evizi-kit

Multi-agent development kit supporting Claude Code, Cursor, GitHub Copilot, and more. Install pre-configured agents and skills into your project with a single command.

Overview

evizi-kit provides a collection of AI agent configurations and shared skills organized into installable kits. Each kit targets a specific AI coding platform and includes agents (task-specific workflows) and skills (reusable capabilities).

Available Kits

| Kit | Platform | Installs to | Agents | Platform Skills | |-----|----------|-------------|--------|-----------------| | claude | Claude Code | .claude/ | 16 | 2 | | cursor | Cursor | .cursor/ | 16 | 2 | | github | GitHub Copilot | .github/ | 16 | 2 | | agent | VS Code Agent | .agent/ | — | 2 | | shared | All platforms | (merged into each kit) | — | 21 |

When you install a kit, shared skills (21) are automatically merged into that platform's skills folder alongside its own platform-specific skills.

Shared Skills

Skills available across all platforms:

• push-code — Push code changes
• self-review — Self-review code before committing
• pr-review — Review pull requests
• fix-automation-tests — Fix failing automation tests
• increase-unit-test-coverage — Improve test coverage
• update-feature-document — Update feature documentation
• web-auto-coding — Web automation coding
• web-auto-test-cases — Generate test cases
• web-auto-ticket-design — Design test tickets
• web-auto-ticket-playbook — Create ticket playbooks
• web-auto-fix-and-run-test — Fix and run tests
• web-auto-assisted-fix-and-run — Assisted fix and run
• web-auto-fe-extract-selectors — Extract frontend selectors
• web-auto-chrome-devtools-mcp-extract-selectors — Extract selectors via Chrome DevTools
• web-auto-playwright-mcp-extract-selectors — Extract selectors via Playwright
• web-auto-extract-lessons-learned — Extract lessons learned
• web-auto-generate-instructions — Generate automation instructions
• web-auto-generate-project-blueprint — Generate project blueprints
• web-auto-generate-best-practices — Generate best practices
• web-auto-update-source-instructions — Update source instructions
• workspace-ai-nav-creator — Create AI navigation configs

Platform-Specific Skills

Each platform kit also includes:

• skill-creator — Create, modify, and benchmark skills
• claude-code-subagent-creator — Create custom subagent profiles

End-User Guide

Installation

# Install the CLI globally
npm install -g evizicli@latest

Quick Start

cd your-project

# Interactive — choose which kits to install
evizi init

# Install a specific kit
evizi init --agent claude

# Install multiple kits
evizi init --agent claude,cursor

# Install all kits
evizi init --all

What Gets Installed

For example, evizi init --agent claude installs:

your-project/
├── .claude/
│   ├── agents/          # 16 agent workflows
│   ├── skills/          # 23 skills (2 platform + 21 shared)
│   └── docs/            # 5 workflow documents
└── .evizi-lock.json     # Tracks installed kits & versions

Upgrading

When a new version of evizi-kit is published:

evizi upgrade

This compares your installed versions (from .evizi-lock.json) with the latest, and updates changed kits. Files listed in .evizi-ignore are preserved.

Protecting Custom Files

Create a .evizi-ignore file in your project root to prevent specific files from being overwritten during upgrades:

# Keep my custom rules
.claude/skills/my-custom-skill/
.cursor/agents/my-custom-agent.agent.md

CLI Reference

evizi init [options]        Install agent kits into your project
  --agent <kits>            Comma-separated kit names (e.g. claude,cursor,github)
  --all                     Install all available kits

evizi upgrade   Upgrade installed kits to latest version

evizi --help                Show help
evizi --version             Show version

Developer Guide

Repository Structure

evizi-kit/
├── kits/                           # All agent kits
│   ├── claude/                     # Claude Code kit
│   │   ├── .claude/
│   │   │   ├── agents/             # Agent workflow definitions (.md)
│   │   │   └── skills/             # Platform-specific skills
│   │   └── manifest.json
│   ├── cursor/                     # Cursor kit
│   │   ├── .cursor/
│   │   │   ├── agents/             # Agent definitions (.agent.md)
│   │   │   └── skills/
│   │   └── manifest.json
│   ├── github/                     # GitHub Copilot kit
│   │   ├── .github/
│   │   │   ├── agents/             # Agent definitions (.agent.md)
│   │   │   └── skills/
│   │   └── manifest.json
│   ├── agent/                      # VS Code Agent kit
│   │   ├── .agent/
│   │   │   └── skills/
│   │   └── manifest.json
│   └── shared/                     # Shared across all platforms
│       ├── skills/                 # 21 shared skills
│       ├── docs/                   # Workflow documentation
│       └── manifest.json
├── cli/                            # evizi-cli source
│   ├── bin/evizi.js                # CLI entry point
│   ├── src/
│   │   ├── commands/
│   │   │   ├── init.js             # evizi init command
│   │   │   └── upgrade.js          # evizi upgrade command
│   │   └── lib/
│   │       ├── resolver.js         # Kit source resolution (npm or local)
│   │       ├── manifest.js         # Manifest loading & dependency resolution
│   │       ├── copier.js           # File copying with dependency mappings
│   │       ├── ignore.js           # .evizi-ignore support
│   │       └── lockfile.js         # .evizi-lock.json read/write
│   └── package.json
├── scripts/
│   └── sync-versions.js            # Auto-sync manifest versions
├── package.json                     # evizi-kit npm package
└── README.md

Kit Manifest (manifest.json)

Each kit declares its metadata, files, dependencies, and how shared resources map into its folder:

{
  "name": "claude",
  "displayName": "Claude Code",
  "version": "1.0.4",
  "description": "Claude Code agent configuration with agents and skills",
  "files": [
    { "src": ".claude/", "dest": ".claude/" }
  ],
  "dependencies": ["shared"],
  "dependencyMappings": {
    "shared": {
      "skills/": ".claude/skills/",
      "docs/": ".claude/docs/"
    }
  },
  "deletions": []
}

| Field | Description | |-------|-------------| | files | Source-to-destination mapping for the kit's own files | | dependencies | Other kits that must be installed alongside (e.g., shared) | | dependencyMappings | Where dependency files get copied into this kit's folder | | deletions | Files to remove during upgrades (for renamed/deprecated files) |

Adding a New Agent Kit

1. Create kits/<name>/ with the platform's config folder
2. Create manifest.json with file mappings and dependencyMappings
3. The CLI auto-discovers new kits from manifests

Example — adding Windsurf:

kits/windsurf/
├── .windsurf/
│   ├── agents/
│   └── skills/
└── manifest.json

{
  "name": "windsurf",
  "displayName": "Windsurf",
  "version": "1.0.4",
  "files": [
    { "src": ".windsurf/", "dest": ".windsurf/" }
  ],
  "dependencies": ["shared"],
  "dependencyMappings": {
    "shared": {
      "skills/": ".windsurf/skills/",
      "docs/": ".windsurf/docs/"
    }
  }
}

Adding a New Shared Skill

1. Create a folder in kits/shared/skills/<skill-name>/
2. Add SKILL.md (required) and optional templates/, references/, scripts/, evals/
3. The skill is automatically available to all platforms on next publish

Publishing

First-time setup

npm login

Publish evizi-kit (the kits package)

# Bump version — this auto-syncs all manifest.json files
npm version patch    # or minor, major

# Publish to npm
npm publish

Publish evizi-cli (the CLI tool)

cd cli
npm version patch
npm publish

Local Development

Test the CLI against your local kits without publishing:

# From any project directory
evizi init --local /path/to/evizi-kit
evizi upgrade --local /path/to/evizi-kit

# Or link the CLI for development
cd cli
npm link

Version Sync

The npm version lifecycle hook automatically runs scripts/sync-versions.js, which updates all kits/*/manifest.json versions to match the root package.json. This ensures kit manifest versions always stay in sync with the published package version.

License

MIT