bmad-code-org__bmad-method/docs/plan/plan-inside-an-organization.md

title, description, sidebar

title	description	sidebar
Plan Inside an Organization	How the full planning path works when several people must agree on the product, several engineers build it in parallel, and someone has to sign off — who owns which document, where approval happens, and how change flows.	order 2

Use this page when the work belongs to an organization rather than to you: more than one person must agree on what the product is, more than one engineer will build it, or someone must sign off before money is spent. In that setting, planning documents are contracts between people first and input to the skills second.

The Scaled-Down Version First

A single builder, or a small team that already agrees, does not need most of what follows. A forged idea, a PRFAQ summary, a brainstorm intent, or a well-written issue goes straight to bmad-spec, and the rest is one spec per epic. Choose a Planning Path covers that route. bmad-spec will tell you if the input is too thin; until it does, no PRD is required.

Reach for the full path when one of these is true:

• People who did not do the thinking must approve what the product is.
• Several epics, teams, or agents will build against the same decisions and must not diverge.
• A regulator, a steering committee, or an enterprise process requires named documents as evidence.

The PRD Is What the Organization Owns

The PRD is the document the organization owns. It is written by bmad-prd, validated by it, and updated through it; nothing else in the chain claims to say what the product is. Brainstorming, Forge Idea, Deep Recon, a product brief, and a PRFAQ exist to get the PRD written well. Everything after the PRD is derived from it:

• bmad-ux writes DESIGN.md and EXPERIENCE.md alongside the PRD.
• bmad-architecture writes a short architecture document (the spine): the decisions that keep independently built epics compatible.
• bmad-spec writes one spec per epic from the PRD, pointing at the spine and the UX documents rather than copying them.
• bmad-create-epics-and-stories and bmad-sprint-planning turn the specs into tracked stories.

Nothing downstream reinterprets the PRD. If a spec needs an answer the PRD does not give, the answer goes into the PRD first; the spec is re-run after.

Bring the Documents You Have

Nothing here asks you to replace the planning system you already run. An organization arrives with a PRD in Confluence or Notion, a backlog in Jira or Linear, and a review cadence, and all of it stays.

• Your PRD is the input. bmad-prd opens with a brain dump and reads any files you point it at, so the first run is "here is our PRD". Ask it to validate and you get a findings report on the document as it stands, with nothing changed. Ask it to create from that input and you get the same requirements in the shape the later skills read, with [ASSUMPTION] tags on anything it had to fill in. After that, the copy your reviewers already edit is the source. When it changes, re-run bmad-prd in Update mode pointing at it and let the skill bring prd.md in line; never edit prd.md by hand to catch up.
• The same holds for design and architecture. A design system, an existing architecture document, or a live codebase is what bmad-ux and bmad-architecture start from. On an existing system the architecture skill reads the code and records the conventions already there rather than proposing new ones.
• Your tracker stays your tracker. Jira remains where the organization plans, reports, and reviews. The skills read and write one file, sprint-status.yaml, and it holds only story status and open action items. There is no automatic sync in either direction; treat that file as the engineering-side view and update it when the tracker changes.
• Your reviews stay your reviews. The five sign-off moments below are where the skills produce something reviewable. Put your existing approvals at those points and the documents the skills write become the material those meetings already needed.

The one thing that does change is where an edit goes: in the PRD the skills read, not in a spec or a story, so that every derived document can be regenerated from it.

Who Owns What

Role	Runs	Owns
Product manager	Brainstorming, Forge Idea, Deep Recon, then bmad-product-brief or bmad-prfaq, then bmad-prd	prd.md and its update cycle; the one-pager the steering committee reads
Designer	bmad-ux	DESIGN.md, EXPERIENCE.md
Tech lead or architect	bmad-architecture	The architecture spine
One engineer, per epic	bmad-spec, Build per story, bmad-retrospective	That epic: SPEC.md, stories.yaml, its verdict
Whoever tracks the whole	bmad-sprint-planning	sprint-status.yaml, open action items

The rows are roles, not headcount. One person can hold several; what matters is that each document has exactly one owner, because each has exactly one skill that writes it. An epic is a handful of Build sessions, usually a day's work for one person. The organization's coordination lives in the PRD and the spine; the epic itself never needs a committee.

Where Sign-Off Happens

The skills give you five moments where a human decision is expected, and each one blocks something specific. Put approvals there rather than inventing new gates.

Moment	What is judged	What it blocks
PRFAQ verdict	Whether the concept is strong enough to resource	Writing the PRD
PRD validate	A findings report on the PRD without changing it	Design and architecture work
Architecture spine review	The decisions every epic will follow, with alternatives weighed	Writing specs for the epics
Readiness gate	Could a developer implement these stories without inventing decisions	Generating sprint tracking
Retrospective verdict	Did the epic meet its own acceptance criteria	Starting the next epic

Every one of these produces a written result, so the approval has something to attach to. In regulated or enterprise settings those documents are the audit trail. PRFAQ and Retrospective can run unattended (-H) when you want the same check without a conversation.

:::note[Reviewers read copies; change the source] People will review the PRD, the spine, and a spec, and they will ask for changes in whichever one they happen to be reading. Apply the change where it belongs: to the PRD if it is about what the product is, to the spine if it is about how epics stay compatible, to the spec only if it is about that epic alone. Then re-run the later skills. Editing a spec to work around a PRD that no longer says the right thing is how the documents stop agreeing. :::

Several Epics at Once

One PRD, one spine, one spec per epic. Several engineers can each take an epic at the same time when the boundaries are explicit; the spine is what makes that safe, because it records the calls two people would otherwise make differently: the API style, how state is changed, who owns shared data. Each epic's spine inherits the parent's decisions and records only what the parent left open. Run integration checks and a retrospective at every epic boundary, not only at the end. Design UX and Architecture shows what goes wrong without the spine.

When Requirements Change Mid-Flight

They will. The path for a change is the same as the path for the original:

1. Run bmad-prd in Update mode with the change signal. It surfaces conflicts with earlier decisions before applying anything.
2. If the change touches a cross-epic decision, update the spine.
3. Re-run bmad-spec for each affected epic. It updates SPEC.md in place and keeps capability IDs stable, so stories that are unaffected stay unaffected.
4. Re-run Story Breakdown or bmad-sprint-planning for the affected epics. Regenerating tracking is safe; finished work stays finished.

For a change large enough to threaten the plan itself, run bmad-correct-course before touching documents.

What You Get

A set of documents that agree with each other because each has one writer and one owner; five named points where the organization can say yes or no with something in hand; and a change path that flows from the PRD outward instead of from whichever document someone happened to edit. Engineering still implements one Build session at a time, exactly as it would for a single-builder change.