Claude Code does not need a giant prompt that says 'build my whole startup'. Give it the project context and a workflow that tells it what decision to make next.
What the method is trying to solve
BMAD-METHOD is an agile planning and development workflow designed to give AI coding agents structured roles and artifacts. It contains workflows for project discovery, product planning, architecture, stories and implementation. The key idea is to turn a broad goal into a sequence of explicit decisions and documents the agent can refer to.
That can be helpful when work has several moving parts, unknown requirements or a team handoff. It can also become ceremony. A weekend script may need a short README, a task list and a coding session, not product strategy documents written for an imagined scale. Use the structure to reduce confusion rather than to make a tiny project look enterprise-ready.
Install the current version, not an old video
Start with the official BMAD-METHOD repository and installation guide. The project evolves, so copy the install command and supported client instructions from the current documentation. Older tutorials may refer to package names or slash commands that have changed. Check the release and installation date, and avoid installing unpinned code you found in a random post.
The official docs currently describe BMad v7 as preview alpha and v6.12 as the stable release. From the project folder, run the Skills CLI install shown in the docs, select Claude Code, and include the bmad skill. Then open Claude Code in that project and ask the bmad skill to run bmad setup. Finish with bmad status so you can see which skills and module versions are installed. The exact version matters because preview workflows and command names change.
The install guide lists Node.js, npm, Git, a supported skills-capable coding tool and uv for its setup scripts. Check those prerequisites before installing. The guide notes that v7 is still a preview and changes frequently, while v6.12 is frozen except for bug fixes. A new developer can follow the current v7 guide if they are comfortable with change, or choose the documented stable branch if predictability matters more.
Claude Code supports project instructions and custom skills or commands. Follow Anthropic's current documentation for how your version loads project-level context. Keep project-specific instructions in the repository and review changes the installer makes. Treat a plugin or workflow package like code: inspect where it writes, what tools it can call, and which instructions it adds.
Use one installation path. Do not combine overlapping BMAD versions and Claude workflows without a reason. Duplicate role names and stale instructions can make an agent choose the wrong process.
npx skills add bmad-code-org/BMAD-METHOD
# Select Claude Code and include the bmad skill
# Open Claude Code from the project folder
# Ask the bmad skill: run bmad setup
# Then verify with: bmad statusWrite a one-page project brief first
Before invoking a long workflow, write the problem, user, success condition, existing stack, known constraints and out-of-scope items. Include the command that runs the project and how you currently verify changes. These facts prevent an agent from inventing a product because the prompt left one empty paragraph.
Ask for discovery or planning before code if the problem is still uncertain. Make the first deliverable concrete: a short architecture proposal, a list of questions, or a vertical slice. Review the result and correct the assumptions before starting implementation. Your job is to keep the agent attached to the real goal, not blindly approve every generated document.
A useful instruction names boundaries and a stopping point. For example: 'Plan a local search feature for markdown notes. First inspect the existing app. Propose the smallest version that supports title and body search. List unanswered questions before modifying files.'
Use agile planning to expose risk, not create paperwork
Break work into tasks that end in something observable. A story that can be run, clicked or checked is easier to review than 'build intelligent workspace'. Keep only artifacts that answer a real question: what does a user need, what are we building, and how do we know it works?
Claude can write the task breakdown, but it cannot verify that the customer need is true. BMAD can help an agent follow a process, but a process file does not create good product judgement automatically. Confirm requirements with the person affected, inspect code for project constraints, and keep implementation decisions reviewable.
For a personal project, a brief, a feature checklist, and an architecture note may be enough. For a larger effort, use BMAD's fuller process, but stop generating artifacts when they no longer change what you do next. That keeps your agile workflow agile.
FOLLOW THE EVIDENCE
Sources & further reading
Process note: AI tools assisted with research and drafting. Sources are linked so you can inspect the evidence. Unless an essay says otherwise, it does not claim hands-on testing or independent reproduction.
