Brite-Nites flow-inventory-interview
Greenfield Socratic inventory generator. Produces docs/product/master-flow-inventory.md with proposed domains + flows + 3-tag scope-priority taxonomy, derived from a 5-phase user interview + pattern-catalog synthesis.
This skill is NOT user-invocable (disable-model-invocation: true, per Q7). The user-facing wrapper is /flow:inventory; /flow:start-project dispatches to this skill when flow-preflight classifies the project as greenfield.
Critical relationship to Q11. Shares Phases 0, 1, 2, 5 verbatim with flow-inventory-codebase-scan via _shared/app-classifier-pattern.md (BC-6955 deliverable). Differs in:
- Skipping Phase 3 (no codebase to scan).
- Phase 4 status taxonomy (3-tag scope-priority for greenfield vs. 4-tag implementation status for retrofit).
- Heavier Phase 1 interview (4 greenfield-only follow-up questions).
The full design rationale lives in docs/design-rationale/fda-plugin-interview.md Q19 (memory:208-222). Q11 (memory:68) is the retrofit twin; Q25 defines the master-inventory schema this skill writes.
1. Phase sequence (Q19.1) --- 5 phases
| Phase | Source | Description |
|---|---|---|
| 0 | shared | PROJECT-INTENT.md priority filter |
| 1 | shared | app-classifier interview (extended with 4 greenfield follow-ups) |
| 2 | shared | pattern-driven candidate generation |
| 3 | --- | SKIPPED (no codebase) |
| 4 | greenfield-specific | synthesis with 3-tag scope-priority |
| 5 | shared | user confirmation |
Skill explicitly logs the Phase 3 skip with rationale for audit trail:
Greenfield mode: no codebase to scan; proceeding to synthesis from interview + pattern signals.
2. Phase 1 Socratic depth (Q19.2)
The shared _shared/app-classifier-pattern.md utility owns the base questions (framework, app category, primary persona shape, scale) --- used by both Q11 and Q19.
Q19 adds 4 greenfield-only follow-ups defined in this skill's body (NOT in _shared/ --- they're Q19-specific):
(a) Domain envisioning
"What top-level domains do you imagine? Examples: SaaS CRM = Contacts/Deals/Pipeline/Reports; installation business = Quotes/Properties/Crew/Routing/Billing."
(b) Flow-density-per-domain
"Which domains will have heavy UI investment (10+ atomic actions) vs admin-only (less than 5)?"
(c) MVP sequencing
"Which 2-3 domains are MVP-essential vs post-launch?"
(d) Persona density
"How many distinct personas access this app? small SaaS = 1-2; multi-tenant CRM = 4-8; complex ops = 10+."
Phase 1 depth is the dominant signal source for Q19 (vs. Q11 where code-evidence dominates).
3. Phase 4 synthesis (Q19.3) --- 3-tag scope-priority taxonomy
Per-flow tag:
| Tag | Definition |
|---|---|
mvp |
Must-have for v1; user flagged the domain MVP-essential + flow is core to the domain's primary user task. |
nice-to-have |
Plausible v1 inclusion if scope permits; pattern-catalog-suggested but not user-flagged. |
post-launch |
Out of v1 scope; user flagged the domain post-launch OR pattern catalog flags as advanced. |
Rendered in the Notes column alongside any pattern context (no new column needed; Status column stays blank per master-flow-inventory.md:18 lock).
Deliberate divergence from Q11 on Notes-column content type
Verified against BriteBase master-flow-inventory.md AUTH section :42-52:
- Q11 (retrofit) uses Notes for code-evidence anchors --- route paths (
/login), component names (User-nav dropdown), scope references (Droidor scope (BC-5988)). - Q19 (greenfield) uses Notes for scope-priority tags because no code-evidence anchor exists yet.
Both modes' Notes carry mode-appropriate context, but the content type differs by design.
4. Output format (Q19.4)
Produces docs/product/master-flow-inventory.md matching the BriteBase schema:
- Front-matter
last_reviewed: <ISO-8601>+ status block. - Top-level groupings.
- Domain sections.
- Per-domain table with columns
# / Flow / Status / Notes.
<PROJECT_NAME> substituted from .flow/config.json linear_project_name.
Status column left blank per existing schema. Status map: TBD and Journey: TBD are populated later by flow-linear-scaffold.
Top-level groupings derivation
Agent infers groupings during Phase 4 synthesis from domain semantics + pattern-catalog conventions:
auth/tenancy/team->PLATFORM FOUNDATIONS- primary product domains ->
CORE WORKFLOWS - admin / reporting ->
OPERATIONS
Defaults are agent-proposed; user reviews + edits during Phase 5 confirmation.
Single-grouping fallback
If user declines all proposed groupings (small project), emit a single ## CORE grouping containing all domains.
5. Idempotency (Q19.5) --- skip / interactive merge / --force overwrite
Three scenarios:
| Scenario | Behavior |
|---|---|
| No existing inventory | Create from scratch (greenfield common case). |
| Existing inventory + skill re-run | AskUserQuestion with 3 options: Skip (preserve existing) / Merge (add only newly-proposed flows under existing domains; preserve existing rows + Status column values) / Force overwrite (regenerate from scratch --- destructive, requires confirmation). |
--force flag |
Bypasses the interactive prompt; overwrites. |
Merge is natural for iterative greenfield discovery --- Phase 1 reveals a new domain after the initial pass.
6. Failure recovery (Q19.6)
| Failure | Behavior |
|---|---|
| Phase 1 interview-loop on clarification failures | Max 2 retries per question. On third unclear answer, flag the question as skipped/needs-revisit + continue. |
| Phase 4 synthesis: insufficient signal to tag a flow | Tag nice-to-have with Notes annotation (unclear from interview --- revisit at user confirmation). |
| Phase 5 reject | Skill exits cleanly; user re-runs after refining intent. |
Do NOT auto-iterate forever.
7. User-confirmation gates (Q19.7) --- Phase 5 = 1 of Q10's 4 greenfield gates
Within-Phase-1-interview AskUserQuestions are interview-cadence interactions, NOT high-stakes orchestrator gates --- they don't count against Q10's budget.
Phase 5 surface (matches Q11)
Preview the proposed inventory rendered as output markdown. User picks:
- Approve as-is.
- Edit inline --- slug overrides per Q11 pushback / move flows between mvp/nice-to-have/post-launch / drop flows.
- Reject --- exit; user refines intent and re-runs.
Worked example
Brand Hub greenfield (runtime-determined domain count per Q34 disambiguation; ~5-7 in this example, NOT pinned to BriteBase's 28 nor the legacy-milestone count):
- Phase 0. Read
docs/product/intent.md--- pull out priority verbs ("self-serve brand asset reuse", "agency tenancy", "audit trail"). - Phase 1. App-classifier interview via
_shared/app-classifier-pattern.md+ 4 greenfield follow-ups. User responds: framework=Next.js, persona-density=multi-tenant (4-8), MVP-essential=brand/library/admin, post-launch=integrations/analytics. - Phase 2. Pattern-catalog generation (WebSearch + agent SaaS knowledge) -> proposes ~31 candidate flows across ~5-7 domains.
- Phase 3. SKIPPED with audit-trail log.
- Phase 4. Synthesis assigns 3-tag scope-priority: 11
mvp, 9nice-to-have, 11post-launch. Top-level groupings: PLATFORM FOUNDATIONS (auth + tenancy), CORE WORKFLOWS (brand + library + admin), OPERATIONS (audit + reporting). - Phase 5. Render proposed inventory; user moves 2 flows from
nice-to-havetomvp, drops 1post-launchflow, approves. - Skill writes
docs/product/master-flow-inventory.md. Status column blank;last_reviewed: 2026-05-15. Setsstate.inventory_complete=truefor orchestrator dispatch.
See also
docs/design-rationale/fda-plugin-interview.mdQ19 --- canonical 7-sub-decision spec.docs/design-rationale/fda-plugin-interview.mdQ11 --- retrofit twin (flow-inventory-codebase-scan).docs/design-rationale/fda-plugin-interview.mdQ25 --- master-inventory schema this skill writes.skills/_shared/app-classifier-pattern.md--- BC-6955 shared utility (Phases 0/1/2/5).skills/flow-inventory-codebase-scan/SKILL.md--- retrofit twin.skills/flow-preflight/SKILL.md--- preceding sub-skill;MODE=greenfieldgates this skill's invocation.skills/flow-linear-scaffold/SKILL.md--- downstream sub-skill that consumes the inventory rows.