AI Unified Process Marketplace

AI Unified Process (AIUP) 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

Why AIUP?

AI-assisted development often jumps from a vague prompt directly to code. AIUP 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.

Workflow

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 →

Choose your plugins

Install aiup-core in every project. Add exactly one stack plugin when AIUP supports the project’s implementation stack.

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.

Quick start

Before installing AIUP, 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.

Claude Code

/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.

Any supported agent through Tessl

tessl init --agent agents
tessl install ai-unified-process/aiup-core
tessl install ai-unified-process/aiup-vaadin-jooq

Tessl 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.

Create the first artifacts

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 →

Generated artifacts

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.

Existing applications

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.

Installation

• 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.json and mcp.json;
• portable Agent Skills under skills/;
• a Tessl package manifest;
• evaluation scenarios used during publication.

CI keeps the three plugin manifest formats and their MCP definitions in sync.

Documentation

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.

Repository layout

marketplace/
├── aiup-core/
├── aiup-vaadin-jooq/
├── aiup-angular-jpa/
├── aiup-blazor-dotnet/
├── aiup-nestjs-nextjs/
├── docs/
└── scripts/

Learn more

Visit unifiedprocess.ai for the broader methodology.

License

Licensed under the Apache License 2.0.