Skip to content

jleeder/scrumflow

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

12 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ScrumFlow

A GitHub Copilot agentic plugin that orchestrates end-to-end feature development using a pipeline of specialized AI agents, enforcing software engineering best practices with human approval gates at every critical milestone.

What It Does

ScrumFlow turns a raw feature idea into production-ready code by running it through a structured 6-phase pipeline:

Feature Idea
  ↓ Product Owner    → [Gate 1] Approve user story
  ↓ Test Author      → [Gate 2] Approve BDD spec
  ↓ Architect        → [Gate 3] Approve task plan
  ↓ Engineer(s)      → parallel TDD implementation
  ↓ Code Reviewer    → [Gate 4] Approve code review
  ↓ Push + PR

You stay in control. No phase runs without your explicit approval.

Prerequisites

  • VS Code with GitHub Copilot (agent mode enabled)
  • A git repository with at least one commit
  • A supported test framework (Jest, Vitest, Pytest, RSpec, Mocha, etc.)

Installation

Install directly from GitHub with Copilot CLI:

copilot plugin install jleeder/scrumflow

Then verify installation:

copilot plugin list

This plugin manifest loads:

  • Root orchestrator from SKILL.md
  • Reusable tool skills from skills/*/SKILL.md
  • Role-scoped agent skills from agents/*/SKILL.md

Usage

Open GitHub Copilot chat in agent mode and describe what you want to build. ScrumFlow activates on natural triggers:

"implement this feature"
"build this with ScrumFlow"
"start a new feature: <description>"
"run the pipeline on this"
"continue my ScrumFlow session"

From there, Copilot will guide you through each phase, pausing at each gate to show you the artifact and ask for approval before proceeding.

Pipeline Phases

Phase 1 — Product Owner (scrumflow-product-owner)

Converts your raw idea into a scoped user story. Surfaces ambiguity through structured questions before any code is written. Produces a STORY-NNN.md artifact for your review.

Gate 1: Approve or refine the story before BDD spec work begins.

Phase 2 — Test Author (scrumflow-test-author)

Reads the approved story and writes a complete BDD specification in Given/When/Then format. No implementation details — just the behavioral contract.

Gate 2: Approve or refine the BDD spec before architecture work begins.

Phase 3 — Architect (scrumflow-architect)

Reads the approved BDD spec and produces a complete technical plan: task decomposition, dependency mapping, parallel vs. sequential execution order, and rich per-task documentation that Engineer subagents will use.

Gate 3: Approve or refine the plan before implementation begins.

Phase 4 — Engineer(s) (scrumflow-engineer)

One subagent per task, running in parallel in isolated git worktrees (managed in .worktrees/). Each Engineer follows a strict Red → Green → Commit TDD cycle:

  1. Convert BDD scenarios to failing tests (scrumflow-red-test)
  2. Run tests, confirm red state (scrumflow-test-runner)
  3. Write minimum production code to pass (scrumflow-green-code)
  4. Run tests, confirm green state (scrumflow-test-runner)
  5. Craft an atomic commit (scrumflow-commit-crafter)

Phase 5 — Code Reviewer (scrumflow-code-reviewer)

Reviews the full implementation against the approved story, BDD spec, and task-docs. Checks correctness, BDD coverage, security, edge cases, and test quality. Produces a structured REVIEW-NNN.md report.

Gate 4: Approve or request changes before the branch is pushed.

Phase 6 — Push + PR

Merges worktrees, pushes the branch, and optionally generates a comprehensive pull request description (scrumflow-pr-writer). Ticket sync (scrumflow-ticket-sync) can post gate summaries to Jira, Linear, GitHub Issues, or Azure DevOps.

State Directory

All pipeline state is tracked in .scrum-flow/ inside your worktree (never committed):

.scrum-flow/
├── state.json          # Current phase, gate approvals, task status
├── stories/            # STORY-NNN.md
├── tests/              # BDD-NNN.md
├── tasks/              # TASK-NNN-docs.md + task list
├── reviews/            # REVIEW-NNN.md
└── drafts/             # Working files during active phases

Add .scrum-flow/ to your .gitignore.

Agents

Agent Skill ID Purpose
Product Owner scrumflow-product-owner Story definition
Test Author scrumflow-test-author BDD specification
Architect scrumflow-architect Technical planning
Engineer scrumflow-engineer TDD implementation (parallel)
Code Reviewer scrumflow-code-reviewer Code review

Skills (Tools)

Skill Skill ID Purpose
Red Test scrumflow-red-test Convert BDD scenarios to failing tests
Test Runner scrumflow-test-runner Execute test suite, report pass/fail
Green Code scrumflow-green-code Write minimum code to pass tests
Commit Crafter scrumflow-commit-crafter Generate atomic conventional commits
PR Writer scrumflow-pr-writer Generate pull request descriptions
Ticket Sync scrumflow-ticket-sync Sync gate approvals to issue trackers

Design Principles

  • Human-in-the-loop: Agents cannot advance past a gate without explicit approval.
  • TDD by default: Code is always written in response to failing tests.
  • Isolation: Each Engineer runs in its own git worktree in .worktrees/ to avoid conflicts.
  • Rich context transfer: The Architect writes detailed task-docs so cheaper Engineer models have full context without needing to re-reason from scratch.
  • Model-cost optimization: Premium models handle planning and review; cheaper models handle implementation and tooling.
  • Atomic commits: One commit per task, linked to story and task IDs.
  • Non-blocking integrations: Ticket sync failures never stop the pipeline.

About

An agentic framework based on scrum

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors