Skills and Agents

Use this page to find out what skills and agents a BMad install gives you and how to start each one. How a skill is used in practice lives on its chapter page, linked from the tables below.

What a Skill Is

A skill is a named command the installer places in your AI tool. Type its name — bmad, for example — and the tool loads it. On some platforms the name takes a / or $ prefix.

A skill does one of three things: loads an agent persona, runs a multi-step workflow, or runs a single task.

Skills vs. Agent Menu Triggers

BMad offers two ways to start work.

Mechanism	How you invoke it	What happens
Skill	Type the skill name (e.g. bmad) in your AI tool	Directly loads an agent, runs a workflow, or runs a task
Agent menu trigger	Load an agent first, then type a short code (e.g. BD)	The agent starts the matching workflow while staying in character

Use a skill when you know which workflow you want. Use a trigger when you are already working with an agent and want to switch tasks without leaving the conversation.

Where Skills Live

The installer writes one directory per skill, each holding a SKILL.md, into a directory that depends on your tool:

Tool	Skills directory
Claude Code	.claude/skills/
Cursor, Windsurf, Codex, Auggie, Amp, and most others	.agents/skills/
Cline	.cline/skills/
IBM Bob	.bob/skills/
Antigravity	.agent/skills/
AdaL	.adal/skills/

Some other tools also use their own directory, and a global install goes to a per-user directory instead. The installer output names the exact path for the tool you chose. The directory name is the skill name: bmad-agent-dev/ registers bmad-agent-dev.

The installed directories are the canonical list

Open your skills directory to see every installed skill with its description. Run bmad for guidance on which to use next.

Agents

The BMad Method module installs six named agents. Load one with its skill ID, then type a code from its menu. Codes are scoped to the agent that shows them: CR is a competitive teardown for the Analyst and a code review for the Developer.

Agent	Skill ID	Codes	Menu
Analyst (Mary)	bmad-agent-analyst	BP, MR, DR, TR, TS, CR, UV, CB, WB, PC	Brainstorm; market, domain, and technical research; technology selection; competitive teardown; user-voice research; product brief; PRFAQ challenge; project context
Product Manager (John)	bmad-agent-pm	PRD	Create, update, or validate a PRD
Product Owner (Sofia)	bmad-agent-po	SPEC, TICKETS, REFINE, BOARD, VALIDATE, REPLAN, STORE	Requirements from a PRD, brief, or notes into a ticket; slice an initiative or epic; refine a ticket; the board; validation; replan a mid-flight change; tracker setup
Architect (Winston)	bmad-agent-architect	CA	Architecture spine
Developer (Amelia)	bmad-agent-dev	BD, QA, CR, ER	Build; QA test generation; code review; epic retrospective
UX Designer (Sally)	bmad-agent-ux-designer	CU	UX design

Where is Paige?

The Technical Writer (Paige) is on hiatus. Project context lives on: use the Analyst’s PC code or invoke bmad-project-context directly.

The Developer’s QA code runs bmad-qa-generate-e2e-tests; the full Test Architect is a separate module. See Test Completed Work.

Each agent is an identity plus a customizable layer. See Customize BMad for how that model works and how to change an agent.

Core Skills

Every installation includes the core module: eight skills that work in any project, any module, any phase. No agent session is required.

bmad

The BMad hub. Ask it a question and it answers from the installed modules and recommends the next skill: it reads the active initiative, lists the <type>-<slug>/ folders already written there, and puts the next steps in priority order with their skill commands. It also runs bmad setup and bmad status for the project’s runtime, bmad migrate method to move a v6 project onto the current layout, and shows, switches, creates, or clears the active initiative.

Input: a question in plain language, or one of the commands above. Output: an answer with the prioritized next steps, or the command’s result.

Example

bmad I have a SaaS idea and know all the features. Where do I start?

bmad-advanced-elicitation

A structured second pass over what the model just produced. Instead of “try again,” you pick a named reasoning method and the model re-examines its own output through that lens. A named method forces a particular angle of attack and surfaces what a generic retry misses.

Use it when:

• Output seems fine but you suspect there is more depth
• You want to stress-test assumptions or find weaknesses
• The content is high-stakes and rethinking is worth the time

How it works:

1. It suggests five methods that fit the target (the most recent output, unless you point it elsewhere)
2. You pick one or more, or reshuffle for different options
3. It applies the method and shows the proposed improvements
4. You accept or discard, then repeat or continue

Dozens of methods are available. Examples: pre-mortem analysis, first principles thinking, inversion, red team vs. blue team, Socratic questioning, constraint removal, stakeholder mapping, and analogical reasoning.

The brief, PRD, and UX skills offer it at their own pauses; you can also run it directly on anything recent in the conversation.

Start with a pre-mortem

Pre-mortem analysis is a good first pick for any spec or plan. It consistently finds gaps that a standard review misses.

bmad-review

Reviews a diff, document, or other artifact through one or more lenses and reports every finding in one shape. Zero findings is a valid outcome; it never pads to look thorough.

Lens	Applies to	Method
Adversarial	Anything	Forced-finding review that looks for what is missing, not only what is wrong
Edge case	Anything	Walks every branching path and boundary condition in content that defines behavior
Verification gap	Code	Finds changed behavior that could regress without reliable verification catching it
Structure	Documents	Proposes cuts, merges, moves, and condensing
Prose	Documents	Copy-edits for issues that impede comprehension; runs on top of the structure findings

The two editorial lenses hold content sacrosanct: they critique only how a document is organized and expressed, and they propose changes rather than making them. The lens set is not fixed; a customize.toml override can add lenses or replace shipped ones.

Input: the content (a diff, branch, uncommitted changes, file, or document) and optionally the lenses to run. Output: findings grouped by lens, as JSON, markdown, or both.

Who runs it

bmad-retrospective runs the code lenses over a completed epic’s diff. The product brief, PRD, UX, and architecture skills run the editorial lenses at their finalize step. bmad-build and bmad-code-review use their own reviewer layers; see Review a Change.

bmad-customize

Writes and verifies customization overrides for installed skills, so you can change an agent’s or workflow’s behavior without hand-authoring TOML. Describe the change in plain language; it selects the right scope, writes the override under _bmad/custom/, and verifies the merged result. See Customize BMad.

bmad-brainstorming

Facilitates a brainstorming session using proven creative techniques, guiding you toward 100 or more ideas before organizing them. It shifts creative domain periodically to prevent clustering.

Input: a topic or problem statement, plus optional context. Output: a self-contained brainstorm.html keepsake and an optional brainstorm-<topic>.md for downstream skills. See Explore and Validate an Idea.

bmad-deep-recon

Researches a topic to support a decision, three ways: drafts a research prompt for the tool you already use, turns a finished report into a cited summary other skills consume, or runs the research here with parallel web searches. Built-in types cover market, domain, technical, competitive, user-voice, and academic literature research, plus choosing between candidates. It also answers questions about your own repository from its code and git history: a question one file answers gets a cited answer in chat, and anything larger a document that cites a line or commit for every claim and that bmad-build reads as evidence.

Output: a cited research-<topic>.md and an optional HTML briefing; codebase research also writes each decision record you approve. See Research a Decision.

bmad-forge-idea

Pressure-tests a half-formed idea in a questioning conversation, one question at a time, with different personas probing its weak points, until you can act on it or drop it with confidence.

Output: a forge-report.html keepsake every run, plus a forge-<slug>.md brief when the idea hardens. See Explore and Validate an Idea.

bmad-party-mode

Puts your installed agents, or custom personas, in one conversation, in character, with you steering. A mode setting decides whether one model voices everyone or separate agents think independently, and you can save custom parties for reuse. See Run Multi-Agent Discussions.

BMad Method Skills

The BMad Method module adds the six agents above and these workflow skills. The linked page covers when to use the skill and what it produces.

Skill	Purpose	See
bmad-product-brief	Create, update, or validate a product brief	Define Requirements and a Specification
bmad-prfaq	Stress-test a product concept with the Working Backwards PRFAQ method	Define Requirements and a Specification
bmad-prd	Create, update, or validate a PRD	Define Requirements and a Specification
bmad-ux	Capture the UX vision as DESIGN.md and EXPERIENCE.md	Design UX and Architecture
bmad-architecture	Record the architecture decisions that keep separately built parts consistent	Design UX and Architecture
bmad-project-context	Set up, refresh, or audit the repository’s agent instructions	Set and Maintain Project Context
bmad-build	Turn a work item into working code, reviewed and verified	Build a Change
bmad-build-auto	Run one iteration of an unattended development loop	Autonomous Development Loops
bmad-code-review	Review code changes with several independent reviewers, then triage the findings	Review a Change
bmad-walkthrough	Guide a human review of a commit, PR, file, or directory, one block at a time	Walk Through a Change
bmad-qa-generate-e2e-tests	Generate automated API and end-to-end tests for implemented features	Test Completed Work
bmad-retrospective	Review a completed epic against its evidence and decide whether to accept it	Finish an Epic

Deprecated Names

Earlier skill IDs, such as bmad-create-prd, bmad-edit-prd, bmad-market-research, bmad-generate-project-context, and bmad-checkpoint-preview, still resolve as forwarders to the current skill. Use the current names in new work.

Naming and Modules

Every skill uses the bmad- prefix followed by a descriptive name: bmad-agent-dev, bmad-prd, bmad-build. Modules add their own skills under the same prefix; see Add Modules.

Troubleshooting

Skills not appearing after install. Some platforms require skills to be enabled in settings. Check your tool’s documentation, then restart it or reload the window.

Expected skills are missing. The skills CLI installs only the skills you named. Run npx skills add bmad-code-org/BMAD-METHOD again with the missing --skill entries, then bmad setup, and check that the skill directories exist.

Skills from a removed module still appear. The installer does not delete old skill directories. Remove the stale directories, or delete the whole skills directory and re-run the installer for a clean set.