CLI + Web + Copilot Skill for AI-Agent-Forward Kaizen transformation at scale.

Operationalizes the AI-Kaizen Framework v2 — an eval-driven, PDCA-native approach to AI transformation that works backwards from business outcomes.

"The only AI transformation framework that treats 'knowing when to kill the project' as a success metric."

Every AI initiative must earn the right to act — progressing through bounded autonomy levels (L0 Inform → L1 Recommend → L2 Oversight → L3 Autonomous) based on measurable eval evidence, not hope.

• Python 3.9+ — python3 --version
• pip (comes with Python)

git clone https://github.com/KevinCrosby/ai-kaizen.git
cd ai-kaizen
python3 -m venv .venv
source .venv/bin/activate        # macOS/Linux
# .venv\Scripts\activate         # Windows
pip install -e .

Verify: ai-kaizen --version && ai-kaizen --help

# Create an initiative
ai-kaizen init "Customer Support Ticket Routing"

# Define the target outcome (work backwards from here)
ai-kaizen outcome set --metric "resolution time" --baseline "45 min avg" --target "15-25 min"

# Score data readiness (0-3 per dimension, interactive)
ai-kaizen assess data-readiness

# Scaffold safety eval tests (generates pytest stubs)
ai-kaizen eval scaffold --level L0 --output-dir ./evals

# Log PDCA activity
ai-kaizen pdca log --phase plan --note "Data readiness 14/18. Labels exist but 30% miscategorized."

# Check the gate — ready to proceed?
ai-kaizen pdca gate

# Portfolio dashboard
ai-kaizen status

ai-kaizen serve                           # → http://localhost:5000
ai-kaizen serve --port 8080               # custom port
ai-kaizen serve --host 0.0.0.0 --debug    # network-accessible + auto-reload

A template healthcare portfolio with 10 initiatives across 8 industries is included:

# Seed the healthcare template (2 fully-detailed + 8 industry initiatives)
bash examples/seed-healthcare-portfolio.sh
ai-kaizen serve --port 5001

Dark-themed Flask web app — 9 templates, 20 route handlers.

Key features:

• Severity class (S0-S3) + transformation type (optimize/redesign/reinvent)
• Define outcomes, scaffold evals, record eval run pass rates
• PDCA entry logging with loop/phase tracking + gate checks
• 6-dimension data readiness assessment with visual bars
• 7-dimension PMO scoring → auto-recommend (Fast-track/Qualified/Conditional/Decline)
• ROI tracking with value created vs captured and confidence progression
• Value event logging (create/capture/cost) with category breakdown
• Governance review tracking per initiative
• A3 one-pager canvas export (markdown download per initiative)
• "Should This Be an Agent?" interactive weighted assessment tool
• JSON and HTML export

Research-backed dashboard tracking 7 metric categories:

Sources: Deloitte State of AI 2026 (3,235 leaders), BCG AI Survey (1,400 C-suite), MIT Sloan/BCG, PwC.

The eval scaffold command generates pytest stubs for each level:

ai-kaizen eval scaffold --level L0 --output-dir ./evals   # → test_eval_l0_safety.py
ai-kaizen eval scaffold --level L1 --output-dir ./evals   # → test_eval_l1_assertions.py

Each file contains test stubs with NotImplementedError — fill them in with your domain-specific eval logic.

An 8-question weighted assessment (CLI + web) to determine if a task is ready for agentic automation:

Scoring: ≥12/15 = STRONG YES, 7-11 = MAYBE (address gaps), <7 = NOT YET

Export any initiative as a markdown A3 one-pager for stakeholder reviews:

ai-kaizen canvas   # prints markdown to stdout

Includes: initiative metadata, outcomes, eval pass rates, PDCA history, gate status, data readiness, PMO score, ROI summary, and value tracking — all on one page.

Also available as a download button on each initiative's web detail page.

ai-kaizen pmo score              # Interactive 7-dimension scoring
ai-kaizen pmo rank               # Prioritized backlog
ai-kaizen roi track --value-created 420000 --value-captured 290000 --tco 180000
ai-kaizen status                 # Portfolio dashboard with Estimated/Actual ROI

7-Dimension Scoring Rubric: Business Value (2×), Baseline Measurability, Data Readiness, Change Readiness, Reversibility, Compliance Burden, Platform Reuse → Score out of 40 → Fast-track / Qualified / Conditional / Decline.

ROI Confidence Progression: Projected (±50%) → Estimated (±30%) → Measured (±15%) → Validated (±10%). Each level requires progressively harder evidence.

Portfolio Dashboard: Shows total Estimated ROI (value created), Actual ROI (captured − TCO), Capture Rate %, and per-initiative breakdown with green/red indicators.

See docs/pmo-framework.md for the complete PMO guide.

Ships as a GitHub Copilot CLI extension with 16 tools. Mention "kaizen", "initiative", "eval", or "transformation" and tools auto-activate.

Install:

• Per-repo: Already in .github/extensions/ai-kaizen/ — works for anyone who clones
• User-wide: Copy to ~/.copilot/extensions/ai-kaizen/

┌─────────────────────────────────────────────────┐
│  LAYER 5: Business Outcome (start here)         │
│  LAYER 4: Eval Criteria (5 levels: L0→L3)       │
│  LAYER 3: Agent Architecture (maturity-labeled)  │
│  LAYER 2: Data + Infra Readiness                │
│  LAYER 1: People + Culture + Governance         │
└─────────────────────────────────────────────────┘
         ↕ Feedback loops between all layers

Execution: Three nested PDCA loops — Discovery (2-4 weeks) → Validation (6-12 weeks) → Scaling (ongoing)

ai-kaizen/
├── src/ai_kaizen/
│   ├── cli.py                  # Click CLI (17 commands, 841 lines)
│   ├── domain/models.py        # Pydantic domain models
│   ├── store/database.py       # SQLite persistence (WAL mode, 15 tables)
│   ├── services/core.py        # Business logic (7 service classes)
│   ├── scaffolds/
│   │   ├── eval_templates.py   # L0-L3 pytest scaffold generators
│   │   ├── initiative_canvas.py # A3 one-pager markdown export
│   │   └── agent_decision_tree.py # 8-question agent readiness check
│   ├── logging_config.py       # Structured logging (JSON optional)
│   ├── metrics.py              # Process metrics (counters, histograms)
│   └── web/
│       ├── app.py              # Flask app factory, CSRF, lazy store
│       ├── routes.py           # 20 route handlers
│       └── templates/          # 9 Jinja2 templates (dark theme)
├── .github/
│   ├── extensions/ai-kaizen/   # Copilot CLI skill (16 tools)
│   └── copilot-instructions.md # DB conventions, test commands, macOS notes
├── docs/
│   ├── index.html              # Static site for file:// browsing
│   ├── framework-v2.md         # Full framework document
│   ├── pmo-framework.md        # PMO portfolio management guide
│   └── research.md             # Foundation research
├── examples/
│   └── seed-healthcare-portfolio.sh  # Template portfolio seed script
├── tests/
│   ├── test_services.py        # 42 service + scaffold tests
│   ├── test_web.py             # 39 web route + CxO dashboard tests
│   ├── test_store.py           # 27 store layer tests
│   └── evals/test_eval_l0.py   # Example L0 safety eval stubs
└── pyproject.toml              # Python 3.9+, Flask, Click, Rich, Pydantic

The demo database includes fully-populated initiatives across 8 industries:

Each has outcomes, eval suites with runs, PDCA history, data readiness, PMO scores, ROI entries, and value events.

AI must earn the right to act. Every agent starts at L0 (inform only) and can only advance to L1 → L2 → L3 autonomy by passing progressively harder eval gates — with operations leadership (not the AI team) making the promotion call based on evidence.

The toolkit enforces this: no skipping levels, no vibes-based promotion, explicit kill criteria at every gate.

• Foundation Research: Kaizen, Gemba, AI/ML
• AI-Kaizen Framework v2

git clone https://github.com/KevinCrosby/ai-kaizen.git
cd ai-kaizen
python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"

# Run tests (108 passing, excludes eval stubs)
python -m pytest tests/ --ignore=tests/evals -q

# Run with debug logging
AI_KAIZEN_LOG_LEVEL=DEBUG ai-kaizen status

# Dev server with auto-reload
ai-kaizen serve --debug

# Process metrics
ai-kaizen metrics --json

All data lives in a single SQLite file (WAL mode, 15 tables). To start fresh: rm ~/.ai-kaizen/kaizen.db

Project-specific database: export AI_KAIZEN_DB=./my-project.db

MIT