Get AOD running on your machine.
Prerequisites
- Node.js (≥ 20 LTS) — Install (opens in new tab). Node 20 LTS or newer. The AOD-Kit scaffold runs on Node; older versions miss the
fetchand ESM features the lifecycle scripts use. - Claude Code CLI or compatible agent (v2.1.16+) — Install (opens in new tab). Claude Code v2.1.16 or later. The slash commands and skills are tuned for Claude Code; other agents work too — the methodology is agent-agnostic.
- Git (≥ 2.40) — Install (opens in new tab). Git 2.40 or newer. The lifecycle uses
ghfor issues and PRs, andghneeds a recent Git underneath to authenticate cleanly.
Five stack packs ship with AOD-Kit today: nextjs-supabase, fastapi-react, fastapi-react-local, knowledge-system, and swiftui-cloudkit. Run /aod.stack list inside your AOD-Kit clone to see them, or browse the stacks directory on GitHub (opens in new tab) for the canonical list. Each pack carries its own conventions, persona loader, and scaffold — pick the one that matches what you're building, or skip the pack and bring your own stack.
Install
git clone https://github.com/davidmatousek/agentic-oriented-development-kit.git
cd agentic-oriented-development-kitYou now have an AOD-Kit clone on disk and your shell is sitting inside it. The next command runs the project scaffold — it asks for project name, GitHub org, AI agent, and stack pack, then writes the scaffolded files into this clone. Run it from this directory.
make initScaffold is written. Pick a stack pack to layer the technology-specific conventions on top — the example below activates the Next.js + Supabase pack. Switch to a different pack any time with /aod.stack use <pack>, or skip stacks entirely if you're bringing your own.
/aod.stack use nextjs-supabaseStack pack active. The repo is now ready to run the lifecycle — five slash commands take a feature idea from capture through delivery in one sitting. The next section walks the full sequence with one sample feature; record your time on the first run, the target is thirty minutes.
Your first lifecycle
/aod.discoverThe agent prompts you for the idea, scores it on impact / confidence / effort, then opens a GitHub Issue with the ICE score in the title. The exit gate is PM validation — the idea has a clear user problem, target user, and measurable outcome. The artifact that leaves the stage is a scored, validated idea on the backlog.
/aod.defineThe agent drafts a PRD from the validated idea, then routes it through the Triad — product-manager, architect, and team-lead each sign off. The exit gate is triple-signed product requirements: what gets built, the technical baseline, the realistic timeline. The artifact that leaves the stage is a PRD the spec phase consumes verbatim.
/aod.planThe agent chains three artifacts in sequence — spec, project plan, tasks — pausing for sign-off between each. The exit gate is triple-signed tasks: product-manager, architect, and team-lead approve before any code is written. The artifact that leaves the stage is an implementation contract bound to wave assignments and agent owners.
/aod.buildThe agent executes the tasks in parallel waves with architect checkpoints at each critical phase — design quality, security scan, code simplification. The architect approves each checkpoint or sends the work back. The exit gate is gate-checked code; the artifact that leaves the stage is implementation that passed every wave's quality bar.
/aod.deliverThe agent verifies the Definition of Done — pushed, tested, user-validated — runs the retrospective, captures learnings as Knowledge Base entries, and marks the draft PR ready for review. The exit gate is the full Triad DoD check. The artifact that leaves the stage is a closed feature with a retrospective on the books.
/aod.documentThe agent runs a quality pass on the merged feature — simplifies what got over-built during the wave, fills missing docstrings, updates the CHANGELOG, refreshes API docs. The exit gate is reviewer sign-off on the cleanup diff. The artifact that leaves the stage is feature-complete code with the documentation surface that future-you will read.
Thirty-minute benchmark — record your time. The target on a clean machine is git clone start to first /aod.deliver completion in thirty minutes. If your first run lands inside the window, the page works; if it overruns, note where the time went and open an issue.