Docker Sandbox

Create and manage Docker sandboxed VM environments for safe agent execution. Use when running untrusted code, exploring packages, or isolating agent workloads. Supports Claude, Codex, Copilot, Gemini, and Kiro agents with network proxy controls.

Install only if you intend to use Docker Desktop sandbox features. When testing untrusted code, mount a disposable or carefully chosen workspace, use deny-by-default network rules where practical, avoid detached runs unless needed, and treat reset/rm as destructive cleanup operations.

Type: OpenClaw Skill Name: docker-sandbox Version: 1.0.0 The OpenClaw AgentSkills skill bundle for 'docker-sandbox' is classified as benign. Its stated purpose is to create and manage isolated Docker sandboxed VM environments for safely executing untrusted code or agent workloads. All capabilities described, such as running arbitrary commands, mounting workspaces, and network controls, are explicitly designed to operate within this isolated environment. The documentation emphasizes security features like network policy controls to mitigate risks from code running inside the sandbox, rather than introducing risks to the host system. There is no evidence of prompt injection against the agent, data exfiltration, malicious execution on the host, or obfuscation within the provided files.