Oz CLI Remote Node

Execute commands or tasks on the Oz Build Node via CLI, with mode-based control and direct bash command support on the remote node.

This skill does what it says—it proxies oz-cli and arbitrary bash commands to a remote node via the 'nodes' tool—but it also asks the agent to read local sandbox indicators and to store run metadata, prompts, and 'Open in Oz' URLs in workspace files. Before installing, consider: 1) Only enable if you trust the remote node and whoever has access to it, because the skill allows running arbitrary commands there. 2) Expect persistent files (TOOLS.md, oz_run.<id>.md) containing prompts and URLs; these may contain sensitive data or tokens—inspect and clean these files regularly or avoid storing them. 3) The skill reads local files like /.dockerenv and /proc/1/cgroup; if you do not want the agent checking local runtime state, do not install. 4) Limit who can invoke the skill, and avoid sending secrets or credentials in prompts. If you want reassurance, ask the author to remove local filesystem checks or to make persistent storage explicitly configurable and documented.

Type: OpenClaw Skill Name: oz-cli-remote-node Version: 1.1.3 The skill acts as a remote execution proxy, allowing arbitrary bash commands and 'oz-cli' prompts to be run on a remote node via the high-privilege 'nodes' tool. While this functionality is the stated purpose of the skill, the capability to execute unconstrained shell commands (via the '!' prefix) on remote infrastructure represents a high-risk behavior. The instructions in SKILL.md guide the agent to manage state and track runs, but the inherent risk of the provided remote access warrants a cautious classification.