AiScrum Pro is the runtime engine for the AI-Scrum Framework — an open-source methodology for human-AI collaboration built on real Scrum principles.

Where the framework defines the what (roles, ceremonies, boundaries, manifesto), AiScrum Pro is the how — a TypeScript engine that orchestrates GitHub Copilot CLI via the Agent Client Protocol to execute sprints autonomously.

You are the Stakeholder. You set direction, drop ideas, and review deliverables. The AI team handles everything else — refinement, planning, parallel execution, quality gates, code review, sprint retros, and continuous process improvement.

What Makes This Different

This isn't a chatbot wrapper or a prompt template. It's a full AI development team with structure, boundaries, and accountability.

| | Ad-hoc AI Coding | AiScrum Pro | |---|---|---| | Planning | None — chat until it works | ICE-scored sprint backlog, milestone grouping | | Execution | One issue at a time, manually | Parallel workers via git worktrees, auto-merge | | Quality | "It should work" | 7 enforced gates: tests, lint, types, build, scope, diff-size, review | | Memory | Lost every session | Sprint logs, velocity tracking, issue comments, ADRs | | Scope control | Feature chasing | Drift detection, sprint lock, escalation model | | Improvement | Static | Every retro improves the process itself |

Built on the AI-Scrum Manifesto: Autonomous execution over constant approval. Verified evidence over claimed completion. Sprint discipline over feature chasing.

The Sprint Cycle

Five ceremonies, fully automated. Start a sprint and come back to finished work.

 ┌─────────┐    ┌──────────┐    ┌───────────┐    ┌──────────┐    ┌─────────┐
 │ Refine  │───▶│   Plan   │───▶│  Execute  │───▶│  Review  │───▶│  Retro  │
 │         │    │          │    │           │    │          │    │         │
 │ Ideas → │    │ ICE score│    │ Parallel  │    │ Velocity │    │ Process │
 │ Issues  │    │ Scope    │    │ Workers   │    │ Metrics  │    │ Improve │
 └─────────┘    └──────────┘    └───────────┘    └──────────┘    └─────────┘
                                  ↕ Quality Gates enforced after every issue

• Refinement — Stakeholder drops ideas, AI researches and decomposes into concrete issues with acceptance criteria
• Planning — AI triages backlog, ICE-scores issues, selects sprint scope, assigns milestones
• Execution — Parallel workers implement issues in isolated git worktrees, each gated by tests + lint + types + build + code review
• Review — Sprint metrics, velocity tracking, deliverable summary for stakeholder acceptance
• Retro — What went well, what didn't, process improvements applied to agents and workflows

Web Dashboard

Real-time sprint control center. Monitor progress, chat with agents, navigate sprint history.

| Sprint Board | Product Backlog | |:---:|:---:| | | |

| Sprint Report | Settings | |:---:|:---:| | | |

| Blocked Issues | Decisions Pending | |:---:|:---:| | | |

| Ideas Inbox | Logs | |:---:|:---:| | | |

9 views: Sprint Board · Sprint Backlog · Product Backlog · Blocked · Decisions · Ideas · Sprint Report · Logs · Settings

Guardrails & Controls

Autonomous execution needs boundaries. AiScrum Pro has them built in.

🔒 Drift Control — Sprint scope is locked. Discovered work goes to backlog, not into the current sprint. If >2 unplanned issues appear, the engine escalates.

⚖️ Escalation Model — The AI decides how, never what. Strategic direction changes, ADR modifications, scope changes, and dependency additions always require stakeholder approval.

🛡️ Quality Gates — 7 checks enforced on every issue: tests exist, tests pass, lint clean, types clean, build passes, scope drift check, diff size limit.

🏛️ Challenger Agent — An adversarial reviewer that challenges assumptions and finds blind spots before sprint review.

📋 Definition of Done — Acceptance criteria before coding. Tests that verify behavior. PR reviewed. CI green. Issue closed with summary. No shortcuts.

Quick Start

Prerequisites

• Node.js ≥ 20
• GitHub Copilot CLI with ACP support — copilot --acp --stdio
• gh CLI authenticated — gh auth login

Install & Run

# Install dependencies + setup git hooks
npm install

# Launch web dashboard (auto-detects sprint from milestones)
npx tsx src/index.ts web

The dashboard opens at http://localhost:9100 with live sprint status, issue tracking, and agent chat.

Test Mode

Run AiScrum Pro against dummy issues without affecting your real backlog:

make test-setup      # Create test data (milestones + issues)
make test-web        # Launch dashboard in test mode
make test-cleanup    # Remove all test artifacts when done

CLI Commands

aiscrum web                             # Launch web dashboard
aiscrum full-cycle --sprint 3           # Run complete sprint: refine → plan → execute → review → retro
aiscrum execute-issue --issue 42        # Execute a single issue
aiscrum plan --sprint 3                 # Sprint planning only
aiscrum check-quality --branch feat/x   # Run quality gates on a branch
aiscrum refine                          # Refine ideas into actionable issues
aiscrum review --sprint 3               # Sprint review ceremony
aiscrum retro --sprint 3                # Sprint retrospective
aiscrum metrics --sprint 3              # Sprint velocity & metrics
aiscrum drift-report                    # Scope drift analysis
aiscrum pause / resume                  # Pause/resume execution
aiscrum status                          # Active worker status

Configuration

Everything is config-driven. One YAML file controls the entire sprint engine. Ready-to-use examples for TypeScript, Python, React, and Go — just copy and go:

cp -r examples/python/.aiscrum .aiscrum    # Pick your stack
$EDITOR .aiscrum/config.yaml               # Set project name

# .aiscrum/config.yaml — Zod-validated at startup
project:
  name: "my-project"
  base_branch: "main"

sprint:
  prefix: "Sprint"
  max_issues: 8
  enable_challenger: true

copilot:
  max_parallel_sessions: 4
  phases:
    planner:  { model: "claude-opus-4.6" }
    worker:   { model: "claude-sonnet-4.5" }
    reviewer: { model: "claude-opus-4.6" }

quality_gates:
  require_tests: true
  require_lint: true
  require_types: true
  require_build: true
  max_diff_lines: 300

git:
  auto_merge: true
  squash_merge: true

Architecture

┌──────────────────────────────────────────────────┐
│                 Web Dashboard                     │
│  Sprint Board · Backlog · Chat · Sessions · Logs  │
└────────┬───────────────────────────┬──────────────┘
         │ WebSocket                 │ REST API
┌────────┴───────────────────────────┴──────────────┐
│              Dashboard Server                      │
│  Event Bridge · Issue Cache · Chat Manager          │
└────────┬───────────────────────────────────────────┘
         │ SprintEventBus
┌────────┴───────────────────────────────────────────┐
│              AiScrum Pro — Sprint Engine            │
│  init → refine → plan → execute → review → retro   │
├─────────────┬──────────────┬───────────────────────┤
│ Ceremonies  │ Enforcement  │ Infrastructure         │
│ · Planning  │ · Quality    │ · ACP Client           │
│ · Execution │ · Drift      │ · Git Worktrees        │
│ · Review    │ · Escalation │ · GitHub API (gh CLI)   │
│ · Retro     │ · Challenger │ · Sprint Docs           │
└─────────────┴──────────────┴───────────────────────┘
         │ Agent Client Protocol (ACP)
┌────────┴───────────────────────────────────────────┐
│          GitHub Copilot CLI (copilot --acp)         │
└────────────────────────────────────────────────────┘

The AI-Scrum Framework

AiScrum Pro implements the AI-Scrum Framework — an open-source methodology for human-AI software development.

Core values from the AI-Scrum Manifesto:

• Autonomous execution over constant approval
• Verified evidence over claimed completion
• Sprint discipline over feature chasing
• Continuous process improvement over static workflows

The operating model is simple:

  ┌─────────────┐                    ┌──────────────────────┐
  │ Stakeholder │◀──── Decisions ────│                      │
  │   (Human)   │──── Direction ────▶│  AI Team (AiScrum)   │
  │             │                    │  · Lead Agent (PO+SM)│
  │ Sets goals  │◀── Deliverables ──│  · Worker Agents     │
  │ Reviews work│                    │  · Challenger Agent  │
  │ Has veto    │                    │  · Reviewer Agent    │
  └─────────────┘                    └──────────────────────┘

The AI decides how to implement. The human decides what to build. Strategic decisions always escalate. Scope never drifts without approval.

Read the full framework: trsdn.github.io/ai-scrum

Development

make check         # Lint + types + tests
make fix           # Auto-fix lint + format
make test          # Run all tests (vitest)
make test-quick    # Fast fail (--bail 1)
make coverage      # Tests with coverage report
make build         # Build TypeScript
make security      # Security scan

Git hooks are installed automatically on npm install:

• pre-commit: format check + lint + typecheck (~15s)
• pre-push: full gate including tests + build (~60s)

Documentation

| Document | Description | |----------|-------------| | AI-Scrum Framework | The conceptual foundation — manifesto, operating model, ceremonies | | Overview | Architecture deep-dive and component documentation | | Deployment | Installation, configuration, and production setup | | Contributing | How to contribute — branching, testing, PR process | | Process Constitution | Full development process — ceremonies, DoD, ICE scoring, labels | | Philosophy | Values and principles | | ADRs | Architectural Decision Records | | Examples | Ready-to-copy .aiscrum/ configs for TypeScript, Python, React, Go | | Changelog | Version history |

License

MIT — Built with ❤️ by the AiScrum Pro team (human + AI, working as one).