BMAD Starter

A pre-bootstrapped BMAD Method install, checked into git so it can be used in environments where the official installer can't run (e.g. corporate proxies blocking npx / install scripts, air-gapped networks).

Instead of running the installer, you clone this repo and you're ready to go.

Modules & versions

Snapshot last updated 2026-05-22 with BMAD installer v6.7.1 (initial install: 2026-05-14 / v6.6.0).

Module	Version	Source	What it does
core	6.7.1	built-in	Cross-cutting skills: brainstorming, party mode, editorial reviews, adversarial review, doc indexing/sharding, customize, distillator, help
bmm	6.7.1	built-in	BMad Method — the planning/build pipeline and agents (Analyst, PM, UX, Architect, Dev, Tech Writer)
tea	v1.19.0	bmad-method-test-architecture-enterprise (npm: bmad-method-test-architecture-enterprise, sha 8734d51)	Test Architect (Murat) — risk-based test strategy, ATDD, framework scaffolding, NFR/trace/test-design workflows
automator	main (channel next)	bmad-automator (npm: bmad-story-automator, sha 3b01cfd)	Story Automator — drives the create → dev → QA → review → retro loop autonomously. New official module name; the legacy baut/ directory (v1.14.2) is still on disk from the previous install but the v6.7.x installer now ships this module under automator/

Configured IDEs: claude-code, codex, github-copilot, cursor. GitHub Copilot agents are now generated into .github/agents/ (analyst, architect, dev, PM, tech-writer, UX designer, TEA).

What's new since the previous snapshot (v6.6.0 → v6.7.1)

• bmad-prd — new unified PRD skill (create / update / validate intents). Supersedes bmad-create-prd, bmad-edit-prd, and bmad-validate-prd, which are now marked deprecated and slated for removal in v7 (aliases still resolve for now).
• bmad-investigate — new forensic investigation skill: evidence-graded case files for bug hunts, incident triage, or building a mental model of unfamiliar code before working on it.
• automator module — Story Automator pulled from the next channel with a refreshed runtime (new runtime_layout / stop_hooks modules, updated orchestration policy and tmux runtime).
• TEA → v1.19.0 — adds the confidence-gate knowledge resource and refreshes the NFR / test-design / trace workflows.
• Skill catalog is up to 56 BMAD skills (from 54).

Full manifest (install timestamps, SHAs, channels) lives in _bmad/_config/manifest.yaml.

What's in here

.
├── _bmad/                  # BMAD configuration + modules (core, bmm, tea, automator, baut*)
│   ├── config.toml         # project config (installer-managed)
│   ├── config.user.toml    # per-user config (installer-managed)
│   └── custom/             # your durable overrides (never touched by installer)
│       ├── config.toml         # team — committed
│       └── config.user.toml    # personal — gitignored
├── _bmad-output/           # BMAD output: planning / implementation / test artifacts
├── .claude/skills/         # 56 BMAD skills for Claude Code
├── .agents/skills/         # same skills for other agent clients (Codex, Cursor)
├── .github/agents/         # GitHub Copilot agent definitions
└── docs/                   # your project knowledge: PRDs, architecture, etc.

* baut/ is the legacy directory for the Story Automator module — see the modules table above.

How to use

1. Fork / push this repo to your own GitHub (org or personal).
2. Clone it as the starting point for your project:

   git clone <your-repo-url> my-project
   cd my-project

3. Customize the config (see below).
4. (Optional) Wipe history for a fresh start:

   rm -rf .git && git init

5. Start working — in Claude Code, type /bmad-help to see available workflows.

What to change

_bmad/config.toml — project-level

Field	What to set
project_name	your project name (currently "bmad-starter")
document_output_language	language for generated artifacts (PRD, architecture, stories) — currently "English"

_bmad/config.user.toml — per-user

Field	What to set
user_name	your name (currently "Dev")
communication_language	language the agent should talk to you in (currently "Polish")
user_skill_level	beginner / intermediate / advanced

⚠️ Both files carry an "installer-managed, regenerated on every install" header. In this no-installer workflow, editing them directly is fine. If you ever re-run the installer, it will overwrite them — use _bmad/custom/ (below) for changes that must survive a reinstall.

_bmad/custom/ — durable overrides

Use these when you want changes to survive a possible future reinstall:

• _bmad/custom/config.toml — team-wide overrides, committed to the repo. Example: tweak an agent's description:

  [agents.bmad-agent-pm]
  description = "Short, bulleted PRDs over prose."

• _bmad/custom/config.user.toml — your personal overrides, gitignored.

Tables deep-merge in order: base config → team custom → user custom (last wins).

Communication language vs document language

Two separate settings, easy to confuse:

• communication_language (config.user.toml) — what the agent talks to you in.
• document_output_language (config.toml) — what the agent writes artifacts in (PRDs, architecture, stories).

A common setup: chat in your native language, write artifacts in English so the rest of the team can read them.

.gitignore

This repo does not ship with a .gitignore yet. At minimum, before your first commit, add:

_bmad/custom/config.user.toml

The rest depends on the tech stack you drop into the project.

Credit

BMAD Method itself: https://github.com/bmad-code-org/BMAD-METHOD. This repo is just a frozen snapshot of a fresh install — all credit for the method, skills, and agents goes there.