Skip to content

Sandboxed Coding Agent

Run Claude Code, OpenCode or Codex in an isolated smolvm machine with your project mounted

Stars
6,901
Licence
Apache-2.0
Files
2 files

What it does

It walks the agent through setting up a persistent smolvm virtual machine that holds a coding agent such as Claude Code, OpenCode or Codex, with your project folder mounted inside. The agent can then run with full permissions, such as skip-permissions mode, and install anything it likes without affecting your host computer, while its file edits still appear in your project and git works on the repo.

Great for

  • Developers who want to run a coding agent in skip-permissions or YOLO mode without giving it free rein over their machine
  • Teams wanting a disposable environment where an agent can install packages and experiment safely

Skip it if

  • Running untrusted code with no network access; the skill itself says to use throwaway-machine for that instead
  • Anyone who expects the sandbox to protect the project files themselves, since the project directory is mounted read-write from the host and agent edits land there directly

Try asking

  • “Sandbox my Claude Code so I can run it with --dangerously-skip-permissions safely”
  • “Set up a smolvm machine with OpenCode and mount my project so it can install anything”
  • “Help me create a Smolfile and start a machine where my agent can use git on the repo”
Step-by-step guide: how to use Sandboxed Coding AgentInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 2 original files, the Apache-2.0 licence and an attribution note.

    coding-agent.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip coding-agent.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip coding-agent.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Use --secret-env so API keys are read from the host at launch and not written to the machine record
  • Set block_private = true under [network] to keep the agent off your local network and the host's ports
  • Remember init runs only once; changing the install list later needs a new machine or running commands with exec

What's inside

  • README.md
  • SKILL.md

Redistributed unmodified from smol-machines/smolvm under the Apache-2.0 licence. The assessment was drafted with AI from the skill's files on 2026-11-02. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.