local-code-interpreter

A zero-dependency, zero-config skill that gives your AI agent the ability to execute Python code and shell commands directly on your local machine.

No AWS. No API keys. No cold starts. Just Python.

---

Features

• ✅ Execute Python code inline or from files
• ✅ Run shell commands natively
• ✅ Full pandas / numpy / matplotlib support out of the box
• ✅ Chinese chart labels (PingFang HK on macOS, WenQuanYi on Linux)
• ✅ Works offline
• ✅ Compatible with OpenClaw / CodeBuddy / Claude Code / Cursor

Requirements

Tool	Version
Python	3.10+
OS	macOS or Linux

No extra Python packages are required at install time — the skill installs missing libraries on demand via pip3 install.

Installation

npx skills add https://github.com/Allen091080/local-code-interpreter -g -y

Or manually copy SKILL.md and scripts/ into your agent's skill directory:

~/.workbuddy/skills/local-code-interpreter/
~/.claude/skills/local-code-interpreter/
~/.agents/skills/local-code-interpreter/

Quick Start

Once the skill is loaded, just ask your agent:

帮我用 pandas 分析这个 CSV 文件
画一个柱状图显示每月营收
用 Python 计算这组数据的标准差

The agent will automatically write and execute the code locally.

Usage — Direct CLI

# Run a code string
python3 scripts/run.py --code "import sys; print(sys.version)"

# Run a file
python3 scripts/run.py --file /tmp/my_analysis.py

# Suppress header output
python3 scripts/run.py --code "print(42)" --no-header

Example: Chart with Chinese Labels

import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt

plt.rcParams['font.family'] = 'PingFang HK'   # macOS
months = ['1月','2月','3月','4月','5月','6月']
revenue = [12, 19, 15, 25, 22, 30]

plt.figure(figsize=(8, 4))
plt.plot(months, revenue, 'o-', color='#2196F3', linewidth=2)
plt.title('上半年营收趋势（万元）')
plt.tight_layout()
plt.savefig('/tmp/chart.png', dpi=150, bbox_inches='tight')
print('Saved: /tmp/chart.png')

Repository Structure

local-code-interpreter/
├── SKILL.md              # Agent skill definition
├── scripts/
│   └── run.py            # Core execution script
├── README.md             # This file
├── LICENSE               # MIT License
└── .github/
    └── workflows/
        └── test.yml      # CI — smoke test on push

Security Notes

• Code runs with your user's permissions — the same as if you ran it in a terminal.
• There is no sandboxing. Only use this skill in contexts you trust.
• Do not expose this skill to untrusted input sources.

License

MIT © 2026