Swarm Control Feishu

一键配置飞书智能体集群，支持多 agent 并行、语音服务、全权限控制

Do not run deploy.sh or start-voice-service.sh blindly. Key risks: - This skill intentionally disables sandboxing and grants agents full exec and filesystem access, and enables 'elevated' actions from Feishu/webchat — meaning remote users could cause arbitrary commands to run on the host. - deploy.sh attempts to add a passwordless sudoers entry for user 'lehua' (lehua ALL=(ALL) NOPASSWD: ALL). That elevates privilege permanently and is unsafe. - start-voice-service.sh will write a Python service to /tmp, install packages via pip globally (no virtualenv), enable CORS=* and run an HTTP endpoint — all increasing exposure. - The documentation suggests pushing configs to GitHub; openclaw-config.json contains API keys and secrets and would leak them if published. What to do before installing: 1. Review every file line-by-line (especially scripts/deploy.sh and files/openclaw-config.json). Fix obvious bugs (the deploy script has a malformed line) and remove or modify unsafe defaults. 2. Never enable sandbox: off, fs.workspaceOnly=false, exec.ask=off, or tools.profile=full unless you fully accept the host compromise risk. Prefer least-privilege settings. 3. Do NOT apply the sudoers change. If you need privileged actions, use carefully-scoped sudoers entries for your specific username or use an alternative elevation mechanism, and avoid NOPASSWD for broad commands. 4. Run the voice service in an isolated environment (container or virtualenv) and avoid global pip installs. Restrict CORS and bind to localhost if not needed externally. 5. Replace hard-coded username 'lehua' with your actual username or make the script prompt for the correct user; verify file paths are correct for your environment. 6. Keep secrets out of repo/config files; use secure secret storage or environment variables with appropriate metadata declared. Do not publish the filled config to GitHub. 7. If you want to test, do so inside a disposable VM or container to limit blast radius. If you are not comfortable auditing and mitigating these issues, avoid installing this skill in production systems.

Type: OpenClaw Skill Name: swarm-control-feishu Version: 2.0.0 The skill bundle configures an extremely permissive environment by disabling all sandboxing, allowing unrestricted command execution without user confirmation (exec.security: full, ask: off), and granting full filesystem access (fs.workspaceOnly: false) in files/openclaw-config.json. Most critically, scripts/deploy.sh attempts to grant passwordless sudo privileges to a specific user by modifying /etc/sudoers.d/lehua. Furthermore, files/AGENTS.md explicitly instructs the AI agents that they have 'FULL ACCESS' and should execute system commands autonomously via curl and exec tools. While these actions are framed as features for managing a Feishu agent swarm, the combination of privilege escalation and lack of oversight represents a significant security risk.