Tool Registry

工具注册与发现系统。基于Token匹配的工具路由，支持权限控制和子代理工具白名单。 当用户说"工具有哪些"、"搜索工具"、"查找技能"、"工具路由"时触发。

This skill implements a registry that includes tools able to read/write local files and run shell commands. While that capability can be legitimate for a tool router, there are two red flags you should address before installing or enabling it for agents: (1) The runtime register includes dangerous tools (bash, write_file) available to broad agent types (the code registers bash for 'general'), and (2) the execute() method does not enforce the documented permission or agent-type checks, so callers can invoke dangerous tools without proper filtering. Recommended actions: review and/or modify the provided scripts so execute() enforces user/agent permissions, restrict dangerous tools to tightly scoped agent types, add input validation/sanitization and sandboxing for command execution, and test in an isolated environment. If you cannot audit or harden the code, avoid enabling the skill for autonomous agents or only allow it for highly-trusted, manual invocations.

Type: OpenClaw Skill Name: tool-registry Version: 1.0.0 The skill bundle implements a tool registry that includes pre-configured, high-risk capabilities in `scripts/tool-registry.mjs`, specifically arbitrary shell command execution (the `bash` tool using `child_process.spawn`) and direct file system read/write access (`read_file` and `write_file`). While these tools are aligned with the stated purpose of providing a discovery and execution framework for an AI agent, the inclusion of functional RCE and filesystem manipulation tools without external sandboxing or robust input validation constitutes a significant security risk. No evidence of intentional data exfiltration or hidden backdoors was found, but the inherent power of the provided tools warrants a suspicious classification.