█████╗  ██████╗ ███████╗███╗   ██╗████████╗██╗ ██████╗    ███████╗██████╗ ██╗      ██████╗
██╔══██╗██╔════╝ ██╔════╝████╗  ██║╚══██╔══╝██║██╔════╝    ██╔════╝██╔══██╗██║     ██╔════╝
███████║██║  ███╗█████╗  ██╔██╗ ██║   ██║   ██║██║         ███████╗██║  ██║██║     ██║
██╔══██║██║   ██║██╔══╝  ██║╚██╗██║   ██║   ██║██║         ╚════██║██║  ██║██║     ██║
██║  ██║╚██████╔╝███████╗██║ ╚████║   ██║   ██║╚██████╗    ███████║██████╔╝███████╗╚██████╗
╚═╝  ╚═╝ ╚═════╝ ╚══════╝╚═╝  ╚═══╝   ╚═╝   ╚═╝ ╚═════╝    ╚══════╝╚═════╝ ╚══════╝ ╚═════╝

A portable, multi-LLM framework that runs the full software lifecycle through a crew of specialized AI agents.

---

Overview

Agentic SDLC turns the seven phases of a Software Product Lifecycle into autonomous AI agents. Point the orchestrator at any codebase — or at an empty folder for greenfield work — and a coordinated crew of ~30 specialized agents handles requirements, design, code, tests, review, release, and operations.

Zero dependencies. One file (sdlc-cli.mjs) drives the entire pipeline across Claude Code, OpenAI Codex, Aider, or your clipboard for any other tool.

The seven phases

PREPARE → DEFINE → DESIGN → DEVELOP → VERIFY → RELEASE → OPERATE
                                       ▲                    │
                                       └────── feedback ────┘

#	Phase	Output
1	Prepare	Codebase analysis, CLAUDE.md cheat sheet
2	Define	Functional and non-functional requirements, user stories
3	Design	Architecture overview, component specs, ADRs (Nygard format)
4	Develop	Per feature: implementation plan → code → tests → review
5	Verify	Coverage, independent review, static analysis, UAT gate
6	Release	Version bump, changelog, release notes
7	Operate	Issue triage, dependency monitoring, incident response

Every non-functional requirement is tagged with a focus track: Quality · Security · Eco-efficiency · AI.

Quick start

git clone https://github.com/orchestratedbyalex/agentic-sdlc.git
cd agentic-sdlc
node sdlc-cli.mjs

The CLI auto-detects whether the folder contains source code (existing project) or is empty (greenfield), persists progress in docs/requirements/sdlc-metadata.yml, and dispatches each phase to the AI tool of your choice. After every phase, simply /exit the AI tool — the menu reappears with the next phase ready to run.

Project structure

.
├── sdlc-cli.mjs        # zero-dependency orchestrator
├── docs/agents/        # seven phase playbooks + bootstrap prompts
├── AUTHORS.md
└── LICENSE

When the framework is applied to a project, it grows the following structure under docs/:

docs/
├── requirements/        # FR-*.md, NFR-*.md, US-*.md, traceability matrix
└── design/              # architecture, ADRs, component specs, implementation plans

Design principles

• Read the rule book first. Every author agent reads CLAUDE.md and existing ADRs before producing anything. Missing information triggers an explicit AMBIGUITIES block — never a guess.
• Reviewer ≠ author. Phase 5's Independent Code Reviewer is always a fresh agent instance, enforcing Microsoft SDL separation of duties.
• Plans are contracts. Each Phase 4 run persists a PLAN-NNN-<slug>.md file as the binding agreement between architect, code author, and test author.
• ADRs are guardrails. Architecture Decision Records (Nygard format) are written once and treated as constraints by every later phase.

Standards anchored

ISO/IEC/IEEE 12207 · IEEE 1012 / 1016 · ISO/IEC 25010 · ISO/IEC 27001 · Microsoft SDL · ITIL 4 · Nygard ADRs · Google SRE / DORA

License & authors

Released under the PolyForm Noncommercial License 1.0.0 — free for research, education, and personal use. Commercial use is prohibited.

Designed and orchestrated by @orchestratedbyalex. See AUTHORS.md for full credits.