Empire Teas CRM — Decomposition Readiness Checklist
Used by
/forge-decompose(Gate 3). Customize per project — drop items that don’t apply, add items that do.Each item is pass / fail / N/A. Failures are logged with the specific feature(s) at issue.
Slicing
- [ ] Slicing principle stated (by module / by user journey / by capability / by phase)
- [ ] Slicing principle applied consistently across the breakdown (don’t mix journey-sliced and module-sliced units silently)
Coverage
- [ ] Every in-scope item from the PRD maps to at least one feature spec
- [ ] No feature spec exists that doesn’t map back to the PRD (no scope creep at decomposition time)
Separation
- [ ] No two feature specs claim ownership of the same surface
- [ ] Shared concerns (auth, data model, design system) live in their own foundation specs, not duplicated across feature specs
Dependencies
- [ ] Dependency graph between features is drawn (mermaid block in
.forge/features.md) - [ ] No cycles
- [ ] Critical path identified — what blocks what
- [ ] Parallelisable work identified — features with no mutual dependency are marked safe to run concurrently, so the graph reads both ways: what must be sequential (and in which order) and what can start beside it. This is the work-division plan across the developers
- [ ] Foundation features (auth, schema, design system) sequenced before features that depend on them
Sizing
- [ ] Feature sizing audit (FR-1) run — three signals computed for every candidate (cross-layer spread / substrate-slice consumers / multi-cluster capability); tripped features either accepted via FR-2 cut proposal (substrate-cut / cluster-cut / read-vs-write-cut) or kept-as-one with FR-4 one-sentence justification recorded in
engagement-gate-runs.md - [ ] If reslice mode was used (
/forge-decompose --reslice), the dry-run audit + precondition reports were reviewed before any--applypass; FR-7.a preconditions honored for every applied feature - [ ] Each feature spec is independently implementable in one agent run (rough heuristic: a single backend + frontend + tests cycle, not a multi-week epic)
- [ ] Features that don’t fit the size heuristic are split or flagged for further decomposition
Per-feature spec readiness
- [ ] Each stub feature spec follows
_TEMPLATE-spec.md - [ ] Each one is self-contained enough that an agent reading only that spec + the PRD can implement it
Delivery Phases (optional — skip for projects under ~10 features)
- [ ] Delivery phases defined (count, titles, themes) — captured in
tracker.yamldelivery.phasesand mirrored infeatures.md## Delivery Plan - [ ] Each phase has a stakeholder-facing theme (one sentence — what does this phase deliver?)
- [ ] Every in-scope feature is assigned a
delivery_phase(or explicitly leftnullwith reasoning innotes) - [ ] Phase ordering respects the dependency graph — features in Phase N only depend on features in Phase ≤ N
- [ ] Exactly one phase is at
status: in-progressat Gate-3 sign-off; the rest arelocked;current_phasematches - [ ] Each phase is independently demonstrable — sealing the phase produces something the team can show stakeholders