Skip to content
| Marketplace
Sign in
Visual Studio Code>AI>Docker Sandbox BrowserNew to Visual Studio Code? Get it now.
Docker Sandbox Browser

Docker Sandbox Browser

triozillion

|
4 installs
| (0) | Free
Chat-style panel for Docker Sandboxes: pick an sbx instance, browse slash commands, and drive Claude Code from the sidebar.
Installation
Launch VS Code Quick Open (Ctrl+P), paste the following command, and press enter.
Copied to clipboard
More Info

Docker Sandbox Browser

Chat with your Docker Sandboxes from the VS Code sidebar. Pick a sandbox, browse every available slash command, send prompts, and watch the agent's reply stream back — without leaving the editor.

Docker Sandbox Browser panel


Why

Docker Sandboxes (sbx) run coding agents in isolated containers. Normally you drive them from a terminal, which means remembering command names and switching windows. This extension puts that in a proper chat panel:

  • every slash command is discoverable and searchable
  • one command runs at a time, with a visible queue and a Stop button
  • the agent's answer appears in the panel, not just the terminal
  • each sandbox keeps its own conversation

Quick start

1. Install the prerequisites

You need the sbx CLI on your PATH and at least one sandbox:

sbx create claude .

Check it is visible:

sbx ls

2. Open the panel

Click the Docker Sandbox Browser icon in the Activity Bar, or run SBX: Open Docker Sandbox Browser from the Command Palette (Ctrl+Shift+P).

The panel beside the agent terminal

3. Pick a sandbox

Choose an instance from the dropdown at the top. The dot shows its state:

Dot Meaning
Green Running — ready to chat
Amber Stopped — sending a message starts it

Selecting a stopped sandbox never starts it silently. You stay in control — it only starts when you send your first message. Press Refresh to re-read the current status.

Selecting a sandbox

4. Send a prompt or a command

Type a message and press Enter, or type / to browse commands.

Slash command menu


Using commands

Press / in the composer to open the command menu. Commands are grouped by where they came from:

Group Source
Built-in Standard agent commands such as /help, /status, /context
Project commands .claude/commands/*.md in the sandbox workspace
Project skills .claude/skills/*/SKILL.md in the sandbox workspace
Plugin skills Skills installed inside the running sandbox

Selecting a command inserts it into the composer so you can add arguments — it does not send immediately. Press the send button (or Enter with the menu closed) when you are ready.

/speckit-specify add a login page

Composer keys

Key Action
Enter Send
Shift + Enter New line
/ Open the command menu
Up / Down Move through commands
Tab / Right / Enter Insert the highlighted command
Esc Close the menu

While a command runs

The send button turns into a Stop button. The reply streams into the transcript as the agent writes it, and the Activity section shows the queue and local logs.

Command running with live output

  • Commands are serialized — one runs at a time, the rest queue up.
  • Stop interrupts the current command and moves to the next.
  • A command stays running until the agent actually goes idle, so long tasks are not marked finished early.

Conversations

Button What it does
New chat Archives the current conversation and starts a fresh one
History Reopens a past session for this sandbox
Refresh Re-reads commands and logs (never starts a sandbox)
@ Attaches files from the workspace to your next message

Conversations are kept per sandbox and persist across restarts.


Settings

Setting Default Description
dockerSandboxBrowser.sbxExecutable sbx Path or name of the sbx CLI
dockerSandboxBrowser.preferredTerminalName SBX Claude Terminal name prefix
dockerSandboxBrowser.runScriptPath (empty) Optional Node launcher (node <script> <sandbox>). Empty uses sbx run --name <sandbox>
dockerSandboxBrowser.kitLogDirectory (empty) Optional workspace-relative folder of setup logs
dockerSandboxBrowser.completionPollMs 2000 How often to check whether a command finished
dockerSandboxBrowser.completionIdleMs 6000 Idle time before a command counts as finished
dockerSandboxBrowser.completionMaxMs 900000 Maximum wait before auto-completing

How it works

Prompts and commands are sent to a dedicated SBX Claude: <sandbox> integrated terminal, started with sbx run --name <sandbox>. The agent CLI owns the real session and history, so nothing is duplicated or lost. The panel reads the agent's transcript inside the sandbox to display output and to detect when a command has finished.


Troubleshooting

The dropdown is empty Run sbx ls in a terminal. If it works there but not in the panel, VS Code may not have inherited your PATH — set dockerSandboxBrowser.sbxExecutable to the full path of the sbx executable, then press Refresh.

A command looks stuck on running The panel waits for the agent to stop writing before marking a command complete. Increase dockerSandboxBrowser.completionIdleMs if your tasks pause for long periods, or press Stop.

Commands from my project are missing Project commands are read from the selected sandbox's workspace. Confirm .claude/commands exists there, then press Refresh.


License

MIT

  • Contact us
  • Jobs
  • Privacy
  • Manage cookies
  • Terms of use
  • Trademarks
  • Your Privacy Choices
  • Consumer Health Privacy
© 2026 Microsoft