Auto BMAD

Auto-BMAD runs BMAD 6.6 execution pipelines across Claude Code, Codex, and shared Agent Skills hosts. It does not generate an app from one prompt. It orchestrates existing BMAD skills with checkpoints, reviews, reports, and resumable sprint state.

Fork of stefanoginella/auto-bmad, updated for BMAD-METHOD v6.6.0 with quick/full modes, sprint automation, diagnostics, and flattened agent architecture.

Requirements

Auto-BMAD runs on top of an existing BMAD project. Before installing Auto-BMAD, set up the required BMAD modules from their own documentation:

• BMAD-METHOD v6.6.0 - required for every Auto-BMAD workflow
• TEA v1.15.1 - required only for full BMM mode
• GDS v0.2.2 - required only for game-development workflows

Quick mode only needs BMAD-METHOD. Auto-BMAD assumes BMAD skills and config files already exist in the target project.

Install Auto-BMAD

Claude Code

Fast path:

npx @bramvera/auto-bmad

Or install manually inside Claude Code:

/plugin marketplace add bramvera/claude-code-plugins
/plugin install auto-bmad@bramvera-plugins --scope user
/reload-plugins

Smoke test:

/auto-bmad-check

For unattended sprint runs, start Claude Code with claude --dangerously-skip-permissions in a repo you trust. Without it, Claude will prompt for approval on most actions.

Codex and Shared Agent Skills Hosts

Use this path for Codex, Pi, and other hosts that read project-local .agents/skills:

cd /path/to/your/project
npx @bramvera/auto-bmad init

Smoke tests:

Host	Command
Codex	$auto-bmad-check
Pi / slash-skill hosts	/skill:auto-bmad-check

See Installation for local checkout installs, updates, uninstall commands, and host-specific notes.

First Run

Start with quick mode unless you already know you need the heavier full pipeline.

Step	Claude Code	Codex	Shared Agent Skills
Check install	/auto-bmad-check	$auto-bmad-check	/skill:auto-bmad-check
Create/update plan	/auto-bmad-plan <product description or @file>	$auto-bmad-plan <product description or @file>	/skill:auto-bmad-plan <product description or @file>
Run first quick sprint	/auto-bmad-sprint-quick 1	$auto-bmad-sprint-quick 1	/skill:auto-bmad-sprint-quick 1

Codex and shared Agent Skills installs use the same generated skill names. Only the host prefix changes: $auto-bmad-sprint-quick in Codex, /skill:auto-bmad-sprint-quick in slash-skill hosts.

Sprint Wizard Resume And Reset

The sprint wizard is for selecting multiple epics and optional steps, then resuming from auto-bmad-artifacts/sprint-plan.yaml.

Common Codex forms:

$auto-bmad sprint wizard
$auto-bmad sprint wizard autonomous
$auto-bmad sprint wizard reset

• sprint wizard shows existing plan status, remaining tasks, capability warnings, and asks what to do.
• sprint wizard autonomous resumes the saved plan without routine prompts.
• sprint wizard reset backs up and archives the existing plan, then asks the wizard questions again.
• sprint wizard reset autonomous still shows the rebuilt wizard questions first, then asks whether to proceed autonomously.
• The wizard asks whether to run BMAD steps through fresh workers/subagents or in the current session. Codex only spawns subagents when that option, or an explicit subagent/worker request, is present.
• In the wizard, choose A for recommended runnable epics or type explicit epic numbers. Completed epics are excluded from the runnable plan; hard startovers require an intentional project/git reset outside the wizard.
• Wizard prompts use explicit answers instead of relying on blank input: A for recommended epics, optional step letters such as c for E2E or n for none, 1 for fresh workers/subagents or 2 for current session, and y/n for confirmations.
• Wizard time estimates are rough. For GPT-5.5 planning, use about 60 output tokens/sec as a baseline, then account for tool calls, tests/builds, hidden reasoning, retries, and review fixes.

Reset creates timestamped safety files before rebuilding:

_bmad-output/auto-bmad-artifacts/sprint-plan-backup-before-reset-<datetime>.yaml
_bmad-output/auto-bmad-artifacts/sprint-plan-archived-<datetime>.yaml

Wizard status also reports whether your selected steps match the installed capabilities. Quick mode needs only BMAD-METHOD. E2E requires bmad-qa-generate-e2e-tests. Full/TEA steps require _bmad/tea/config.yaml and matching bmad-testarch-* skills.

Choose a Mode

	Quick Mode	Full Mode
BMAD modules	BMAD-METHOD only	BMAD-METHOD + TEA
Per story	Create, develop, code review	Create, adversarial review, ATDD, develop, edge-case hunt, 3x review, trace, automate
Best for	Prototypes, side projects, familiar domains, token control	Production systems, complex domains, brownfield risk, traceability
Typical sprint	Workload-dependent; often a few hours for a small quick epic, with wall time affected by host/model speed and reasoning effort	Workload-dependent; slower because TEA quality gates add more steps and review/test work

Game-development pipelines use GDS. See GDS Tutorial.

What Auto-BMAD Runs

Auto-BMAD automates BMAD execution after your analysis and planning inputs are ready:

• planning pipelines for BMM and GDS projects
• quick story and sprint execution
• full story and sprint execution with TEA quality gates
• brownfield change specs and change development
• progress files, reports, git checkpoints, rollback, and resume

It intentionally does not automate interactive discovery work such as brainstorming, party-mode debates, product discovery, or human approval checkpoints. See Concepts for the reasoning.

Documentation

• Installation - install commands, smoke tests, updates, local development installs
• Getting Started - new-user flow from check to first sprint
• Concepts - where Auto-BMAD fits, mode tradeoffs, compatible BMAD skills
• Sprints - sprint lifecycle, resume behavior, failure handling, duration
• Commands Reference - every command mapped to the BMAD skills it calls
• BMM Tutorial - Business Model Method walkthrough
• GDS Tutorial - Game Dev Suite walkthrough
• Codex Tutorial - Codex install and skill command syntax
• FAQ - common questions and troubleshooting

Credits

Built on the original auto-bmad by Stefano Ginella, who designed the core pipeline orchestration concept and the BMM/GDS command structure.

The pipelines are powered by the BMAD Method by bmad-code-org.

License

MIT