The required base pack of BMAD+: four agents from idea to reviewed delivery, six shared skills, install command, eight copyable prompts, outputs, limits.
Core is the base pack of BMAD+ and the only one you cannot leave out: every installation contains it. It gives you four agents that cover the path from an idea to a reviewed delivery, six skills that carry the shared protocols, and one data file for automatic role switching. An agent is a persona with written instructions and a menu of capabilities that your AI coding tool loads; you call it by name in the chat. A skill is a reusable protocol, such as the autopilot pipeline, that an agent opens when it needs it.
The menu codes are shortcuts you can type in the chat after calling the agent, for example Atlas PR for the PRD, Forge CA for the architecture, Sentinel RG for the acceptance gate, Nexus AP for autopilot. They are chat shortcuts, not terminal commands. Plain sentences work as well: the agents accept a code, a menu line or a natural request.
The six Core skills and the data file
Name
What it does
Opened by
bmad-plus-autopilot
Runs the Discovery, Build and Ship pipeline with checkpoints; runs the intake of every pack involved first.
Run the installer in your project folder. Core is always included, so you never list it. In the interactive wizard you pick the AI tools and optional packs.
In your project terminal
npx bmad-plus install
02
Or install without questions
For scripts and CI, name the packs and the tool yourself and add --yes. With --yes and no detected tool, no adapter file is written, so pass --tools.
Restart or open the AI tool in this project so it reads AGENTS.md and its adapter file. Then ask for the roster; bmad-help is a message to the assistant, not a shell command.
In your project terminal
bmad-help
04
Give one agent a concrete task
Start with Atlas for a new idea, with Forge for a change in existing code. Atlas shows its menu first and waits; if only the menu comes back, answer with the idea or a menu code.
In your project terminal
Atlas, help me define a small invoicing app for freelancers. Ask about the users, identify the first useful workflow and write a brief with acceptance criteria.
Use it first, before any code exists. Atlas works as Product Manager here. Atlas stops at activation to show its menu: if that is all you get, answer with the idea or the code CB.
In your AI assistant
Atlas, help me define a small invoicing app for freelancers. Ask about the users, identify the first useful workflow and write a brief with acceptance criteria.
What you should get Atlas asks its questions, then writes the brief (capability CB) in _bmad-output/discovery/; the file name carries the project name, product-brief-<project>.md. Review it before asking Forge to build anything.
Use it to test whether a market is worth entering. The words analyze and market trigger the Analyst role by themselves.
In your AI assistant
Atlas, analyze the market for AI note-taking apps. Compare the main alternatives, name the customer needs they leave open and state which claims you could not source.
What you should get Atlas shows its menu first, because it waits at activation; send the request again if that is all it does. It then announces the switch ("I’m switching to Analyst mode", with the reason, and "Say skip to stay in current mode"), then runs the market research capability (code MR): competitive landscape and customer needs. Whether live sources can be consulted depends on your host tool.
Use it after the brief or PRD is reviewed. Forge works as Architect.
In your AI assistant
Forge, propose an architecture for the app described in the product brief in _bmad-output/discovery/. Record each technical decision with its trade-offs and stop for my review before any code.
What you should get A guided architecture document (capability CA). The documented deliverable is _bmad-output/discovery/architecture.md. In autopilot, the architecture checkpoint waits for your approval here.
Use it for a bounded change in existing code. quick dev is the QD capability in Forge’s menu; the words quick, hotfix or simple are triggers of Nexus’s Quick Flow role, which runs the same capability.
In your AI assistant
Forge, quick dev: add a loading spinner to the invoice list. Keep the change small, run the checks that cover it and list what you did not run.
What you should get The change is made in the project’s own style (capability QD). Forge’s Developer role maps each acceptance criterion to a check it actually ran, reports failed, skipped and unrun checks separately, and does not reset, stage or commit unless you ask.
Use it on a finished change. If the BMAD+ command line is installed, the review leaves sealed files; otherwise it is a plain review.
In your AI assistant
Sentinel, review this change against the acceptance criteria. If the BMAD+ CLI is available, seal the scope with bmad-plus review scope, quote the code of every finding, account for every selected file, then report what bmad-plus review gate answers.
What you should get Files under _bmad-output/review/<id>/ (scope.json, checklist.md, then findings.json and coverage.json). The gate answers incomplete, findings or clean; Sentinel reports that word without upgrading it, and a stopped review is never called clean.
Use it when a version reaches your test environment and someone must check it on screen. The words recette, UAT or "what should I check" switch Sentinel to its Recette role.
In your AI assistant
Sentinel, write the recette (UAT acceptance recipe) for this delivery on our test environment. One step per gesture, one verifiable fact per line, on-screen labels quoted from the code, and mark the steps that write for real. Build the page, then tell me the link, how long it takes and what I must not skip.
What you should get A recipe spec under _bmad-output/uat/specs/, a self-contained page under _bmad-output/uat/pages/, then the link, the duration and the steps that write. Later, Sentinel reads the run, classifies each failure (product, recipe, data or undecided) and computes the gate; an undecided failure keeps the gate red.
Use it when the users and the problem are described. Nexus acts as Scrum Master and delegates.
In your AI assistant
Nexus, plan the first usable release of my appointment-booking app. Have Atlas define scope, Forge propose the smallest implementation and Sentinel define acceptance checks. List dependencies and stop at the plan review before implementation.
What you should get A prioritised plan with a small first story and explicit review points. Files come when you ask for the next steps: ES writes epics.md in _bmad-output/discovery/, SP the sprint plan sprint-status.yaml and CS one file per story in _bmad-output/build/.
Use it for a whole new project when you accept to review at each checkpoint. Say the word autopilot, then give the idea.
In your AI assistant
autopilot. The project: a billing SaaS for small businesses with quote management. Stop at every checkpoint and tell me what you need me to approve.
What you should get With the default settings Nexus meets four checkpoints. It runs Discovery (brainstorming, brief, PRD, UX spec) and waits at the PRD and UX spec (discovery); Build (architecture, then epics, sprint plan and each story implemented and tested) and waits at the architecture (architecture), then reports each finished story without stopping (story is notify_only in autopilot mode, require_approval otherwise); Ship (code review, UX review, documentation, acceptance recipe, retrospective) and waits at the final delivery (delivery). The checkpoints in _bmad/config.yaml change which stops block.
The installer writes _bmad/config.yaml. These keys decide how much Nexus asks you. require_approval stops and waits for you, notify_only reports and continues, auto continues and logs. The full file is described on the configuration page.
Excerpt of _bmad/config.yaml (the comments are added here)
Prerequisite: Node.js 20 or newer for the installer. Core needs no Python, no API key and no extra tool. Your AI coding tool supplies the model, the permissions and any parallel agents.
BMAD+ has no telemetry. The Core commands that can send data off your machine all need your action: update-check (opt-out through update-policy --mode off), update --latest (opt-in), nexus launch and nexus verify (they run the commands you declared in a plan) and the acceptance page of uat (served on 127.0.0.1 only unless you publish it yourself).
Installing Core writes instruction files. Nothing runs in the background and BMAD+ includes no model. Parallel agents, browsing and command execution exist only if your host tool provides them; when it does not, Nexus runs the tasks one after the other and says so.
Warning· No pack ships a pipeline yet
The autopilot skill can replace its generic phases with a pipeline that a pack declares, but in 0.23.0 no pack declares one, so autopilot always follows Discovery, Build and Ship. The same holds for swarms: the runtime exists, and no pack ships a swarm definition.
Tip· Core and Dev Studio can coexist
Dev Studio has personas that overlap with Atlas, Forge, Sentinel and Nexus. Install both if you like: choose the persona whose workflow you want. The Design pack goes further: when installed, its rules govern visual design and accessibility, and Sentinel’s UX review hands those over.