ralph-deep-init

name	ralph-deep-init
description	Deep project initialization using the Architect-Builder pattern. Use for large projects (20+ tasks, multiple domains) where a single task breakdown would be unwieldy or exceed context limits.

Ralph Deep Init Skill

Load this skill when running /ralph-deep-init. Splits initialization across three phases to manage scope and context.

When to Use

Use deep init instead of standard init when:

• The project has 20+ distinct tasks
• Work spans 4+ independent domains (auth, API, UI, infra, etc.)
• A single prd.json would be difficult to reason about in one pass

Phase 1: Architect

Analyze the full project scope. Identify exactly 6 distinct functional groups that together cover the entire project without overlap.

Good group examples: Authentication & Authorization, API Layer, Database Schema & Migrations, Frontend UI, DevOps & Deployment, Testing Infrastructure

Create architecture.json:

{
  "project": "Project name",
  "description": "What this project builds",
  "groups": [
    {
      "id": "group-001",
      "name": "Authentication & Authorization",
      "description": "User login, JWT, RBAC, session management",
      "estimated_tasks": 6
    },
    {
      "id": "group-002",
      "name": "API Layer",
      "description": "REST endpoints, validation, error handling",
      "estimated_tasks": 8
    }
  ]
}

Report: "Architecture defined: [list all 6 groups with estimated task counts]"

Phase 2: Builders

For each group in architecture.json, create a partial task file.

Filename: prd-partial-{group-id}.json

{
  "group_id": "group-001",
  "group_name": "Authentication & Authorization",
  "tasks": [
    {
      "id": "task-001",
      "title": "Set up JWT token generation and validation",
      "priority": "high",
      "status": "pending",
      "dependencies": [],
      "acceptance_criteria": [
        "JWT tokens are generated with configurable expiry",
        "Token validation rejects expired and malformed tokens",
        "Unit tests pass for both generation and validation paths"
      ]
    }
  ]
}

Cross-group dependencies: Use format group-002/task-005 — these will be resolved in Phase 3.

Create all 6 partial files before proceeding to Phase 3.

Phase 3: Assembly

1. Read all prd-partial-*.json files
2. Merge all tasks arrays into a single list
3. Renumber task IDs sequentially: task-001, task-002, ...
4. Resolve all cross-group dependency references to final sequential task IDs
5. Sort tasks: high priority first within dependency constraints
6. Write final prd.json
7. Create AGENTS.md — use the ralph-init skill structure
8. Create progress.log with header
9. Delete all prd-partial-*.json files
10. Delete architecture.json

Report:

Deep init complete.

• AGENTS.md — ready
• prd.json — [N] total tasks ([N] high, [N] medium, [N] low priority)
• progress.log — ready

Run /ralph-loop to start.