OpenAI Agents

Give an OpenAI Agents SDK agent sandbox tools with sandboxTools from @opencoredev/sandbox-sdk/openai. A strict tool array for new Agent.

sandboxTools(box) from @opencoredev/sandbox-sdk/openai returns a tool array for the OpenAI Agents SDK. Pass it as tools to new Agent and the agent can run commands, read and edit files, and start dev servers in the sandbox.

For the tools themselves and how to limit them, see Give your agent a sandbox.

Install

npm install @opencoredev/sandbox-sdk @openai/agents zod

pnpm add @opencoredev/sandbox-sdk @openai/agents zod

yarn add @opencoredev/sandbox-sdk @openai/agents zod

bun add @opencoredev/sandbox-sdk @openai/agents zod

Needs @openai/agents 0.20 and Zod 4.6. Some cloud providers also need their own SDK; Install lists them.

Run an agent

import { Agent, run } from "@openai/agents";
import { createSandbox } from "@opencoredev/sandbox-sdk";
import { e2b } from "@opencoredev/sandbox-sdk/e2b";
import { sandboxTools } from "@opencoredev/sandbox-sdk/openai";

await using box = await createSandbox(e2b());

const coder = new Agent({
  name: "Coder",
  instructions: "You work inside a Linux sandbox. Verify every change by running it.",
  tools: sandboxTools(box),
});

const result = await run(coder, "Find every TODO in /workspace and list them by file.");
console.log(result.finalOutput);

This uses OPENAI_API_KEY and the Agents SDK’s default model. The agent searches with the run tool, for example grep -rn TODO, and answers in result.finalOutput. Allow the optional search tool to give it a dedicated search.

Keep a sandbox across runs

Give the sandbox a key and later runs pick up where the last one stopped:

import { Agent, run } from "@openai/agents";
import { createSandbox } from "@opencoredev/sandbox-sdk";
import { e2b } from "@opencoredev/sandbox-sdk/e2b";
import { sandboxTools } from "@opencoredev/sandbox-sdk/openai";

const userId = "user_123";

await using box = await createSandbox(e2b(), {
  key: `agent:${userId}`,
  timeoutMs: 2 * 60 * 60_000,
});

const coder = new Agent({
  name: "Coder",
  instructions: "Your project is in /workspace/app.",
  tools: sandboxTools(box, { root: "/workspace/app" }),
});

const result = await run(coder, "Add a README that explains how to run the project.");
console.log(result.finalOutput, box.created);

With a key, await using closes the handle and leaves the sandbox running. box.created is false on every run after the first. timeoutMs makes E2B delete it two hours after it was created.

What the Agents SDK gets

Each tool is a strict function tool with a Zod schema, so the model’s arguments are validated before the tool runs. A tool returns its compact text result: a verdict line and capped output. Aborting the run aborts the command inside the sandbox too.