piperhex/remoteai/cua-sandboxes
Claude Code and agent skill by piperhex/remoteai
- community
Checking that the SKILL.md file is still in the repository.
Create, use and clean up cua sandboxes (disposable Linux or macOS computers) on this computer with the `cua` CLI or the cua SDK, and browse the web inside one. Use when a task needs an isolated machine to run code, test an app, drive a desktop GUI, browse or test a website, fill a web form, take a web page screenshot, or reproduce something without touching the user's own computer.
Install
npx vskill@latest install piperhex/remoteai/cua-sandboxesSKILL.md
cua sandboxes
A sandbox is a disposable computer: a container or VM on this machine. The cua CLI and the cua SDK (Python, TypeScript, Swift, Kotlin, Rust) share one runtime and one list of sandboxes.
Before you start
Local sandboxes need no account. Delete them when you are done.
Create
--on is where (local, direct:<addr>), --kind what (auto, container, vm), --runtime which engine (auto, gvisor, runc, qemu, lume). An impossible combination fails with invalid placement and lists the valid values. Local runtimes are set up on first use (cua runtime setup does it explicitly). Omit --name for a generated name. cua sb ls prints each sandbox's ref (local:dev, direct:host:port); every NAME argument takes a ref or a bare name that is unique across locations (--local narrows it). Add --json to any command for machine-readable output. exec, shell, cp, screenshots and GUI control go through cua-spacesd, so use an image that ships it (the linux alias does). Plain images still get lifecycle, logs and port-forward.
Use
For GUI work, select the sandbox once and use cua do (see the gui-automation skill):
Clean up
From code (Python)
The TypeScript, Swift and Kotlin bindings expose the same objects (Cua, sandboxes(), Sandbox.spacesd()).
From the cua MCP server
When the cua MCP server is connected, prefer its tools: images_list (which images exist), sandbox_list, sandbox_create, sandbox_get, sandbox_delete, then the computer_* tools (screenshot, click, type, shell, file) against the sandbox.
Browse the web
Browse inside a sandbox, never in the user's own browser. Everything below is cua MCP tool calls; no other browser tool or setup is needed.
- Pick an image:
images_list {"browser": true}(the canonical
ghcr.io/trycua/linux:24.04 ships Chromium).
- Create it with a browser: `sandbox_create {"browser": true, "url":
"https://example.com"}. It returns id, session and browser.target_id / browser.tab_id`.
- Drive the browser with `call_tool {"space": <id>, "tool": <tool>,
"arguments": {"session", "target_id", "tab_id", ...}}`:
- read:
get_browser_state {"snapshot_format": "semantic_v2"}returns
an outline and refs (p1:0); add "include_screenshot": true for an image;
browser_navigate {"url"}, `browser_click {"ref", "delivery_mode":
"foreground"}, browser_type {"ref", "text", "replace": true} (text "\n" with "mode": "keystrokes" presses Enter), browser_pointer` (hover, scroll, drag);
- read again after every page change: refs are per snapshot.
To test a local app, run it in the sandbox (space_write, space_bash) and open http://127.0.0.1:<port>.
- Logged-in sites: only when the user asks.
teleport_browser_session
with the exact sites first returns what would move and a consent requirement; show it to the user and continue only after they approve. Never move a session without that approval.
- Clean up:
sandbox_delete {"name": <id>}.
From a terminal: cua images ls --browser, cua sb create --browser --open https://example.com, cua sb view <id> to watch.
Rules
- Never run destructive commands on the user's own machine when a sandbox
would do.
- Delete the sandboxes you created as soon as the task ends.
- Report the sandbox name you used so the user can inspect or remove it.
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingpackages/cua/vendor/libs/cua/skills/cua-sandboxes/SKILL.md
Popularity
- Repository stars
- 1,077
- Repository forks
- 41
- 7-day trend score
- 0
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/piperhex/remoteai/cua-sandboxes)