Subagent Collaboration

多子代理协作分析与构建技能。自动分析 OpenClaw 中多个子代理的能力、使用模式和协作关系，智能推荐协作模式（并行/串行/分层/竞争/会诊/接力），生成协作流程设计和安全配置。适用于复杂任务分解、多专家会诊、工作流编排等场景。

This skill appears to do what it claims (analyze subagents, generate workflows, run security checks) but it reads and writes files in a hardcoded workspace path (/Users/nanlin/.openclaw/workspace) which may contain session history, configs, or other sensitive data. Before installing or running: 1) Inspect the three Python scripts yourself (they are included); 2) Modify the scripts to use a configurable WORKSPACE (or pass a safe path) instead of the hardcoded path; 3) Run them in a sandboxed or non-privileged account (or container) first to see what files they read/write; 4) Remove the unnecessary node requirement or confirm why node is needed; 5) Check outputs for any unexpected external network calls (the scripts shown do not make network requests, but generated workflows could spawn agents that do). If you cannot review or sandbox them, consider not installing or request an author-provided version that parameterizes workspace path and documents file access.

Type: OpenClaw Skill Name: subagent-collaboration Version: 1.0.0 The subagent-collaboration skill bundle provides a framework for designing and managing multi-agent workflows in OpenClaw. It includes Python scripts for analyzing existing sub-agent roles (analyze_subagents.py), generating collaboration workflows and JavaScript code (generate_workflow.py), and a security validator (security_check.py) that enforces RSG v2.2.0 safety rules such as mandatory sandboxing, timeout limits, and detection of sensitive keywords or command execution. The code is well-documented, lacks obfuscation, and focuses on promoting secure configuration practices for sub-agents.