Agentic Swarm Prompts for TDD Software Development

A collection of specialized system prompts designed to orchestrate a swarm of AI coding agents for robust, test-driven software development.

Primarily designed and tested with the RooCode agentic coding tool, the principles and structure can be adapted for other multi-agent AI development environments.

Overview

Coordinating multiple AI agents to build software effectively requires clear roles, defined responsibilities, and a structured workflow. This project provides a set of fine-tuned system prompts for distinct agent roles, enabling them to collaborate efficiently on software projects.

The core focus is on Test-Driven Development (TDD), ensuring quality from the outset. It also emphasizes security, performance, user experience, rapid feedback via MVPs, and creating code optimized for future AI interaction (Agentic Coding).

Core Philosophy & Principles

These prompts are built upon the following development principles:

1. Test-Driven Development (TDD): The "Red -> Green -> Refactor" cycle is central. Tests define requirements and are written first.
2. Security by Design: Security considerations are integrated from the architecture phase onwards.
3. Performance Awareness: Efficiency is considered during design and implementation.
4. User Experience (UX) Focus: The end-user's needs guide requirements and validation.
5. MVP & Pareto (80/20): Prioritize the core 20% of features delivering 80% of the value for rapid feedback and iteration.
6. Agentic Coding: Code should be modular, clear, well-documented (docstrings, why comments), and token-efficient for maintainability by both humans and AI agents.
7. Concise Dev Log: Maintain essential context (activeDevelopment.md) for seamless handoffs and session resumption.
8. Trunk-Based Development: Utilize short-lived feature branches merging frequently into the main trunk.
9. Minimal & Efficient Artifacts: Maintain a defined set of project documents, avoiding redundancy and optimizing for clarity and token efficiency.

The Agent Roles

The swarm consists of the following specialized agents:

• 👑 Team Lead (Orchestrator): Manages the overall workflow, delegates tasks, monitors progress, and ensures adherence to principles. Initiates foundational setup (like Git repo) early.
• 🏗️ Software Architect: Designs the system architecture, defines the technical plan, and collaborates with UX/Security. Creates systemDesign.md.
• 🎨 UX Designer: Champions the user experience, defines user flows, and validates implementation against UX goals. Contributes primarily to projectOverview.md.
• 🛡️ Security Consultant: Integrates security practices into the design, defines security requirements, and reviews implementation for vulnerabilities. Contributes to systemDesign.md.
• 🧪 Software Engineer in Test (SET): Writes failing tests first (Red phase) based on requirements, defining "done". Maintains the test suite (part of Source Code).
• 💻 Software Engineer: Writes the minimum code to pass tests (Green phase), then refactors code and tests for quality and maintainability. Updates activeDevelopment.md and .aiderrules.
• 🚀 DevOps Engineer: Manages the codebase (Git), CI/CD pipelines, environments, deployments, and generates user-facing CHANGELOG.md release notes. Sets up the repository early and maintains techEnvironment.md.
• 📋 Audit: Undertakes an audit of the project looking at document consistency, compliance with code management plan, and alignment with development standards.

The Workflow

The intended workflow emphasizes TDD and early setup:

mermaid
graph TD
    A[Start Project] --> B{Design & Plan};
    B --> C[Architect Defines Plan];
    %% DevOps setup starts early/in parallel %%
    A --> D(DevOps Sets Up Repo & CI Stub);
    C --> E[Tech Writer Documents Design in systemDesign.md];
    D --> F[Repo Ready];
    E --> G{Write Tests (SET) -> Commit to Repo};
    F --> G;
    G --> H{Write Code (Engineer) -> Commit to Repo};
    H --> I[Update Docs (Writer/Eng) -> Commit to Repo & Merge];
    I --> J{Build & Deploy (DevOps)};
    J --> K[Release Notes (DevOps) -> Update CHANGELOG.md];
    K --> L[Cycle Complete/Next Feature];