roude/coding-agent-sandbox repository overview

⁠coding-agent-sandbox

A Dockerfile to be used as a sandbox for your coding agent. This is a highly personal setting and might be chaning on my personal preferences.

⁠included tools

Comes setup with:

• antigravity-cli
• claude-code
• mise
• context-mode⁠
• various unix tools needed for day-to-day work

⁠Usage

I have a shell function to start the docker container with pre-defined mounts:

agent-sandbox() (
  docker run -ti --rm \
    -v ~/.config/gemini:/home/ubuntu/.gemini \
    -v ~/.config/claude:/home/ubuntu/.claude \
    -v ~/.cache/mise:/home/ubuntu/.cache/mise \
    -v ~/.config/mise:/home/ubuntu/.config/mise \
    -v ~/.local/share/mise:/home/ubuntu/.local/share/mise \
    -v ~/.local/state/mise:/home/ubuntu/.local/state/mise \
    -v ./:/workdir \
    coding-agent-sandbox
)

which adds my gemini and claude setup (custom dir), mise caches and the current working directory.

⁠Claude

claude code stores settings in two places. The .claude directory and the ~/.claude.json file. The later one is important and tracks the initial onboarding as well as the authentication. To avoid getting reprompt for this on every start be sure to also place this file in the container.

⁠mise tasks

⁠actions:lint

• Usage: actions:lint

Lint GitHub Actions workflows and actions with ghalint

⁠build

• Usage: build

Builds the docker container

⁠build:ghcache

• Usage: build:ghcache

Builds the docker container

⁠docs

• Usage: docs
• Aliases: doc

Update tasks in README

⁠run

• Usage: run

Runs the docker container for manual testing