A knowledge-accumulating workflow framework for Claude Code

Work on real tasks. Let the knowledge build itself.

日本語版 README

With Claude Code, do you find yourself repeating the same explanations every time you switch projects? Project-specific rules can be handled with CLAUDE.md. But industry expertise, deep technical knowledge, and hard-won lessons from past mistakes — these should accumulate across projects, and there's currently no built-in mechanism for that.

Kaizen-CLI doesn't ask you to run a separate improvement process. Instead, it embeds a knowledge-accumulation loop into your everyday work. You focus on your actual tasks — building features, fixing bugs, running analyses. When you hit friction during that work — an unexpected error, a technique you had to look up, a pattern that turned out to work well — that friction becomes the raw material for updating your shared knowledge base. Over time, the knowledge that accumulates from real work accelerates all your future work across every project.

┌─────────┐    ┌─────────┐    ┌─────────┐    ┌─────────┐
│  Init   │───▶│ Suggest │───▶│ Execute │───▶│ Reflect │
│         │    │  Next   │    │(= your  │    │         │
│/kaizen  │    │/kaizen  │    │ actual  │    │/kaizen  │
│-init    │    │-suggest │    │  work)  │    │-reflect │
│-project │    │-next    │    │         │    │-learning│
└─────────┘    └─────────┘    └─────────┘    └─────────┘
                    ▲                              │
                    │  Friction from work feeds     │
                    │  knowledge/ which accelerates │
                    │  the next cycle               │
                    └──────────────────────────────┘

• No extra process: You don't need to think about improvements separately. Friction that naturally arises during real work becomes the input for knowledge updates
• Knowledge persists: Lessons learned are written to knowledge/ files and available across all your projects
• Cross-project learning: Shared knowledge via symlinks means one project's insights benefit all projects
• Signal stays high: Accumulation is paired with periodic curation, so knowledge/ grows as a resource — not as noise
• Faster over time: Accumulated knowledge means less repeated explanations, fewer regressions, and fewer known pitfalls

# 1. Clone and set up Kaizen-CLI (one-time)
git clone https://github.com/tomfook/kaizen-cli.git
bash kaizen-cli/setup.sh

# 2. Initialize your project
cd your-project
claude  # Start Claude Code, then:
        # /kaizen-init-project    — set up knowledge/ symlink and project config

# 3. Start the Kaizen cycle
        # /kaizen-suggest-next       — get next step suggestions
        # /kaizen-reflect-learning   — capture lessons learned
        # /kaizen-update-docs        — sync project docs and inventory auto memory
        # /kaizen-unregister-project — remove a project from the registry
        # /kaizen-audit-knowledge   — audit knowledge files for quality

setup.sh creates the shared knowledge directory ($KAIZEN_KNOWLEDGE_DIR) and links the kaizen-init-project skill. Run /kaizen-init-project in each project to set up kaizen commands and symlinks. The knowledge/ in each project is a symlink to a registry within this shared directory — knowledge accumulated in one project is automatically available to all others sharing the same registry. You can create multiple registries (e.g., work and personal) to keep knowledge separated by context.

For a detailed walkthrough, see docs/QUICKSTART.md.

kaizen-cli/
├── docs/                    # Methodology and guides
│   ├── CONCEPT.md           # The Kaizen-CLI philosophy
│   ├── DESIGN_PRINCIPLES.md # Design principles
│   ├── QUICKSTART.md        # 5-minute getting started guide
│   └── CUSTOMIZATION.md     # How to adapt for your domain
│
├── framework/               # Copy-and-use template set
│   ├── bin/                 # Helper shell scripts
│   ├── templates/           # Locale-separated templates
│   │   ├── en/              # English templates
│   │   └── ja/              # Japanese templates
│   └── .claude/
│       ├── commands/        # Workflow slash commands
│       └── skills/          # Auto-invoked skill definitions
│
└── examples/                # Domain-specific examples
    ├── data-analysis/       # Data analysis workflow
    └── web-development/     # Web development workflow

• CONCEPT.md — The philosophy behind Kaizen-CLI
• DESIGN_PRINCIPLES.md — Design principles and patterns
• QUICKSTART.md — Get started in 5 minutes
• CUSTOMIZATION.md — Adapt Kaizen-CLI for your domain

Great fit:

• Individual developers or analysts juggling multiple small projects
• Claude Code users who want knowledge gained in one project to carry over to all other projects
• Anyone interested in structured AI-assisted development workflows

Not a fit:

• Single long-running project with no need for cross-project knowledge sharing
• Teams not using Claude Code

Contributions are welcome! Please open an issue or pull request.

MIT