claude-code-java

Agent Skills for Java projects, following the open Agent Skills specification

This project is not affiliated with Anthropic.

What is this?

A collection of reusable skills (structured markdown files that give an AI agent domain knowledge and workflows), plus project templates, MCP server configurations, and setup scripts.

The skills follow the Agent Skills specification, an open format read by a growing number of agents. They are developed and tested with Claude Code, and every skill is validated against the spec in CI.

Who is this for? Java developers who want consistent, high-quality AI assistance for common tasks like code reviews, testing, commits, and architecture decisions.

Routing is tested

An agent picks a skill from its name and description and nothing else. With eighteen of them the descriptions compete, and the failure is quiet: the wrong skill loads and answers plausibly anyway, so nobody notices.

scripts/eval-routing.sh runs a set of prompts against the same list an agent receives and reports where one lands somewhere other than expected. It found a real defect the first time it ran: "this class does too much, split it" reached clean-code rather than solid-principles. Sharpening three descriptions took the set from 17 of 18 prompts routed correctly to 20 of 20.

Every skill has at least one case, enforced by scripts/validate-skills.sh so the cases cannot fall behind the skills. The check runs against any OpenAI-compatible endpoint, including a local model. See docs/SCRIPTS.md.

Purpose

AI-powered development workflows with focus on:

• Token efficiency - Designed for fewer iterations and lower token usage
• Reproducible patterns - Reusable workflows across projects
• Java ecosystem - Tailored for Java/Maven development
• Incremental adoption - Start small, expand as needed

Quick Start

1. Clone this workspace

git clone https://github.com/decebals/claude-code-java.git ~/projects/claude-code-java
cd ~/projects/claude-code-java
chmod +x scripts/*.sh

2. Setup your Java project

./scripts/setup-project.sh ~/projects/your-java-project

This creates .claude/ with symlinked skills, generates CLAUDE.md, and configures settings.

Prefer manual setup? Just copy or symlink the skills you want:

mkdir -p your-project/.claude/skills

# Copy specific skills
cp -r ~/projects/claude-code-java/skills/java-code-review your-project/.claude/skills/

# Or symlink all skills
ln -s ~/projects/claude-code-java/skills/* your-project/.claude/skills/

3. Use with Claude Code

cd ~/projects/your-java-project
claude

# Skills load automatically based on context, or invoke directly:
> /git-commit
> /java-code-review

Available Skills (18)

Skills are automatically loaded by Claude Code based on context.

Workflow

Code Quality

Architecture & Design

Framework & Data

See skills/README.md for full documentation and docs/SCRIPTS.md for setup script options.

Project Structure

claude-code-java/
├── README.md                    # This file
├── LICENSE                      # MIT license
├── .gitignore                   # Git ignore rules
├── .claude/
│   └── skills/                  # 18 reusable skills (see Available Skills above)
├── docs/                        # Guidelines and best practices
│   ├── DESIGN_PRINCIPLES.md     # Core philosophy
│   ├── RED_FLAGS.md             # Warning signs to watch for
│   ├── SAFE_WORKFLOWS.md        # Step-by-step safe workflows
│   ├── SCRIPTS.md               # Scripts documentation
│   ├── SKILL_GUIDELINES.md      # How to create new skills
│   └── TESTING.md               # Testing strategy
├── templates/
│   ├── CLAUDE.md.template       # Template for projects
│   ├── mcp-config.json.template # MCP configuration template
│   ├── MCP_CONFIG.md.template   # MCP documentation template
│   └── settings.json.template   # Claude Code settings (pre-approved commands)
└── scripts/
    ├── setup-project.sh         # Full project setup (orchestrator)
    ├── link-skills.sh           # Symlink skills to project
    ├── generate-claude-md.sh    # Generate CLAUDE.md
    ├── configure-mcp.sh         # Configure MCP servers
    └── test-all.sh              # Run all tests

Typical Workflow

1. Link skills to your Java project
2. Start Claude Code in project directory
3. Load skill relevant to current task
4. Execute workflow with natural language
5. Measure results (tokens used, time saved)

Success Metrics

Track these to validate effectiveness:

• Token reduction: Track your improvement vs manual workflows
• Time savings: Measure before/after per task
• Reusability: Number of projects using skills
• Quality: Code review feedback, test coverage

Requirements

• An agent that reads Agent Skills. Developed and tested with the Claude Code CLI
• Java 11+ projects (Java 17+ recommended)
• Git for version control
• Maven or Gradle build tool
• (Optional) GitHub MCP server for issue management

What's Included

• 18 skills (workflow, code quality, architecture, frameworks)
• Setup automation scripts
• Project templates
• Agent Skills spec compliance, enforced in CI by scripts/validate-skills.sh and the reference validator

Used in automated code review

These skills are not just for code generation — they are also used as the single source of truth for automated code review via skill-review, a reusable GitHub Actions workflow that evaluates pull requests against the same skills that Claude Code uses during development.

Same skills. From generation to review. See skill-review-sandbox for a working example.

Contributing

Skills are evolving based on real-world usage. Try them, open issues, share what works.

1. Try the skills in your projects
2. Open issues for suggestions
3. Share token savings/improvements

Documentation

See docs/ for detailed guides:

• DESIGN_PRINCIPLES.md - Core philosophy
• SAFE_WORKFLOWS.md - Recommended workflows
• RED_FLAGS.md - Warning signs to watch for
• SKILL_GUIDELINES.md - How to create new skills
• RELEASING.md - Versioning and how a release is cut

License

MIT License - Use freely, modify as needed.