openclaw 使用技巧

原创 openclaw VPN

1. 一些常用命令

• 查看版本：openclaw --version
• 查看整体状态：openclaw status
• 重启 openclaw：systemctl --user restart openclaw-gateway
• 查看详细状态：openclaw status --deep
• 打开控制面板：openclaw dashboard
• 查看 Gateway 状态：openclaw gateway status
• 查看模型状态：openclaw models status
• 重新运行初始化向导：openclaw onboard
• 读取配置项：openclaw config get gateway.auth.token

2. session 交互

• 进入交互式聊天界面：openclaw tui
• 使用指定 session：openclaw tui --session <session_name>
  • 如果 session 已存在，就是进入
  • 如果 session 不存在，就是新建并进入
• 列出所有 session：openclaw sessions
• 重制 session：进入到session中
  • /reset or /new
• 本地直接发一条消息：openclaw agent --agent main --message "Reply with exactly: OK" --local
• 删除 session：key 就是你要删除的 session，见下面例子

3. 模型设置

• 查看系统的模型：openclaw models list
• 查看所有支持模型：openclaw models list --all
• 按供应商筛选：openclaw models list --provider github-copilot
• 添加模型：openclaw models set github-copilot/gpt-4o
• .openclaw/openclaw.json 中的配置文件
  • agents.defaults.model.primary 决定默认模型
  • agents.defaults.models 如果被设置，就会变成允许列表
  • models list 显示的是当前可见/允许使用的模型。 

4. agent 交互

• 查看所有 agent：openclaw agents list

• 创建新 agent：openclaw agents add <agent_name> --workspace <workspace_path>

  • 例如：openclaw agents add test --workspace ~/.openclaw/workspace-test

• 创建新 agent 并指定模型：openclaw agents add <agent_name> --workspace <workspace_path> --model <provider/model>

  • 例如：openclaw agents add test --workspace ~/.openclaw/workspace-test --model github-copilot/gpt-4o

• 删除 agent：openclaw agents delete <agent_name>

• 切换 agent：openclaw tui ctrl+G
  • 按 Ctrl+G：切换 agent

• 查看 agent 绑定关系：openclaw agents bindings

• 查看某个 agent 的绑定：openclaw agents bindings --agent <agent_name>

5. 日志相关

• 查看实时日志：openclaw logs --follow
• 查看 Gateway 的 systemd 日志：journalctl --user -u openclaw-gateway.service -n 200 --no-pager
• 重启 Gateway 服务：systemctl --user restart openclaw-gateway

6. skills

在 OpenClaw 里，skill 本质上就是一个目录，其中包含 SKILL.md 文件。OpenClaw 中 skill 有两种常见放置位置

当前 agent 专用 skill，这类 skill 只对当前 workspace 对应的 agent 生效

~/.openclaw/workspace/skills/
~/.openclaw/workspace/skills/hello_world

如果要让本机所有 agent 都可以使用这个 skill，放到这里，这样所有agent都能看到

~/.openclaw/skills/
~/.openclaw/skills/hello_world

1. 创建一个最简单的 Skill

下面以一个最简单的 hello_world skill 为例。假设这个 skill 是共享 skill

mkdir -p ~/.openclaw/skills/hello_world

2. 创建 SKILL.md

以通用 skill 为例

cat > ~/.openclaw/skills/hello_world/SKILL.md <<'EOF'
---
name: hello_world
description: A simple skill that says hello.
---

# Hello World Skill

When the user asks for a greeting, reply with:

Hello from your custom skill!
EOF

创建完 skill 后，需要让 OpenClaw 重新加载。重启 gateway，就可以看到新添加的 skill 了

➜ ~ systemctl --user restart openclaw-gateway
➜ ~ sleep 3
➜ ~ openclaw skills list --eligible
🦞 OpenClaw 2026.2.23 (b817600) — I'm basically a Swiss Army knife, but with more opinions and fewer sharp edges.

Skills (7/7 ready)
┌──────────┬──────────────────┬────────────────────────────────────────────────────────────────┬────────────────────┐
│ Status   │ Skill            │ Description                                                    │ Source             │
├──────────┼──────────────────┼────────────────────────────────────────────────────────────────┼────────────────────┤
...
│          │                  │ and assets.                                                    │                    │
│ ✓ ready  │ 🌤️ weather       │ Get current weather and forecasts via wttr.in or Open-Meteo.   │ openclaw-bundled   │
│          │                  │ Use when: user asks about weather, temperature, or forecasts   │                    │
│          │                  │ for any location. NOT for: historical weather data, severe     │                    │
│          │                  │ weather alerts, or detailed meteorological analysis. No API    │                    │
│          │                  │ key needed.                                                    │                    │
│ ✓ ready  │ 📦 hello_world    │ A simple skill that says hello.                                │ openclaw-managed   │
└──────────┴──────────────────┴────────────────────────────────────────────────────────────────┴────────────────────┘

3. 如何使用 skill

自然语言触发

直接在 TUI、Telegram 或本地消息中描述需求。如果 skill 写得足够清楚，模型会自动使用它

请用 hello_world skill 跟我打个招呼
跟我打个招呼

用 /skill 显式调用

/skill hello_world
/skill hello_world greet me

对话中的常用命令

• /status：查看当前状态

相关例子

1. 删除 session：key 就是你要删除的 session，见下面例子

cp ~/.openclaw/agents/main/sessions/sessions.json ~/.openclaw/agents/main/sessions/sessions.json.bak.$(date +%Y%m%d-%H%M%S)

python3 - <<'PY'
import json
from pathlib import Path

p = Path.home() / ".openclaw/agents/main/sessions/sessions.json"
key = "agent:main:newsession"

data = json.loads(p.read_text(encoding="utf-8"))
if key in data:
    del data[key]
    p.write_text(json.dumps(data, ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
    print(f"Deleted: {key}")
else:
    print(f"Key not found: {key}")
PY

删完以后记得重启 systemctl --user restart openclaw-gateway

2. 添加模型：openclaw models set github-copilot/gpt-4o

➜ ~ openclaw models list

🦞 OpenClaw 2026.2.23 (b817600) — I'll butter your workflow like a lobster roll: messy, delicious, effective.

Model                                      Input      Ctx      Local Auth  Tags
github-copilot/gpt-5-mini                  text+image 125k     no    yes   default,configured
➜ ~ openclaw models set github-copilot/gpt-4o
🦞 OpenClaw 2026.2.23 (b817600)
   If it's repetitive, I'll automate it; if it's hard, I'll bring jokes and a rollback plan.

Config overwrite: /home/yonghuiwang/.openclaw/openclaw.json (sha256 5db3d5fc1e007d8ecff4f5f6c40eba7bb779082204582bb441d1b0909493fd6a -> 121a640420b0e5a763d8ede1f88a7a6a772ffd07cc2fb03a5ae6ba419ef4df1e, backup=/home/yonghuiwang/.openclaw/openclaw.json.bak)
Updated ~/.openclaw/openclaw.json
Default model: github-copilot/gpt-4o
➜ ~ openclaw models set github-copilot/gpt-4o
➜ ~ openclaw models list                     
🦞 OpenClaw 2026.2.23 (b817600) — I speak fluent bash, mild sarcasm, and aggressive tab-completion energy.

Model                                      Input      Ctx      Local Auth  Tags
github-copilot/gpt-4o                      text+image 63k      no    yes   default,configured
github-copilot/gpt-5-mini                  text+image 125k     no    yes   configured
➜ ~ 

一些问题

1. telegram 无法输入图片

Telegram 文本消息正常，但 Telegram 图片下载失败，这和最近 OpenClaw 社区里报告的 Telegram 媒体下载代理问题一致：普通 Bot API 请求能通，但媒体文件下载链路不会可靠地走环境变量代理，解决办法通常是显式配置 channels.telegram.proxy，必要时再加 ipv4first。

➜ ~ openclaw logs --follow                   

🦞 OpenClaw 2026.2.23 (b817600) — Your terminal just grew claws—type something and let the bot pinch the busywork.

Log file: /tmp/openclaw/openclaw-2026-03-13.log
09:00:52 error gateway/channels/telegram {"subsystem":"gateway/channels/telegram"} Telegram command sync failed: HttpError: Network request for 'setMyCommands' failed!
..................
09:33:24 info gateway/channels/telegram {"subsystem":"gateway/channels/telegram"} telegram sendMessage ok chat=1907660075 message=40
09:41:42 warn telegram-auto-reply {"module":"telegram-auto-reply"} {"chatId":1907660075,"error":"MediaFetchError: Failed to fetch media from https://api.telegram.org/file/bot8744486052:AAESKZbOuudvaPOoXnOKjMQ_axJmFQOwAWU/photos/file_3.jpg: TypeError: fetch failed"} media fetch failed
09:42:46 info gateway/reload {"subsystem":"gateway/reload"} config change detected; evaluating reload (channels.telegram.network, channels.telegram.proxy)

修改也比较简单，如下，加入 proxy 和 network 这两行即可

➜ ~ vim ~/.openclaw/openclaw.json

"channels": {
  "telegram": {
    "enabled": true,
    "dmPolicy": "pairing",
    "botToken": "8744486052:xxxxxxxxxxxx",
    "groupPolicy": "allowlist",
    "streaming": "off",
    "proxy": "http://127.0.0.1:10809",
    "network": {
      "dnsResultOrder": "ipv4first",
      "autoSelectFamily": false
    }
  }
}

2. 主要 api 性能排行

排行网站：https://pinchbench.com/

本文由 Yonghui Wang 创作，采用 知识共享署名4.0 国际许可协议进行许可
本站文章除注明转载/出处外，均为本站原创或翻译，转载前请务必署名
最后编辑时间为: Mar 14, 2026 03:28 pm