DocsFrom your agent

Build on Cloudex

Use Cloudex from your coding agent

Let Claude Code, Codex or another coding agent on your computer create and drive Cloudex sandboxes, with a skill, an AGENTS.md snippet and the docs as plain text.

A coding agent on your computer can use the cloudex CLI the way you do: start a sandbox for a risky or long job, run commands and servers there, and give you a link to what it built. Every command prints JSON when its output isn’t a terminal, so agents read it easily. Each of these tells an agent how.

The Claude Code skill

mkdir -p ~/.claude/skills/cloudex
curl -fsSL https://raw.githubusercontent.com/lucataco/cloudex/main/skills/cloudex/SKILL.md \
  -o ~/.claude/skills/cloudex/SKILL.md

Claude Code loads it when a task calls for a clean Linux machine, a job away from your computer, a link to a running app, or copies of one environment working in parallel. It covers creating, stopping and forking sandboxes, exec, preview links, agents in a sandbox, and the costs to keep in mind. Other agents that read SKILL.md files can use the same file.

An AGENTS.md snippet

For Codex, Cursor and other agents that read AGENTS.md or a similar file, add:

## Cloudex sandboxes

`cloudex` runs Linux sandboxes (Ubuntu with sudo, Docker, Node, Python) in my Cloudflare account. Use one for long or risky jobs and for anything I should see running. Commands print JSON when piped; `cloudex <command> --help` lists the flags.

- `cloudex new --name <name>` creates one and makes it current; `--repo <https url>` clones a repo.
- `cloudex exec <name> -- '<shell command>'` runs a command (up to 1 hour with `--timeout 1h`). Start servers in the background (`nohup ... &`), then `cloudex host <name> <port>` gives me an HTTPS link.
- `cloudex stop <name>` saves the disk and stops; `cloudex resume <name>` starts it again. Stop sandboxes you no longer need; ask before `cloudex delete`.

Reporting problems

When cloudex fails in a way that looks like a bug in Cloudex, the skill tells your agent to say so, and with your OK, to report it with cloudex feedback: the agent shows you the report first, and sends it only once you agree. For other agents, add to the snippet above:

- If `cloudex` fails in a way that looks like a bug in Cloudex, tell me. With my OK, run `cloudex feedback --last-error "what you were doing"`: it sends nothing, and prints a report and a `cloudex feedback --send <id>` command. Show me the report, and run that command only if I agree.

The docs as text

cloudex.build/llms.txt lists every docs page, and llms-full.txt holds them all in one file. Each page is also Markdown at its URL plus .md.

Agents inside a sandbox

Claude Code, Codex and OpenCode running in a sandbox already know how to manage it from inside, with the sandbox’s own cloudex command; see Inside the sandbox.

For agents and LLMs: this page as Markdown,llms.txt and all docs in one file.