ARCH protocol skill — structured human+AI software development based on Toyota Production System principles. Enforces atomic tasks, pre-flight validation, context declaration, and kaizen retrospectives.
Use when the user wants to analyze accumulated ARCH LOGs, detect failure patterns, and get proposed improvements to SKILL.md or CLAUDE.md
Use when the user invokes ARCH, mentions the ARCH protocol, or wants to apply Toyota Production System discipline to a software development task
"The best prompt is the one you never have to write."
AI chaos isn't fixed with better AI. It's fixed with better process.
ARCH (Autonomous Routing & Context Hierarchy) is a discipline protocol for AI-assisted software development. It's inspired by the Toyota Production System (TPS) and translates its principles into the language of Git, prompts, and AI agents.
The problem it solves: AI coding assistants are fast but chaotic. They write code without knowing which files exist, start massive refactors without a Git checkpoint, and leave no trace of what changed or why. You end up with fast output and slow debugging.
The solution: ARCH imposes a 7-step workflow that Claude follows on every task — regardless of how urgent it feels. It's not a better prompt. It's a protocol that makes any prompt work better.
| Step | What it forces Claude to do |
|---|---|
| GATE | Claude must state in its own words the goal, the files, and the constraints before writing any code. If it doesn't understand, it asks. This is a hallucination filter. |
| ANCHOR | Confirms a Git commit exists as a restore point before starting. No safety net, no change. |
| ATOM | If the task is large (more than 5 files or 3 responsibilities), Claude suggests splitting it into smaller pieces (S/M/L sizing). |
| PULL | Explicitly declares which files it will read before writing. No implicit context. |
| Generate | Produces one logical change only. One commit per task. |
| EYES | Reminds you to review the git diff and not trust the AI's summary. Final responsibility is yours. |
| LOG | Closes every task with a 3-line kaizen retrospective: what worked, what failed, what you'd change. This builds a learning record that improves future tasks and gives you data to evolve your own process. |
Most problems with AI-assisted development aren't model problems — they're process problems. The same mistakes repeat: lost context, no Git safety net, tasks too large to review, no institutional memory.
ARCH borrows TPS's answer to the same problem in manufacturing: stop the line before defects multiply.
The protocol is deliberately resistant to pressure. When you say "just write the fix, I have a demo in 2 hours", ARCH doesn't comply — it runs GATE first, because that's exactly the moment when skipping it causes the most damage.
"I built ARCH because I was tired of spending more time debugging AI-generated code than writing it. After 6 months of daily use, I've reduced my context-loss errors by 80%. This is the protocol I wish I'd had from day one."
— Valentín Liñeiro, creator of ARCH.
Add the marketplace, then install the plugin:
/plugin add-marketplace https://github.com/valentinlineiro/arch-protocol
/plugin install arch-protocol@arch-protocol
Then invoke it:
"Let's work on X using ARCH"
Once installed, ARCH runs in the background. You'll notice Claude starts asking for context before writing code, and every task ends with a short retrospective. You don't have to think about ARCH — it just makes your AI assistant feel more… professional.
"The best punch is the one never thrown. The best prompt is the one never written.
ARCH is not a better prompt. It's a protocol that makes any prompt work better."
ARCH is open and community-driven.
Contributions and forks are welcome.
Designed by Valentín Liñeiro.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
npx claudepluginhub valentinlineiro/arch-protocol --plugin arch-protocolComprehensive skill pack with 66 specialized skills for full-stack developers: 12 language experts (Python, TypeScript, Go, Rust, C++, Swift, Kotlin, C#, PHP, Java, SQL, JavaScript), 10 backend frameworks, 6 frontend/mobile, plus infrastructure, DevOps, security, and testing. Features progressive disclosure architecture for 50% faster loading.
Develop, test, build, and deploy Godot 4.x games with Claude Code. Includes GdUnit4 testing, web/desktop exports, CI/CD pipelines, and deployment to Vercel/GitHub Pages/itch.io.
Feature development with code-architect/explorer/reviewer agents, CLAUDE.md audit and session learnings, and Agent Skills creation with eval benchmarking from Anthropic.
Comprehensive feature development workflow with specialized agents for codebase exploration, architecture design, and quality review
Production-grade engineering skills for AI coding agents — covering the full software development lifecycle from spec to ship.