Code Interpreter

Code Interpreter

A general-purpose code execution environment powered by AWS Bedrock AgentCore Code Interpreter. Run code, execute shell commands, and manage files in a secure sandbox.

Available Tools

• execute_code(code, language, output_filename): Execute Python, JavaScript, or TypeScript code.
• execute_command(command): Execute shell commands.
• file_operations(operation, paths, content): Read, write, list, or remove files in the mounted session workspace.

Tool Parameters

execute_code

execute_command

file_operations

tool_input Examples

execute_code — text output

{
  "code": "import pandas as pd\ndf = pd.DataFrame({'A': [1,2,3], 'B': [4,5,6]})\nprint(df.describe())",
  "language": "python"
}

execute_code — generate chart

{
  "code": "import matplotlib\nmatplotlib.use('Agg')\nimport matplotlib.pyplot as plt\nimport numpy as np\nx = np.linspace(0, 10, 100)\nplt.figure(figsize=(10,6))\nplt.plot(x, np.sin(x))\nplt.title('Sine Wave')\nplt.savefig('sine.png', dpi=300, bbox_inches='tight')\nprint('Done')",
  "language": "python",
  "output_filename": "sine.png"
}

execute_command — install a package

{
  "command": "pip install yfinance"
}

execute_command — check environment

{
  "command": "python --version && pip list | head -20"
}

file_operations — write a file

{
  "operation": "write",
  "content": [{"path": "config.json", "text": "{\"key\": \"value\"}"}]
}

file_operations — list files

{
  "operation": "list",
  "paths": ["."]
}

file_operations — read a file

{
  "operation": "read",
  "paths": ["output.csv"]
}

When to Use This Skill

Use code-interpreter as a sandbox for testing and prototyping code. For production tasks (creating documents, charts, presentations), prefer specialized skills.

Do NOT use for:

• Formatting or displaying code examples (respond directly with markdown code blocks)
• Explaining code or algorithms (respond directly with text)
• Simple calculations you can do mentally (just provide the answer)
• Any task that doesn't require actual code execution

Code Interpreter vs Code Agent

Workspace Integration

The chat session has a persistent filesystem mounted at /mnt/workspace. Relative file paths used by Code Interpreter tools resolve inside this directory. Files written there are scratch files and remain available when the interpreter session is restarted. They are not user-downloadable artifacts unless output_filename is supplied.

The mount is required. If it cannot be configured or attached, Code Interpreter returns an error instead of starting an isolated non-persistent session.

Create persistent files directly:

{
  "tool": "execute_code",
  "code": "from pathlib import Path\nPath('/mnt/workspace/results.json').write_text('{\"ok\": true}')"
}

Use output_filename whenever a generated file must appear in Generated Files or be downloadable by the user. The tool publishes and verifies that file before returning success. Do not create Markdown links to /mnt/workspace or describe raw workspace paths as download links; the application renders the file action.

Uploaded files:

Files uploaded by the user are available in the mounted workspace without manual loading or base64 transfer under /mnt/workspace/inputs. JSON, JSONL, and NDJSON attachments may be represented by a bounded text excerpt in the conversation; use the mounted file when the full dataset is needed.

Use file_operations for scratch-file inspection. Published files are surfaced by the application and should be referenced by their displayed filename.

Environment

• Languages: Python (recommended, 200+ libraries), JavaScript, TypeScript
• Shell: Full shell access via execute_command
• File system: /mnt/workspace persists across Code Interpreter restarts for the chat session
• Session state: Variables persist within one interpreter session; files persist in the mounted workspace
• Network: Internet access available (can use requests, urllib, curl)

Supported Languages

• Python (recommended) — 200+ pre-installed libraries covering data science, ML, visualization, file processing
• JavaScript — Node.js runtime, useful for JSON manipulation, async operations
• TypeScript — TypeScript runtime with type checking

Pre-installed Python Libraries

Data Analysis & Visualization

Machine Learning & AI

Mathematical & Optimization

File Processing & Documents

Image & Media

Data Storage & Formats

Web & API

Utilities

For the full list of 200+ libraries with versions, run: execute_command(command="pip list")

Usage Patterns

Pattern 1: Data Analysis

import pandas as pd
import numpy as np

df = pd.DataFrame({
    'date': pd.date_range('2024-01-01', periods=100),
    'revenue': np.random.normal(1000, 200, 100),
    'costs': np.random.normal(700, 150, 100),
})
df['profit'] = df['revenue'] - df['costs']

print("=== Summary Statistics ===")
print(df.describe())
print(f"\nTotal Profit: ${df['profit'].sum():,.2f}")
print(f"Profit Margin: {df['profit'].mean() / df['revenue'].mean() * 100:.1f}%")

Pattern 2: Visualization (with output_filename)

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

fig, axes = plt.subplots(2, 2, figsize=(14, 10))

categories = ['Q1', 'Q2', 'Q3', 'Q4']
values = [120, 150, 180, 210]
axes[0,0].bar(categories, values, color='#2196F3')
axes[0,0].set_title('Quarterly Revenue')

x = np.linspace(0, 10, 50)
axes[0,1].plot(x, np.sin(x), 'b-', linewidth=2)
axes[0,1].set_title('Trend')

sizes = [35, 30, 20, 15]
axes[1,0].pie(sizes, labels=['A','B','C','D'], autopct='%1.1f%%')
axes[1,0].set_title('Market Share')

x = np.random.normal(50, 10, 200)
y = x * 1.5 + np.random.normal(0, 15, 200)
axes[1,1].scatter(x, y, alpha=0.5, c='#FF5722')
axes[1,1].set_title('Correlation')

plt.tight_layout()
plt.savefig('dashboard.png', dpi=300, bbox_inches='tight')
print('Dashboard saved')

Pattern 3: Machine Learning

from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import classification_report
from sklearn.datasets import load_iris

iris = load_iris()
X_train, X_test, y_train, y_test = train_test_split(
    iris.data, iris.target, test_size=0.3, random_state=42
)

model = RandomForestClassifier(n_estimators=100, random_state=42)
model.fit(X_train, y_train)
y_pred = model.predict(X_test)

print(classification_report(y_test, y_pred, target_names=iris.target_names))

Pattern 4: SQL with DuckDB

import duckdb
import pandas as pd

orders = pd.DataFrame({
    'order_id': range(1, 101),
    'customer': [f'Customer_{i%20}' for i in range(100)],
    'amount': [round(50 + i * 3.5, 2) for i in range(100)],
})

result = duckdb.sql("""
    SELECT customer, COUNT(*) as cnt, ROUND(SUM(amount), 2) as total
    FROM orders GROUP BY customer
    HAVING COUNT(*) >= 3 ORDER BY total DESC LIMIT 10
""").df()
print(result.to_string(index=False))

Pattern 5: Fetch Data from Web

import requests
import pandas as pd

response = requests.get("https://api.example.com/data")
data = response.json()
df = pd.DataFrame(data)
print(df.head())

Pattern 6: Multi-step Workflow (session state persists)

Call 1: execute_code → load and clean data, store in variable `df`
Call 2: execute_code → analyze `df`, generate chart, save as PNG
Call 3: execute_code → export results to CSV
Call 4: file_operations(operation="read") → download the CSV

Variables (df) and files persist across calls in the same session.

Important Rules

1. matplotlib.use('Agg') before import matplotlib.pyplot — sandbox has no display.
2. Use print() for text output — stdout is how results are returned.
3. output_filename must match exactly — the filename in plt.savefig() or wb.save() must match the output_filename parameter.
4. Use execute_command for shell tasks — ls, pip install, curl, etc.
5. Use file_operations for file management — read/write/list/remove files explicitly.
6. Session state persists — variables and files remain across calls. Use this for multi-step workflows.

Common Mistakes to Avoid

• Forgetting matplotlib.use('Agg') before import matplotlib.pyplot as plt
• Using plt.show() instead of plt.savefig() — there is no display
• Typo in output_filename — must match the file saved by the code exactly
• Using execute_code for shell tasks — use execute_command instead
• Writing binary files via file_operations — use execute_code to generate binary files, then download with output_filename