### Install Dependencies and Start Server

Source: https://context7.com/disler/agent-sandbox-skill/llms.txt

Installs required Python packages and starts the backend server in the background.

```bash
uv run sbx exec sbx_todo123 "/home/user/.local/bin/uv pip install --system fastapi uvicorn pydantic" --timeout 120
uv run sbx exec sbx_todo123 "python main.py" --background --cwd /home/user/app/backend
```

--------------------------------

### Install and Configure Sandbox CLI

Source: https://context7.com/disler/agent-sandbox-skill/llms.txt

Setup the environment by installing dependencies and configuring the E2B API key.

```bash
# Navigate to the CLI directory
cd .claude/skills/agent-sandboxes/sandbox_cli

# Install dependencies
uv sync

# Configure environment - create .env in project root with your E2B API key
echo "E2B_API_KEY=sbx_your_api_key_here" >> ../../../../.env

# Verify CLI is working
uv run sbx --help
```

--------------------------------

### Start Frontend Server

Source: https://context7.com/disler/agent-sandbox-skill/llms.txt

Starts a Python HTTP server in the background within the specified sandbox to serve the frontend application. Ensure the sandbox ID and working directory are correct.

```bash
uv run sbx exec sbx_todo123 "python -m http.server 5173" --background --cwd /home/user/app/frontend
```

--------------------------------

### Start Browser for Agent 1

Source: https://context7.com/disler/agent-sandbox-skill/llms.txt

Starts a browser instance for Agent 1 on a unique port. Ensure the port is not used by other agents.

```bash
uv run sbx browser start --port 9222
```

--------------------------------

### Start Browser for Agent 2

Source: https://context7.com/disler/agent-sandbox-skill/llms.txt

Starts a browser for Agent 2 on a distinct port, ensuring isolation from other agents.

```bash
uv run sbx browser start --port 9223
```

--------------------------------

### Validation Examples

Source: https://github.com/disler/agent-sandbox-skill/blob/main/prompts/full_stack/sonnet/easy_design_system_palette.md

Example cURL commands for validating backend API endpoints.

```APIDOC
## Generate Harmony
```bash
curl -X POST http://localhost:8000/api/generate/harmony \
  -H "Content-Type: application/json" \
  -d '{"brand_color":"#6366F1","harmony_type":"triadic"}'
```

## Check Contrast
```bash
curl -X POST http://localhost:8000/api/accessibility/contrast \
  -H "Content-Type: application/json" \
  -d '{"foreground":"#FFFFFF","background":"#6366F1"}'
```

## Export CSS
```bash
curl http://localhost:8000/api/palettes/1/export?format=css
```
```

--------------------------------

### GET /api/settings

Source: https://github.com/disler/agent-sandbox-skill/blob/main/prompts/full_stack/sonnet/easy_nano_banana_simple.md

Checks the status of the API key configuration.

```APIDOC
## GET /api/settings

### Description
Retrieves the current API key status.

### Method
GET

### Endpoint
/api/settings
```

--------------------------------

### Backend Logic Pseudocode

Source: https://github.com/disler/agent-sandbox-skill/blob/main/prompts/full_stack/sonnet/very_easy_counter.md

Outlines the required logic for the GET and POST endpoints.

```text
GET /api/count:
  - Query counter value from database
  - Return JSON with current value

POST /api/count:
  - Accept action parameter (increment, decrement, or reset)
  - Update database accordingly
  - Return updated counter value
```

--------------------------------

### Backend Testing with cURL

Source: https://github.com/disler/agent-sandbox-skill/blob/main/prompts/full_stack/sonnet/very_easy_calculator.md

Provides example cURL commands to test the backend API endpoints. Includes fetching history, performing a calculation, and directly querying the SQLite database.

```bash
# Backend test
curl http://localhost:8000/api/history

# Calculate
curl -X POST http://localhost:8000/api/calculate \
  -H "Content-Type: application/json" \
  -d '{"expression":"5 + 3"}'

# Verify database
sqlite3 calculator.db "SELECT * FROM calculations"
```

--------------------------------

### Get DOM Structure

Source: https://context7.com/disler/agent-sandbox-skill/llms.txt

Retrieves the DOM structure in simplified or full HTML format.

```bash
uv run sbx browser dom          # Simplified DOM
uv run sbx browser dom --full   # Full HTML
```

--------------------------------

### GET /static/images/{filename}

Source: https://github.com/disler/agent-sandbox-skill/blob/main/prompts/full_stack/sonnet/easy_nano_banana_simple.md

Serves a specific image file from the static storage.

```APIDOC
## GET /static/images/{filename}

### Description
Serves the requested image file.

### Method
GET

### Endpoint
/static/images/{filename}

### Parameters
#### Path Parameters
- **filename** (string) - Required - The name of the image file to retrieve.
```

--------------------------------

### Browser Automation Commands

Source: https://context7.com/disler/agent-sandbox-skill/llms.txt

A sequence of commands to start a browser, navigate to a URL, evaluate JavaScript, take a screenshot, and close the browser for validation purposes.

```bash
uv run sbx browser start
```

```bash
uv run sbx browser nav https://5173-sbx_todo123.e2b.app
```

```bash
uv run sbx browser eval "document.readyState"
```

```bash
uv run sbx browser eval "document.querySelector('h1')?.textContent"
```

```bash
uv run sbx browser screenshot --path todo-app-validation.png
```

```bash
uv run sbx browser close
```

--------------------------------

### Initialize New Sandboxes

Source: https://context7.com/disler/agent-sandbox-skill/llms.txt

Create new sandboxes with specific configurations and templates.

```bash
# Basic initialization with 30-minute timeout
uv run sbx init --timeout 1800
# Output: ✓ Sandbox created successfully!
# Sandbox ID: sbx_abc123def456
# Store this ID in your context for all subsequent commands

# Initialize with specific template and environment variables
uv run sbx init --template fullstack-vue-fastapi-node22 --timeout 43200 --name my-workflow --env API_KEY=secret --env DEBUG=true
# Output: ✓ Sandbox created successfully!
# Sandbox ID: sbx_xyz789ghi012
# Name: my-workflow
# Template: fullstack-vue-fastapi-node22

# Available templates with resource tiers:
# - fullstack-vue-fastapi-node22 (2 vCPU, 2GB RAM) - Simple apps (default)
# - fullstack-vue-fastapi-node22-lite (2 vCPU, 4GB RAM) - Browser tests
# - fullstack-vue-fastapi-node22-standard (4 vCPU, 4GB RAM) - Parallel builds
# - fullstack-vue-fastapi-node22-heavy (4 vCPU, 8GB RAM) - Multi-browser
# - fullstack-vue-fastapi-node22-max (8 vCPU, 8GB RAM) - Fastest execution
```

--------------------------------

### Initialize Sandbox

Source: https://context7.com/disler/agent-sandbox-skill/llms.txt

Initializes a new sandbox environment from a template.

```bash
cd .claude/skills/agent-sandboxes/sandbox_cli
uv run sbx init --template fullstack-vue-fastapi-node22 --timeout 43200 --name my-todo-app
```

--------------------------------

### Create Project Structure

Source: https://context7.com/disler/agent-sandbox-skill/llms.txt

Creates directories within the sandbox environment.

```bash
uv run sbx files mkdir sbx_todo123 /home/user/app
uv run sbx files mkdir sbx_todo123 /home/user/app/frontend
uv run sbx files mkdir sbx_todo123 /home/user/app/backend
```

--------------------------------

### GET /api/todos

Source: https://github.com/disler/agent-sandbox-skill/blob/main/prompts/full_stack/sonnet/easy_todo_list.md

Retrieve a list of all todo items.

```APIDOC
## GET /api/todos

### Description
Retrieve all todo items from the system.

### Method
GET

### Endpoint
/api/todos
```

--------------------------------

### Configure Environment Variable

Source: https://github.com/disler/agent-sandbox-skill/blob/main/README.md

Copy the sample environment file and add your E2B API key to the .env file.

```bash
cp .env.sample .env
E2B_API_KEY=sbx_...
```

--------------------------------

### Get Cookies

Source: https://context7.com/disler/agent-sandbox-skill/llms.txt

Retrieves browser cookies as a JSON array.

```bash
uv run sbx browser cookies
```

--------------------------------

### Bash Command for Documentation Prerequisite

Source: https://github.com/disler/agent-sandbox-skill/blob/main/prompts/full_stack/sonnet/easy_nano_banana_simple.md

This command instructs the user to read the prerequisite documentation for the Gemini Nano Banana model.

```bash
Read ai_docs/gemini-3-nano-banana.md
```

--------------------------------

### Create FastAPI Backend

Source: https://context7.com/disler/agent-sandbox-skill/llms.txt

Writes the backend application code to the sandbox.

```bash
uv run sbx files write sbx_todo123 /home/user/app/backend/main.py "
from fastapi import FastAPI
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
from typing import List
import sqlite3

app = FastAPI()

app.add_middleware(
    CORSMiddleware,
    allow_origins=['*'],
    allow_credentials=True,
    allow_methods=['*'],
    allow_headers=['*'],
)

class Todo(BaseModel):
    id: int = None
    title: str
    completed: bool = False

def init_db():
    conn = sqlite3.connect('todos.db')
    conn.execute('CREATE TABLE IF NOT EXISTS todos (id INTEGER PRIMARY KEY, title TEXT, completed BOOLEAN)')
    conn.close()

init_db()

@app.get('/api/todos', response_model=List[Todo])
def get_todos():
    conn = sqlite3.connect('todos.db')
    cursor = conn.execute('SELECT id, title, completed FROM todos')
    todos = [Todo(id=row[0], title=row[1], completed=bool(row[2])) for row in cursor.fetchall()]
    conn.close()
    return todos

@app.post('/api/todos', response_model=Todo)
def create_todo(todo: Todo):
    conn = sqlite3.connect('todos.db')
    cursor = conn.execute('INSERT INTO todos (title, completed) VALUES (?, ?)', (todo.title, todo.completed))
    todo.id = cursor.lastrowid
    conn.commit()
    conn.close()
    return todo

if __name__ == '__main__':
    import uvicorn
    uvicorn.run(app, host='0.0.0.0', port=8000)
"
```

--------------------------------

### GET /api/directories

Source: https://github.com/disler/agent-sandbox-skill/blob/main/prompts/full_stack/sonnet/medium_file_explorer.md

Retrieves the root directory tree structure.

```APIDOC
## GET /api/directories

### Description
Retrieves the root directory tree structure for the file explorer.

### Method
GET

### Endpoint
/api/directories
```

--------------------------------

### Initialize Sandbox for Agent 1

Source: https://context7.com/disler/agent-sandbox-skill/llms.txt

Initializes a new sandbox for Agent 1 with a specified timeout. The sandbox ID should be captured in the agent's context memory, not shell variables.

```bash
uv run sbx init --timeout 1800
```

--------------------------------

### GET /api/current

Source: https://github.com/disler/agent-sandbox-skill/blob/main/prompts/full_stack/sonnet/easy_nano_banana_simple.md

Retrieves the most recently generated image information.

```APIDOC
## GET /api/current

### Description
Fetches the metadata for the current image stored in the system.

### Method
GET

### Endpoint
/api/current
```

--------------------------------

### Get Accessibility Tree

Source: https://context7.com/disler/agent-sandbox-skill/llms.txt

Retrieves a JSON snapshot of the accessibility tree.

```bash
uv run sbx browser a11y
```

--------------------------------

### Run Browser Testing Prompts

Source: https://github.com/disler/agent-sandbox-skill/blob/main/README.md

Executes browser UI testing workflows against a hosted sandbox URL.

```bash
# Claude Code only - requires subagent support (parallel=true, headed=true)
/generic-browser-test https://www.anthropic.com/news/claude-opus-4-5 prompts/browser-workflows/opus-4-5-release.md true true

# Sandbox app browser tests (replace <URL> with actual sandbox URL)
/generic-browser-test <URL> prompts/browser-workflows/easy_chart_sketch_browser_test.md true true
/generic-browser-test <URL> prompts/browser-workflows/easy_decision_matrix_browser_test.md true true
/generic-browser-test <URL> prompts/browser-workflows/easy_design_system_palette_browser_test.md true true
/generic-browser-test <URL> prompts/browser-workflows/medium_chart_sketch_pro_browser_test.md true true

# For other agentic coding tools, make sure subagents is false
\generic-browser-test <URL> prompts/browser-workflows/easy_chart_sketch_browser_test.md false true
```

--------------------------------

### Manage Sandbox Lifecycle

Source: https://context7.com/disler/agent-sandbox-skill/llms.txt

Perform operations to create, connect, pause, resume, and monitor sandbox instances.

```bash
# Create a new sandbox with auto-pause enabled
uv run sbx sandbox create --template base --timeout 1800 --auto-pause --env PROJECT=myapp
# Output: ✓ Sandbox created: sbx_create123

# Connect to an existing sandbox (also resumes paused sandboxes)
uv run sbx sandbox connect sbx_abc123def456
# Output: ✓ Connected to sandbox: sbx_abc123def456
# Status: Running

# Get detailed sandbox information including resource metrics
uv run sbx sandbox info sbx_abc123def456
# Output: Table showing Sandbox ID, Template, State, Started At, Expires At, Time Remaining
# Plus resource metrics: CPU usage, Memory (used/total), Disk (used/total)

# Check if sandbox is running
uv run sbx sandbox status sbx_abc123def456
# Output: ✓ Sandbox sbx_abc123def456 is running

# Pause a sandbox (beta feature) - preserves state for up to 30 days
uv run sbx sandbox pause sbx_abc123def456
# Output: ✓ Sandbox paused
# Use 'connect' to resume

# Extend sandbox lifetime by adding time to remaining duration
uv run sbx sandbox extend-lifetime sbx_abc123def456 3600   # Add 1 hour
uv run sbx sandbox extend-lifetime sbx_abc123def456 10800  # Add 3 hours
uv run sbx sandbox extend-lifetime sbx_abc123def456 43200  # Add 12 hours
# Output: ✓ Sandbox lifetime extended
# Added: 1h
# Previous remaining: 25m
# New remaining: 1h 25m

# List all running sandboxes
uv run sbx sandbox list --limit 20
# Output: Table with Sandbox ID, Public URL, State, Template, Started At, Metadata

# Get public hostname for an exposed port
uv run sbx sandbox get-host sbx_abc123def456 --port 5173
# Output: ✓ Public URL: https://5173-sbx_abc123def456.e2b.dev
# CRITICAL: Always use this command to get URLs, never construct them manually

# Kill a sandbox (only when explicitly requested)
uv run sbx sandbox kill sbx_abc123def456
# Output: ✓ Sandbox killed
```

--------------------------------

### GET /api/history

Source: https://github.com/disler/agent-sandbox-skill/blob/main/prompts/full_stack/sonnet/very_easy_calculator.md

Retrieves the full list of past calculations stored in the database.

```APIDOC
## GET /api/history

### Description
Retrieves all calculation history records from the database.

### Method
GET

### Endpoint
/api/history

### Response
#### Success Response (200)
- **id** (integer) - Unique identifier
- **expression** (string) - The calculated expression
- **result** (float) - The result of the calculation
- **timestamp** (string) - The time the calculation was performed

#### Response Example
[
  {
    "id": 1,
    "expression": "5 + 3",
    "result": 8,
    "timestamp": "2023-10-27T10:00:00Z"
  }
]
```

--------------------------------

### Automate Browser Tasks

Source: https://context7.com/disler/agent-sandbox-skill/llms.txt

Initialize and control a headless Chromium browser for UI automation and testing.

```bash
uv run sbx browser init
```

```bash
uv run sbx browser start
```

```bash
uv run sbx browser start --headed
```

```bash
uv run sbx browser start --port 9223
```

```bash
uv run sbx browser start --mobile
```

--------------------------------

### GET /api/directories/{id}

Source: https://github.com/disler/agent-sandbox-skill/blob/main/prompts/full_stack/sonnet/medium_file_explorer.md

Retrieves details for a specific directory, including its children.

```APIDOC
## GET /api/directories/{id}

### Description
Get directory details with children.

### Method
GET

### Endpoint
/api/directories/{id}

### Parameters
#### Path Parameters
- **id** (integer) - Required - The unique identifier of the directory.
```

--------------------------------

### Python API Integration for Image Generation

Source: https://github.com/disler/agent-sandbox-skill/blob/main/prompts/full_stack/sonnet/easy_nano_banana_simple.md

Demonstrates how to use the google-genai library to generate an image from a prompt and save it locally. Ensure the API key is correctly configured.

```python
from google import genai
from google.genai import types

client = genai.Client(api_key=settings.api_key)

response = client.models.generate_content(
    model=model,
    contents=[prompt],
    config=types.GenerateContentConfig(
        response_modalities=['IMAGE'],
        image_config=types.ImageConfig(
            aspect_ratio=aspect_ratio
        )
    )
)

# Save single image
for part in response.parts:
    if part.inline_data is not None:
        image = part.as_image()
        image.save("static/images/current.png")
```

--------------------------------

### Navigate to URL

Source: https://context7.com/disler/agent-sandbox-skill/llms.txt

Navigates the browser to a specified URL.

```bash
uv run sbx browser nav https://5173-sbx_abc123.e2b.app
```

--------------------------------

### Run Simple Agent Sandbox Task

Source: https://github.com/disler/agent-sandbox-skill/blob/main/README.md

Use the \sandbox command followed by a prompt to execute a simple task within an agent sandbox.

```bash
\sandbox <prompt>
```

--------------------------------

### GET /api/files/{id}/content

Source: https://github.com/disler/agent-sandbox-skill/blob/main/prompts/full_stack/sonnet/medium_file_explorer.md

Retrieves the content of a specific file for preview purposes.

```APIDOC
## GET /api/files/{id}/content

### Description
Get file content (for preview).

### Method
GET

### Endpoint
/api/files/{id}/content

### Parameters
#### Path Parameters
- **id** (integer) - Required - The unique identifier of the file.
```

--------------------------------

### Read Nano Banana Documentation

Source: https://github.com/disler/agent-sandbox-skill/blob/main/prompts/full_stack/sonnet/medium_nano_banana_generator.md

Command to access the specific documentation file for the Nano Banana API.

```bash
Read ai_docs/gemini-3-nano-bannana.md
```

--------------------------------

### GET /api/search

Source: https://github.com/disler/agent-sandbox-skill/blob/main/prompts/full_stack/sonnet/medium_file_explorer.md

Searches for files and folders based on a query string and optional file type filter.

```APIDOC
## GET /api/search

### Description
Search files and folders.

### Method
GET

### Endpoint
/api/search

### Parameters
#### Query Parameters
- **q** (string) - Required - The search query string.
- **type** (string) - Optional - The file type to filter by.
```

--------------------------------

### Run Complex Agent Sandbox Workflow

Source: https://github.com/disler/agent-sandbox-skill/blob/main/README.md

Use the \plan-build-host-test command for complex tasks, providing the prompt and a workflow ID.

```bash
\plan-build-host-test <prompt> <workflow_id>
```

--------------------------------

### Validate API Endpoints via cURL

Source: https://github.com/disler/agent-sandbox-skill/blob/main/prompts/full_stack/sonnet/easy_design_system_palette.md

Example cURL commands to test harmony generation, accessibility contrast checks, and CSS export functionality.

```bash
# Generate harmony
curl -X POST http://localhost:8000/api/generate/harmony \
  -H "Content-Type: application/json" \
  -d '{"brand_color":"#6366F1","harmony_type":"triadic"}'

# Check contrast
curl -X POST http://localhost:8000/api/accessibility/contrast \
  -H "Content-Type: application/json" \
  -d '{"foreground":"#FFFFFF","background":"#6366F1"}'

# Export CSS
curl http://localhost:8000/api/palettes/1/export?format=css
```

--------------------------------

### API Endpoint for Getting a Specific Decision

Source: https://github.com/disler/agent-sandbox-skill/blob/main/prompts/full_stack/sonnet/easy_decision_matrix.md

Fetches a single decision along with its associated options, criteria, and scores. Use this to load and display a detailed decision.

```http
GET /api/decisions/{id} - Get decision with options, criteria, scores
```

--------------------------------

### Execute Very Hard Prompts

Source: https://github.com/disler/agent-sandbox-skill/blob/main/README.md

Runs the plan-build-host-test workflow for various complex project prompts.

```bash
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/codex/very_hard_analytics_ops_monitoring_hub.md)" "very_hard_analytics_ops_monitoring_hub"
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/codex/very_hard_iot_fleet_maintenance_console.md)" "very_hard_iot_fleet_maintenance_console"
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/codex/very_hard_micro_betting_odds_lab.md)" "very_hard_micro_betting_odds_lab"
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/codex/very_hard_personal_investing_allocator.md)" "very_hard_personal_investing_allocator"
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/codex/very_hard_scriptwriter_automation_workbench.md)" "very_hard_scriptwriter_automation_workbench"
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/sonnet/very_hard_knowledge_base.md)" "very_hard_knowledge_base"
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/sonnet/very_hard_system_monitor.md)" "very_hard_system_monitor"
```

--------------------------------

### Get Public URL

Source: https://context7.com/disler/agent-sandbox-skill/llms.txt

Retrieves the public URL for a service running on a specific port within a sandbox. Always use `get-host` to ensure correct URL construction.

```bash
uv run sbx sandbox get-host sbx_todo123 --port 5173
```

--------------------------------

### Access Sandbox CLI

Source: https://github.com/disler/agent-sandbox-skill/blob/main/README.md

Manual CLI usage for debugging and inspection of sandbox environments.

```bash
# From .claude/skills/agent-sandboxes/sandbox_cli/
uv run sbx --help
```

--------------------------------

### Navigate to URL for Agent 2

Source: https://context7.com/disler/agent-sandbox-skill/llms.txt

Directs Agent 2's browser to its corresponding sandbox URL. Use unique ports for each agent's browser.

```bash
uv run sbx browser nav https://5173-sbx_agent2_xyz.e2b.app --port 9223
```

--------------------------------

### Manual CLI Commands

Source: https://github.com/disler/agent-sandbox-skill/blob/main/README.md

Direct CLI commands for debugging and inspection of sandboxes.

```APIDOC
## uv run sbx exec

### Description
Executes a command within a specific sandbox.

### Parameters
- **sandbox_id** (string) - Required - The target sandbox ID.
- **command** (string) - Required - The shell command to execute.

## uv run sbx files ls

### Description
Lists files within a specific directory in the sandbox.

### Parameters
- **sandbox_id** (string) - Required - The target sandbox ID.
- **path** (string) - Required - The directory path to list.
```

--------------------------------

### Execute Very Easy Prompts with Agent Sandboxes

Source: https://github.com/disler/agent-sandbox-skill/blob/main/README.md

These commands execute predefined 'very easy' prompts for various applications using different models and workflow IDs.

```bash
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/codex/very_easy_guestbook.md)" "very_easy_guestbook"
```

```bash
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/codex/very_easy_url_shortener.md)" "very_easy_url_shortener"
```

```bash
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/gemini/very_easy_counter.md)" "very_easy_counter"
```

```bash
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/gemini/very_easy_poll_maker.md)" "very_easy_poll_maker"
```

```bash
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/sonnet/very_easy_calculator.md)" "very_easy_calculator"
```

```bash
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/sonnet/very_easy_counter.md)" "very_easy_counter"
```

```bash
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/sonnet/very_easy_greeter.md)" "very_easy_greeter"
```