Code Runner
Execute code snippets safely. Run Python, JavaScript, and more from your agent.
Details
Code Runner skill lets your agent execute code in sandboxed environments. For calculations, data transforms, and quick scripts.
When to use
Use when the agent should run code (math, parsing, scripts). Use sandboxing; good for power users and data tasks.
How Code Runner fits into your OpenClaw setup
An assistant is only as useful to an engineer as the tools it can actually operate. Development skills like Code Runner are how OpenClaw earns a place in a software workflow: they let the assistant act on repositories, tooling, and build output instead of just talking about them. The teams that get the most from skills in this category treat the assistant as a junior collaborator with real responsibilities, not a novelty.
What working with it looks like
A typical session starts in chat: you describe what you need in plain language, and the assistant works out that Code Runner is the right tool for the job. It runs the work, watches the output, and replies with the result or a follow-up question. Because the conversation happens in Discord, Telegram, or whichever channel you have connected, the full history of what was asked and what was done stays searchable alongside the rest of your team's discussion.
Installing Code Runner
Installation follows the standard ClawHub flow. From your OpenClaw directory, run the install command below, then restart your assistant so the new capability registers. If the skill needs credentials or API access, its README will say so — set those up before the first use.
Then test it. Ask the assistant for something small that exercises the new skill end to end. Skills fail in two ways — not registered, or registered but misconfigured — and a quick first request tells you immediately which situation you are in, if either.
clawhub install code-runnerGetting the most out of it
No skill does its best work alone. The assistant routinely chains Code Runner with other installed capabilities — gathering context with one skill, acting with another, reporting with a third — without being asked to. That is why it pays to choose skills as a toolkit rather than one at a time: each addition makes the others more capable.
Setting this up well takes judgment as much as effort — which skills, in what configuration, integrated with which of your tools. That is the work we do for clients: full OpenClaw setup, skills like Code Runner tuned to your workflow, and custom development where nothing off the shelf fits.