Skip to content

Clarify

Clarify the Idea

Use this page when you are not yet sure what to build, or whether it is worth building. Clarify answers two questions: why build it, and is it worth it? Its tools are independent. Pick the ones the gap calls for, in any order. No skill here writes tickets.

Clarify
You start withA transcript, a thread, a pile of notes, a half-formed idea, or a decision you cannot make yet
You produceA short document that states the idea, or a decision not to build it
It is done whenYou can say what should be true when the work is done, what must not change, and what is out of scope
Skip it whenThe intent is already written down and agreed: a good issue, an existing PRD, a clear paragraph
What comes nextPlan the Work, or Build a Change when the idea turns out to be one small change

Start with Mary, the Business Analyst (bmad-agent-analyst). She runs brainstorming and research for you, and writes the brief or PRFAQ. Each skill also runs on its own.

/bmad-agent-analyst I have an idea for check-in at meeting-room doors; help me work out whether it is worth building
You haveAskYou get
A recording, a transcript, a thread, or a pile of notesbmad-ingestA note where every item cites its source
A topic and no idea yetMary, or bmad-brainstormingFar more options than you would find alone, and the few you keep
One idea you are unsure ofbmad-forge-ideaA verdict: hardened, killed, or clearer
A decision that depends on the market or a vendorMary, or bmad-deep-reconA cited research summary with a recommendation
A question about how your own code worksbmad-deep-recon on your repositoryA cited account of how it works and why, with no recommendation
Doubts that the concept deserves money or peopleA PRFAQ from Mary, or a prototypeA verdict on the concept, or something real to show people
A startup idea, and the business around itNadia, the Founder Advisor, from the optional Founder moduleA startup diagnostic that ends in one dated assignment, or a validation experiment with its pass mark set first

Planning skills work best on conclusions, not on a recording of how people reached them. Give bmad-ingest a transcript, a chat or email thread, a voice memo, or a long document, and it writes a note a later session can act on. Every item points back to where it was said: a timestamp, a section, or a message’s date and speaker. Owners, dates, and conditions appear only as the source states them, and anything it leaves unsettled goes under a last section, “Unclear in the source”.

The original is kept unchanged: a file from outside the project is copied to _bmad-output/raw/, and the note goes to _bmad-output/notes/ unless you name another folder. It picks a template, such as a meeting or a user interview, and says why. Ask for one combined note when several sources belong together, for example every stakeholder conversation about one decision, so that conflicting accounts sit side by side.

With no idea yet, ask Mary to brainstorm, or run bmad-brainstorming with a topic and a goal. Pick a stance: Facilitator (every idea is yours), Creative Partner (it trades ideas with you), or Ideate for me (it runs the session itself). When a batch stops producing, choose another technique, converge, or wrap up. At the end you choose what to keep: an HTML keepsake of the session and a short summary of the chosen ideas that later skills read.

With one idea you are unsure of, run bmad-forge-idea with the idea in a sentence or two. It asks your goal (clarify it, test whether it holds, or make it better) and whether the idea changes an existing product. Then it argues with you one question at a time; say “attack this”, “defend this”, or “switch roles” to change how a point is argued. Every session ends on a stamped report: Hardened, Killed, or Clarified. Killing an idea here is a good result, and it costs little. Every session also keeps forge-<slug>.md, its running record; for a hardened idea it holds the decisions and the options rejected.

Ask Mary to research, or run bmad-deep-recon with the decision, not the topic: “help me choose a door tablet: an iPad in kiosk mode or an Android room-display panel”. It offers three services you can combine. Run does the research here, after you approve a plan of the questions and the effort. Draft writes a prompt for your own research tool, and Process turns that tool’s report into the same summary. Whichever you use, you get research-<topic>.md, every claim with its source and date. A choice between candidates ends with a pick, a runner-up, and the strongest argument against the pick. Ask it later to refresh the claims that age fastest, or to deepen one area.

Point it at your own repository and it researches the code and its git history instead of the web, which helps before changing an unfamiliar system. You get a cited account of how the code works and why, with no recommendation: what to change stays with you.

A PRFAQ from MaryA prototype
How it answersBy arguing the concept and researching its market claimsBy building something people can try
Use it whenYou are about to commit real money or people, or you lead with a solution instead of a customer problemYou need to know whether people want it, whether it can be built, or where the effort is
Ends withA press release, hard customer and internal FAQs, and a verdict, in prfaq-<slug>/prfaq-<slug>.md with a press-release page in HTML beside itNotes on what it taught you, and a decision to keep it, throw it away, or drop the idea

Ask Mary, the Business Analyst (bmad-agent-analyst), for a PRFAQ:

/bmad-agent-analyst create a PRFAQ for check in or lose the room

She shows you its four parts and takes them in order, finishing each before the next: the press release, the customer FAQ, the internal FAQ on feasibility, viability, resources, and risk, and the verdict, written last. Her role changes with the part: an editor for the press release, a skeptical customer, then the stakeholder panel. The verdict is narrative, not a score: what is solid, what needs more work before a PRD, and what could sink the concept, each weakness with what it would take to fix, so you can deal with what could sink it before you plan.

If you already believe in the product and want it written down, ask Mary for a brief instead; see Document Templates.

A throwaway prototype needs no skill: prompt your agent directly. John, the Product Manager (bmad-agent-pm), writes a prototype plan when you want the run recorded, its Outcome noting whether the idea held; for thresholds set in advance, use the experiment card. Once the prototype plan is final, John offers to build it with bmad-build-auto, then asks you to keep it, throw it away, or drop the idea. Make that choice on purpose. If you throw the prototype away, its notes are some of the best input Sofia can get. If you keep it, it is an existing codebase and needs a spec like any other.

For a funding decision, ask Mary for a business case; to judge a problem before any solution, an opportunity assessment; for a leadership strategy, a strategy narrative.

The next stage reads what you hand it, so hand it the conclusion, not the pile. A clear intent says what should be true when the work is done, what must not change, and what is out of scope: complete enough that someone else could build it without guessing, and no longer than that. Where it came from does not matter.

Keep it short. Turn a pile of raw input into a note with bmad-ingest. Past about 50,000 characters, run bmad-shard-pro on a document you will keep and edit, such as an old spec; agents read long files badly. It splits the document or adds a contents list, and you hand on the main file; see Prepare Long Documents for Agents. To improve any draft before you hand it on, point bmad-advanced-elicitation at it: it reads the draft independently and returns one revision for you to accept or reject.