SDLC Agent-Skill Wiki

bmad

7 min read

BMAD-METHOD

Workflow — the primary skill per SDLC stage this framework runs, top to bottom (folded and off-stage steps omitted).

flowchart TB
    A["bmad-analyst<br/><i>Align</i>"] --> S["bmad-pm<br/><i>Specify</i>"] --> P["bmad-architect<br/><i>Plan</i>"] --> I["bmad-dev<br/><i>Implement</i>"] --> R["bmad-code-review<br/><i>Review</i>"] --> L["bmad-retrospective<br/><i>Learn</i>"]

BMAD — “Breakthrough Method for Agile AI Driven Development” (the docs also gloss it “Build More Architect Dreams”) by bmad-code-org — is the wiki’s most lifecycle-complete and most role-oriented framework: an AI-driven agile method built around a cast of named expert personas (Analyst, PM, Architect, Developer…) who facilitate the human through a scale-adaptive pipeline from idea to shipped code. Install: npx bmad-method install (Node.js 20.12+). ~50K★, MIT. Works in Claude Code / Cursor / Copilot, with planning also available as Gemini Gems / ChatGPT GPTs (“web bundles”).

Its stated stance is the sharpest articulation in the wiki of agent-as-collaborator: “traditional AI tools do the thinking for you, producing average results. BMad agents and facilitated workflows act as expert collaborators who guide you through a structured process.” Concretely, its workflows enforce a facilitator rule — “YOU ARE A FACILITATOR, not a content generator; NEVER generate content without user input” — so planning is a grilling dialogue (see pattern-grilling), not autogeneration.

What makes BMAD different

Every other framework here exposes commands or skills; BMAD is the first to organize the whole lifecycle around distinct named personas with persistent character — Mary the analyst, John the PM, Winston the architect, Amelia the engineer — each carrying its own influences and principles. That is its signature contribution (pattern-persona-agents), and it enables Party Mode, where the personas debate a decision as a roundtable. Two more distinctions:

  • Scale-adaptive ceremony. BMAD sizes the process to the work — a one-line fix skips straight to code (Quick Flow), an enterprise system runs the full PRD → architecture → epics → stories pipeline. The planning depth is a dial, not a fixed gate (pattern-scale-adaptive-planning).
  • Context-engineered story files. Each unit of work is a self-contained story file that “embeds architectural decisions, dependency context, and project conventions directly into the implementation artifact,” so the dev agent never loses context (pattern-context-engineering).

The four-phase lifecycle (+ Quick Flow)

“The BMAD Method follows 4 distinct phases in sequence,” plus a parallel Quick Flow track for small work. Each phase’s output “becomes context for the next.” The source tree mirrors the phases (src/bmm-skills/{1-analysis,2-plan-workflows,3-solutioning,4-implementation}/):

PhaseOwning persona(s)Stage
1. Analysis (optional) — explore & validate the ideabmad-analyst · bmad-tech-writerstage-align
2. Planning — define what to build & for whombmad-pm · bmad-ux-designerstage-specify
3. Solutioning — decide how, break into storiesbmad-architect · bmad-pmstage-plan
4. Implementation — build it, one story at a timebmad-devstage-implement · stage-review
Quick Flow (parallel) — skip 1–3 for small workbmad-devstage-implement

Like speckit, BMAD core ships no deploy/ship step; it closes each epic with a retrospective (a learning close-out, see stage-release) rather than a release to production.

Capabilities

Named agents (personas) — one page each

The six BMM personas (“BMad ships six named agents, each anchored to a phase”). Each is a skill (bmad-agent-*) invoked by name; each dispatches a menu of workflow skills:

  • bmad-analyst — Mary 📊; Analysis. Brainstorming, research, product briefs, PRFAQ, brownfield docs.
  • bmad-tech-writer — Paige 📚; Analysis. Project documentation, diagrams, doc validation.
  • bmad-pm — John 📋; Planning. PRD, epics & stories, readiness, course-correction.
  • bmad-ux-designer — Sally 🎨; Planning. UX design specifications.
  • bmad-architect — Winston 🏗️; Solutioning. The architecture spine.
  • bmad-dev — Amelia 💻; Implementation. Story creation, dev, code review, sprints, retros (absorbed the classic SM + QA roles).

Conflict (version drift): classic v4/v5 BMAD shipped separate Scrum Master (Bob), Product Owner (Sarah), BMad Master, and BMad Orchestrator agents. v6 folds these in — SM/QA duties → the Developer (bmad-dev), epic breakdown → the PM (bmad-pm), orchestration → the Party-Mode skill. Some secondary sources (e.g. DeepWiki) still list a sm/”Bob” agent; the v6 docs are canonical at “six named agents.” This wiki follows v6.

Workflow skills — one page each

Grouped by phase; each implements: a canonical stage:

Core utility skills (mechanisms & tooling — catalogued, not paged)

Cross-phase core-skills/ that back the paged capabilities rather than performing a lifecycle stage themselves:

SkillRole
bmad-party-modeSummon multiple personas into one roundtable to debate a tradeoff — the pattern-persona-agents mechanism
bmad-advanced-elicitationPush the LLM to reconsider its output (socratic / first-principles / pre-mortem / red-team)
bmad-review-adversarial-general · bmad-review-edge-case-hunter · bmad-review-verification-gapThe parallel review layers behind bmad-code-review (pattern-adversarial-review, pattern-parallel-persona-review)
bmad-checkpoint-preview · bmad-sprint-statusHuman-in-the-loop review aid; read-only sprint/risk summary
bmad-generate-project-contextEmit project-context.md (AI rules) — the pattern-context-engineering rules file
bmad-qa-generate-e2e-testsGenerate E2E tests for existing features (dispatched by bmad-dev)
bmad-dev-autoUnattended one-iteration variant of bmad-quick-dev for an orchestrator
bmad-shard-docSplit a large PRD/architecture by level-2 headings into focused files (sharding)
bmad-help · bmad-customize · bmad-index-docsRouting, override authoring, doc indexing

Optional modules (expansion packs — catalogued, not paged)

Only BMM (+ shared core) installs by default; v6 “separated core functionality from domain-specific modules for independent versioning.” The rest are optional:

ModuleCodeWhat it adds
BMad BuilderbmbMeta-module: build custom agents / workflows / modules “from a conversation”
Creative Intelligence Suitecis6 ideation personas (Carson, Dr. Quinn, Maya, Victor, Caravaggio, Sophia); SCAMPER, TRIZ
Test ArchitectteaMurat — risk-based testing, ATDD, NFR audits, release gate decisions, requirement-to-test traceability (the quality-review depth BMM core lacks)
Game Dev Studiogds / BMGDGame personas for Unity / Unreal / Godot / Phaser; GDD generation; 21+ game types
Whiteport Design StudiowdsDesign-first UX methodology (Saga, Freya, Mimir)
BMad Loopbmad-loopDeterministic Python unattended dev loop with adversarial review (no personas)

The TEA module’s Murat is BMAD’s answer to a dedicated quality reviewer — it clusters with addy-security-auditor / addy-test-engineer on the stage-review quality-gate side, but ships outside core, so it is catalogued here rather than paged.

Artifacts produced

Patterns applied

See Also

Graph View

Backlinks