Tool Master

工具使用大师 - 基于关键词映射表的实用工具查找系统

This skill is a keyword→command helper and is internally consistent, but it encourages executing system, network, and OpenClaw admin commands. Before installing: 1) Review tool_keyword_map.md and remove or sandbox any destructive or admin commands (rm -rf, chown, gateway/skills install/uninstall, dd, direct network downloads). 2) Configure the agent so the skill is user-invocable only (disable self-driven/autonomous runs) or require explicit, auditable confirmation before executing any command that modifies system state or uses network. 3) Run the skill in a restricted environment (container or VM) if you plan to let it execute commands. 4) If you accept it, add safeguards: whitelist safe commands, log all executions, and prevent access to sensitive paths (e.g., ~/.openclaw, /etc, home directories) or credential stores. If you want a lower-risk alternative, use the skill only to suggest commands and require the human to copy-and-run them manually.

Type: OpenClaw Skill Name: tool-master Version: 0.1.1 The bundle defines a 'Tool Master' system that encourages an AI agent to autonomously find and execute a wide array of system commands and Python snippets based on a keyword mapping table (tool_keyword_map.md). While the stated intent is productivity, the bundle provides the agent with high-risk capabilities including network exfiltration tools (curl, wget, python-requests), system configuration access (openclaw config), and process manipulation (kill), without providing the actual implementation scripts (quick_find.sh, tool_finder.sh) for security review. The 'Self-driven workflow' instructions in SKILL.md increase the risk of the agent performing unauthorized actions or being leveraged for 'living off the land' attacks if influenced by malicious user prompts.