Code Runner

Code Runner skill gives your Claude agent the ability to write and execute code in a secure sandboxed environment, then use the output to inform further reasoning. This transforms the agent from a code suggester into a code executor — it can run calculations, process data, test algorithms, and iterate based on actual results rather than predictions.

Supported runtimes: Python 3.11 (with NumPy, Pandas, Matplotlib, and Requests pre-installed), Node.js 20, Bash, Ruby, Go, and Rust. Each execution runs in an isolated container with a 30-second timeout and 256MB memory limit, preventing runaway processes and protecting system resources. File I/O within the sandbox is permitted — scripts can read input files and write output files that persist to a configured directory.

A typical workflow: your agent generates a data processing script, runs it against a sample of your data in the sandbox, observes errors or incorrect output, fixes the script, and re-runs until the result is correct — all autonomously. This closed-loop iteration is far more reliable than one-shot code generation.

Visualization support: Matplotlib plots are rendered as PNG files and returned as base64 images for display or saving. The skill also handles package installation within the sandbox (pip install, npm install) for common libraries not included in the base environment. Data scientists, researchers, and developers use Code Runner to build autonomous data analysis agents that handle real computation tasks end-to-end.

Installation

clawhub install code-runner

Install: clawhub install code-runner