Bmm Skills

BMad Method (BMM) — Master router

This is the top-level entry point for the BMad Method framework installed under .claude/skills/bmm-skills/. It does not contain workflow logic. Its job is to route the agent to the correct sub-skill folder based on the user's intent and to expose the four-phase mental model.

The framework is organized in four sequential phases. Each phase produces artifacts the next phase consumes:

1-analysis/        → discovery, research, problem framing
2-plan-workflows/  → PRDs, UX specs (the "what")
3-solutioning/     → architecture, epics & stories (the "how")
4-implementation/  → story execution, review, sprint tracking (the "ship")

Module-level configuration (output folders, user skill level, full agent roster with personas) lives in module.yaml at the root of this folder. Read it once at the start of any BMad session to load shared variables ({output_folder}, {planning_artifacts}, {implementation_artifacts}, {project_knowledge}).

How to use this index

1. Identify the user's intent (use the routing tables below).
2. Open the matching sub-skill's SKILL.md and follow its instructions.
3. Each sub-skill self-resolves its own customize.toml on activation — you don't need to do that here.
4. Most sub-skills consume artifacts from earlier phases. If the user asks for phase 3 work but phase 2 artifacts are missing, surface that gap first.

Don't dump the whole framework into context. Read only the sub-skill that matches the current request, plus any artifacts it needs.

Phase 1 — Analysis (1-analysis/)

Discovery, research, and problem framing. Outputs feed the planning phase.

Typical entry points: new green-field idea → bmad-product-brief or bmad-prfaq; existing codebase the team needs to onboard → bmad-document-project.

Phase 2 — Planning (2-plan-workflows/)

Turn the analysis into a Product Requirements Document and UX specs. Owned by John (PM) and Sally (UX).

Pre-requisites: a product brief or PRFAQ from phase 1 (not strictly required, but the PRD workflow asks for one).

Phase 3 — Solutioning (3-solutioning/)

Translate the PRD + UX into technical architecture, epics and stories, and AI-agent context. Owned by Winston (architect).

Gate to phase 4: bmad-check-implementation-readiness is the last stop before development. If it surfaces gaps, loop back to phase 2 or 3 to close them.

Phase 4 — Implementation (4-implementation/)

Execute approved stories, review, track sprints, retrospect. Owned by Amelia (dev).

Quick-dev vs. full pipeline: bmad-quick-dev is the catch-all for any code change request; it's a hardened single-pass workflow that produces a reviewable artifact. Use it when there is no formal story file. Use bmad-create-story → bmad-dev-story when the user wants the full BMad pipeline (story spec then implementation).

Agent quick map

The framework defines six named personas. Each lives in a bmad-agent-* sub-skill that establishes identity + activation. When the user names an agent, route to the persona skill first; the persona then takes over and may invoke phase workflows.

Full persona descriptions and module configuration (output folders, language settings, user skill level prompts) are in module.yaml.

Conventions used by every sub-skill

All BMad sub-skills follow the same path conventions. Note these once and they apply everywhere:

• Bare paths (e.g. references/guide.md) resolve from the sub-skill's root.
• {skill-root} resolves to the sub-skill's installed directory (where customize.toml lives).
• {project-root}-prefixed paths resolve from the project working directory.
• {skill-name} resolves to the sub-skill directory's basename.

On activation, every sub-skill resolves its customize.toml (with optional team and user overrides under {project-root}/_bmad/custom/). If the resolver script is unavailable, the sub-skill's SKILL.md explains how to merge the layers manually (base → team → user, scalars override, tables deep-merge, arrays of tables key-merge, other arrays append).

Routing rules (when in doubt)

1. Agent name in the prompt? → load that agent's persona skill, full stop. The persona handles the rest.
2. Action verb without an agent? → match the verb to the trigger tables above.
3. "BMad" / "BMM" / "the framework" without specifics? → ask one clarifying question: which phase or which artifact (brief, PRD, architecture, story, sprint)?
4. Code change without ceremony? → 4-implementation/bmad-quick-dev/SKILL.md — it's the framework's catch-all.
5. Multiple matches? → prefer the deepest, most specific phase. If user says "create a PRD and validate it", read bmad-create-prd first; the user can chain bmad-validate-prd after.

Don't preload more than one sub-skill at a time. The framework's progressive disclosure depends on it.