Sprint Planning

Creates structured task files with YAML frontmatter for agent coordination and automated progress tracking.

Bash
# Read PRD and Architecture documents
# Create sprint structure with individual task files
# Each task = separate .md file with YAML frontmatter for agent dispatch

Core Principle: Read PRD + Architecture → Create structured task files → Track progress through JSON

Progress:

• Phase 1: Read project documents (PRD, Architecture, CLAUDE.md)
• Phase 2: Create sprint directory structure
• Phase 3: Extract tasks from PRD into individual files
• Phase 4: Organize tasks into sprint
• Phase 5: Initialize tracking files

Phase 1: Read Project Documents

Required reading before creating tasks:

1. PRD document (~/Documents/*-PRD.md) - Extract user stories, features, acceptance criteria
2. Architecture document (~/Documents/*-Architecture.md) - Technical components, dependencies
3. CLAUDE.md - Available agents, tech stack
4. Existing agents (.claude/agents/) - Map capabilities to task types

Phase 2: Create Sprint Structure

[project]/.claude/sprints/
├── claude-progress.txt          # Session bridging (CRITICAL!)
├── features.json                # Machine-parseable progress
├── backlog/                     # Individual task files
│   ├── AUTH-001.md
│   ├── AUTH-002.md
│   └── AVATAR-001.md
├── sprint-001/                  # Active sprint
│   ├── sprint.md               # Sprint overview
│   └── [task files moved from backlog]
├── completed/                   # Archived sprints
└── templates/
    └── task.md

Phase 3: Task File Format

Each task = separate file with YAML frontmatter:

Markdown
---
task_id: AUTH-001
title: Setup Supabase Auth
agent: backend-developer
sprint: 1
priority: P0
points: 3
status: pending
dependencies: []
---

# AUTH-001: Setup Supabase Auth

• Configure auth in shared/api/supabase
• Create server/client instances
• Add RLS policies

• Email/password auth working
• Session management implemented
• Tests passing

• src/shared/api/supabase/
• src/features/auth/

### Phase 4: Agent Assignment

| Task Type | Agent |
|-----------|-------|
| UI, Components, Pages | frontend-developer |
| API, DB, Webhooks | backend-developer |
| Tests | test-engineer |
| Stripe | payments-specialist |
| AI APIs (HeyGen, OpenAI) | ai-integrations |
| Telegram, Social | messaging-specialist |

### Phase 5: Progress Tracking Files

**claude-progress.txt** (Session bridging):
```text
=== Session 2026-01-20 10:30 ===
Agent: backend-developer
Task: AUTH-001
Status: in_progress
Progress: Created Supabase client configuration
Remaining: Create useAuth hook, write tests
Blockers: None
Next: Continue AUTH-001

features.json (Machine-parseable):

JSON
{
  "version": "1.0",
  "sprints": {"current": 1, "total": 1},
  "features": [
    {
      "id": "AUTH-001",
      "title": "Setup Supabase Auth",
      "agent": "backend-developer",
      "sprint": 1,
      "priority": "P0",
      "points": 3,
      "status": "pending",
      "dependencies": []
    }
  ]
}

Example 1: Planning from PRD Input: PRD with "User can login with email/password" user story Output: AUTH-001.md task file with Supabase auth setup requirements

Example 2: Agent Dispatch Input: claude "Execute task from .claude/sprints/sprint-001/AUTH-001.md" Output: Agent reads YAML metadata, executes task, updates status

Example 3: Sprint Transition Input: All sprint 1 tasks have status: completed Output: Automatic transition to sprint 2, archived sprint 1 to completed/

• One task per file - Enables agent dispatch via file path
• YAML frontmatter required - Contains all metadata for parsing
• Extract from PRD, don't invent - Tasks based on actual requirements
• Update claude-progress.txt after each session - Critical for session bridging
• Use story point estimation - Base on PRD complexity (1-8 points)
• Status values: pending, in_progress, completed, blocked

• ❌ Creating tasks without reading PRD - Tasks must come from requirements
• ❌ One file with all tasks - Agent dispatch breaks
• ❌ Markdown checkboxes for status - Use YAML frontmatter for machine parsing
• ❌ Ignoring claude-progress.txt - Next session starts blind
• ❌ Day/Week-based tracking - Use session-based progress instead
• ❌ Agent tags in titles like [FE] - Put agent in YAML frontmatter
• ❌ Missing dependencies - Extract from architecture document