Skip to content

Plan

Write and Slice the Initiative

Use Sofia, the Product Owner (bmad-agent-po), to write the initiative: the ticket that is the spec for the whole piece of work. She lifts what your inputs say into requirement lines with stable ids, names the gaps, and then cuts the initiative into epics. Every later skill reads the initiative instead of your original documents.

Sofia sizes new work first: one story or bug becomes a standalone ticket, several stories stay one epic, and more than one epic becomes an initiative. She proposes one epic unless a distinct outcome, or a part people can use early, splits it.

Say what you want and name your inputs: “write the initiative from prd-check-in/”, “distill this PRD”, or “write the requirements from these notes”. A PRD, a brief, a PRFAQ, cited notes, UX files, or an idea explained in a paragraph all work, and she takes more than one. With no initiative active, she creates initiative-<slug>/ and records it as the active initiative.

She writes the initiative’s Requirements, one line each under a stable R<n> id naming its source, such as R1 (FR-1), and its Constraints, each citing its source id, such as (RULE-1). From a PRD, each capability becomes a Requirements line and each rule a Constraints line, citing its id. The input’s non-goals go to Boundaries. On a new initiative she also writes its Description, Outcome, and Done when: three to six checks at the level of your source’s ids. Read them as closely as the R lines. Tables, diagrams, and documents other skills wrote stay beside the ticket and are named in its References rather than copied.

She extracts rich input without asking questions. With sparse input she names the gaps and walks them with you, Requirements then Constraints. Say you would rather move on, and each gap becomes an Open question: line. To work an idea out before she writes it, she offers John for a PRD. For something lighter, ask Mary for a brief yourself. A long document goes in as the main file bmad-shard-pro writes; see Prepare Long Documents for Agents.

After each run she reports the requirement count, the companions (files later skills read beside the ticket), and the verdict of her own check. The check tests each line against her writing rules, such as a testable result per requirement, and confirms that every source claim that matters reached the ticket or a companion. Then she lists the Assumption: and Open question: lines. Walk them with her. An answer can update the source, the ticket, or both; say which. Each answer you confirm turns its Assumption: into a dated Decision: line in Notes. Read the R lines yourself too: now, before anything is cut from it, change any line you would not defend to a stakeholder.

Sofia is the only writer of these sections, so don’t edit them by hand. When an input changes, run her again with the change: she keeps every id and names the stories whose requirements changed.

Ask for the cut, and tell her the split if you already have one in mind:

/bmad-agent-po split this initiative into epics

Before cutting epics, she asks you to confirm the scope. She proposes the epics, their build order, and what each needs from the ones before, and records the agreed order in the initiative’s tickets/tickets.toml. Each epic is written thin: its description, outcome, boundaries, and done-when as far as they are known, marked refined: false. Its requirements come later, when you refine it.

An epic is one capability, or a tightly coupled pair, delivered and deployed together, and typically eight to twelve stories once planned. A service or module is an epic only when it is also a deployment boundary. A unit the work only consumes or configures gets no epic: Sofia lists it in the initiative’s Boundaries as a touch point, with the epic that owns it. If your team cuts epics by its own rule, tell Sofia and she offers to save it to _bmad/custom/bmad-agent-po.toml.

She then lists the decisions more than one epic must adopt, such as a contract, a data format, or a shared value list, and offers bmad-architecture to settle them. Decline, and each becomes a story in the opening epic that the others wait on. See Design UX and Architecture. She records each slicing choice, and each suggestion you decline, as a dated Decision: line in the initiative’s Notes.

  • initiative-<slug>/initiative-<slug>.md, the spec for the whole initiative, with any companions beside it
  • initiative-<slug>/tickets/tickets.toml, the epics in build order
  • One folder per epic under tickets/, each holding a thin epic file

When the work is more than one initiative’s worth, Sofia proposes a parent initiative instead of nesting deeper: an ordinary initiative whose breakdown lists child initiatives in build order, each in its own folder beside it.