The harness that turns AI coding into a system.
CodeMySpec is a requirements-driven development harness for building real software with AI. It runs as a plugin inside your AI coding agent and walks that agent through a full lifecycle for every piece of your app: specification, tests, implementation, review, and QA, in the right order.
Your specifications live as plain markdown and your tests as standard ExUnit, both in your own repository. You bring your own model and API keys. There is no token markup and nothing to lock you in. Building is free.
how it works
One task at a time, in dependency order.
CodeMySpec models your project as a graph of stories, components, and requirements, then hands the agent the single next piece of work whose prerequisites are already met. Every piece runs the same loop, and the work is verified before the agent moves on.
/next starts the next requirement,
/implement loops the whole graph, and
/qa runs QA.
where it sits
It doesn't replace your agent. It runs inside it.
Cursor, Claude Code, and Copilot are the agent that writes the code. CodeMySpec is the harness on top that adds specs, architecture, verification, and QA. The tools it is genuinely an alternative to are the app builders. Lovable builds the demo; CodeMySpec builds the app you can still maintain in two years.
| Category | Tools | What they do | Relationship |
|---|---|---|---|
| Coding agents | Cursor, Claude Code, Copilot | Generate and edit code in your editor | Runs inside them |
| Full app builders | Lovable, Replit, Bolt, v0 | Build an app for someone who won't read the code | The real alternative |
| Spec-only tools | Kiro, GitHub Spec Kit, OpenSpec | Generate a spec and a plan, then hand off | Carries the spec through tests, verification, and QA |
no lock-in
It doesn't sit between you and the model. It sits around it.
The harness is a local server on your machine, and each plugin is a thin adapter that points your agent at it. Your requests to the model are never proxied, metered, or marked up. That is also why supporting a new agent is a small adapter rather than a rewrite.
what holds it together
AI accelerates code generation. Without boundaries, it accelerates entropy at the same rate.
Boundaries the agent can't cross.
CodeMySpec validates the dependency graph for circular references and boundary violations, and generates architecture views the agent reads as context before it writes code.
It knows what a context is.
Contexts, schemas, repositories, LiveViews, GenServers, controllers and more are first-class types, each with its own design rules. The harness knows a context is a boundary and a GenServer is stateful, so it builds idiomatic OTP.
Every line traces to a reason.
A Three Amigos process turns acceptance criteria into named rules, Gherkin scenarios, and executable ExUnit specs. Code traces back to a story and a criterion.
proving it works
"Tests pass" is not the same as "it works."
A QA agent drives your running application in a real browser — per story, per user journey, or across the whole app — and records an audited trail of every attempt. Unit tests miss integration failures, broken flows, and UI regressions that look correct in the code and break in front of a user. Agentic QA catches that class of failure.
what it costs
Building is free. You pay when it's running.
The harness and the local server cost nothing to run with your own AI keys. The hosted Operate and Grow layer — customer email on your own domain, live chat, content publishing, and team collaboration — is $100 per month per user. A done-with-you build engagement starts at $500, quoted. You never pay CodeMySpec for model tokens.
questions
Straight answers.
What is CodeMySpec?
CodeMySpec is a requirements-driven development harness for building real software with AI. It runs as a plugin inside your AI coding agent and walks that agent through a full lifecycle for every piece of your app — specification, tests, implementation, review, and QA, in the right order. Your specs are plain markdown and your tests are standard ExUnit, both in your own repository.
Does CodeMySpec replace Cursor, Claude Code, or Copilot?
No — CodeMySpec runs inside them. Those are the coding agent that writes the code; CodeMySpec is the harness on top that adds specs, architecture enforcement, verification, and QA. The tools CodeMySpec is genuinely an alternative to are the app builders like Lovable and Replit: it is the engineer's version of that promise, where the result is code you can still maintain.
What AI agents does CodeMySpec work with?
CodeMySpec works with Claude Code, OpenAI Codex, and Google Antigravity. Claude Code has the deepest integration and is the flagship; Codex and Antigravity are early. It works across agents because the harness is not the plugin — it is a local server on your machine, and each plugin is a thin adapter that points your agent at it.
Is CodeMySpec only for Phoenix and Elixir?
No. The core lifecycle — the requirement graph, specs, BDD scenarios, and browser QA — works on any stack through your AI agent. Phoenix and Elixir are the deepest integration, where CodeMySpec understands 16 component types and builds idiomatic, OTP-aware code.
Does CodeMySpec lock me in?
No. Your specifications are plain markdown and your tests are standard ExUnit, both in your own repository, and you bring your own model and API keys with no markup on tokens. There is no proprietary spec format, no custom test runner, and no DSL to learn. If you stopped using CodeMySpec tomorrow, you would keep every artifact of value.
How does CodeMySpec catch bugs that tests miss?
A QA agent drives your running application in a real browser — per story, per user journey, or across the whole app — and records an audited trail of every attempt. Passing tests are not the same as working software: unit tests miss integration failures, broken flows, and UI regressions that look correct in the code and break in front of a user.
What does CodeMySpec verify before code ships?
Every time the agent stops, CodeMySpec compiles the code, runs Credo, runs the tests, and runs the Spex specs, and it can block the agent from continuing on failure. Blocking is configurable per check: off, block on any failure, block only on changed files, or never block. The agent cannot quietly leave the build broken or skip the tests.
What does CodeMySpec cost?
Building is free. The harness and the local server cost nothing to run with your own AI keys, and CodeMySpec never marks up model tokens. The hosted Operate and Grow layer is $100 per month per user, and a done-with-you build engagement starts at $500, quoted.
Who is CodeMySpec for?
CodeMySpec is for engineers who ship and maintain the code they build — senior Phoenix and Elixir engineers, Elixir agencies, and solo technical founders shipping a real product. It is not for throwaway prototypes. If you are throwing it away, prompt directly; if you are shipping it, you need the harness.
Try it on one real feature. Building is free.
Install the plugin, run /init to link your project,
then /next to start the next requirement.