agent- manager

Manage OpenClaw Agents by creating new agents, configuring workspaces, setting up Feishu bot integrations, and verifying multi-bot routing.

Security Scan

The skill's purpose (managing agents and Feishu bindings) matches the included scripts, but there are several coherence and privilege concerns — notably missing declared requirements, direct modification of local config (including storing secrets), and the ability to create agents and restart the gateway which could be abused if the skill is auto-invoked.

Purpose & Capability

The name/description match what the code and SKILL.md do: create agents, configure workspaces, generate Feishu authorization links, and write bindings. However the skill metadata claims no required binaries or env vars while the runtime expects python3, the 'openclaw' CLI, network access, and writes to ~/.openclaw/openclaw.json — the missing declared requirements is an inconsistency that reduces transparency.

Instruction Scope

SKILL.md instructs the agent to run local scripts (under ~/.openclaw/.../scripts/...), produce and send authorization URLs to users, poll Feishu, write client_id/app_secret into the user's ~/.openclaw/openclaw.json, remove .git directories in workspaces, and restart the OpenClaw gateway. Those actions go beyond simple read-only management: they modify local configuration, persist secrets, and control a system service (gateway restart). The instructions also require the agent to present clickable authorization links produced by the script — i.e., relay external URLs to users — which is expected for OAuth flows but should be explicit in metadata.

Install Mechanism

No install spec (instruction-only) so nothing is downloaded at install time; the code files are bundled with the skill. This is lower risk than arbitrary remote installs. Still, bundled scripts will be executed and perform filesystem/network operations at runtime.

Credentials

The skill declares no required environment variables or primary credential, yet the code obtains and persists Feishu credentials (app_secret) into ~/.openclaw/openclaw.json. It reads/writes user config in the home directory and stores transient device codes in /tmp. Requesting and storing secrets is proportionate to binding a bot, but the lack of declared credentials and explicit permission in the metadata is a transparency issue. Also removing .git directories is a filesystem-modifying action that may be unexpected.

Persistence & Privilege

always:false (good) but the skill can create new agents programmatically, write credentials to local config, and restart the OpenClaw gateway (openclaw gateway restart). Because model invocation is allowed, an agent could autonomously create agents or alter config if given permission during a session. That combination increases blast radius and could enable 'self-propagation' if the skill is invoked without strict human approval.

What to consider before installing

This skill appears to implement exactly what it claims, but proceed with caution. Before installing or enabling automatic use: 1) confirm you trust the skill's author (owner ID is unknown); 2) review the bundled scripts yourself (they are included) and ensure you are comfortable with them writing to ~/.openclaw/openclaw.json and removing .git directories; 3) be aware the skill will collect Feishu app credentials and persist them locally — back up openclaw.json before use; 4) ensure the 'openclaw' CLI and python3 exist and run the scripts manually once to verify behavior; 5) restrict autonomous invocation or require explicit human confirmation for creation/restart steps (especially gateway restarts); and 6) if you do not want the agent to create other agents or restart system services, do not enable the skill for autonomous actions and require manual steps for app secret handling and gateway restarts.

---

name: agent-manager description: Manage OpenClaw Agents. Create new agents, configure workspaces, manage Feishu bot integrations (QR code scanning, bindings), and verify multi-bot routing.

Agent 管理专家 (Agent Manager)

本技能用于帮助用户快速创建、配置和管理 OpenClaw Agent 及其飞书绑定。

核心流程（必须严格遵守）

1. 需求确认 (Triage) - ⚠️ 关键原则：信息收集完才能发确认

在执行任何操作前，你需要收集齐以下所有信息：

• Agent 的中文名称（如果用户说了主要用途或者功能描述，你可以帮忙拟名字）
• Agent 的主要用途/功能描述（用户自己说，或者你根据用户提供的信息去补充）
• 是否需要绑定飞书机器人（默认需要创建新的机器人）

收集规则： 信息都收集齐全后，使用下面的模板向用户发送确认卡片。

最终确认模板：

📋 创建计划确认

请确认以下信息：
- **Agent 中文名**：[用户提供的名称]
- **Agent ID**：[自动推断的英文ID，例如 sales-assistant]
- **工作区路径**：`~/.openclaw/workspace-[ID]`
- **功能描述**：[用户提供的功能描述]
- **飞书机器人**：[新建（默认）/绑定已有/暂不绑定]
- **机器人权限**：[全部放开（默认）/需要认证]

确认请回复"确认"或"是"，如有修改请说明。

2. 飞书快捷创建流程 (Feishu Quick Setup Flow)

如果用户选择了“新建”飞书机器人，在确认计划后，你需要执行以下交互式步骤：

1. 获取授权链接：运行以下脚本生成飞书授权 URL：

   python3 ~/.openclaw/workspace/skills/agent-manager/scripts/feishu_auth.py generate
   

2. 发送快捷授权卡片给用户：脚本运行后会输出 URL=https://...。请使用 Markdown 的链接语法包装成一个醒目的按钮/卡片形式发送给用户，并提示他们点击授权。

   必须使用的回复格式示例：

   请点击下方按钮完成飞书机器人授权：

   🔗 👉 点击这里快捷创建飞书机器人 👈

   (授权完成后，请回复“好了”或“已完成”)

3. 轮询结果：当用户回复“好了”之后，运行轮询脚本获取 app_id 和 app_secret：

   python3 ~/.openclaw/workspace/skills/agent-manager/scripts/feishu_auth.py poll
   

3. 执行创建 (Execution)

在完成确认（和飞书凭证获取）后，使用提供的脚本创建 Agent：

# 如果不需要绑定飞书
python3 ~/.openclaw/workspace/skills/agent-manager/scripts/manage_agent.py create \
  --name "销售助手" \
  --id "sales-assistant" \
  --description "负责处理销售数据和客户跟进"

# 如果需要绑定飞书，且权限为【全部放开（默认，open）】
python3 ~/.openclaw/workspace/skills/agent-manager/scripts/manage_agent.py create \
  --name "销售助手" \
  --id "sales-assistant" \
  --description "负责处理销售数据和客户跟进" \
  --app-id "cli_xxx" \
  --app-secret "yyy"

# 如果需要绑定飞书，且用户明确要求【需要认证（pairing）】
python3 ~/.openclaw/workspace/skills/agent-manager/scripts/manage_agent.py create \
  --name "销售助手" \
  --id "sales-assistant" \
  --description "负责处理销售数据和客户跟进" \
  --app-id "cli_xxx" \
  --app-secret "yyy" \
  --require-pairing

4. 收尾工作 (Cleanup & Restart)

必须按顺序严格执行以下步骤：

1. 执行完毕后，先向用户汇报：“✅ Agent [中文名] 及其相关配置已经创建完毕！为了避免 Git 嵌套冲突，我已经自动清理了它的 .git 目录。”
2. 询问是否重启：向用户询问：“一切准备就绪，需要现在为您重启 OpenClaw 网关以使新机器人上线吗？”
3. 等待用户回复：只有当用户明确回复“重启”、“好的”或类似同意的指令后。
4. 执行重启：
   • 发消息：“🔄 正在重启网关，请稍候（约30秒）...”
   • 执行命令：openclaw gateway restart
   • 检查状态：openclaw gateway status
   • 发消息告知：“✅ 网关重启成功！[中文名]已上线，可以开始使用了。”

附带资源 (Bundled Resources)

• scripts/manage_agent.py: 核心创建脚本（自动创建 Agent，写入配置并清理工作区）。
• scripts/feishu_auth.py: 飞书快捷授权链接生成与轮询脚本。
• scripts/verify_bindings.py: 验证飞书账号与 Agent 的绑定关系。