Python Use Agent

智能执行用户描述的Python任务，自动生成、执行代码并反馈结果，无需额外配置，兼容旧命令用法。

This skill implements what it claims (AI-driven Python execution) but there is a meaningful implementation mismatch: config.json and executor.py describe sandboxing and safety checks, yet index.js directly writes a temp .py file and runs it with child_process.exec without applying those checks or honoring sandbox settings. That allows arbitrary Python to run with whatever environment and filesystem access the agent process has. Before installing or using the skill, consider: 1) Do not pass secrets or sensitive file paths as 'context' or task text — executed code could read environment variables and files. 2) Prefer running this skill only in an isolated environment (container or VM) or with network/file-system access disabled. 3) Ask the author/maintainer whether index.js should call the Python executor (which would apply safety checks) or whether sandbox enforcement will be added; request explicit runtime enforcement of blocked imports/patterns and network/file-system restrictions. 4) If you must use it in a sensitive environment, review and modify index.js to enforce config.json security settings or route execution through a verified sandbox. If you cannot get these assurances, treat the skill as risky and avoid supplying any credentials or private data to it.

Type: OpenClaw Skill Name: python-use-agent Version: 1.0.0 The skill is designed to generate and execute arbitrary Python code based on AI prompts, which is an inherent Remote Code Execution (RCE) risk. While it includes basic security configurations and safety checks (e.g., `_check_safety` in `executor.py` and `blocked_imports` in `config.json`), the execution mechanism in `index.js` uses `child_process.exec` on temporary files without a robust, hardened sandbox. There is no evidence of intentional malice or data exfiltration, but the broad execution capabilities and the 'No Agents, Code is Agent' philosophy present a significant attack surface if the AI is prompted to generate harmful code.