Scrum — AI-Adapted Agile Workflow

Lightweight Scrum framework adapted for AI agents. No meetings, no overhead — just the mechanics that drive autonomous delivery.

Quick Start

1. Initialize: node scripts/init.mjs (creates data/scrum/ structure)
2. Add work: /scrum backlog add "Build user authentication" --points 5
3. Plan sprint: /scrum sprint plan
4. View board: /scrum board
5. Agent heartbeat auto-progresses sprint items

Data Layout

data/scrum/
├── backlog.json          # Product backlog (all items not in a sprint)
├── current-sprint.json   # Active sprint metadata + items
├── sprints/              # Sprint history
│   └── archive/          # Completed sprints (SPR-001.json, ...)
└── metrics/
    └── velocity.json     # Rolling velocity data

Core Concepts

Work Item Hierarchy

Item Schema (JSON)

{
  "id": "S-001",
  "type": "story",
  "title": "Real-time chat between lobsters",
  "description": "As a user, I want lobsters to chat in real-time",
  "epic": "E-001",
  "status": "todo",
  "priority": "P1",
  "points": 3,
  "tasks": ["T-001", "T-002"],
  "sprint": "SPR-002",
  "created": "2026-03-23",
  "updated": "2026-03-23",
  "labels": ["backend", "websocket"],
  "acceptance": ["Messages delivered <500ms", "100 concurrent users"],
  "assignee": "forge"
}

Kanban Statuses

backlog → todo → in_progress → review → done → archived

• backlog — In product backlog, not committed to sprint
• todo — Committed to current sprint, not started
• in_progress — Actively being worked on
• review — PR/MR created, awaiting review
• done — Merged and verified
• archived — Sprint closed, moved to history

Story Points (Fibonacci)

Commands

All commands use /scrum prefix. See references/commands.md for full details.

Board Output Format

Strict template — agents must follow exactly:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🏃 SPR-{id} "{name}"
📅 Day {n}/{total} | ⏰ Deadline: {date}
📊 {done_pts}/{total_pts} pts | {percent}% complete
▓▓▓▓▓▓▓▓░░ {percent}%
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

📋 TODO ({count})
  ○ {id} {title} [{pts}pt] {labels}

🔨 IN PROGRESS ({count})
  ◉ {id} {title} [{pts}pt] → {context}

👀 REVIEW ({count})
  ◎ {id} {title} [{pts}pt] → PR #{n}

✅ DONE ({count})
  ✓ {id} {title} [{pts}pt]

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🏔️ Epic {id} — {percent}% | 📦 Backlog: {n} items
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Rules:

1. Progress bar: ▓ (done) + ░ (remaining), 10 chars total
2. Empty columns: skip entirely (don't show)
3. Labels: 🏷️ label1, label2 (only if present)
4. Separator: ━ × 35 characters

Workflows

Sprint Planning

1. Read backlog.json — sort by priority
2. Calculate capacity (default: 20 pts per 5-day sprint)
3. Pull highest-priority stories up to capacity
4. Create current-sprint.json with status todo
5. Break stories into tasks if needed

Sprint Execution (Heartbeat-driven)

Each heartbeat cycle:

1. Pick highest-priority todo → set in_progress → execute
2. Check in_progress → update status from task runner
3. Check review → poll PR status → done if merged
4. Sprint end date reached → trigger Sprint Review

Sprint Review & Retro

On sprint completion:

1. Calculate velocity (completed story points)
2. Move incomplete stories back to backlog
3. Generate summary (shipped, missed, blockers)
4. Analyze patterns (estimate accuracy, bottlenecks)
5. Archive to sprints/archive/SPR-xxx.json
6. Update metrics/velocity.json

Duty Decomposition

Recurring duties generate stories automatically:

Duty: "blog-study" (weekly)
  → Epic: E-100 "Blog Knowledge Pipeline"
    → Story: S-150 "Process new blog post" (auto on detection)
      → Task: T-300 "Fetch and parse"
      → Task: T-301 "Write bilingual summary"

Integration Points

• Heartbeat — Sprint execution runs during heartbeat cycles
• Goals — Goals auto-create backlog items; scrum tracks execution detail
• Task dispatch — in_progress items map to background task dispatches
• Duties — Duties generate recurring stories/tasks