rai-cli 2.0.0a8

RaiSE CLI - Reliable AI Software Engineering governance framework

RaiSE

Reliable AI Software Engineering — Ship quality software at AI speed.

Raise your craft, feature by feature.

---

What is RaiSE?

RaiSE is a methodology + toolkit for professional developers who use AI assistants. It solves the problem of AI-generated code that's fast but inconsistent: governance that works naturally, validation at every step, and memory that persists across sessions.

The RaiSE Triad:

        RaiSE Engineer
        (You - Strategy, Judgment, Ownership)
              │
              │ collaborates with
              ▼
           Rai
   (AI Partner - Execution + Memory)
              │
              │ governed by
              ▼
           RaiSE
    (Methodology + Toolkit)

Rai is your AI collaborator — not a generic assistant, but a partner trained in the discipline of reliable AI software engineering. Rai remembers your patterns, calibrates to your velocity, and maintains coherence across sessions.

---

Developer Onboarding

Prerequisites

• Python 3.12+
• uv (recommended) or pip
• Claude Code CLI installed and configured

Quick Install (from PyPI)

# Current release is alpha — requires --pre flag
pip install --pre rai-cli

# Or with uv:
uv pip install --prerelease=allow rai-cli

# Verify
rai --help

Development Setup

# 1. Clone and checkout the development branch
git clone https://gitlab.com/humansys-demos/product/raise1/raise-commons.git
cd raise-commons
git checkout v2

# 2. Install in development mode
uv pip install -e ".[dev]"

If error "No virtual environment found."
run `uv venv` to create an environment.

# 3. Verify installation
rai --help

Windows WSL (Ubuntu/Debian).

# 1 — Use pipx

sudo apt update
sudo apt install pipx -y
pipx ensurepath


# 2 Close and new terminal WSL.

# 3 Install:

pipx install rai-cli

Onboarding with Rai

Once installed, open Claude Code in the project directory and run:

/rai-welcome

This single command will:

• Detect your situation (new developer, returning developer, etc.)
• Create your profile (~/.rai/developer.yaml) with your name and pattern prefix
• Build the knowledge graph so Rai has project context
• Scaffold CLAUDE.local.md for your personal Claude Code instructions
• Optionally customize communication preferences (language, style)
• Verify everything works

After welcome completes, start working:

/rai-session-start

This loads your context, memory, and proposes focused work.

What You Get

Shared (committed)	Personal (gitignored)
Patterns (.raise/rai/memory/patterns.jsonl)	Session history (.raise/rai/personal/sessions/)
Governance docs	Session state (.raise/rai/personal/session-state.yaml)
Skills, methodology	Calibration data (.raise/rai/personal/calibration.jsonl)
Work artifacts	Knowledge graph (.raise/rai/memory/index.json)

Each developer builds their own personal context through working sessions. Pattern IDs are developer-prefixed (e.g., PAT-A-001 for Alice, PAT-B-001 for Bob) to prevent collisions in shared repositories.

---

Usage

Initialize RaiSE on Your Project

# Navigate to your existing project
cd your-project

# Initialize RaiSE governance structure
rai init --detect

# Open Claude Code and run onboarding
/rai-welcome

This scaffolds the .raise/ directory, detects your project's conventions (language, testing framework, linting), and builds the knowledge graph.

Daily Workflow

A typical session follows this pattern:

1. /rai-session-start          # Load context, see what's pending
2. /rai-story-start             # Create branch, define scope
3. /rai-story-design            # Design the approach (recommended)
4. /rai-story-plan              # Break into atomic tasks
5. /rai-story-implement         # TDD execution with validation gates
6. /rai-story-review            # Retrospective, capture patterns
7. /rai-story-close             # Merge, cleanup
8. /rai-session-close           # Persist learnings for next session

You don't need to complete all steps in one session — Rai remembers where you left off.

What Rai Remembers

• Patterns — Reusable insights learned from your work (e.g., "always validate config at boundaries")
• Calibration — Your velocity, strengths, growth edges
• Session history — What you worked on, decisions made, items deferred
• Coaching corrections — Mistakes Rai made and learned from

Each session builds on the last. Over time, Rai becomes a more effective collaborator for your specific codebase and working style.

---

Available Skills

Skills are structured processes that guide AI-assisted development. Run them as /skill-name in Claude Code.

Session Lifecycle

Skill	Purpose
/rai-welcome	One-time developer onboarding
/rai-session-start	Begin a session with memory and context
/rai-session-close	End a session, persist learnings

Story Lifecycle

Skill	Purpose
/rai-story-start	Initialize a story with branch and scope
/rai-story-design	Create lean specs for complex stories
/rai-story-plan	Decompose into atomic tasks
/rai-story-implement	Execute with TDD and validation gates
/rai-story-review	Retrospective and learnings
/rai-story-close	Merge, cleanup, tracking

Epic Lifecycle

Skill	Purpose
/rai-epic-start	Initialize an epic with branch
/rai-epic-design	Design multi-story epics
/rai-epic-plan	Sequence stories into plans
/rai-epic-close	Epic retrospective and merge

Other Skills

Skill	Purpose
/rai-research	Epistemologically rigorous research
/rai-debug	Root cause analysis (5 Whys, Ishikawa)
/rai-docs-update	Sync architecture docs with code
/rai-discover-start	Initialize codebase discovery
/rai-discover-scan	Extract and describe components

---

CLI Commands

The rai CLI provides deterministic operations:

# Build Rai's knowledge graph from project artifacts
rai memory build

# Query governance concepts
rai memory context mod-session

# Query Rai's memory
rai memory query "velocity patterns"

# Validate the memory graph (structural + completeness)
rai memory validate

# Visualize the memory graph as interactive HTML
rai memory viz                    # Opens in browser
rai memory viz --output graph.html  # Custom output path

# List releases and their associated epics
rai release list

# Start a session (creates profile on first run)
rai session start --name "YourName" --project "$(pwd)" --context

# Close a session
rai session close --state-file /tmp/session-output.yaml --project "$(pwd)"

---

Repository Structure

raise-commons/
├── .claude/skills/      # Claude Code skills (24 skills)
│
├── framework/           # Public textbook (concepts, reference)
│   ├── reference/       #   Constitution, glossary, philosophy
│   ├── concepts/        #   Core concepts (katas, gates, artifacts)
│   └── getting-started/ #   Greenfield/brownfield guides
│
├── .raise/              # Framework engine
│   ├── rai/             #   Rai's memory and personal data
│   │   ├── memory/      #     Patterns, knowledge graph (shared)
│   │   └── personal/    #     Sessions, calibration (per-developer, gitignored)
│   ├── katas/           #   Process definitions
│   ├── gates/           #   Validation criteria
│   ├── templates/       #   Artifact scaffolds
│   └── skills/          #   Legacy skill definitions
│
├── governance/          # Project governance
│   ├── architecture/    #   Module docs, system design
│   └── solution/        #   Vision, guardrails, business case
│
├── src/rai_cli/         # CLI toolkit (Python)
│
├── work/                # Work in progress
│   └── stories/         #   Story artifacts (scope, design, plan, retro)
│
└── dev/                 # Framework maintenance
    ├── decisions/       #   ADRs (Architecture Decision Records)
    └── parking-lot.md   #   Ideas and tangents for later

---

Branch Model

main (stable releases)
  └── v2 (development)
        └── epic/e{N}/{name}
              └── story/s{N}.{M}/{name}

• Work on v2 (development branch)
• Stories branch from and merge back to their epic or v2
• main receives releases from v2

---

Core Concepts

Concept	Description
RaiSE Engineer	You — the human who directs AI-assisted development
Rai	AI partner with memory, calibration, and accumulated judgment
Skill	Structured Claude Code prompt for a methodology phase
Validation Gate	Quality checkpoint with specific criteria
Guardrail	Constraint that guides AI behavior
ShuHaRi	Mastery levels (beginner → practitioner → master) that adapt Rai's verbosity

See the full Glossary for canonical terminology.

---

Key Principles

From the Constitution:

1. Humans Define, Machines Execute — Specs are source of truth
2. Governance as Code — Standards versioned in Git
3. Validation Gates — Quality checked at each phase
4. Observable Workflow — Every decision traceable
5. Jidoka — Stop on defects, don't accumulate errors

---

Status

This is a pre-release (v2.0.0-alpha). The framework is being used in production but the API may change.

We value your feedback:

• Questions? Open an issue
• Found a bug? Open an issue with reproduction steps
• Ideas? We want to hear them — open an issue or reach out directly

---

License

Apache-2.0

---

RaiSE — Reliable AI Software Engineering Neither is complete alone.