Back to ArcKit

Choose your AI platform

1

Install Claude Code

macOS, Linux & WSL:

curl -fsSL https://claude.ai/install.sh | bash

Windows (PowerShell):

irm https://claude.ai/install.ps1 | iex

Already have claude on your PATH? Make sure you're on the latest:

claude install latest
ArcKit v4.14.0+ requires Claude Code v2.1.121 or later. claude install accepts stable, latest, or a specific version. The plugin warns at session start if you're below the supported floor. See setup docs for Homebrew and WinGet alternatives.
2

Install ArcKit Plugin

/plugin marketplace add tractorjuice/arc-kit

Then install from the Discover tab. The plugin provides the ArcKit command set, autonomous research agents, automation hooks, and bundled MCP servers. Updates are automatic via the marketplace.

Lighter clone: to skip the other AI-assistant distributions in the monorepo (~100 MB), add the marketplace from your shell with sparse-checkout:

claude plugin marketplace add tractorjuice/arc-kit --sparse .claude-plugin arckit-claude
3

Open Your Project

cd my-project
claude
4

Get Oriented

/arckit.start

Shows project status, connected tools, command decision tree, and recommended next steps.

5

Initialize Project Structure

/arckit.init

Creates the projects/ directory structure for architecture artifacts.

1

Install Gemini CLI

npm (requires Node.js 20+):

npm install -g @google/gemini-cli

Homebrew (macOS & Linux):

brew install gemini-cli
2

Install ArcKit Extension

gemini extensions install https://github.com/tractorjuice/arckit-gemini

Zero-config: ArcKit commands, templates, scripts, and bundled MCP servers. Update with gemini extensions update arckit.

3

Open Your Project

cd my-project
gemini
4

Get Oriented

/arckit:start

Shows project status, connected tools, command decision tree, and recommended next steps.

5

Initialize Project Structure

/arckit:init

Creates the projects/ directory structure for architecture artifacts.

1

Install Codex CLI

npm (requires Node.js 22+):

npm install -g @openai/codex

Homebrew:

brew install codex
2

Install ArcKit Plugin

Add the ArcKit Codex marketplace:

codex plugin marketplace add tractorjuice/arckit-codex

Enable lifecycle hooks in your user or project Codex config:

# ~/.codex/config.toml or .codex/config.toml
[features]
hooks = true
plugin_hooks = true

Restart Codex, run /plugins, choose ArcKit Plugins, then install and enable ArcKit. The plugin bundles 120 skills, 10 agent configs, MCP config, templates, schemas, scripts, and hooks/hooks.json.

Prefer a project-scoped copy instead? Use arckit init my-project --ai codex from the ArcKit CLI.

3

Open Your Project

cd my-project
codex
4

Get Oriented

$arckit-start

Shows project status, connected tools, command decision tree, and recommended next steps.

5

Initialize Project Structure

$arckit-init

Creates the projects/ directory structure for architecture artifacts.

1

Install OpenCode CLI

Install script (recommended):

curl -fsSL https://opencode.ai/install | bash

npm:

npm install -g opencode-ai

Homebrew (macOS & Linux):

brew install anomalyco/tap/opencode
2

Install ArcKit CLI & Initialize Project

Install the ArcKit CLI, then initialize your project:

# Install with pip
pip install git+https://github.com/tractorjuice/arc-kit.git

# Or with uv (recommended)
uv tool install arckit-cli --from git+https://github.com/tractorjuice/arc-kit.git

# Initialize project
arckit init my-project --ai opencode

This copies ArcKit commands, templates, and scripts into your project.

3

Open Your Project

cd my-project
opencode
4

Get Oriented

/arckit.start

Shows project status, connected tools, command decision tree, and recommended next steps.

5

Initialize Project Structure

/arckit.init

Creates the projects/ directory structure for architecture artifacts.

1

Install VS Code & GitHub Copilot

Install VS Code and the GitHub Copilot extension.

Requires a GitHub Copilot subscription (Individual, Business, or Enterprise).
2

Install ArcKit CLI & Initialize Project

Install the ArcKit CLI, then initialize your project:

# Install with pip
pip install git+https://github.com/tractorjuice/arc-kit.git

# Or with uv (recommended)
uv tool install arckit-cli --from git+https://github.com/tractorjuice/arc-kit.git

# Initialize project
arckit init my-project --ai copilot

This scaffolds 67 prompt files, 9 research agents, templates, and scripts into your project.

3

Open Your Project in VS Code

code my-project
4

Use Commands in Copilot Chat

Open Copilot Chat (Ctrl+Shift+I / Cmd+Shift+I) and type / to see all arckit-* commands in autocomplete.

/arckit-start

Shows project status, command decision tree, and recommended next steps.

5

Initialize Project Structure

/arckit-init

Creates the projects/ directory structure for architecture artifacts.

Fast path

Vibe Start — the 3-prompt super prompt

Skip the decision tree. Three prompts take you from empty repo to a fully-populated project. ArcKit chains commands via the handoffs: metadata in each command's frontmatter.

Use your platform syntax: Claude Code/OpenCode use /arckit.command, Gemini CLI uses /arckit:command, Codex CLI uses $arckit-command, and GitHub Copilot uses /arckit-command.

Claude Code / OpenCode example:

/arckit.init This is a project for {one-line description of what you're building}.
/arckit.principles

Codex CLI example:

$arckit-init This is a project for {one-line description of what you're building}.
$arckit-principles
Now create all the artifacts. Take your time. Do not stop until complete.

When to use it: greenfield projects, demos, proofs of concept, vibe-coding sessions where you want a complete first-pass set of artifacts to react to.

When not to use it: heavily regulated work (Secure by Design, MOD, EU AI Act) where each artifact needs reviewer sign-off; existing projects with artifacts already on disk (run /arckit.navigator first); token-constrained sessions.

If the run stalls: resume with "Continue from where you stopped. Do not stop until complete." Then run your platform's health command (/arckit.health, /arckit:health, $arckit-health, or /arckit-health) to spot anything skipped. Full guide: Getting Started with ArcKit.

Faster path (Claude Code only)

The GDS Harness — /arckit.build

New in v4.13.0. Run one command and the whole architecture builds itself: requirements, stakeholder analysis, ADRs, risk register, business case, designs, Secure-by-Design assessment, DPIA, diagrams, traceability matrix, plus the post-build health check. Reads a YAML recipe, dispatches subagents in parallel waves, commits each wave atomically.

/arckit.build 001 --plan
/arckit.build 001
/arckit.build 002 --recipe uk-mod-sovereign
/arckit.build 003 --recipe uae-federal-ai
/arckit.build 004 --recipe ca-federal-fitaa

Four built-in recipes: uk-saas (38 targets, UK civilian SaaS), uk-mod-sovereign (38 targets, MOD / air-gapped with JSP 936 and MOD Secure by Design), uae-federal-ai (48 targets, UAE Cabinet agentic AI decree compliance with all 12 UAE community commands), and ca-federal-fitaa (Canada Federal FITAA-class compliance suite — all 12 ca-* commands plus core architecture governance).

Every recipe ships a research wave by default. An upstream ORG_RESEARCH target researches the target organisation once per repo (output to projects/000-global/research/), then per-project technology research runs in parallel (general market, AWS, Azure, GCP, plus GOV_REUSE for UK projects). Skip what you don't need with --exclude AZURE_RESEARCH etc.

Customize: copy a recipe to .arckit/recipes/{name}.yaml and edit. The harness reads project overrides first, falling back to the plugin defaults — your customisations survive plugin updates.

Resumable: state persists to projects/{P}/.arckit/state.json. If a wave fails, fix the underlying issue and run /arckit.build 001 --resume to pick up exactly where it stopped.

Codex note: this command is not generated for Codex. The build harness depends on Claude Code's parallel agent dispatch and plugin hook behaviour, so there is no supported $arckit-build skill.

When to use it: end-to-end governance builds where you want a full template-driven artefact set committed in atomic, reviewable waves. Full guide: Build Harness Guide.

6

Follow GDS Phases

Commands shown use Claude Code / OpenCode syntax (/arckit.command). For Gemini CLI use /arckit:command. For Codex CLI use $arckit-command. For GitHub Copilot use /arckit-command.

Discovery

/arckit.stakeholders Analyze stakeholders for appointment system
/arckit.risk Create risk register
/arckit.sobc Generate Strategic Outline Business Case

Alpha

/arckit.requirements Define requirements
/arckit.principles Establish architecture principles
/arckit.wardley Create Wardley Map for build vs buy decisions

Beta

/arckit.hld-review Review High-Level Design
/arckit.diagram Create C4 architecture diagrams
/arckit.analyze Comprehensive gap analysis
Warning AI generates first drafts following templates. Always review, validate, and refine AI output with human expertise.

Next Steps

Command Reference

Browse the ArcKit command set with filters and dependency matrix.

Command Guides

Step-by-step guides with prerequisites and real-world examples.

DDaT Role Guides

Find commands tailored to your UK Government DDaT role.

Example Projects

See AI-generated architecture documentation from real-world scenarios.