AI Development Framework

Reusable multi-agent framework for GitHub Copilot and Copilot coding agents.

Overview

This repository provides reusable planning, implementation, review, and documentation workflow assets that can be used across repositories with different technology stacks.

Features

• Planning Agent
• Implementation Agent
• Code Review Agent
• Documentation Agent
• Repository context template
• Reusable instruction set
• Plan/review artifact conventions
• Optional framework validation workflow
• Optional automated issue development orchestration (Actions controller + watchdog)
• Safe installation script for existing repositories

Repository Structure

• .github/agents/ reusable agents
• .github/instructions/ reusable workflow/security/testing/documentation rules
• .github/plans/ implementation plan artifacts
• .github/reviews/ review artifacts
• .github/scripts/ controller script and unit tests for automated issue orchestration
• .github/workflows/ optional validation, controller, and watchdog workflows
• templates/ repository customization templates
• scripts/install-ai-framework.ps1 installer
• docs/ framework documentation

Installation

1) New repository from template

Mark this repository as a GitHub template, then create a new repository from it.

2) Existing repository installation

./scripts/install-ai-framework.ps1 -RepositoryPath "C:\Git\MyProject"

Optional:

• -IncludeWorkflows
• -Force

Customization

In the consuming repository:

1. Create .github/repository-context.md from templates/repository-context.md.
2. Customize repository Copilot instructions from templates/copilot-instructions.md.
3. Add technology-specific instruction files as needed.

Agent Workflow

Plan → Implement → Review → (Fix → Review)* → Document

Default max fix/review iterations: 3, then require human intervention.

Automated Issue Development Orchestration (Optional)

An opt-in GitHub Actions controller can run this same lifecycle for one GitHub issue at a time. It starts only when the exact ai-development label is present on issue creation or is later added, calls the existing four agents directly through the Copilot Agent Tasks API against one ai/issue-<number>-<slug> branch and draft pull request, and never merges, auto-merges, or approves that pull request — a human remains responsible for that decision. See docs/automated-issue-workflow.md for prerequisites, required label/secret setup, the state model, and troubleshooting.

Plans and Reviews

• Plans: .github/plans/
• Reviews: .github/reviews/

Use predictable names:

• YYYYMMDD-<short-task-slug>.plan.md
• YYYYMMDD-<short-task-slug>.review.md

Automation Support

• Fully supported: Manual agent-driven workflow with plan/review artifacts.
• Partially supported: Generic framework validation workflow; optional automated issue development orchestration (public-preview Agent Tasks API, opt-in per issue).
• Manual: Merging the pull request produced by any workflow; no workflow in this repository merges or approves on an agent's behalf.

Security

• No credentials or secrets should be stored in framework files.
• Use least-privilege workflow permissions.
• Do not encode production access or environment secrets.

Versioning

Semantic versioning (MAJOR.MINOR.PATCH):

• MAJOR: breaking changes
• MINOR: backward-compatible features
• PATCH: backward-compatible fixes

Use Git tags/releases and publish upgrade notes.

Contributing

Contributions should preserve framework/repository-knowledge separation, remain technology-agnostic in core files, and include validation evidence.

License

No license file is currently included. Choose a license intentionally based on intended reuse and governance model before broad distribution.