BuilderOS

An operating system for builders. BuilderOS is a collection of agent skills that give product builders a repeatable system for ideating, designing, and building software products with AI — turning coding agents from autocomplete into disciplined collaborators that ship reviewed, tested work.

Why

AI coding agents are fast, but speed without discipline produces features that "compile" rather than features that work. BuilderOS encodes the habits of strong product teams — plan-driven work, mandatory review, end-to-end testing, honest reporting — as skills your agent follows automatically.

How it works

Each skill is a SKILL.md file that teaches an AI coding agent a workflow. Install the skill for the agent you use, then trigger it with a natural prompt — the agent picks up the workflow and follows it until the work is genuinely done.

skills/
├── Build Loop - Claude Code/   # for Anthropic's Claude Code
├── Build Loop - Codex/         # for OpenAI's Codex CLI
└── Build Loop - Cursor/        # for Cursor

Skills

Build Loop

A quality-gated feature workflow: nothing ships on "it compiles" — every increment is built → reviewed → tested end to end → fixed before the agent reports "done." Available in three flavors, adapted to each agent's native review tooling:

Skill	Agent	Review mechanism
Build Loop – Claude Code	Claude Code	/review, plus /security-review for sensitive surfaces (auth, payments, user input, data access)
Build Loop – Codex	OpenAI Codex CLI	/review on uncommitted changes, plus a custom security-focused review pass for sensitive surfaces
Build Loop – Cursor	Cursor	Cursor's /review

What the loop does:

1. Build — works the first unchecked task from your plan file (or builds from your prompt, confirming scope and success criteria first). Surgical changes, no speculative scope, matches project conventions.
2. Review — runs the agent's code review and fixes every finding in scope: bugs, security issues, edge cases, performance. Checks UI changes against your design system spec so nothing bypasses design tokens. Re-runs review until clean.
3. Test end to end — runs the full test suite, adds tests for new logic, then exercises the feature as a real user would — including empty, loading, and error states.
4. Fix — anything testing surfaces goes back through the loop. A failing task is never marked complete.
5. Continue — checks the task off the plan and moves to the next, until the requested scope is done.
6. Report — tells you what was built, what review found and fixed, how it was verified, and what needs your attention — honestly, including anything flaky or only partially verified.

Trigger it with: "run the build loop", "build the next task", "continue the plan", or "build this feature properly".

Getting started

1. Clone this repo or copy the skill folder for your agent into your project's skills directory (e.g. .claude/skills/ for Claude Code).
2. Optionally add a plan file to your repo — any markdown task list with - [ ] checkboxes works.
3. Ask your agent to "run the build loop."

Roadmap

BuilderOS covers the full product-building loop. Build is here today; skills for the earlier phases are coming:

• Ideate — finding, validating, and pressure-testing product ideas
• Design — turning ideas into specs, design systems, and buildable plans
• Build — disciplined, quality-gated implementation ✅

Changelog

See CHANGELOG.md.