M2C1 - Fully Autonomous Development Framework

A meta orchestration framework for Claude Code that converts brain dumps into fully built, tested, and deployed software - autonomously.

What It Does

M2C1 is a 12-phase workflow that takes your idea from concept to production with minimal human intervention. You provide a brain dump, and the system orchestrates research, planning, implementation, and testing through coordinated subagents.

The 12 Phases

Phase	What Happens
0. Setup	Creates orchestration folder structure
1. Brain Dump to PRD	Synthesizes your idea into a structured Product Requirements Document
2. First Research Wave	Parallel subagents research every domain your project touches
3. Discovery Questions	Asks you detailed questions to fully clarify scope and decisions
4. Second Research Wave	Implementation-focused research filtered through your answers
5. Tool Setup	Configures MCP servers, API keys, and external services
6. Tool Verification	Tests every tool and integration before building
7. Skill Creation	Creates project-level skills for each tool and domain
8. Context Compact	Prompts context refresh (all state is in files)
9. PHASES.md Creation	Master implementation plan with tasks, dependencies, and testing
10. Task Sharding	Expands each task into a detailed, self-contained prompt for execution agents
11. Synergy Review	Cross-phase coherence check for contradictions and gaps
12. Final Artifacts	Creates PROGRESS.md, START.md, and updates CLAUDE.md

After setup, run /start and the orchestrator autonomously executes every task, tests every feature, and runs comprehensive end-to-end testing on the deployed software.

Key Principles

• Every artifact has a template - Agents read templates before creating anything
• Parallel by default - Subagents run in background wherever independent
• Multi-angle testing at every level - Task-local, phase regression, final comprehensive e2e
• Human-emulating testing - Agents test as users would via Playwright + simulated assets
• Tool-aware - Researches and configures MCP servers, CLIs, external services
• Agent-autonomous - Agents do "human steps" in browser, only truly manual steps require user
• Generalizable - Works for any software type (SaaS, API, CLI, mobile, etc.)

Installation

As a Claude Code Skill

Copy the entire folder to your Claude Code skills directory:

cp -r m2c1/ ~/.claude/skills/m2c1/

Claude Code will auto-discover the skill via the SKILL.md description.

Prerequisites

• Claude Code installed and configured
• Playwright MCP must be available (browser automation for testing and tool setup)

Usage

1. Start a Claude Code session in your project directory
2. Describe your project idea (brain dump style is fine)
3. Claude will recognize the M2C1 skill and begin the 12-phase workflow
4. Answer discovery questions when prompted
5. After Phase 12 completes, run /start to begin autonomous execution

File Structure

m2c1/
├── SKILL.md                           # Skill entry point (auto-discovered)
├── orchestration-workflow.md           # Complete 12-phase protocol
└── artifact-templates/
    ├── prd.md                         # Product Requirements Document template
    ├── phases.md                      # Implementation plan template
    ├── progress.md                    # Progress tracker template
    ├── research-file.md               # Research output template
    ├── task-file.md                   # Execution task template
    ├── discovery.md                   # Discovery Q&A template
    ├── claude-md-section.md           # CLAUDE.md orchestration section template
    └── start.md                       # Orchestrator protocol template

How It Works Under the Hood

Planning Phase (Phases 0-8): The main agent coordinates research subagents, asks you discovery questions, configures tools, and creates project-level skills. All decisions are stored in DISCOVERY.md - the top authority document.

Sharding Phase (Phases 9-12): The master plan (PHASES.md) gets broken into individual task files - each one a complete, self-contained prompt for an execution agent. These are reviewed for cross-phase coherence before execution begins.

Execution Phase (Post-Setup): The orchestrator reads PROGRESS.md, finds the next pending task, spawns a subagent with the task file, and moves to the next task after completion. Every phase ends with a regression test. The final phase is dedicated entirely to comprehensive e2e testing on the live software.

Failure Handling: 3-tier escalation - subagent self-recovery, orchestrator intervention, user escalation (last resort).

License

MIT