What it is
BMAD (Breakthrough Method for Agile AI-Driven Development) runs an agile team as agents. Product, architecture, UX, development and testing roles produce briefs, PRDs, architecture documents and stories, and the Build skill implements from them. Since v6.12, Build decides how much process a change needs after investigating it, so a small fix gets a two-section spec rather than the full sequence.
The project ships on two lines. Stable v6.12 installs with npx bmad-method install and is frozen apart from bug fixes. V7 is a preview alpha on the main branch, installed as skills or as Claude Code and Codex plugins; the README gave 16 October 2026 as its release target, and no v7 release had been published by 23 October. Official modules cover test architecture, game development, creative work, skill building and an unattended epic runner called BMad Loop.
Our review
BMAD is the heaviest spec-driven method in this category, and deliberately so: it produces the brief, PRD, architecture and stories a product owner can approve before an agent writes code. The repo has about 54,500 stars. Stable v6.12.4 shipped on 14 October 2026, while v7 remains a preview alpha that missed the 16 October target its own README set.
The last two minor releases answered the loudest complaint. v6.11 cut the core skills from fourteen to eight and made Build the one official way BMAD implements code. v6.12 lets Build size the spec after looking at the change, and review triage now logs a verdict and evidence for every finding, so nothing is dropped silently.
Speed still counts against it. Reenbit reports one informal run in which the same CRM dashboard took 5.5 hours with BMAD, 90 minutes with Spec Kit and 12 minutes with OpenSpec. That was a single run on an older version, but it matches other reviewers. Churn is the other cost: v6.12 alone renamed a review variable, emptied the auto-loaded project facts and made legacy shims opt-in, so tutorials and custom overrides go stale quickly.
Windows deserves a specific warning. Many of the open bug reports concern Windows: setup failing on read-only files or UTF-8 byte-order marks, and one in which migrate_v6.py --clean deletes customisations inside a shared folder linked by a directory junction. On a team that shares a _bmad folder, a backup before migrating is cheap insurance.
The code is MIT, but BMad and BMAD-METHOD are trademarks of BMad Code, LLC, so a company forking it for internal use has to rename it. Against Spec Kit, the difference is mostly about the audience: Spec Kit writes for the engineers reviewing a pull request, BMAD also writes for the product owner upstream of them.
Where it earns its place
A product owner and four engineers starting a greenfield B2B app
Get a signed-off PRD and architecture before agents write code
Run the v6.12 planning skills, review the PRD and architecture as pull requests, then let bmad-build work through the stories. The software is free; the cost is agent usage and, on Reenbit's informal test, hours rather than minutes per feature.
A consultancy delivering fixed-scope builds for clients
Hand over auditable planning documents with the code
The brief, PRD and architecture files become deliverables, and since v6.11 retrospectives must cite a source artefact for every finding. Pin v6.12 for the length of an engagement, because v7 changes skill names and install paths.
A platform engineer rolling BMAD out to 20 developers on Windows laptops
Install one shared configuration without losing customisations
Install v6.12 per project and keep overrides in _bmad/custom/config.toml rather than a junctioned shared folder. Open issues report setup and migrate_v6.py mishandling junctions and read-only files on Windows, so back up _bmad before upgrades.
Before you commit
- Setup
- Stable: npx bmad-method install in the project (Node.js 20.12+, Python 3.10+ and uv), then invoke bmad-build or bmad-help in your coding agent. V7 alpha: npx skills add bmad-code-org/BMAD-METHOD --skill '*' or the bmad-plugins marketplace in Claude Code or Codex.
- Integrations
- Installs into Claude Code, Codex and other agents (6.12 added Polytoken, Grok and ZCode). Official modules: Test Architect, Game Dev Studio, Creative Intelligence Suite, BMad Builder and BMad Loop for unattended epics.
- Data and privacy
- Runs locally and writes Markdown and config into _bmad and your repo; your agent sends that content to its own model provider. The README does not mention telemetry and there are no published GitHub security advisories. Code is MIT; the BMad names are trademarks of BMad Code, LLC.
- Cost at scale
- No paid edition; the README says no workflow is paywalled and asks companies for sponsorship instead. The real costs are agent tokens across several planning documents per feature and the reviewer time to read them.
- Limits
- 78 open issues and pull requests, many Windows-specific; v6.11 and v6.12 both renamed or removed skills and variables; v7 changes often and has no release date now that 16 October has passed.
What's good
- Full PRD, architecture and story documents
- Build sizes the spec to the change since v6.12
- Stable line frozen apart from bug fixes
- No paywalled workflows
What's not
- Hours per feature on jobs Spec Kit does in minutes
- Breaking renames in recent minor releases
- v7 alpha missed its October release target
- Open Windows bugs in setup and migration
Best for
Teams on larger projects that want PRDs, architecture and stories generated and reviewed.
Alternatives
BMAD Method: common questions
- Is BMAD Method worth it?
- BMAD earns its paperwork on projects with stakeholders who sign things off, and charges for it in hours on everything smaller. We rate it 3.8 out of 5.
- How much does BMAD Method cost?
- Free and open source (MIT). "BMad" and "BMAD-METHOD" are trademarks of BMad Code, LLC.
- What are the downsides of BMAD Method?
- Hours per feature on jobs Spec Kit does in minutes; breaking renames in recent minor releases; v7 alpha missed its October release target; open Windows bugs in setup and migration.
- Who is BMAD Method best for?
- Teams on larger projects that want PRDs, architecture and stories generated and reviewed.
- How much does BMAD Method cost for a team or in production?
- No paid edition; the README says no workflow is paywalled and asks companies for sponsorship instead. The real costs are agent tokens across several planning documents per feature and the reviewer time to read them.
- Is BMAD Method safe for company data?
- Runs locally and writes Markdown and config into _bmad and your repo; your agent sends that content to its own model provider. The README does not mention telemetry and there are no published GitHub security advisories. Code is MIT; the BMad names are trademarks of BMad Code, LLC.
- What are the best BMAD Method alternatives?
- The closest alternatives we have reviewed are Spec Kit, Superpowers and Everything Claude Code (ECC).
Answers checked Oct 23, 2026.