Agents & OrchestrationSûr100100/100
Sandbox SDK
Build sandboxed applications for secure code execution. Load when building AI code execution, code interpreters, CI/CD systems, interactive dev environments, or executing untrusted code. Covers Sandbox SDK lifecycle, commands, files, code interpreter, and preview URLs. Biases towards retrieval from Cloudflare docs over pre-trained knowledge.
ou envoie-le directement à ton agent.
Installer dans ton projet
$ npx arboris-cli@latest install sandbox-sdkCloudflare
affaan-m/ECC
Contenu à copier
---
name: sandbox-sdk
description: Build sandboxed applications for secure code execution. Load when building AI code execution, code interpreters, CI/CD systems, interactive dev environments, or executing untrusted code. Covers Sandbox SDK lifecycle, commands, files, code interpreter, and preview URLs. Biases towards retrieval from Cloudflare docs over pre-trained knowledge.
metadata:
origin: Cloudflare
---
# Cloudflare Sandbox SDK
Build secure, isolated code execution environments on Cloudflare Workers.
## FIRST: Verify Installation
```bash
npm install @cloudflare/sandbox
docker info # Must succeed - Docker required for local dev
```
## Retrieval Sources
Your knowledge of the Sandbox SDK may be outdated. **Prefer retrieval over pre-training** for any Sandbox SDK task.
| Resource | URL |
|----------|-----|
| Docs | https://developers.cloudflare.com/sandbox/ |
| API Reference | https://developers.cloudflare.com/sandbox/api/ |
| Examples | https://github.com/cloudflare/sandbox-sdk/tree/main/examples |
| Get Started | https://developers.cloudflare.com/sandbox/get-started/ |
When implementing features, fetch the relevant doc page or example first.
## Required Configuration
**wrangler.jsonc** (exact - do not modify structure):
```jsonc
{
"containers": [{
"class_name": "Sandbox",
"image": "./Dockerfile",
"instance_type": "lite",
"max_instances": 1
}],
"durable_objects": {
"bindings": [{ "class_name": "Sandbox", "name": "Sandbox" }]
},
"migrations": [{ "new_sqlite_classes": ["Sandbox"], "tag": "v1" }]
}
```
**Worker entry** - must re-export Sandbox class:
```typescript
import { getSandbox } from '@cloudflare/sandbox';
export { Sandbox } from '@cloudflare/sandbox'; // Required export
```
## Quick Reference
| Task | Method |
|------|--------|
| Get sandbox | `getSandbox(env.Sandbox, 'user-123')` |
| Run command | `await sandbox.exec('python script.py')` |
| Run code (interpreter) | `await sandbox.runCode(code, { language: 'python' })` |
| Write file | `await sandbox.writeFile('/workspace/app.py', content)` |
| Read file | `await sandbox.readFile('/workspace/app.py')` |
| Create directory | `await sandbox.mkdir('/workspace/src', { recursive: true })` |
| List files | `await sandbox.listFiles('/workspace')` |
| Expose port | `await sandbox.exposePort(8080)` |
| Destroy | `await sandbox.destroy()` |
## Core Patterns
### Execute Commands
```typescript
const sandbox = getSandbox(env.Sandbox, 'user-123');
const result = await sandbox.exec('python --version');
// result: { stdout, stderr, exitCode, success }
```
### Code Interpreter (Recommended for AI)
Use `runCode()` for executing LLM-generated code with rich outputs:
```typescript
const ctx = await sandbox.createCodeContext({ language: 'python' });
await sandbox.runCode('import pandas as pd; data = [1,2,3]', { context: ctx });
const result = await sandbox.runCode('sum(data)', { context: ctx });
// result.results[0].text = "6"
```
**Languages**: `python`, `javascript`, `typescript`
State persists within context. Create explicit contexts for production.
### File Operations
```typescript
await sandbox.mkdir('/workspace/project', { recursive: true });
await sandbox.writeFile('/workspace/project/main.py', code);
const file = await sandbox.readFile('/workspace/project/main.py');
const files = await sandbox.listFiles('/workspace/project');
```
## When to Use What
| Need | Use | Why |
|------|-----|-----|
| Shell commands, scripts | `exec()` | Direct control, streaming |
| LLM-generated code | `runCode()` | Rich outputs, state persistence |
| Build/test pipelines | `exec()` | Exit codes, stderr capture |
| Data analysis | `runCode()` | Charts, tables, pandas |
## Extending the Dockerfile
Base image (`docker.io/cloudflare/sandbox:0.7.0`) includes Python 3.11, Node.js 20, and common tools.
Add dependencies by extending the Dockerfile:
```dockerfile
FROM docker.io/cloudflare/sandbox:0.7.0
# Python packages
RUN pip install requests beautifulsoup4
# Node packages (global)
RUN npm install -g typescript
# System packages
RUN apt-get update && apt-get install -y ffmpeg && rm -rf /var/lib/apt/lists/*
EXPOSE 8080 # Required for local dev port exposure
```
Keep images lean - affects cold start time.
## Preview URLs (Port Exposure)
Expose HTTP services running in sandboxes:
```typescript
const { url } = await sandbox.exposePort(8080);
// Returns preview URL for the service
```
**Production requirement**: Preview URLs need a custom domain with wildcard DNS (`*.yourdomain.com`). The `.workers.dev` domain does not support preview URL subdomains.
See: https://developers.cloudflare.com/sandbox/guides/expose-services/
## OpenAI Agents SDK Integration
The SDK provides helpers for OpenAI Agents at `@cloudflare/sandbox/openai`:
```typescript
import { Shell, Editor } from '@cloudflare/sandbox/openai';
```
See `examples/openai-agents` for complete integration pattern.
## Sandbox Lifecycle
- `getSandbox()` returns immediately - container starts lazily on first operation
- Containers sleep after 10 minutes of inactivity (configurable via `sleepAfter`)
- Use `destroy()` to immediately free resources
- Same `sandboxId` always returns same sandbox instance
## Anti-Patterns
- **Don't use internal clients** (`CommandClient`, `FileClient`) - use `sandbox.*` methods
- **Don't skip the Sandbox export** - Worker won't deploy without `export { Sandbox }`
- **Don't hardcode sandbox IDs for multi-user** - use user/session identifiers
- **Don't forget cleanup** - call `destroy()` for temporary sandboxes
## Detailed References
- **[references/api-quick-ref.md](references/api-quick-ref.md)** - Full API with options and return types
- **[references/examples.md](references/examples.md)** - Example index with use casesColle ce Markdown dans ton agent ou utilise les boutons ci-dessus pour l’écrire dans ton projet.
Ce que fait Sandbox SDK
`getSandbox()` returns immediately - container starts lazily on first operation
Containers sleep after 10 minutes of inactivity (configurable via `sleepAfter`)
Use `destroy()` to immediately free resources
Same `sandboxId` always returns same sandbox instance
Comment utiliser Sandbox SDK
1
Copie le prompt
Un clic copie le prompt packagé (ou l'envoie à ton agent).
2
L'agent installe le skill
Il ajoute le SKILL.md et ses ressources à ton projet.
3
Activation automatique
Le skill s'active dès que le contexte correspond.
Déclencheurs pour Sandbox SDK
Dis simplement à ton agent quelque chose comme :
Applique le skill sandbox-sdk à cette tâche
Utilise sandbox-sdk pour améliorer cette implémentation
Passe en revue ce sujet avec sandbox-sdk
Skills liés à Sandbox SDK
Acceleration (Microsoft)
Compare synthetic acceleration options and clearly separate scenario value from forecast commitments.Account Overview (Microsoft)
Create a synthetic Acme account overview and identify evidence that merits seller attention.Acroform Writer (Microsoft)
Use this skill whenever the user wants values written INTO an existing PDF form that lives in the connected SharePoint knowledge source — filling out, completing, populating, or submitting a fillable PDF (application, intake sheet, contract, government form) from data they supply, a spreadsheet row, or the conversation. Triggers include "fill out the job application form", "complete the intake form for [person]", "populate our standard NDA template", and "make this ready to send / non-editable" (flatten). The source form should be found via knowledge search; only ask the user to upload it if knowledge search turns up nothing plausible. different task), and do NOT use it on scanned/image-only PDFs that have no real fillable fields — those need OCR or manual overlay instead, which this skill explicitly detects and reports rather than silently failing.