Code Sandbox

Secure sandbox for executing Node.js, Python, Go, and Rust code with timeout, CPU, and isolated temporary directory constraints.

This package is a prototype sandbox that runs user code by writing files and executing them via shell commands. Don't run untrusted code with it on any machine that holds secrets — the code intentionally (and by default) passes the host environment to child processes and has no network or filesystem isolation or enforced memory limits. If you consider installing or using it: - Do not enable autonomous invocation for agents that have this skill without strict controls. - Audit and remove exposure of sensitive environment variables before running (or run the sandbox with a deliberately empty env for child processes). Specifically, change child process env to a minimal whitelist (not ...process.env). - Prefer running in a container/VM (Docker or isolated VM) with network disabled and resource quotas if you need to execute untrusted code. - Review and harden executeWithTimeout behavior (exec spawns a shell; killing it may not kill grandchildren). Consider spawn with no shell, or use container runtime and process supervision. - Because the SKILL.md itself warns this is not suitable for untrusted code, treat the project as experimental. If you cannot accept the risk of environment leakage or arbitrary process execution, do not install or run this skill on sensitive hosts.

Type: OpenClaw Skill Name: yuyonghao-code-sandbox Version: 0.1.0 The skill implements a 'Code Sandbox' that lacks meaningful isolation, executing code directly on the host via child_process.exec in src/sandbox.js. A critical security flaw exists where the NodeExecutor explicitly passes the parent process's environment variables (process.env) to the executed code, exposing sensitive API keys and secrets to the 'sandbox' environment. While SKILL.md includes safety warnings and a roadmap for Docker integration, the current implementation provides a false sense of security and functions as a high-risk Remote Code Execution (RCE) wrapper.