Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
An agent skill for managing isolated execution environments
| Date | Stars |
|---|---|
| 2026-07-31 | 378 |
| 2026-08-06 | 378 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# Agent Sandbox Skill
An agent skill for managing isolated execution environments using [E2B Sandboxes](https://e2b.dev/).
This skill enables AI agents (Gemini CLI, Claude Code, Codex CLI) to safely execute code, build full-stack applications, and perform arbitrary engineering tasks in a secure, isolated sandbox.
Watch the [Gemini 3 Demo](https://youtu.be/V5IhsHEHXOg) or the newer [Claude Opus 4.5](https://youtu.be/3kgx0YxCriM) Demo to understand what this codebase/skill can do for your agentic coding.
<img src="images/sandboxes.png" alt="Agent Sandboxes" width="800">
## Why Use Agent Sandboxes? The **Value Proposition**
>
> Agent Sandboxes unlock 3 key capabilities for your agentic engineering:
- **Isolation**: Each agent fork runs in a fully isolated, gated E2B sandbox, this means no matter what your agent does, it's secure and safe from your local filesystem and production environment.
- **Scale**: You can run as many agent forks as you want, each fork is independent and has its own sandbox. This is a very literal way to scale your compute to scale your impact.
- **Agency**: Your agents have full control over the sandbox environment, they can install packages, modify files, run commands, etc. This means they can handle more of the engineering process for you.
## 🚀 Features
* **Isolated Execution**: Run untrusted code, tests, and binaries safely.
* **Full-Stack Development**: Scaffold, build, and host Vue + FastAPI + SQLite apps.
* **Browser Automation**: Built-in Playwright integration for visual validation.
* **Agent-First Design**: Optimized for CLI agents with structured prompts and robust error handling.
* **Persistent Context**: Tools to manage sandbox lifecycles across agent turns.
## 🛠️ Setup
### Prerequisites
* Python >= 3.12
* `uv` package manager (recommended)
* E2B Account & API Key
### Installation
1. **Clone the repository** (if not already done).
2. **Configure Environment**:
Create a `.env` file in the project root:
```bash
cp .env.sample .env
```
Add your E2B API key:
```env
E2B_API_KEY=sbx_...
```
*(Get your key from [E2B Dashboard](https://e2b.dev/dashboard/keys))*
3. **Open your Agentic Coding Tool**:
Open your Agentic Coding Tool and start prompting!
```bash
claude
or
gemini
or
codex
```
Then for one of simple tasks run:
```bash
\sandbox <prompt>
```
For complex tasks run
```bash
\plan-build-host-test <prompt> <workflow_id>
```
## Starter Prompts
The `prompts/{claude,gemini,codex}/<difficulty>_<task>.md` contain full stack application prompts you can
Try these prompts out with a SOTA Model for the best results.
Run these prompts in your Agentic Coding Tool.
Start with the Very Easy Prompts to get a feel for the tool. Keep in mind token usage increases as the complexity of the prompt increases.
### Very Easy Prompts
```bash
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/codex/very_easy_guestbook.md)" "very_easy_guestbook"
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/codex/very_easy_url_shortener.md)" "very_easy_url_shortener"
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/gemini/very_easy_counter.md)" "very_easy_counter"
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/gemini/very_easy_poll_maker.md)" "very_easy_poll_maker"
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/sonnet/very_easy_calculator.md)" "very_easy_calculator"
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/sonnet/very_easy_counter.md)" "very_easy_counter"
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/sonnet/very_easy_greeter.md)" "very_easy_greeter"
```
### Easy Prompts
```bash
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/gemini/easy_api_mock_studio.md)" "easy_api_mock_studio"
\agent-sandboxes:plan-build-host-test "$(cat prompts/full_stack/gemini/easy_codeExcerpt of 13,429 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:20f6a0703be426c3, llm:description: 'An agent skill for managing isolated execution environments'
matched fp:20f6a0703be426c3, llm:description: 'An agent skill for managing isolated execution environments'