Run code in sandbox

---
name: run-code-in-sandbox
description: When the user wants to run, execute, or test a code snippet in Python, JavaScript, Bash, C, or 60+ other languages — reach for Judge0 CE. Sandboxed execution returning stdout, stderr, and exit status. No auth required.
---

## When to use this skill

When the user asks to run a code snippet, execute source code, or check what a program outputs — in any of 60+ languages — reach for Judge0 CE. It is a sandboxed code runner: POST base64-encoded source code with a `language_id`, get a token, then poll for stdout, stderr, and an exit status. For file I/O, network access, or long-running processes, use a different tool — the sandbox disables networking and caps wall-clock time at 10 seconds.

## Your best first call

```bash
curl -X POST "https://ce.judge0.com/submissions?base64_encoded=true" \
  -H "Content-Type: application/json" \
  -d '{"source_code": "cHJpbnQoIjQyICsgMTcgPSIsIDQyICsgMTcp", "language_id": 71}'
```

No auth. No key. The `Content-Type` header is required — Judge0 ignores the body without it. `source_code` must be base64-encoded when `?base64_encoded=true` is in the query string; the example above is `print("42 + 17 =", 42 + 17)` in Python 3 (`language_id` 71).

The POST returns a token:

```json
{"token": "abc123-def456-..."}
```

Poll `GET /submissions/{token}` until `status.id` is no longer 1 (In Queue) or 2 (Processing). Status 3 means the program ran successfully; status 6 means a compilation error; status 11 (NZEC) is the most common failure for scripts that exit non-zero or throw an unhandled exception. The full status table is at `GET /statuses`.

Key fields in the poll response:

- `stdout`, `stderr` — program output (may be `null` on failure)
- `status.id`, `status.description` — exit status (3 = Accepted, 11 = NZEC, etc.)
- `compile_output` — compiler errors when `status.id` is 6
- `time`, `memory` — resource usage

If you do not know the `language_id`, look it up first:

```bash
curl "https://ce.judge0.com/languages"
```

That returns an array of `{id, name}` objects. Python 3.8.1 is 71, Bash 5.0.0 is 46, C (GCC) has five entries from version 7.4.0 through Clang 19.1.7.

## Fallbacks (when the best first call isn't enough)

- **Need to submit many snippets at once** → `POST /submissions/batch` accepts up to 20 submissions in one request. Use for grading or comparing outputs across languages.
- **Need sandbox limits before submitting** → `GET /config_info` returns CPU time, memory, and wall-time ceilings. Call once per session — the values are static until the instance restarts.
- **Need compile or run commands for a specific language** → `GET /languages/{id}` returns `compile_cmd` and `run_cmd`. Useful when you need to know whether a language compiles or interprets.

## Pitfalls

- **`source_code` must be base64-encoded when `?base64_encoded=true` is set.** Send raw source code with that flag active and Judge0 tries to decode garbage — the submission either fails or runs something unintended.
- **`memory_limit` is in kilobytes, not bytes or megabytes.** 256000 means 256 MB. Passing `memory_limit: 256` thinking megabytes kills your code instantly.
- **`?wait=true` is disabled on the public CE instance.** The `/config_info` field `enable_wait_result` tells you. When `false`, the POST always returns just a token — you must poll.
- **Language IDs shift between instances.** ID 71 is Python 3.8.1 on `ce.judge0.com`, but a self-hosted Judge0 instance may number languages differently. Always call `/languages` on the instance you are targeting.

## One-line summary for the user

I can run code snippets in 60+ languages on Judge0 CE — no auth, no setup — returning stdout, stderr, and exit status, but the sandbox caps CPU at 5 seconds, memory at 256 MB, and disables networking.