docker/sbx-kit-claude-sbx-statusline repository overview

⁠claude-sbx-statusline

A two-line Claude Code status line⁠ for Docker Sandboxes. Compose this mixin onto the built-in claude agent and every session renders a compact dashboard of where you are and what the session is costing:

• Line 1 — 🐳 Docker Sandboxes · sandbox host · working directory · git branch (with a * dirty marker).
• Line 2 — model · context-window used % (colour-coded) · memory used/total · 1-minute load average · session cost in USD.

The kit ships the script to ~/.claude/statusline.sh and registers it under statusLine in ~/.claude/settings.json — merging into the file so the claude base image's other settings are preserved. Only the statusLine key is written; if the file already had one, it is replaced with this kit's script (that's the point of installing the kit), and every other key is left as-is.

⁠What you get

🐳🏖️  Docker Sandboxes · my-sandbox · /home/agent/workspace (main*)
Claude Opus 4.8 · ctx 32%/200k · mem 1.2/8.0G · load 0.41 · $0.87

Segments are omitted when there's nothing to show (e.g. the git segment is blank outside a repo, the context segment is blank early in a session). Memory reads the cgroup v2 limit so it reflects the sandbox's allocation, not the host's.

⁠Quick start

Pair the mixin with the claude agent via --kit, from its published OCI artifact on Docker Hub:

sbx run claude --kit "docker.io/docker/sbx-kit-claude-sbx-statusline:latest" .

Or pull it straight from this repo over git (pinned by ref):

sbx run claude --kit "git+https://github.com/docker/sbx-kits-contrib.git#dir=claude-sbx-statusline" .

Or with a local clone of this repo:

sbx run claude --kit ./claude-sbx-statusline .

⁠How it works

• files/home/.claude/statusline.sh is baked into the kit's overlay at /home/agent/.claude/statusline.sh, so it arrives with the image rather than being written per sandbox. (Under v2 it was packed from the files/home/ tree and copied in at create time; v3 has no files/ convention — a mixin's layers are its content — so the recipe⁠ COPYs it. The source stays in place so the mapping is still legible.) The install hook below still chmod +xes it, which is belt-and-braces now that COPY carries the mode — without the executable bit Claude Code silently renders no status line. The script receives the session JSON on stdin and prints the two lines.

• The install hook (run as root) merges the statusLine block into ~/.claude/settings.json with jq:

  {
    "statusLine": { "type": "command", "command": "~/.claude/statusline.sh" }
  }
  

  Copy

  It creates the file if missing and leaves every other key untouched — only statusLine is set (replacing a prior one if present) — then chowns back to the agent user exactly the paths it touched: ~/.claude, ~/.claude/settings.json, and ~/.claude/statusline.sh. The chown is deliberately not recursive — ~/.claude holds runtime-managed content this kit doesn't own, so a chown -R over the parent would claim ownership of paths outside the kit's control. Re-running is idempotent. The temp file is created inside ~/.claude so the final mv is an atomic same-filesystem rename rather than a cross-device copy.

• Note that the script does not perform any checks to verify that you are indeed inside a sandbox. Care should be taken to not put this status line to your host's Claude Code installation as it would then incorrectly state that you are inside a sandbox when you are not.

⁠Requirements

The script and the install merge both use jq, which is present on the claude-code base image. It also uses git, awk, and hostname — all standard on the base image.