claude-code-job

A Docker image that packages Claude Code CLI with tooling for running automated jobs. The image provides the runtime environment and entrypoint — skills and prompts are injected at deployment time.

Skills are maintained separately in thurbeen-skills.

Quick start

docker pull ghcr.io/thurbeen/claude-code-job:latest

docker run --rm \
  -e CLAUDE_CODE_OAUTH_TOKEN \
  -e CLAUDE_PROMPT="Hello, what can you help me with?" \
  ghcr.io/thurbeen/claude-code-job:latest

What's included

The Docker image (node:24-bookworm base) bundles:

Tool	Purpose
claude	Claude Code CLI
gh	GitHub CLI
git	Version control
kubectl	Kubernetes CLI
curl, jq, yq	HTTP requests & data processing
ripgrep, fd-find, fzf	Fast file search
python3, uv	Python runtime & package manager
build-essential	Build toolchain

Entrypoint

The image uses entrypoint.sh as its entrypoint. It handles:

• Skill cloning — set SKILLS_REPO + SKILL_NAME to clone a skill from GitHub at startup
• Skill injection — mount a CLAUDE.md file at CONFIG_DIR to inject project-level instructions
• MCP config — mount mcp-config.json at CONFIG_DIR for MCP server auto-discovery
• Prompt — mount prompt.txt at CONFIG_DIR or set CLAUDE_PROMPT env var
• Session persistence — mount PERSIST_DIR for conversation continuity across runs
• OAuth refresh — automatically refreshes Google OAuth tokens if configured

Environment variables

Variable	Required	Default	Description
CLAUDE_CODE_OAUTH_TOKEN	Yes	—	Claude Max OAuth token
CONFIG_DIR	No	/etc/claude-code-job	Config mount path
PERSIST_DIR	No	/var/lib/claude-code-job	Persistent storage path
CLAUDE_PROMPT	No	—	Fallback prompt if no prompt.txt mounted
CLAUDE_MODEL	No	—	Override Claude model
ALLOWED_TOOLS	No	Bash,Read,Glob,Grep,Edit,Write	Override allowed tools
SKILLS_REPO	No	—	GitHub repo to clone skills from (e.g. owner/repo)
SKILL_NAME	No	—	Skill directory to install from SKILLS_REPO
SKILLS_REF	No	main	Git ref/tag to checkout from SKILLS_REPO

Example with skill from GitHub

docker run --rm \
  -e CLAUDE_CODE_OAUTH_TOKEN \
  -e GH_TOKEN \
  -e SKILLS_REPO=Thurbeen/thurbeen-skills \
  -e SKILL_NAME=bump-pr-fixer \
  -e CLAUDE_PROMPT="Run the bump-pr-fixer skill" \
  -v claude-data:/var/lib/claude-code-job \
  ghcr.io/thurbeen/claude-code-job:latest

Example with mounted config

docker run --rm \
  -e CLAUDE_CODE_OAUTH_TOKEN \
  -e GH_TOKEN \
  -v ./my-job-config:/etc/claude-code-job:ro \
  ghcr.io/thurbeen/claude-code-job:latest

Where my-job-config/ contains:

• CLAUDE.md — project-level instructions for Claude
• prompt.txt — the prompt to run
• mcp-config.json — (optional) MCP server configuration

Persistent sessions

Mount a volume at PERSIST_DIR to resume conversations:

docker run --rm \
  -e CLAUDE_CODE_OAUTH_TOKEN \
  -e CLAUDE_PROMPT="Continue monitoring" \
  -v claude-data:/var/lib/claude-code-job \
  ghcr.io/thurbeen/claude-code-job:latest

Development

Build locally

docker build -t claude-code-job .

Pre-commit hooks

pre-commit install
pre-commit install --hook-type commit-msg

CI/CD

• CI runs on every push and PR to main: Dockerfile linting (hadolint), shell linting (shellcheck), secret detection (gitleaks), and Docker build.
• CD runs on push to main: uses cocogitto to auto-bump semver from conventional commits, creates a git tag, builds and pushes the image to ghcr.io/thurbeen/claude-code-job, and publishes a GitHub Release.

License

Apache-2.0