Analysis
Explore the problem space and validate the idea before committing to a plan.
BMad Method
BMad is an open-source, customizable set of agents and workflows built on agile practice. Clarify the work, approve the plan, build it, and verify the result.
v6.12.0Community snapshot · live from npm and GitHub
Everything on this page - the phases, the agents, the artifacts, the gate - exists to carry one argument: AI is the facilitator and the human is the expert. The spec is the work, agile evolved rather than being abandoned, and the rigour should match the work in front of you.

They are effective at implementation. What they cannot do is notice the decision nobody made: the compliance constraint never written down, the non-goal that was assumed, the interface contract that lived in someone's head.
BMad makes those decisions explicit before implementation starts, and preserves them as durable context for the work that follows.

Enter at Analysis, Plan, or Implement according to how much is known. Verification closes the loop. Rigor scales to the work, not the org chart.

Explore the problem space and validate the idea before committing to a plan.
Define what to build, for whom, and how it holds together: requirements, UX, architecture, epics and stories.
Every path converges on one workflow. Planning artifacts add context; they never select a different path.
Discharge what the spec promised. Review, tests mapped back to requirements, a retrospective that feeds the next plan.
Bring in product, architecture, UX, development, and testing expertise when it helps. Agents hand work off through files, not chat history, and teams can extend the roster.
Business Analyst
Brainstorming, research, and the project context everything downstream inherits.
Product Manager
Requirements, epics and stories, and the readiness gate into implementation.
UX Designer
The design spine pair, so experience decisions are recorded rather than assumed.
System Architect
Architecture decisions with stable numbers that epics inherit as binding constraints.
Senior Software Engineer
Specs, code, tests, review and retrospectives, across both sides of the implementation boundary.
Master Test Architect
Test-to-requirement traceability and risk prioritization. Ships in the Test Architect module.
Every artifact is plain text, version-controlled beside your code, reviewable in a pull request, and inheritable by whoever picks the work up next. There is no BMad-hosted storage of any kind.

Why, Capabilities, Constraints, Non-goals, Success signal. Derived from the plan and never hand-edited, so it cannot drift from what was agreed.
Every decision, constraint, assumption and open question, one line each, in the order it happened. Never edited, never reordered.
Capabilities and architecture decisions carry stable identifiers, so a requirement can be followed forward into a diff and back again. Orphans are flagged in both directions.
BMad installs markdown and configuration into your repository, then your approved coding tool executes it. No BMad runtime ever sees your code, prompts, or data.
No telemetry, no analytics, no account, no runtime service. Outbound calls happen at install time only: the npm registry for a version lookup, the GitHub API to resolve release tags, and git against the remote you named if you install an external module.
Artifacts, configuration and skills are all version-controlled alongside your code. Nothing is stored anywhere else, because there is nowhere else.
Before any sprint tracking exists, one adversarial question: could a developer implement these epics without inventing decisions nothing records? Gaps are listed with where each one lives.
Every skill is a markdown file in your repository. What an agent is going to do can be read, diffed and approved in a pull request before anyone runs it.
Install only the modules you need. Team and personal overrides live outside installed skills, survive updates, and avoid forks.

Core and BMad Method ship with the installer. Test Architect, Builder, Creative Intelligence and Loop are separate repositories, resolved at install time and pinnable to an explicit version for CI.
Personal preferences beat team settings beat the defaults a skill ships with. Personal files are gitignored; team files are committed and inherited by every engineer who pulls the repo.
Required fields on an artifact, your own PRD template pinned org-wide, an internal compliance reference the workflow consults before drafting. Injected as persistent context at the step where it matters, rather than left to whoever remembers.
You can add and replace, but there is no removal mechanism. That stops a local override from quietly stripping a step somebody else relies on.
Get started
Node.js 20.12+, uv, and Git for external modules. Nothing to procure, nothing to provision, nothing to sign.
$ npx bmad-method install