Collo.dev: Scrum Master Github AI Agent Template

Why AI Needs a Scrum Master

AI coding agents are powerful — they can plan features, write code, and run tests. But without coordination, they become chaotic. Multiple agents working on the same codebase need someone to break down work, assign tasks, track progress, and ensure nothing falls through the cracks.

That's exactly what a human Scrum Master does for a development team. This project replaces that role with an AI Scrum Master that runs entirely on GitHub:

• Breaks down your ideas into discrete, prioritized backlog tickets
• Assigns work to specialized AI agents (planner, developer, tester)
• Tracks progress on a Kanban board that updates automatically
• Keeps humans in the loop at every decision point

The result: you describe what you want to build, and a coordinated team of AI agents builds it — with full visibility and human oversight at every stage.

Why GitHub-Native? (vs. OpenClaw & Generalist Agents)

Recently, general-purpose AI agents (like OpenClaw) have become popular for personal, single-player automation. While they are great for drafting personal emails or fetching data via a chat interface, enterprise teams—across engineering, finance, marketing, and operations—require sophisticated security, auditability, and structured workflows. This template leverages GitHub's enterprise-grade infrastructure to run AI teams securely. Here is how our GitHub-native approach differs from generalist local agents:

Feature	This Template (GitHub-Native Enterprise)	Generalist Agents (e.g., OpenClaw)
Security & Privacy	Enterprise RBAC: Operates under GitHub's strict permission model. The agent only accesses specific, authorized repositories. No shadow IT, and fully auditable by your security team.	Root Access Risks: Often requires broad, un-sandboxed access to a local machine or VPS filesystem, creating significant data privacy and security vulnerabilities.
Architecture	Serverless & Ephemeral: Every task runs in a sterile, isolated cloud container. Zero idle costs, no leaked data between runs, and no corrupted environments.	Stateful & Always-On: Runs continuously on local hardware. Prone to state corruption, dependency conflicts, and accidental data exposure.
Methodology	Structured Agile Workflows: Orchestrates specialized roles (e.g., Project Manager, Analyst, Reviewer, Dev) with distinct prompts, creating natural checks and balances for any complex project.	Solo Generalist: A single "brain" trying to do everything at once via a chat window, often leading to context-loss and hallucinations on large tasks.
Data & UI	Version-Controlled & Auditable: Work is tracked in Kanban boards, Issues, and standard Pull Requests. Whether it is a code diff, a marketing campaign, or a financial report, every change is structured and reviewable.	Chat Interface: Typically relies on messaging apps (WhatsApp, Discord). Extremely difficult for a team to review complex documents, track progress, or audit history in a chat thread.
Reliability	Cloud-Native Scheduling: Powered by GitHub Actions. Scheduled workflows (like weekly reports or daily data syncing) run reliably in the cloud with built-in retry logic and logs.	Local Cron Jobs: Relies on the host laptop or server being awake, authenticated, and connected to the internet to trigger scheduled tasks.

How It Works

Four specialized Claude AI agents collaborate through GitHub Actions, coordinated by the Scrum Master:

You describe a feature → Scrum Master creates tickets → Planner designs it →
Developer builds it → Tester validates it → You review & merge

Agent	Role
Scrum Master	Manages Kanban board, creates tickets, assigns work, tracks status
Planner	Analyzes feature requests, creates detailed implementation plans
Fullstack Dev	Implements features (frontend + backend), writes tests, creates PRs
QA Tester	Reviews PRs, runs lint/tests/build, reports bugs or approves

Work flows through a Kanban board (a GitHub issue) with stages:

Backlog → Planning → Developing → Testing → Human Review → Done

The Scrum Master moves tickets between stages automatically as work progresses. You can see the full status of your project at any time by checking the Kanban board issue.

The Workflow in Detail

1. You comment on the Kanban issue: @scrum-master Create backlogs for user auth and dashboard
2. Scrum Master breaks it into tickets, adds them to the Backlog
3. You approve a ticket: approve User authentication
4. Scrum Master creates a feature issue, triggering the Planner
5. Planner analyzes the request and posts an implementation plan
6. You review the plan and approve it
7. Scrum Master moves the ticket to Developing, Fullstack Dev starts coding
8. Dev pushes a feature branch, opens a PR to develop
9. Scrum Master moves the ticket to Testing, QA Tester reviews the PR
10. Tester runs lint, tests, build — reports bugs or approves
11. You merge the PR, Scrum Master marks the ticket as Done

Your Role as a Human

You stay in control at key checkpoints:

When	What you do
Backlog created	Review tickets, approve <ticket> the ones you want built
Plan created	Review the implementation plan, add approved-plan label
Tests pass	Review the PR, merge to develop
Release ready	Merge develop → main for production

Everything else is automated. You can monitor progress on the Kanban board issue at any time.

Quick Start: Use This Template

Step 1: Fork or clone this repo

git clone https://github.com/YOUR_USERNAME/nextjs-app-sample.git
cd nextjs-app-sample

Or click "Use this template" on GitHub to create your own repo from this template.

Step 2: Add your Claude API token

You can use either of these authentication methods:

Option A: Claude Code OAuth Token (if you are Pro or Max user, generate this by running claude setup-token locally)

1. Go to your GitHub repo → Settings → Secrets and variables → Actions
2. Click New repository secret
3. Name: CLAUDE_CODE_OAUTH_TOKEN
4. Value: Your Claude Code OAuth token
5. Click Add secret

Option B: Anthropic API Key

1. Go to your GitHub repo → Settings → Secrets and variables → Actions
2. Click New repository secret
3. Name: ANTHROPIC_API_KEY
4. Value: Your API key from Anthropic Console
5. Click Add secret
6. Update the workflow files in .github/workflows/ to replace CLAUDE_CODE_OAUTH_TOKEN with anthropic_api_key in the claude-code-action config

Step 3: Install the Claude GitHub App

1. Go to https://github.com/apps/claude
2. Click Install
3. Select your repository (or all repositories)
4. This enables Claude to interact with your issues and PRs directly

Step 4: Set workflow permissions

1. Go to Settings → Actions → General
2. Under Workflow permissions, select Read and write permissions
3. Check Allow GitHub Actions to create and approve pull requests
4. Save

Step 5: Create the required labels

Go to Issues → Labels and create these:

Label	Color suggestion
kanban	blue
feature-request	teal
awaiting-human-review	yellow
needs-revision	light yellow
approved-plan	green
bugs-found	red
tests-passed	green
ready-for-merge	green
needs-clarification	yellow
priority-high	red
priority-medium	yellow
priority-low	green

Step 6: Create your Kanban board

Click the link below to auto-create the Kanban board issue (replace YOUR_ORG/YOUR_REPO with your actual repo):

Create Kanban Board Issue

Or manually create a new issue with title Kanban Board, label kanban, and this body:

## Kanban Board

### Backlog

### Planning

### Developing

### Testing

### Human Review

### Done

Step 7: Customize for your project

1. Edit CLAUDE.md — Update the project overview, tech stack, and coding standards to match your app. This is the main context file that all agents read before doing any work.
2. Edit agent configs in .claude/agents/ if you want to adjust agent behavior, models, or communication patterns.

Step 8: Start building!

Comment on your Kanban board issue to tell the Scrum Master what to build:

@scrum-master Create backlogs for a task management app with user authentication,
dashboard, CRUD for tasks, drag-and-drop reordering, and search/filter

The Scrum Master will break this down into backlog tickets on the board. Then:

1. Approve a ticket: Comment approve <ticket-title> on the Kanban issue
2. Review the plan: The Planner creates a plan on the feature issue — review it, then add approved-plan label and comment @fullstack-dev please implement
3. Watch it build: The Fullstack Dev implements the feature and opens a PR
4. Auto-tested: The QA Tester reviews the PR, runs tests, and reports results
5. Merge: When tests pass, review the PR and merge to develop

The Kanban board updates automatically at every stage.

Stack-Agnostic Design

This template works with any tech stack. The agents read CLAUDE.md to discover your project's:

• Framework and language
• Build, test, and lint commands
• Project structure and conventions
• Dependencies and how to install them

Just edit CLAUDE.md to describe your stack — whether it's Next.js, Django, Rails, Flutter, Go, Rust, or anything else — and the agents will adapt.

Git Workflow

• main — Production (stable, deployed)
• develop — Development (integration branch)
• feature/* — Feature branches (created by AI from develop)

All PRs target develop. Only humans merge develop → main.

Customization

Change the tech stack

Edit CLAUDE.md to describe your actual tech stack. The agents read this file before every task, so keeping it accurate ensures they write code that fits your project.

Adjust agent behavior

Agent configs live in .claude/agents/:

• scrum_master.md — Board management and coordination
• product_manager.md — Feature analysis and planning
• fullstack_dev.md — Code implementation
• tester.md — Code review and testing

Change AI models

Each workflow specifies a model in claude_args. You can change these in the .github/workflows/ files:

• claude-opus-4-6 — Most capable, used for planning and coordination
• claude-sonnet-4-6 — Fast and cost-effective, used for implementation and testing

Add more agents

To add a new agent (e.g., a dedicated API developer):

1. Create .claude/agents/your_agent.md with YAML frontmatter
2. Create a workflow in .github/workflows/ that triggers the agent
3. Update CLAUDE.md to document the new agent's role

Documentation

• CLAUDE.md — Project context, tech stack, and coding standards
• WORKFLOW_SETUP.md — Detailed setup and architecture guide
• COMMUNICATION_PROTOCOL.md — How agents communicate and hand off work

Cost

Each workflow run consumes Claude API tokens. To manage costs:

• Opus is used sparingly (planning, Scrum Master coordination)
• Sonnet handles implementation and testing (lower cost)
• Adjust --max-turns in workflow files to limit conversation length
• Monitor usage at console.anthropic.com

---

Don't want to configure GitHub Actions and labels manually? We're building Collo.dev — a fully hosted GitHub App that installs this entire AI Scrum team in one click with zero setup. Join the waitlist.