for create subagent

子Agent创建工具 - 用于批量创建和管理OpenClaw子Agent 包含完整的子Agent创建流程：目录结构、配置文件、飞书集成等 用于快速创建具有独立workspace的子Agent

This skill appears to be a procedural guide for creating OpenClaw subagents and integrating them with Feishu, but there are important risks to consider before using it: - Sensitive-credentials handling: The guide asks you to place Feishu appId/appSecret into the global openclaw.json. Do not paste production secrets into that file on a shared machine without understanding who/what can read it. Prefer using secure secret storage or scoped credentials. - Data leakage risk: The SOUL.md rules explicitly require sending every message (including task contents, progress, etc.) to Feishu and says 'do not filter'. That can leak sensitive user data or internal secrets to an external chat. Remove or tighten this rule before deploying agents connected to real data. - Global config edits: The instructions modify the platform-wide openclaw.json and require restarting the gateway. Back up that file before editing and test changes in an isolated environment to avoid breaking other agents. - Skill copying: The process copies 'skills' directories into the new agent workspace. Review the code and permissions of any skills you copy — a malicious or buggy skill replicated into multiple agents increases blast radius. - Least privilege & testing: Create and test subagents with minimal permissions and in a sandbox or test workspace first. Limit Feishu bot scopes to the minimum necessary and rotate credentials if you suspect exposure. If you want a clearer safety posture, ask the skill author (or provider) for: an explicit list of required secrets, justification for forwarding all messages, and recommended secure storage practices for app secrets. If those are absent, treat this as risky for production use.

Type: OpenClaw Skill Name: subagent-creator Version: 1.0.0 The subagent-creator skill bundle provides templates and instructions for automating the creation of sub-agents within the OpenClaw framework. It includes directory structures, configuration snippets for 'openclaw.json', and integration steps for Feishu (Lark) messaging. While it utilizes hardcoded Windows paths (e.g., 'C:\Users\Administrator') and includes instructions for sub-agents to synchronize all messages to an external platform, these behaviors are consistent with its stated purpose of agent management and communication integration. No evidence of intentional malice, unauthorized data exfiltration, or harmful prompt injection was identified.