code-runner - SKILL.md Agent Skill

name: code-runner description: "Execute Python code snippets in a sandboxed environment. Supports data analysis, visualization, and quick scripts." version: 1.0.0 metadata: echo: tags: [Python, Code, Execute, Sandbox, DataAnalysis]

Code Runner

Safe Python code execution with resource limits and import restrictions.

Usage

python3 scripts/safe_exec.py "print(sum(range(100)))"
python3 scripts/safe_exec.py --file script.py
python3 scripts/safe_exec.py --timeout 10 "import time; time.sleep(5); print('done')"

Security

Best-effort guards (NOT a true sandbox — code can still read the filesystem):

Timeout: default 30 seconds, configurable
Memory limit: 256MB on Linux (via resource.setrlimit); not enforced on macOS
Blocked patterns: os.system, subprocess, shutil.rmtree, __import__('os')
AST check: scans code for dangerous import patterns before execution
Isolated working dir: runs in a fresh tmpdir, cleaned up after execution
Minimal env: only PATH/HOME/LANG passed to subprocess

Limitations: This is NOT a security sandbox. The child process can read arbitrary files on the host filesystem. For untrusted code, use a container-based executor (e.g., docker-manage skill) instead.

Allowed Libraries

Safe for use (common data/analysis):

math, statistics, decimal, fractions
json, csv, re, datetime, collections
pandas, numpy (if installed)
matplotlib (saves to file, no display)

Blocked Patterns

BLOCKED = [
    "os.system", "os.exec", "os.popen", "os.remove",
    "subprocess", "shutil.rmtree", "importlib",
    "__import__", "eval(", "exec(",
    "open('/etc", "open('/root",
]

Example Workflows

Data analysis:

import pandas as pd
df = pd.read_csv("/tmp/data.csv")
print(df.describe())
print(df.groupby("category")["amount"].sum())

Quick plot (saved to file):

import matplotlib.pyplot as plt
plt.plot([1,2,3,4], [1,4,2,3])
plt.savefig("/tmp/plot.png")
print("Plot saved to /tmp/plot.png")