# AtomizerSpec v2.0 - Quick Start Guide

## TL;DR

**Goal**: Replace 4+ config formats with one unified `atomizer_spec.json` that Canvas, Backend, Claude, and Optimization Engine all use.

**Effort**: ~50 tasks across 4 phases (~12 weeks)

**Key Files**:
- Design: `docs/plans/UNIFIED_CONFIGURATION_ARCHITECTURE.md`
- Tasks: `docs/plans/UNIFIED_CONFIG_EXECUTION_PLAN.md`
- Schema: `optimization_engine/schemas/atomizer_spec_v2.json`
- Prompts: `docs/plans/UNIFIED_CONFIG_RALPH_PROMPT.md`

---

## Start Autonomous Implementation

### 1. Open Claude CLI
```bash
claude --dangerously-skip-permissions
```

### 2. Paste Initial Prompt
```
You are implementing the AtomizerSpec v2.0 Unified Configuration Architecture for Atomizer.

Read these documents before starting:
1. `docs/plans/UNIFIED_CONFIGURATION_ARCHITECTURE.md` - Master design
2. `docs/plans/UNIFIED_CONFIG_EXECUTION_PLAN.md` - Task list (50+ tasks)
3. `optimization_engine/schemas/atomizer_spec_v2.json` - JSON Schema

Implement tasks in order (P1.1 → P4.12). Use TodoWrite to track progress. Commit after each logical group. Reference the design document for code examples.

Begin by reading the documents and loading tasks into TodoWrite. Start with P1.1.
```

### 3. Let It Run
Claude will work through the task list autonomously, committing as it goes.

---

## Phase Summary

| Phase | Tasks | Focus | Key Deliverables |
|-------|-------|-------|------------------|
| **1** | P1.1-P1.17 | Backend | SpecManager, REST API, Migration |
| **2** | P2.1-P2.16 | Frontend | SpecRenderer, WebSocket Sync |
| **3** | P3.1-P3.14 | Claude | MCP Tools, Custom Functions |
| **4** | P4.1-P4.12 | Polish | Tests, Migration, Docs |

---

## Key Concepts

**AtomizerSpec**: Single JSON file containing everything needed for optimization AND canvas display.

**SpecManager**: Backend service that validates, saves, and broadcasts spec changes.

**SpecRenderer**: Frontend component that renders canvas directly from spec.

**Real-time Sync**: WebSocket broadcasts changes to all connected clients.

**Custom Functions**: Python code stored in spec, loaded dynamically during optimization.

---

## Success Metrics

- [ ] Canvas loads/saves without data loss
- [ ] Claude can modify specs via MCP tools
- [ ] All existing studies migrated
- [ ] WebSocket sync works between clients
- [ ] Custom extractors execute in optimization
- [ ] All tests pass

---

## If Resuming

```
Continue implementing AtomizerSpec v2.0.

Reference:
- `docs/plans/UNIFIED_CONFIG_EXECUTION_PLAN.md`

Check current todo list, find last completed task, continue from there.
```

---

*Full details in the referenced documents.*