agentleFS
casooncasoon/ai-agent-config/skills/sandbox-next/SKILL.md

sandbox-next

Skill1 starsChanged 47 days agoMITView on GitHub

Use when building or changing Cloudflare Sandbox apps on @cloudflare/sandbox@next (Sandbox SDK 1.0 preview)—code execution, AI runners, interpreters, CI-like jobs, terminals, files, mounts, tunnels, preview URLs, lifecycle, or errors. Not for the default stable package (use sandbox-stable) or for porting stable to @next (use sandbox-migrate-to-next).

13 other projects use a version of this, 7 of them edited.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

6 rules. 3 are common, 0 are distinctive, and 3 are very specific to this project.

  • 3

    Common

    Most of the 8 edited versions we read keep these.

  • 0

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Common:

    Before writing code, inspect the app:

    line 16
  2. Very specific:

    Never mix an @next Worker package with a stable container image (or the reverse).

    line 29
  3. Very specific:

    No implicit shell. Shell syntax needs an explicit shell, e.g.

    line 37
  4. Very specific:

    Do not invent removed stable APIs (gitCheckout on core, string-exec completion, session execution, sandbox.terminal(request)).

    line 44
  5. Common:

    Do not use one retry loop for every error (see Errors docs).

    line 45
  6. Common:

    No live secrets in sandbox env

    line 90

“Common” is measured: kept by at least 6 in 10 of the 8 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Sandbox SDK — @next (1.0 preview)

Isolated Linux environments on Cloudflare Containers, driven from Workers.

Prefer preview docs and installed @next types over memory. APIs change; this skill is a gate, a contract, and a retrieval map—not a full manual.

We recommend new projects on this line. Apps still on the default package use sandbox-stable. Port only when asked, via sandbox-migrate-to-next.

1. Gate — confirm the package line

Before writing code, inspect the app:

CheckMust match
npm dependency@cloudflare/sandbox@next (or another preview tag)
Container imageSame line (e.g. cloudflare/sandbox:next, next-python)
If you find…Action
Default @cloudflare/sandbox (no @next)Stop. Load sandbox-stable. Do not apply this skill’s APIs.
User wants to port stable → @nextStop. Load sandbox-migrate-to-next.
Self-deployed bridge onlyBridge is not on the 1.0 preview line yet. Keep bridge on stable package + image. Bridge (stable)

Never mix an @next Worker package with a stable container image (or the reverse).

Skills install: Agent setup · cloudflare/skills

2. Contract — non-negotiables

  • sandbox.exec(argv) takes an argv list and resolves when the process starts. It returns a handle, not a finished command result.
  • Collect results with handle methods: output(), logs(), waitForExit(), waitForPort(), waitForLog(), kill(signal?).
  • No implicit shell. Shell syntax needs an explicit shell, e.g. ["/bin/bash", "-lc", script].
  • Each launch is independent. A cd / export in one exec is not visible to the next. Pass cwd and env per launch, or one shell script.
  • Process handles have no stdin. Interactive use → terminals (createTerminal + connect).
  • Local wait timeout / AbortSignal cancel the wait only. They do not kill the process. Use kill or exec’s remote timeout.
  • getProcess / listProcesses / getTerminal / listTerminals do not start a container; they return null / [] when none is up.
  • Process and terminal IDs belong to the current container, not forever to a sandbox ID. For work that must survive replace, store the full job (argv, cwd, env, app state)—not only an id.
  • Non-secret config only in setEnvVars / launch env. Live credentials stay in the Worker; use outbound handlers when the sandbox calls external APIs.
  • Do not invent removed stable APIs (gitCheckout on core, string-exec completion, session execution, sandbox.terminal(request)).
  • Do not use one retry loop for every error (see Errors docs).

Minimal shape:

import { getSandbox, proxyToSandbox, Sandbox } from "@cloudflare/sandbox";

export { Sandbox };

const sandbox = getSandbox(env.Sandbox, "user-123");
const process = await sandbox.exec(["python3", "-c", "print(2 + 2)"]);
const result = await process.output({ encoding: "utf8" });
// result.stdout, result.exitCode

Optional non-exhaustive cheatsheet (process/terminal/interpreter only): references/api-quick-ref.md
Examples index (next branch): references/examples.md

3. Retrieve — open the doc for the task

Fetch the page before implementing. Installed @next types win over guesses.

You need to…Open
Orient / choose preview1.0 preview overview
First Worker, wrangler, DockerfileGet started
exec, handles, readiness, durabilityProcess execution
Process API signaturesProcesses API
Sandbox ID vs container vs sleep/destroyLifecycle
cwd / env / setEnvVarsEnvironment
Interactive PTY / browser terminalTerminals · Terminals API
Python/JS code interpreterInterpreter · Interpreter API
Extensions modelExtensions
Error classes and recoveryErrors · Errors API
Common failuresTroubleshooting
API hubAPI reference
Files, mounts, backups, ports, tunnels, proxyToSandboxMain docs for shared surfaces (ignore stable-only session/transport/sandbox.terminal): Files · Storage / mounts · Ports · Tunnels · Backups · Outbound traffic · Expose services · Production
Example appsexamples on next
Still on stable packagesandbox-stable · Main Sandbox docs
Porting an existing stable appsandbox-migrate-to-next · Migrate

4. Before you ship

  • Lockfile and Dockerfile on the same @next line
  • Typecheck against installed @next types
  • No live secrets in sandbox env
  • Production preview hostnames need wildcard DNS on a custom domain when using those URL patterns

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.