Catalog Discovery — Spec & Slice Map (design/UX-first)

Source: devlog/_plan/260704_catalog_discovery/00_spec.md

Date: 2026-07-04

Goal: design/UX-first Catalog Discovery interview mode in the canonical PABCD skills.

Predecessor design: devlog/260704_catalog_discovery_design.md (3-employee discussion synthesis).

Loop archetype: spec/repair (a verifier defines *done* — the catalog + interview

section must satisfy schema + design-first ordering checks; not open-ended optimization).

Trigger: user goal — add catalog_discovery mode, design/UX leading.
Goal (user-visible outcome): an author/agent running the PABCD Interview on a novice

("사주 앱 만들고 싶어") is guided design/UX-FIRST through a versioned option catalog,

and downstream (security/data/ops/feature) questions are DERIVED from the design+domain

choices — not asked in a flat undifferentiated list.

Non-goals: no host-CLI code, no runtime engine, no GUI build, no downstream repo edits

(this pass is canonical-source only; ports tracked separately in backlog).

Verifier: (a) YAML parses (python3 -c "import yaml,sys; yaml.safe_load(...)"),

(b) schema-consistency check — every implies/conflicts/derived_from id resolves to a

real entry; design axis entries exist for all 6 dials; axis_order lists design first,

design-first ordering; (d) saju dry-run doc shows design questions asked before

security/data questions. Measures: referential integrity + ordering correctness, not just

"file exists".

Stop condition: all 3 work-phases pass their verifier; goal objective's 3 deliverables exist.
Memory artifact: this folder + devlog/260704_catalog_discovery_design.md.
Expected terminal states: DONE (all deliverables + verifiers green) | BLOCKED (schema

can't express design-first derivation cleanly) | NEEDS_HUMAN (business ambiguity in axis semantics).

Escalation: if design-first derivation needs a runtime engine to be meaningful (i.e.

static YAML can't encode it), stop and surface — that would change scope to code.

The user's hint is the load-bearing constraint: **design/UX is the most important axis; the

backend (security/data/ops) asks its questions BASED ON the design+domain answers.** This is

encoded three ways so it survives prompt drift:

stage 1: design → stage 2: domain(app-type) → stage 3: derived (feature/data/security/ops/cost).

Design is stage 1, always first.

derived_from (data): stage-3 entries carry derived_from: [design.*, domain.*] so the

configurator can PRIORITIZE/FILTER them from earlier answers instead of dumping a flat list.

Interview prose (skill): the catalog_discovery sub-mode explicitly asks design/UX first,

wired to dev-uiux-design Product-Personality-Selection as the step-1 methodology.

Phase	Outcome unit	Deliverable	Verifier
0 (done)	Design synthesis	`devlog/260704_catalog_discovery_design.md`	committed (82ab348)
1	Versioned catalog exists, design-first	`skills/dev-pabcd/references/catalog-discovery.yaml`	YAML parse + schema-integrity script
2	Interview uses catalog design-first	`skills/dev-pabcd/SKILL.md` Interview section: catalog_discovery + configurator sub-modes	prose references catalog + design-first ordering; agent-neutral (no host-CLI)
3	Design-first flow proven	`devlog/_plan/260704_catalog_discovery/30_saju_dryrun.md`	walkthrough shows design Qs before security/data Qs

Phase docs: 10_phase1_catalog.md, 20_phase2_interview.md, 30_saju_dryrun.md.

Each phase's D records what did NOT get proven and what would falsify the design-first claim.

Other files in this unit

Phase 1 Plan Audit — record (05_audit.md)
Phase 1 — Catalog data structure (design/UX-first) (10_phase1_catalog.md)
Phase 2 — Interview section update (design/UX-first catalog_discovery mode) (20_phase2_interview.md)
Phase 3 — Saju Persona Dry-Run Walkthrough (30_saju_dryrun.md)