Agent Toolkit

A collection of production-quality skills, tools, and patterns for AI-powered development workflows with Claude Code.

Overview

Agent Toolkit provides a curated set of resources to enhance Claude Code's capabilities:

• Skills: Domain-specific knowledge modules that Claude loads on-demand
• Patterns: Cross-language architectural patterns and best practices
• Workflows: Integrated development workflows (TDD, security scanning, code review)

Prerequisites

Tested on: macOS (Darwin). Linux should work; Windows may require adjustments for shell-based tools.

Some skills include helper scripts that require external tools. Install via Homebrew:

# Required: Python package manager (for running skill scripts)
brew install uv

# Required: Python with PyYAML (for skill validation)
uv tool install pyyaml

# Optional: Language-specific tools
brew install go              # For golang skill
brew install shellcheck      # For bash skill
brew install terraform       # For terraform skill
brew install trivy           # For security scanning
brew install node            # For web-automation (Playwright)

Python dependencies for skill scripts:

Skill	Dependencies	Install
skill-creator	PyYAML	cd skills/skill-creator && uv sync

Quick Start

Installation

# Copy skills to Claude Code's skills directory
mkdir -p ~/.claude/skills
cp -r skills/* ~/.claude/skills/

# (Optional) Copy the example CLAUDE.md to customize your workflow
cp CLAUDE.md-example ~/.claude/CLAUDE.md
# Then edit ~/.claude/CLAUDE.md to personalize

Verify Installation

# List installed skills
ls ~/.claude/skills/

# Test a skill (in Claude Code)
claude "/python"

Repository Structure

agent-toolkit/
├── README.md                 # This file
├── LICENSE                   # MIT License
├── CLAUDE.md-example         # Example CLAUDE.md configuration
├── scripts/                  # Helper scripts
│   ├── statusline.sh         # Custom status line script
│   └── settings-statusline.json  # Settings snippet for status line
└── skills/                   # Claude Code skills
    ├── design-patterns/      # Cross-language architectural patterns (skill)
    ├── ast-grep/             # AST-aware code search patterns
    ├── bash/                 # Bash scripting skill
    ├── golang/               # Go development skill
    ├── python/               # Python development skill
    ├── make/                 # Makefile skill
    ├── terraform/            # Infrastructure as Code skill
    ├── scm/                  # Source control (Git) skill
    ├── trivy/                # Security scanning skill
    ├── gemini-review/        # Code review with Gemini
    ├── web-automation/       # Playwright E2E testing
    ├── spec-driven-dev/      # Spec-driven development workflow
    ├── skill-creator/        # Create new skills
    └── improve-skill/        # Improve existing skills

Available Skills

Language Skills

Skill	Description	Triggers
bash	Production-quality bash scripts with ShellCheck compliance	bash script, shell script, .sh
golang	Go 1.26 idiomatic development with quality gates	go code, golang, go patterns
python	Modern Python 3.11+ with type hints and Pydantic	pythonic, python patterns, .py
make	Idiomatic Makefiles with safety and self-documentation	makefile, make target, .mk
terraform	Infrastructure as Code with Terraform/Terragrunt	terraform, terragrunt, iac, .tf

Workflow Skills

Skill	Description	Triggers
scm	Git best practices, Conventional Commits, PR workflows	git, commit, branching
trivy	Security vulnerability scanning before commit	trivy, vulnerability scan, security
gemini-review	Design and code review with Gemini	code review, gemini review
web-automation	E2E testing and browser automation	e2e test, playwright
spec-driven-dev	Spec-driven development with iterative refinement	/spec.plan, /spec.refine, /spec.tasks, /spec.run

Architecture Skills

Skill	Description	Triggers
design-patterns	Cross-language patterns (DI, errors, config, testing)	dependency injection, anti-pattern, design pattern
ast-grep	AST-aware code search and refactoring	ast-grep, find function, structural search

Meta Skills

Skill	Description
skill-creator	Create and validate new Claude Code skills
improve-skill	Analyze sessions to improve existing skills

CLAUDE.md Configuration

The CLAUDE.md-example file provides a starting point for configuring Claude Code's behavior. It includes:

• Decision Framework: Tiered autonomy (autonomous, collaborative, ask permission)
• Code Philosophy: TDD, KISS, YAGNI, Boy Scout Rule
• File Standards: ABOUTME headers, naming conventions
• Workflow Integration: Gemini review triggers, skill invocation rules

To use:

cp CLAUDE.md-example ~/.claude/CLAUDE.md
# Edit the file to customize for your workflow

Custom Status Line

A custom status line script that displays model info, working directory, git branch, session cost, and context usage.

The status line shows:

• Model: Current Claude model (e.g., Opus 4.5)
• Directory: Current working directory (with ~ for home)
• Git Branch: Active branch when in a git repository
• Cost: Session cost in USD
• Context: Token usage and percentage of context window

Installation

# Copy the status line script
mkdir -p ~/.claude/scripts
cp scripts/statusline.sh ~/.claude/scripts/
chmod +x ~/.claude/scripts/statusline.sh

Configuration

Add the following to your ~/.claude/settings.json:

{
  "statusLine": {
    "type": "command",
    "command": "~/.claude/scripts/statusline.sh",
    "padding": 0
  }
}

Or merge the provided snippet:

# If you don't have a settings.json yet
cp scripts/settings-statusline.json ~/.claude/settings.json

# If you already have one, manually merge the statusLine section

Skill Usage

Skills are invoked automatically when Claude Code detects relevant keywords:

# Automatic invocation (keyword detection)
claude "write a bash script to backup my files"

# Manual invocation
claude "/bash"
claude "/terraform"

Integrated Workflows

Secure Commit Workflow

The toolkit includes an integrated security scanning workflow:

git commit → Detect Changes → Trivy Scan → Remediation (if needed) → Commit

See skills/scm/references/commit-workflow.md for details.

TDD Workflow

Skills enforce Test-Driven Development:

1. Red: Write failing test
2. Green: Minimal code to pass
3. Refactor: Clean up
4. Commit: After each phase

Code Review Workflow

Automatic Gemini review triggers:

• More than 100 lines changed
• More than 3 files with logic changes
• Security or performance code

Creating New Skills

Use the skill-creator skill:

# Initialize a new skill
uv run ~/.claude/skills/skill-creator/scripts/init_skill.py \
    my-new-skill \
    --path ~/.claude/skills

# Validate a skill
uv run ~/.claude/skills/skill-creator/scripts/validate_skill.py \
    ~/.claude/skills/my-new-skill

Skill Structure

skill-name/
├── SKILL.md              # Main skill definition (required)
├── scripts/              # Executable helpers (optional)
├── references/           # Additional documentation (optional)
└── resources/            # Templates and assets (optional)

SKILL.md Format

---
name: skill-name
description: >-
  What the skill does. Triggers on "keyword1", "keyword2".
allowed-tools: Read, Write, Edit, Bash
---

# ABOUTME: Brief description of file purpose
# ABOUTME: Key context or dependencies

# Skill Title

[Skill content...]

Design Principles

• Production-Ready: Skills enforce patterns that work in real codebases
• Opinionated: Best practices are baked in, not optional
• Integrated: Skills work together (security + commit + review)
• Progressive Disclosure: Core info in SKILL.md; details in references

Security

Skills can execute scripts on your system via Claude Code. Before using third-party skills:

1. Audit all scripts in scripts/ directories before execution
2. Review SKILL.md files for any suspicious tool permissions
3. Check allowed-tools in frontmatter (especially Bash permissions)

The skills in this repository have been reviewed, but you should always verify code that runs on your machine.

# List all executable scripts in a skill
find ~/.claude/skills/<skill-name>/scripts -type f -executable

# Review a script before running
cat ~/.claude/skills/<skill-name>/scripts/<script>.py

Contributing

Contributions welcome! Please ensure:

1. Skills follow the established structure
2. ABOUTME headers present on all files
3. Trigger phrases included in descriptions
4. Code passes validation: validate_skill.py

Acknowledgements

This project is inspired by and builds upon the work of Massimo Aroffo and the claude-forge repository. Many of the architectural patterns and skill design principles originated from that project.

License

MIT License - see LICENSE for details.

Use, fork, and modify freely; attribution required (keep the copyright notice).

Resources

• Claude Code Documentation
• Anthropic Documentation