AI Unified Process is a requirements-first workflow for taking software from a product vision to reviewed specifications, implementation, and traceable tests. This repository distributes the workflow as portable Agent Plugins and Agent Skills for Claude Code, OpenAI Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode, and other compatible coding agents.
Get started · Understand the workflow · Choose a plugin · Installation guides
AI-assisted development often jumps from a vague prompt directly to code. AI Unified Process inserts durable, human-reviewable artifacts between intent and implementation:
- requirements with stable identifiers;
- an explicit domain entity model;
- use cases that define user goals and behavior;
- test journeys that trace back to those use cases;
- stack-specific implementation and tests built from the reviewed specifications.
The workflow is inspired by the phases of the Rational Unified Process, adapted for coding agents and plain-text artifacts that live with the source code.
Inception Elaboration Construction
───────────── ─────────────────────────────────── ─────────────────────────────────
/requirements → /entity-model → /use-case-diagram → /use-case-spec → migration
↘ implementation
↘ tests
aiup-core owns the stack-independent path from vision to specifications. One stack plugin continues from those
specifications into migrations, application code, and tests.
Each step reads the files produced by earlier steps. You can review or edit an artifact before continuing, and every later result remains traceable to the corresponding requirement or use case.
Read about phases, artifacts, traceability, and change handling →
Install aiup-core in every project. Add exactly one stack plugin when AI Unified Process supports the project's implementation
stack.
| Plugin | Stack and responsibility | Construction skills |
|---|---|---|
aiup-core |
Stack-independent analysis and specifications | /requirements, /entity-model, /use-case-diagram, /use-case-spec, /test-case, /reverse-engineer |
aiup-vaadin-jooq |
Vaadin Flow or Hilla with jOOQ | Flyway, implementation, Browserless or Karibu, Playwright, coverage check |
aiup-angular-jpa |
Angular with Spring Boot and JPA | Flyway, implementation, Spring Boot tests, Vitest, Playwright |
aiup-blazor-dotnet |
C# and Blazor on .NET 10 with EF Core | EF migrations, Vertical Slices, bUnit, xUnit, Playwright |
aiup-nestjs-nextjs |
NestJS and Drizzle with Next.js App Router | Drizzle migrations, implementation, Vitest, Supertest, React Testing Library, Playwright |
aiup-vaadin-jooq additionally ships the read-only uc-coverage
sub-agent. It reports which parts of the specification have no code or no test behind them, and which code has no
specification behind it. The /coverage-check skill runs that
audit on demand and judges implementation and tests together; the implementation and testing skills end by handing
off to it rather than running it themselves, so a construction round stays fast.
Use only aiup-core when working with another implementation stack. The methodology ends at a documented boundary,
so the specifications can feed a custom implementation workflow.
Before installing AI Unified Process, create docs/vision.md in the target project. It should describe the mission, target users,
goals, scope, and constraints. A copy-ready vision template is available.
/plugin marketplace add ai-unified-process/marketplace
/plugin install aiup-core
/plugin install aiup-vaadin-jooq
Replace aiup-vaadin-jooq with the selected stack plugin.
tessl init --agent agents
tessl install ai-unified-process/aiup-core
tessl install ai-unified-process/aiup-vaadin-jooqTessl installs versioned skills and maps their MCP servers into the configured agent layout.
Replace agents with a host-specific identifier such as claude-code, cursor, gemini, codex, copilot, or
copilot-vscode when you do not want the vendor-neutral layout.
Run the core skills in the target project:
/requirements
/entity-model
/use-case-diagram
/use-case-spec UC-001
Agents that do not expose skills as slash commands can invoke them by intent, for example: "Create the requirements
catalog from docs/vision.md" or "Specify UC-001".
Follow the complete getting-started guide →
The stack-independent workflow creates a shared documentation contract:
docs/
├── vision.md # maintained by the team
├── requirements.md # /requirements
├── entity_model.md # /entity-model
├── use_cases.puml # /use-case-diagram
├── use_cases/
│ └── UC-001-*.md # /use-case-spec
└── test_cases/
└── TC-001-*.md # /test-case
Stack plugins consume these files and place generated code and tests according to the conventions of the target project. Their READMEs document the supported layouts and exact prerequisites.
Start an undocumented application with:
/reverse-engineer
The skill inspects entry points, domain models, schema, authentication, and integrations, then proposes an entity model, use case diagram, and individual use case specifications. It reports behavior it cannot classify so the team can review gaps before adopting the result as a baseline.
- Claude Code — add the marketplace and install plugins directly.
- Tessl — install pinned plugins for one or more supported agents.
- Agent Plugins and manual setup — direct packages, skill locations, MCP examples, and host-specific caveats.
Every plugin directory contains:
- a Claude Code manifest and MCP configuration;
- an Agent Plugins v1.0.0
plugin.jsonandmcp.json; - portable Agent Skills under
skills/; - sub-agent definitions under
agents/, where the plugin ships one; - a Tessl package manifest;
- evaluation scenarios used during publication.
CI keeps the three plugin manifest formats and their MCP definitions in sync, and verifies that every
skill and document carries its copyright header and that each plugin ships LICENSE and NOTICE.
| Guide | Contents |
|---|---|
| Getting started | Select plugins, prepare a vision, install AI Unified Process, and create the first artifacts |
| Workflow and artifacts | Phases, traceability, artifact ownership, changes, and legacy projects |
| Project setup | Recommended project documentation tree and agent guidance |
| Vision template | Starting point for docs/vision.md |
| CLAUDE.md template | Stack-neutral repository instructions for Claude Code |
| Complete tutorial | Book Library example: stack-neutral analysis followed by Vaadin/jOOQ construction |
Detailed skill behavior is documented in each plugin's skills/*/SKILL.md. Those files are the authoritative source
for inputs, outputs, safety constraints, and execution steps; READMEs provide navigation and concise summaries.
marketplace/
├── aiup-core/
├── aiup-vaadin-jooq/
├── aiup-angular-jpa/
├── aiup-blazor-dotnet/
├── aiup-nestjs-nextjs/
├── docs/
└── scripts/
Visit unifiedprocess.ai for the broader methodology.
Licensed under the Apache License 2.0.
Copyright 2025-2026 Simon Martinelli and the AI Unified Process contributors. The skills and documentation are licensed under Apache 2.0. "AI Unified Process" identifies the original methodology by Simon Martinelli (unifiedprocess.ai). You may fork and modify this work under the license terms, but derived works must keep the NOTICE file and may not present themselves as the official AI Unified Process. If you build on it, please say so and link to this repository.