Breakdown
Turn a PRD into actionable vertical-slice tickets.
$ARGUMENTS — PRD URL/path, spec URL/path, doc path, or empty (search configured docs for recent PRDs/specs).
Prerequisite: load wystack-agent-kit:workspace first for configured task/doc providers.
Workflow
- Load PRD + spec. PRD = what to build; spec = key decisions and architecture. Extract user goals, system boundaries, and the decisions that inform how to slice.
- Explore codebase. Affected modules, integration points, existing patterns, test infrastructure — informs where slice boundaries fall.
- Slice vertically. Each slice cuts through all layers (UI, service, data) as a thin complete feature: independently implementable, testable end-to-end on its own, proves cross-layer integration. Order the thinnest end-to-end path first; later slices add depth. Inter-slice dependencies must be explicit.
- Phase into shippable increments. Group slices into phases. Phase 1 = thinnest end-to-end slice (proves integration across all layers); later phases add depth, edge cases, polish. Each phase ships.
- Create tickets. Each slice → one ticket referencing the PRD (which stories it satisfies) and the spec (which design applies). Include: what the slice does, ACs, suggested verification approach, slice dependencies, and a
Source documents / Related docs section linking PRD + spec with relevant requirement IDs. Delegate creation to wystack-agent-kit:task-manager; require the actual ticket URLs/IDs back.
- Back-link tickets into source docs (mandatory). Delegate to
wystack-agent-kit:wiki-librarian:
- Add/update an
Implementation tickets section in the PRD, grouped by user story / requirement ID.
- Add/update an
Implementation tickets section in the spec, grouped by phase / architectural area / decision anchor.
- Only one source doc exists → update it, note the missing counterpart.
- Verify by fetching the updated pages — don't report success from write receipts alone.
- Can't update automatically → report a setup gap: name what blocked the write (missing adapter capability, permissions) and the fix that automates it next time. Surface the gap; don't hand the user the edits as a chore.
Splitting techniques (SPIDR)
When a slice is still too big:
- Spike — extract unknowns into research tickets first.
- Paths — split by alternative flows (happy path first, error handling later).
- Interfaces — split by input/output method.
- Data — split by data variations (simple case first, complex later).
- Rules — split by business rules (basic first, edge cases later).
Ticket quality (INVEST)
- Independent — workable in any order; minimize dependencies.
- Negotiable — not over-specified; room for implementation judgment.
- Valuable — visible user value, not just technical infrastructure.
- Estimatable — clear enough to size.
- Small — completable in one sprint.
- Testable — acceptance criteria writable immediately.
Principles
- Vertical, not horizontal — "login form → API → DB → response", not "all UI first, then all backend".
- Each slice is verification-ready — clear enough to name the right proof: strategic test, runtime check, screenshot, typecheck, or lint.
- Stories, not tasks — "user can see workflow progress", not "build the progress tracking service".
Reference
wystack-agent-kit:task-manager / wystack-agent-kit:wiki-librarian — use provider adapters; never call provider APIs directly unless the adapter says so.