Software development and architecture guidance for the modern era — where AI is your pair programming partner.
This repository contains practical guidance for software developers and architects, with a particular focus on effective workflows when collaborating with generative AI tools like Claude Code, GitHub Copilot, and similar assistants.
The patterns here emerged from real-world project experience, not theory. They address the actual challenges developers face when integrating AI into their daily workflow.
AI-assisted development is powerful but introduces new challenges:
- Context limitations — AI tools have finite memory and lose context
- Session continuity — Work spans multiple sessions; AI doesn't remember
- Decision traceability — "Why did we do it this way?" gets lost
- Knowledge transfer — Onboarding AI to your codebase takes effort
Traditional development practices don't fully address these issues. This repo provides patterns that do.
Choose the guide that matches your tooling:
| Tool / Environment | Guide | Best For |
|---|---|---|
| Claude Code CLI | Claude Code Workflow Guide | Terminal-first development, automatic hooks, integrated workflow |
| GitHub Copilot + Visual Studio 2026 | Copilot Workflow Guide | Windows development, .NET/C# projects, Azure DevOps integration |
Both workflows provide 100% functional equivalence — same patterns, same results, different tools.
| Guide | Description |
|---|---|
| AWS Serverless Backend Guide | Let AI build your API: Lambda, API Gateway, DynamoDB, custom domains, hardening |
| Architecture Decision Records | When and how to document design decisions with ADRs |
- Context management — Strategies for working within AI context limits
- Session summaries — Maintaining continuity across sessions
- Work documentation — Patterns for preserving decisions and progress
- Project organization — Folder structures that work with AI tools
- Hooks and automation — Reducing manual overhead
# Clone and start
git clone https://github.com/SangeetAgarwal/devarch.git
cd devarch
# Read the guide
cat docs/claude-code-workflow-guide.md
# Set up hooks (optional)
# See guide for .claude/settings.json configuration# Clone repository
git clone https://github.com/SangeetAgarwal/devarch.git
cd devarch
# Create a new feature
.\scripts\New-Feature.ps1 -FeatureName "your-feature"
# Read the comprehensive guide
code docs\copilot-workflow-guide.md
# See all available scripts
Get-ChildItem .\scripts\*.ps1Prerequisites:
- PowerShell 5.1+ (pre-installed on Windows)
- Git
- Visual Studio 2026 with GitHub Copilot extensions
- Claude Opus 4.5 model (select in Copilot settings)
Once you've reviewed this repository, you can apply the GitHub Copilot workflow to any of your projects:
# 1. Create your project repository
mkdir my-project && cd my-project
git init
# 2. Copy the workflow files from devarch
# Option A: Manual copy
# Copy these to your project:
# - scripts/ (all PowerShell scripts)
# - .github/copilot-instructions.md
# - .copilotignore
# - docs/context/.session-template.md
# Option B: Using git (recommended)
git remote add devarch https://github.com/SangeetAgarwal/devarch.git
git fetch devarch
git checkout devarch/copilot/adapt-cli-workflow-for-copilot -- scripts/ .github/copilot-instructions.md .copilotignore docs/context/.session-template.md
git remote remove devarch
# 3. Customize for your project
# Edit .github/copilot-instructions.md:
# - Update "Overview" section with your project description
# - Update "Current Work" section
# - Update "Project Structure" to match your codebase
# Edit .copilotignore:
# - Add your project-specific build outputs
# - Add framework-specific folders to ignore
# 4. Create required directories
mkdir -p docs/context docs/architecture/adrs docs/work
# 5. Start your first feature
.\scripts\New-Feature.ps1 -FeatureName "initial-setup"
# This creates the branch, work folder, and initial documentation
# 6. Begin working with the workflow
# The generated prompt is already in your clipboard - paste into Copilot Chat# 1. Navigate to your existing repository
cd path/to/your-existing-repo
# 2. Copy the workflow files (same as above - Option A or B)
git remote add devarch https://github.com/SangeetAgarwal/devarch.git
git fetch devarch
git checkout devarch/copilot/adapt-cli-workflow-for-copilot -- scripts/ .github/copilot-instructions.md .copilotignore docs/context/.session-template.md
git remote remove devarch
# 3. Customize for your existing project
# Edit .github/copilot-instructions.md with your:
# - Project overview and architecture
# - Current conventions and patterns
# - Existing folder structure
# - Build/test commands
# Edit .copilotignore for your build artifacts
# 4. Create documentation structure (if not exists)
mkdir -p docs/context docs/architecture/adrs docs/work
# 5. Start tracking your current work
.\scripts\New-Session.ps1
# Paste the generated prompt into Copilot Chat to beginAfter copying the workflow files, customize these for your specific project:
-
.github/copilot-instructions.md(IMPORTANT)- Replace the devarch-specific content with your project details
- Update the "Overview", "Current Work", and "Project Structure" sections
- Add your project's specific conventions, patterns, and architecture decisions
- Update build/test commands
-
.copilotignore- Add your project-specific build directories (e.g.,
target/for Java,build/for Gradle) - Add framework-specific folders (e.g.,
venv/for Python,vendor/for PHP) - Keep the common entries already present
- Add your project-specific build directories (e.g.,
-
docs/context/.session-template.md(optional)- Modify the template structure if you need different sections
- Most projects can use it as-is
The workflow creates this structure in your project:
your-project/
├── .github/
│ └── copilot-instructions.md # Project instructions for Copilot
├── .copilotignore # Context optimization
├── scripts/ # 8 PowerShell automation scripts
│ ├── New-Session.ps1
│ ├── Resume-Session.ps1
│ ├── End-Session.ps1
│ ├── New-WorkSummary.ps1
│ ├── New-Feature.ps1
│ ├── New-ADR.ps1
│ ├── Context-Check.ps1
│ ├── Update-ImplementationPlan.ps1
│ └── README.md
├── docs/
│ ├── context/ # Session summaries
│ │ ├── .session-template.md
│ │ └── session-*.md
│ ├── architecture/
│ │ └── adrs/ # Architecture decisions
│ └── work/
│ └── {branch-name}/ # Work folders (match branch names)
│ ├── README.md
│ ├── implementation-plan.md
│ └── context/ # Work summaries
└── [your existing code]
The scripts work in any repository - they just manage documentation and generate Copilot prompts. Your actual code stays in your existing structure.
- Read the Copilot Workflow Guide for detailed usage
- Run
.\scripts\New-Session.ps1to start your first session - Use
.\scripts\Context-Check.ps1regularly to monitor context - Generate work summaries after significant milestones with
.\scripts\New-WorkSummary.ps1
- Documentation is memory — What you write down survives; what stays in AI context doesn't
- Progressive capture — Document as you go, not at the end
- Branch = Work folder — Keep related artifacts together
- Explicit over implicit — AI works better with clear structure and instructions
This repository supports two complementary workflows:
- Automation: Automatic hooks for session management
- Integration: Native CLI commands (
write work summary,/context) - Best for: Terminal-based development, quick automation
- Automation: PowerShell scripts for session management
- Integration: Visual Studio 2026, Azure DevOps pipelines
- Best for: Windows development, .NET ecosystem, enterprise environments
Both workflows share the same core principles:
- Session summaries for context continuity
- Work folders matching branches
- ADRs for architectural decisions
- Implementation plans for progress tracking
- Progressive documentation throughout development
Found a pattern that works well? Have improvements to suggest? Contributions are welcome.
MIT
These patterns evolved from building Sharpee, a parser-based interactive fiction engine, using AI-assisted development.