tfplan2md is developed using a multi-agent AI workflow where specialized agents handle different phases of the development lifecycle
This project uses an agent-based workflow for feature development, inspired by best practices from GitHub Copilot agents and modern software engineering principles. Each agent is a specialized AI assistant with a clear responsibility in the development process.
The workflow is coordinated by a human Maintainer who manages handoffs between agents and provides clarifications as needed. Agents produce artifacts as markdown files in the repository, creating a traceable development history.
Each agent has a single, well-defined role in the workflow
All decisions and changes are documented in markdown files
Standardized workflow ensures quality and completeness
Leverages GitHub Copilot's multi-model capabilities
The diagram below shows the complete agent workflow from requirements to release. Each agent produces artifacts that are consumed by the next agent in the sequence.
Agents produce and consume artifacts. Solid arrows show artifact creation and consumption. Dashed arrows indicate rework/feedback loops.
Gathers and clarifies requirements for new features
Investigates bugs and technical issues
Designs solutions and documents decisions
Defines test plans and acceptance criteria
Implements features and tests
Updates and maintains documentation
Reviews code quality and standards
Validates user-facing features
Prepares and executes releases
Identifies improvement opportunities
Improves the workflow itself
Maintains the project website
The Maintainer identifies a need (new feature, bug fix, or workflow improvement) and starts with the appropriate entry agent
Each agent produces artifacts (markdown documents) that serve as inputs for the next agent in the workflow
All decisions, requirements, and changes are documented in versioned artifact files in the repository
Code Reviewer and UAT Tester validate changes before Release Manager creates the pull request
Retrospective agent analyzes the process and provides feedback to the Workflow Engineer for improvements
The workflow supports both local (VS Code) and cloud (GitHub) execution modes, enabling both interactive development and automated workflows.
This page provides a high-level overview of the AI workflow. For complete details including artifact formats, handoff criteria, and best practices, see the full documentation.