A framework for structured software development using multiple AI agents. This repo codifies a repeatable SDLC where each agent has a defined role, clear inputs/outputs, and mandatory quality gates before deployment.
git clone <this-repo> && cd ai-sdlc-framework
chmod +x bootstrap.sh
./bootstrap.shThe script asks for your project name and a one-sentence description, then prints a ready-to-paste prompt for Claude.ai that onboards you into the framework with your project context baked in.
- Architect/Implementer split: Planning and analysis happen in a conversational AI (e.g. Claude.ai). Implementation happens in an agentic coding tool (e.g. Claude Code, Cursor, Copilot Workspace).
- Named agents, defined roles: Each agent has a specific job. No agent does everything.
- Mandatory gates: Code cannot be deployed without passing CHECKIN. CHECKIN requires both a Code Review pass and a Security pass.
- Templates over improvisation: Prompts are structured and repeatable, not ad-hoc.
- Copy this repo.
- Fill in all {{CUSTOMIZE}} sections with your project specifics.
- Document your domain rules in
agents/code-review.md. - Use the templates in
templates/as your starting point for every implementation prompt you send to your Implementer agent.
| Agent | Tool | Role |
|---|---|---|
| Architect | Conversational AI | Design, analysis, prompt crafting |
| Orchestrator | Conversational AI | Routing, gate enforcement |
| Implementer | Agentic coder | File writes, test execution |
| Code Review | Agentic coder | R1 Skeptic, R2 Tester, R3 Domain Expert |
| Security | Agentic coder | 4-lens security review |
| Monitor | Agentic coder | Post-deploy validation and telemetry |