Getting Started

Archon Protocol is a constraint system that turns AI coding agents into reliable architects. It provides agents, skills, and constraints that guide AI behavior during development.

Prerequisites

• An AI coding tool: Cursor, Claude Code, Codex, VS Code with Copilot, or any tool supporting SKILL.md
• A project with source code

Quick Install

bash archon-protocol/templates/install.sh

This deploys:

• Agents → .cursor/agents/, .claude/agents/
• Skills → .cursor/skills/, .claude/skills/, .codex/skills/
• Config → archon.config.yaml

First Run

In your AI tool, type:

/archon-init

The init command scans your project, detects the tech stack, and confirms everything is deployed correctly.

Daily Usage

Feature development

/archon-demand Add dark mode toggle to settings

This triggers the full delivery pipeline: implement → performance audit → 6-dimension self-audit → fix → test sync → knowledge evolution → commit.

Health check

/archon-audit

Read-only project-wide audit with a scored report (0-100).

Architecture planning

/archon-refactor

Generates a progressive refactoring plan. Future /archon-demand calls automatically align with the plan.

How It Works

You say one line → Agent does everything

/archon-demand "add user settings page"
  │
  ├─ Stage 0: Read refactor plan (if exists)
  ├─ Stage 1: Implement (under constraint skills)
  ├─ Stage 1.5: Linter verification
  ├─ Stage 2: Performance audit
  ├─ Stage 3: 6-dimension self-audit
  ├─ Stage 4: Fix all issues
  ├─ Stage 5: Update refactor progress
  └─ Stage 6: Commit

Choosing the Right Mode

Not every task needs the full pipeline. Use opt-out flags to match the task:

Scenario	Command	What runs
Complex feature	/archon-demand add user settings	All stages (full pipeline)
Quick hotfix	/archon-demand quick fix typo in header	Stages 0, 1, 1.5, 3.1–3.4, 4, 6
Exploration (no commit)	/archon-demand no-commit try new layout	All stages except commit
Styling only	/archon-demand quick skip-tests update button colors	Stages 0, 1, 1.5, 3.1–3.3, 4, 6

Decision tree:

Is this a complex feature or refactor?
  → Yes: use full pipeline (no flags)
  → No:
      Is this a quick fix or small change?
        → Yes: add `quick`
      Do you want to review before committing?
        → Yes: add `no-commit`
      Is this purely visual (no logic change)?
        → Yes: add `skip-tests`

Tool Compatibility

Tool	What you get
Cursor	Agents (primary) + Skills
Claude Code	Agents (primary) + Skills
Codex, Copilot, VS Code	Skills only
Any SKILL.md tool	Skills only

Agents provide isolated context windows. Skills provide the same workflows in a portable format.