Start (1 min)
+## Quick Start
 
 ```bash
-# Install
+# Clone
 git clone https://github.com/dtzp555-max/ocm.git
 cd ocm
 
-# Update (existing clone)
-git pull --ff-only
-
 # Run (no npm install needed)
 bash start.sh        # macOS / Linux
 start.bat            # Windows
+
+# Update
+git pull --ff-only
 ```
 
-Open [http://localhost:3333](http://localhost:3333).
+Open [http://localhost:3333](http://localhost:3333) in your browser. For remote access, use `bash start.sh --host 0.0.0.0`.
-
-## Why OCM
-
-- **Single-file + zero dependency**: one `openclaw-manager.js`, no build step
-- **Fast daily operations**: agent/model/auth/cron/backup in one local UI
-- **Built-in terminal**: run `openclaw` commands with streaming output and presets
-
-## Screenshots
-
-### Landing Page
-Choose between Sub-agent mode and Multi-agent mode. Switch language (EN / 中文) anytime.
-
-![Landing Page](screenshots/landing.png)
-
-### Agent Management
-View all agents (main + sub-agents) at a glance. Each card shows model, group binding, and workspace path. Switch models inline, browse files, or delete with auto-backup.
-
-![Agents](screenshots/agents.png)
-
-### Built-in CLI Terminal
-Run any openclaw command with real-time streaming output. The terminal panel sits at the bottom of the page and expands on demand.
-
-![CLI Overview](screenshots/cli1.png)
-
-<details>
-<summary>More screenshots</summary>
-
-### New Subagent Wizard
-4-step guided wizard to create a sub-agent: basic info → model → personality & memory → confirm. Fully bilingual.
-
-![New Subagent](screenshots/subagents.png)
-
-### Model Selector
-Rich model dropdown grouped by provider — GitHub Copilot, Anthropic, OpenAI, Google, DeepSeek, Kimi, Groq, Mistral, Together, plus your custom models.
-
-![Models](screenshots/models.png)
-
-### Actions Menu
-Quick access to gateway operations, backups, health checks, and directory management. Use **"Switch OpenClaw Dir"** to point OCM to your OpenClaw config on first setup.
-
-![Actions](screenshots/actions.png)
-
-### Cron Job Management
-View, add, enable/disable, and manually trigger openclaw-related cron tasks. Integrates with NAS backup schedules.
-
-![Cron](screenshots/cron.png)
-
-Preset command menu with built-in commands and your personal favorites — one click to run.
-
-![CLI Presets](screenshots/cli3.png)
-
-Command output streams in real-time. Star frequently used commands for quick access, or use the Manage button to organize favorites.
-
-![CLI Detail](screenshots/cli2.png)
-</details>
 
 ## Features
 
-- **Agents & Workspace** — Create sub-agents, switch models inline, and browse/edit workspace files
+**Screenshots** — Redacted UI walkthrough (no personal paths, no Telegram IDs):
-- **Models & Auth** — Manage primary/fallback models and follow provider auth guides
+
-- **Ops Panel** — Restart gateway, view logs, run health checks, and switch OpenClaw directory
+![](docs/annotated-screenshots/dashboard.jpg)
-- **Stats & Cron** — Token/cost stats from `gateway.log` plus cron task management
+
-- **Backups** — Local backup/rollback and NAS backup (SFTP/rsync)
+![](docs/annotated-screenshots/agents.jpg)
-- **Built-in CLI** — Real-time command terminal with presets, favorites, and tab completion
+
+![](docs/annotated-screenshots/channels.jpg)
+
+![](docs/annotated-screenshots/models.jpg)
+
+![](docs/annotated-screenshots/auth.jpg)
+
+![](docs/annotated-screenshots/stats.jpg)
+
+![](docs/annotated-screenshots/cron.jpg)
+
+
+**Agent Management** — Add main agents and sub-agents through a guided setup flow. View all agents in a tree structure with model selection, workspace browsing, and inline configuration.
+
+**Usage Statistics** — Real token usage data parsed directly from OpenClaw session files. Breakdown by model, agent, and day with a visual chart.
+
+**Model & Auth** — Configure models from your registered provider list. Follow built-in guides for provider authentication setup.
+
+**Built-in CLI** — Run any OpenClaw CLI command with real-time streaming output. Includes tab completion and saved favorites.
+
+![](docs/annotated-screenshots/cli.jpg)
+
+**Ops Panel** — Restart gateway, view logs, run health checks, manage backups (local + NAS via SFTP/rsync), and handle cron tasks.
+
+**Bilingual UI** — English and Chinese interface with one-click language switching.
+
+## What's New (v0.7.x)
+
+- **Dashboard redesign**: circular CPU/RAM/DISK gauges + system/gateway/agent overview, with optional auto-refresh.
+- **Sub-agent creation flow**: step-by-step Telegram guide (BotFather → group → get group ID → fill form), with explicit safety warnings.
+- **Telegram allowlist helper**: optional “Your Telegram User ID” field can auto-append to `channels.telegram.allowFrom`.
+- **Model dropdown = real CLI output**: selectors are sourced from `openclaw models list` (with a warning if parsing fails).
+- **Cache-busting**: response headers + version checks help prevent stale frontend after updates.
+- **Setup page EN localization**: first-run directory selection page is fully bilingual.
+
+
+## Telegram Workflow & Safety
+
+OCM is designed primarily for **Telegram-based OpenClaw workflows**:
+
+- Bind Telegram groups to one or more main agents, then create sub-agents under each tree
+- Keep each agent's context isolated via independent `workspace`, `SOUL.md`, and `MEMORY.md`
+- OCM mainly helps you **safely update `openclaw.json`** via UI instead of manual editing
+- Recommended for users who already have basic OpenClaw CLI experience (`onboard`, auth, gateway logs)
+
+Critical Telegram settings:
+
+- In BotFather, keep **Allow Groups = ON**
+- In BotFather, set **Group Privacy = OFF**
+- For each agent group, keep it private: **only you + that agent/sub-agents**
+- Do **not** invite other people to these groups (cost and security risk)
 
 ## Requirements
 
-- Node.js >= 18 ([download](https://nodejs.org/))
+- Node.js 18+ ([download](https://nodejs.org/))
 - A working [OpenClaw](https://github.com/anthropics/openclaw) installation
 
-Install/update/run commands are in **Quick Start (1 min)** above.
-
-### First Run
-
-On first launch, the start script auto-detects `~/.openclaw` and creates a `manager-config.json` for you. If your OpenClaw config is elsewhere, use **Actions → Switch OpenClaw Dir** in the UI to point to the correct path.
-
-### macOS Double-click
-
-Double-click `OpenClaw Manager.app` in Finder. On first launch, you may need to allow it in System Settings → Privacy & Security.
-
-### Custom Port
-
-```bash
-bash start.sh --port 8080
-```
-
 ## Configuration
 
-The app locates your OpenClaw config directory in this order:
+On first launch, OCM auto-detects `~/.openclaw` and creates a local `manager-config.json`. If your OpenClaw config is elsewhere, use **Actions → Switch OpenClaw Dir** or specify it directly:
+
+```bash
+bash start.sh --dir /path/to/.openclaw --port 8080
+```
+
+Detection priority:
 
 | Priority | Method |
 |----------|--------|
@@ -125,9 +110,17 @@ You can also create `manager-config.json` manually:
 { "dir": "~/.openclaw" }
 ```
 
-`~` expands to your home directory on all platforms. `manager-config.json` is in `.gitignore` and won't be committed.
+`manager-config.json` is gitignored and won't be committed.
 
-## Optional: Shell Alias
+## Remote Access
+
+By default OCM binds to `0.0.0.0` (all interfaces). Access it from another device on your LAN using the IP shown at startup. To restrict to localhost only:
+
+```bash
+bash start.sh --host 127.0.0.1
+```
+
+## Shell Alias (Optional)
 
 ```bash
 # Add to ~/.zshrc or ~/.bashrc
@@ -138,13 +131,11 @@ alias ocm="bash ~/path/to/ocm/start.sh"
 
 ```
 ocm/
-├── openclaw-manager.js          ← The entire app (server + frontend, single file)
+├── openclaw-manager.js    ← Entire app (server + frontend, single file)
-├── start.sh                     ← macOS/Linux start script (auto-detect Node, port, config)
+├── start.sh               ← macOS/Linux launcher (auto-detect, port conflict handling)
-├── start.bat                    ← Windows start script
+├── start.bat              ← Windows launcher
 ├── OpenClaw Manager.app/  ← macOS Finder double-click launcher
-├── screenshots/                 ← README screenshots
 ├── manager-config.json    ← Local config (gitignored)
-├── package.json
 ├── DEVLOG.md              ← Development log
 └── README.md
 ```
@@ -155,12 +146,6 @@ ocm/
 - **Frontend**: Vanilla HTML/CSS/JS embedded in a template literal — no build step, no framework
 - **Architecture**: Single-file, zero-dependency, runs anywhere Node.js runs
 
-## Support / Donate
-
-If OCM saves you time, consider starring the repo or sponsoring development.
-
-- GitHub Sponsors: *(add link when ready)*
-
 ## License
 
 MIT
@@ -169,94 +154,54 @@ MIT
 
 ## 中文说明
 
-### 简介
+OpenClaw Manager 是一个零依赖的本地 Web 管理界面，用于可视化管理 [OpenClaw](https://github.com/anthropics/openclaw) AI 智能体。所有代码在一个 `.js` 文件中，只需 Node.js 18+，不需要 `npm install`。
 
-OpenClaw Manager 是一个零依赖的本地 Web 管理界面，用于可视化管理 [OpenClaw](https://github.com/anthropics/openclaw) AI 智能体。所有代码合并在一个 `.js` 文件中，只需要 Node.js 18+，不需要 `npm install`。
+### 快速开始
-
-### 1 分钟上手
 
 ```bash
-# 首次安装
 git clone https://github.com/dtzp555-max/ocm.git
 cd ocm
-
-# 已安装后更新
-git pull --ff-only
-
-# 运行（无需 npm install）
 bash start.sh        # macOS / Linux
 start.bat            # Windows
 ```
 
-访问 [http://localhost:3333](http://localhost:3333)。
+访问 [http://localhost:3333](http://localhost:3333)。远程访问请使用 `bash start.sh --host 0.0.0.0`。
 
-### 界面预览
+### 功能
 
-#### 模式选择
+- **Agent 管理** — 添加主 Agent 和子 Agent，树状结构查看，内联切换模型，浏览工作区文件
-![Landing](screenshots/landing.png)
+- **使用统计** — 从 OpenClaw 会话文件解析真实 Token 用量，按模型/Agent/日期维度展示
+- **模型与认证** — 管理注册模型列表，内置 Provider 认证引导
+- **内置终端** — 实时流式输出，预设命令，收藏夹，Tab 补全
+- **运维面板** — 重启网关、查看日志、健康检查、本地/NAS 备份、Cron 任务管理
+- **双语界面** — 中英文一键切换
 
-#### Agent 管理
+### Telegram 场景与安全提示
-![Agents](screenshots/agents.png)
 
-#### 内置 CLI 终端
+OCM 主要面向 **Telegram 场景**：
-底部终端面板，支持实时流式输出、预设命令、收藏夹、Tab 补全。
 
-![CLI](screenshots/cli1.png)
+- 通过群组绑定主 Agent，并在每条树下管理多个 Sub-Agent
+- 让每个 Agent 拥有独立 `workspace`、`SOUL.md`、`MEMORY.md`
+- 通过可视化方式更新 `openclaw.json`，减少手动改配置风险
+- 建议使用者已具备基础 OpenClaw CLI 操作经验（如 `onboard`、认证、查看网关日志）
 
-<details>
+关键安全设置（务必确认）：
-<summary>更多截图</summary>
 
-#### 新建向导
+- BotFather 里 **Allow Groups = ON**
-![New Subagent](screenshots/subagents.png)
+- BotFather 里 **Group Privacy = OFF**
+- 每个 Agent 群组只保留“你自己 + 对应 Agent/Sub-Agent”
+- 不要邀请其他人进组（会带来安全和 API 费用风险）
 
-#### 模型选择 & 操作菜单
+### 配置
-首次安装后，通过 **Actions → Switch OpenClaw Dir** 指向你的 OpenClaw 配置目录即可开始使用。
 
-![Models](screenshots/models.png)
+启动脚本自动检测 `~/.openclaw` 目录。如需指定其他路径：
-![Actions](screenshots/actions.png)
 
-#### Cron 定时任务
+```bash
-![Cron](screenshots/cron.png)
+bash start.sh --dir /path/to/.openclaw --port 8080
+```
 
-![CLI Presets](screenshots/cli3.png)
+也可手动创建 `manager-config.json`：
-![CLI Detail](screenshots/cli2.png)
-</details>
-
-### 功能一览
-
-- **Agent 与 Workspace** — 新建子智能体、内联切换模型、浏览/编辑工作区文件
-- **模型与认证** — 管理主模型/Fallback 链，按引导完成各 Provider 认证
-- **运维操作面板** — 重启网关、查看日志、健康检查、切换 OpenClaw 目录
-- **统计与计划任务** — 从 `gateway.log` 汇总 Token/费用，并管理 Cron 任务
-- **备份能力** — 本地备份回滚 + NAS 远程备份（SFTP/rsync）
-- **内置 CLI 终端** — 实时输出、预设命令、收藏、Tab 补全
-
-安装/更新/运行命令见上方 **1 分钟上手**。
-
-### 首次运行
-
-启动脚本会自动检测 `~/.openclaw` 目录并创建 `manager-config.json`。如果你的 OpenClaw 配置在其他位置，进入界面后点击 **Actions → Switch OpenClaw Dir** 切换即可。
-
-### 指定 OpenClaw 目录
-
-程序按以下优先级查找配置：
-
-1. `--dir` 命令行参数
-2. `OPENCLAW_DIR` 环境变量
-3. 同目录下的 `manager-config.json`
-4. `~/.openclaw`（默认）
-
-也可以手动创建 `manager-config.json`：
 
 ```json
 { "dir": "~/.openclaw" }
 ```
-
-`~` 在所有平台自动展开为用户主目录。
-
-### Shell 别名（可选）
-
-```bash
-# 加到 ~/.zshrc 或 ~/.bashrc
-alias ocm="bash ~/path/to/ocm/start.sh"
-```

docs/USAGE_GUIDE.en.md

@@ -0,0 +1,275 @@
+# OCM Usage Guide (with screenshots)
+
+> Audience: People who already have OpenClaw running locally (you can run `openclaw status` and read `openclaw gateway logs`).
+>
+> Goal: Use **OpenClaw Manager (OCM)** to make agent/sub-agent setup and day-to-day ops **visual, safer, and easier to audit**.
+
+---
+
+## 0) The pain: why OCM exists
+
+Creating and maintaining OpenClaw agents is powerful, but the “default workflow” is often messy:
+
+1) **Editing `openclaw.json` is risky**
+- Deep/nested config, many fields.
+- One accidental overwrite can break multiple agents (tokens, bindings, accountIds).
+
+2) **Sub-agent topology is hard to see**
+- With multiple bots + multiple groups, it’s not obvious which agent is bound to which Telegram peer.
+
+3) **Model IDs are easy to misconfigure**
+- A wrong `provider/model-id` can silently degrade performance, cause timeouts, or produce “typing…” with no reply.
+
+4) **Usage and troubleshooting take time**
+- Without tooling, you’re digging through JSONL sessions and gateway logs to answer: “who used tokens?”, “what failed?”, “which provider?”
+
+OCM’s role: **a zero-dependency local UI** that helps you view and update OpenClaw configuration safely, keep changes auditable, and centralize common ops.
+
+---
+
+## 1) Why Telegram groups are the best way to build sub-agents
+
+Think of it as:
+
+> **One Telegram group = one agent boundary** (context + purpose + workspace).
+
+### Benefits
+
+1) **Clear isolation**
+- Each agent speaks only inside its own group.
+- Less cross-contamination between tasks.
+
+2) **Controllable context**
+- The group chat is the agent’s input stream.
+- You can keep an agent “clean” by keeping the group focused.
+
+3) **Natural delegation**
+- A main agent can coordinate; sub-agents can specialize (tech / travel / finance / translation, etc.).
+
+4) **Auditable**
+- You can see exactly which message triggered which response.
+
+### Safety / cost: keep groups private
+
+Recommended rule:
+
+- **Each agent group should contain only you + the bot** (and optionally your second account).
+- Don’t invite other people:
+  - cost: anyone can trigger token usage
+  - safety: higher risk of prompt-injection / accidental tool actions
+
+---
+
+## 2) Start OCM (macOS)
+
+From the repo directory:
+
+```bash
+bash start.sh
+```
+
+Then open: <http://localhost:3333>
+
+Restrict to localhost only:
+
+```bash
+bash start.sh --host 127.0.0.1
+```
+
+---
+
+## 3) Dashboard: check health first
+
+The Dashboard is your “sanity check”:
+
+- CPU / RAM / Disk
+- gateway running status
+- agent counts
+
+Screenshot:
+
+![](annotated-screenshots/dashboard.jpg)
+
+---
+
+## 4) Create a main agent (with its own Telegram bot)
+
+> Your `main` agent usually exists already. This section is for creating an additional **root agent** that owns its own Telegram bot/account.
+
+### 4.1 BotFather prerequisites
+
+1) Create a bot
+- In Telegram, open **BotFather**
+- Run `/newbot` → follow prompts → copy the bot token
+
+2) Allow groups
+- BotFather → your bot → Bot Settings
+- **Allow Groups = ON**
+
+3) Disable group privacy (critical)
+- BotFather → your bot → Bot Settings → Group Privacy
+- **Group Privacy = OFF**
+
+If group privacy is ON, the bot can’t see normal group messages. The symptom often looks like:
+- slow replies
+- “typing…” appears frequently
+- sometimes no response at all
+
+### 4.2 Add the agent in OCM
+
+Go to **Agents**:
+
+![](annotated-screenshots/agents.jpg)
+
+Click `+ Add Agent` and fill:
+
+- agent id / name
+- workspace path (recommended: separate folder per agent)
+- model (dropdown is sourced from `openclaw models list`)
+- the BotFather token
+
+OCM writes the agent + Telegram account/binding into `openclaw.json`.
+
+---
+
+## 5) Create a sub-agent (recommended workflow)
+
+### 5.1 Why a new group per sub-agent
+
+- clean context isolation
+- easier debugging
+- easy to “turn off” by muting a group
+
+### 5.2 Step-by-step
+
+In **Agents** → `+ Add Sub-Agent`:
+
+**Step 1 — Create a Telegram group**
+- private group
+- add only: you + the bot
+
+**Step 2 — Get the Group ID (peer id)**
+
+Use gateway logs:
+
+```bash
+openclaw gateway logs --follow
+```
+
+Send a message in the new group. In logs you should see something like:
+
+- `-100xxxxxxxxxx` (Telegram group id)
+
+**Step 3 — Fill the sub-agent form**
+- choose the Parent Agent (which bot/account to share)
+- paste the Group ID
+- set workspace (recommended: its own folder)
+- choose model
+
+**Step 4 — Allowlist (optional but recommended)**
+
+If you use `channels.telegram.allowFrom`, OCM can take “Your Telegram User ID” and append it automatically to the allowlist.
+
+---
+
+## 6) Verify bindings (Channels)
+
+Go to **Channels**:
+
+![](annotated-screenshots/channels.jpg)
+
+Here you can quickly validate:
+- which agent is bound to which Telegram peer
+- whether `main` has the catch-all binding (if you use that pattern)
+
+---
+
+## 7) Model settings (Models)
+
+Go to **Models**:
+
+![](annotated-screenshots/models.jpg)
+
+Key idea:
+- OCM uses the real CLI model list (`openclaw models list`) for dropdowns.
+- This reduces “invalid model-id” mistakes.
+
+---
+
+## 8) Credentials (Auth)
+
+Go to **Auth**:
+
+![](annotated-screenshots/auth.jpg)
+
+Use this page to:
+- see which provider profiles exist
+- troubleshoot expired/missing tokens
+
+---
+
+## 9) Usage stats (Stats)
+
+Go to **Stats**:
+
+![](annotated-screenshots/stats.jpg)
+
+You can break down usage by:
+- model
+- agent
+- time window
+
+OCM parses usage from OpenClaw session JSONL files.
+
+---
+
+## 10) Cron jobs (Cron)
+
+Go to **Cron**:
+
+![](annotated-screenshots/cron.jpg)
+
+You can:
+- see scheduled tasks (backup / update / health)
+- run tasks manually
+- enable/disable entries
+
+---
+
+## 11) Built-in CLI Terminal
+
+OCM includes a built-in terminal panel to run OpenClaw CLI commands directly (for example: `openclaw status`, `openclaw gateway logs --follow`, `openclaw doctor`).
+
+How to open: click **⌨️ Terminal** in the top navigation, or use the bottom CLI panel.
+
+Highlights:
+- **Tab completion** while typing commands
+- **Common commands** dropdown to insert templates quickly
+- **Favorites**: save frequently used commands and run them with one click
+- **Streaming output** (stdout/stderr) for faster debugging
+
+Screenshot:
+
+![](annotated-screenshots/cli.jpg)
+
+## 11) Fast troubleshooting
+
+### “Typing…” but no reply in a Telegram group
+
+Check (in order):
+
+1) BotFather: **Group Privacy = OFF**
+2) `openclaw status` (gateway healthy?)
+3) provider auth / rate limit issues (Auth page + gateway logs)
+
+### Model dropdown doesn’t show a model you expect
+
+- dropdown comes from `openclaw models list`
+- ensure the model is registered/visible in your OpenClaw environment
+
+---
+
+## Notes about screenshots
+
+All screenshots in this guide are **redacted** (no personal paths, no Telegram IDs).
+They live under `docs/annotated-screenshots/`.

docs/USAGE_GUIDE.zh-CN.md

@@ -0,0 +1,267 @@
+# OCM 功能使用说明（含截图）
+
+> 面向：已经在本机安装并跑通 OpenClaw 的用户（至少会用 `openclaw status` / `openclaw gateway logs`）
+>
+> 目标：用 **OpenClaw Manager (OCM)** 把“创建 Agent / Sub-Agent、绑定 Telegram 群、管理模型/认证、看用量、看 cron”这套流程做成可视化、可回溯、低风险。
+
+---
+
+## 0) 先说痛点：为什么要用 OCM
+
+### 0.1 创建/维护 Agent 的典型痛点
+
+1. **改 `openclaw.json` 容易改错**
+   - 字段多、层级深、容易手滑覆盖 token / accountId / bindings。
+   - 一次写错可能导致多个 Agent 全挂。
+
+2. **Sub-Agent 的边界和“谁在用哪个 bot token”不直观**
+   - 多 Bot / 多 agent / 多群组绑定时，光靠肉眼看 JSON 很难确认拓扑结构是否正确。
+
+3. **模型选择很容易“写了一个不存在的 model-id”**
+   - provider/model-id 拼错、或 provider 侧不支持，最后表现就是 Agent 变慢/报错/一直 typing。
+
+4. **统计用量和排查问题耗时**
+   - 需要翻 session JSONL / gateway logs 才能知道谁在烧 token、哪里失败。
+
+OCM 的定位：**把配置变更“可视化 + 可检查 + 可回滚（配合备份）”，并把常用运维入口集中起来。**
+
+---
+
+## 1) 为什么用 Telegram group 来建 Sub-Agent 最合适
+
+> 这里讨论的是 OpenClaw 的一种非常实用的工作方式：**“一个 Telegram 群 = 一个 Agent 的工作空间/人格/上下文边界”**。
+
+### 1.1 Telegram group 方案的优势
+
+1. **边界天然清晰**
+   - 每个 Agent 只在自己的群里说话，和其他 Agent 隔离。
+
+2. **上下文可控**
+   - 一个群的聊天记录就是该 Agent 的输入流；不会和别的任务混在一起。
+
+3. **并发与分工自然**
+   - 主 Agent 负责总控/派工，Sub-Agent 负责专项：技术宅/旅行/理财/翻译……
+
+4. **可审计**
+   - 群里发生了什么一眼可见；出问题也能定位是哪条消息触发。
+
+### 1.2 为什么“群必须私密”（安全与成本）
+
+- 强烈建议：**每个 Agent 群只留你自己 + 对应 bot（+ 必要时的另一个你自己的账号）**。
+- 不要拉别人进来：
+  - 成本：别人一句话就触发 token 消耗
+  - 安全：可能诱导泄露信息/触发危险操作
+
+---
+
+## 2) 启动与进入 OCM
+
+### 2.1 启动（macOS）
+
+在 OCM 目录：
+
+```bash
+bash start.sh
+```
+
+默认访问：<http://localhost:3333>
+
+> 需要限制只能本机访问：`bash start.sh --host 127.0.0.1`
+
+### 2.2 Dashboard（先看健康状态）
+
+建议第一步先看 Dashboard，确认：
+- 机器资源（CPU/RAM/DISK）
+- OpenClaw gateway 是否 Running
+- Agents 数量
+
+截图示例：
+
+![](annotated-screenshots/dashboard.jpg)
+
+---
+
+## 3) 创建主 Agent（Main Agent / 独立 Bot）
+
+> 你的 `main` 一般是默认存在的；这里重点讲“新增一个 **拥有独立 Telegram bot 的主 Agent**”。
+
+### 3.1 BotFather 侧准备（必须）
+
+1) 创建 bot
+- 在 Telegram 找 **BotFather**
+- 发送 `/newbot` → 按提示创建 → 拿到 bot token
+
+2) 允许加群
+- BotFather → 你的 bot → Bot Settings
+- **Allow Groups = ON**
+
+3) 关闭群隐私（非常关键）
+- BotFather → 你的 bot → Bot Settings → Group Privacy
+- **Group Privacy = OFF**
+
+> 如果 Group Privacy 没关，bot 在群里看不到普通消息，会表现成“经常 typing 但没输出 / 没反应”。
+
+### 3.2 在 OCM 添加主 Agent
+
+进入 **Agents** 页面：
+
+![](annotated-screenshots/agents.jpg)
+
+点击 `+ Add Agent`，按表单填写：
+- Agent id / name
+- workspace 路径（建议独立目录）
+- 选择模型（下拉来自 `openclaw models list`，更不容易选错）
+- 填入刚拿到的 bot token
+
+保存后，OCM 会把这颗主 Agent 写进 `openclaw.json`，并把 bot token 绑定到对应 `accountId`。
+
+---
+
+## 4) 创建 Sub-Agent（最常用）
+
+### 4.1 为什么 Sub-Agent 要用“新群”
+
+- 让每个 Sub-Agent 有自己独立的对话上下文、任务边界
+- 出问题直接在该群排查，不污染主 Agent 的对话
+
+### 4.2 Sub-Agent 创建流程（逐步）
+
+进入 **Agents** 页面 → 点击 `+ Add Sub-Agent`，按引导做：
+
+#### Step 1：建 Telegram 群
+- 新建群（私密）
+- 群里只加：你自己 + 对应的 bot
+
+#### Step 2：拿到群 ID（Peer ID / Group ID）
+OpenClaw 通常会在 gateway 日志里打印群的 peer id。
+
+常见拿法：
+```bash
+openclaw gateway logs --follow
+```
+然后在 Telegram 群里随便发一句话，观察日志。
+
+> 你会看到类似 `-100xxxxxxxxxx` 这样的 group id。
+
+#### Step 3：回到 OCM 填表
+- Parent Agent（选择共享哪个 bot/account）
+- 填 Group ID
+- workspace（建议单独目录）
+- 选择模型
+
+#### Step 4：写 allowlist（推荐）
+如果你启用了 Telegram 的 allowlist（例如 `channels.telegram.allowFrom`），
+OCM 的表单里可以填 “Your Telegram User ID”，让它自动把你的 user id 加进去，避免 DM/群消息被拒。
+
+---
+
+## 5) 绑定检查（Channels 页）
+
+进入 **Channels**：
+
+![](annotated-screenshots/channels.jpg)
+
+这里能一眼看到：
+- 哪个 agent 绑定了哪个 Telegram peer（群）
+- main 的“兜底绑定”（any）是否存在
+
+> 如果某个 Sub-Agent 绑定错群/丢 binding，最直观的就是在这里对不上。
+
+---
+
+## 6) 模型管理（Models 页）
+
+进入 **模型**：
+
+![](annotated-screenshots/models.jpg)
+
+关键点：
+- 主模型（primary）和 fallback 链
+- 下拉列表来自 `openclaw models list`（能减少“选了不存在模型”的问题）
+
+---
+
+## 7) 认证管理（Auth 页）
+
+进入 **认证**：
+
+![](annotated-screenshots/auth.jpg)
+
+这里主要用来：
+- 看当前有哪些 provider profile
+- 排查某个 provider 是否过期/缺 token
+
+---
+
+## 8) 用量统计（Stats 页）
+
+进入 **Stats**：
+
+![](annotated-screenshots/stats.jpg)
+
+你可以按：
+- 按模型聚合
+- 按 Agent 聚合
+- 按天趋势
+
+> OCM 是从 OpenClaw 的 session JSONL 里解析真实 usage（比“看日志猜”靠谱）。
+
+---
+
+## 9) CLI 终端（内置）
+
+OCM 内置一个 CLI 面板，用来直接运行 OpenClaw 命令（例如：`openclaw status`、`openclaw gateway logs --follow`、`openclaw doctor`）。
+
+入口：点击顶部导航栏的 **⌨️ 终端**，或使用页面底部的 CLI 区域。
+
+功能要点：
+- **Tab 补全**：输入命令时按 Tab 进行补全/提示（适合 `openclaw ...` 子命令）
+- **常用命令**：可从下拉选择，一键填充常用命令模板
+- **收藏（Favorites）**：把常用命令保存为收藏，后续一键执行
+- **实时输出**：执行后会流式显示 stdout/stderr，方便排错
+
+截图示例：
+
+![](annotated-screenshots/cli.jpg)
+
+## 9) Cron（定时任务）
+
+进入 **Cron**：
+
+![](annotated-screenshots/cron.jpg)
+
+用途：
+- 查看有哪些健康检查/更新/备份任务
+- 手动运行
+- 观察启用/禁用状态
+
+---
+
+## 10) 常见故障排查（快速）
+
+### 10.1 Telegram 群里“一直 typing 但不回”
+
+优先检查：
+1) BotFather 的 **Group Privacy 是否 OFF**
+2) OpenClaw gateway 是否健康（`openclaw status`）
+3) provider 是否限流/超时（Auth 页 + gateway logs）
+
+### 10.2 模型下拉没有你想要的模型
+
+- 现在下拉是 `openclaw models list` 的真实输出。
+- 先确保你在 OpenClaw 里已经注册/可见该模型。
+
+---
+
+## 11) 推荐的“最佳实践”
+
+- 每个 Agent 都用独立 workspace（隔离文件/记忆）
+- 每个 Sub-Agent 都用独立 Telegram 群（隔离上下文）
+- 群里不要加第三方（成本 + 安全）
+- 任何“改配置的大动作”尽量先备份（OCM/脚本/手工都行）
+
+---
+
+## 附：截图说明
+
+本文截图来自 `docs/annotated-screenshots/`，已做脱敏（路径用户名、Telegram 绑定 ID 等均已模糊处理）。

package.json

@@ -1,6 +1,6 @@
 {
   "name": "openclaw-manager",
-  "version": "0.5.0",
+  "version": "0.7.1",
   "description": "A local web UI for managing OpenClaw AI agents — no npm install required",
   "main": "openclaw-manager.js",
   "scripts": {

start.bat

@@ -1,6 +1,6 @@
 @echo off
 :: ================================================================
-:: OpenClaw Manager v0.6.0 — Start Script (Windows)
+:: OpenClaw Manager v0.7.1 — Start Script (Windows)
 ::
 :: Features:
 ::   * Auto-detect Node.js, show install instructions if missing
@@ -19,7 +19,7 @@ set "MIN_NODE_MAJOR=18"
 
 echo.
 echo   ===================================
-echo      OpenClaw Manager  v0.6.0
+echo      OpenClaw Manager  v0.7.1
 echo   -----------------------------------
 echo.
 

start.sh

@@ -1,6 +1,6 @@
 #!/usr/bin/env bash
 # ================================================================
-# OpenClaw Manager v0.6.0 — Start Script (macOS / Linux)
+# OpenClaw Manager v0.7.1 — Start Script (macOS / Linux)
 #
 # Features:
 #   - Auto-detect Node.js, show install instructions if missing
@@ -22,7 +22,7 @@ CYAN='\033[0;36m'; BOLD='\033[1m'; DIM='\033[2m'; RESET='\033[0m'
 
 banner() {
   echo ""
-  echo -e "${CYAN}${BOLD}  🦀 OpenClaw Manager v0.6.0${RESET}"
+  echo -e "${CYAN}${BOLD}  🦀 OpenClaw Manager v0.7.1${RESET}"
   echo -e "${DIM}  ─────────────────────────────${RESET}"
   echo ""
 }