qfai 1.7.9 → 1.7.13

package/assets/init/.qfai/assistant/skills/qfai-sdd/SKILL.md

@@ -6,15 +6,19 @@ argument-hint: "[<spec-id-or-name>] [--auto]"
 allowed-tools: [Read, Glob, Write, TodoWrite, Task, Bash]
 roles:
   [
-    Planner,
-    Architect,
-    RequirementsAnalyst,
-    SpecWriter,
-    TraceabilityBuilder,
-    TestStrategist,
-    QAEngineer,
-    CodeReviewer,
+    orchestrator,
+    delivery-planner,
+    requirements-analyst,
+    solution-architect,
+    test-design-analyst,
+    qa-strategist,
+    completion-reviewer,
+    architecture-reviewer,
+    implementation-reviewer,
+    product-surface-reviewer,
+    qa-gatekeeper,
   ]
+routing-profile: default
 mode: approval-gated
 ---
 
@@ -39,14 +43,27 @@ QFAI Skill Body (SSOT)
   The agent SHOULD preserve structured choice semantics (enumerated options, selection constraints).
   The reason for unavailability MUST be stated.
 
+### Spec Create/Delete Confirmation (Mandatory)
+
+- **When creating a new spec**: approval via AskUserQuestion is mandatory.
+  Example prompt: `"Create new spec-XXXX (${subject}) in category ${category}. Approve?"`
+  Choices: `["Approve", "Reject"]`
+- **When deleting an existing spec**: approval via AskUserQuestion is mandatory.
+  Example prompt: `"Delete spec-XXXX (${subject}). Reason: ${rationale}. Approve?"`
+  Choices: `["Approve", "Reject"]`
+- **When only updating an existing spec**: no confirmation is required.
+- If approval is not granted, skip the operation and record the reason in `delta.md`.
+
 ## FORMAT SSOT (Mandatory)
 
 - Before writing or editing any `.qfai/**` artifact, read and follow:
-  - `.qfai/require/README.md`
+  - `.qfai/discussion/README.md`
   - `.qfai/specs/README.md`
   - `.qfai/contracts/**/README.md`
   - `.qfai/evidence/README.md`
-  - `.qfai/assistant/steering/review-roster.yml`
+  - `.qfai/assistant/steering/agent-catalog.yml`
+  - `.qfai/assistant/steering/agent-routing.yml`
+  - `.qfai/assistant/steering/review-profiles.yml`
   - `.qfai/assistant/skills/qfai-sdd/references/rcp_footer.md`
 - Use skill-local templates as SSOT:
   - `.qfai/assistant/skills/qfai-sdd/templates/report/preflight_summary.md`
@@ -67,7 +84,7 @@ Then read inputs in this order:
 - P1: `.qfai/assistant/instructions/*`
 - P2: `.qfai/assistant/steering/*`
 - P3: existing `.qfai/specs/<spec-id>/**` (if updating)
-- P4: `.qfai/discussion/**`, `.qfai/require/**`, `.qfai/contracts/**`
+- P4: `.qfai/discussion/**`, `.qfai/contracts/**`
 
 ## Sub-agent Delegation (MANDATORY)
 
@@ -104,10 +121,12 @@ Every major artifact in this stage MUST include a `## Work Orders Summary` secti
 
 ### Stage Minimum Roles (MUST)
 
-- Delegate: SpecWriter + TraceabilityBuilder draft shared/spec layered artifacts and edge mappings.
-- Delegate: Architect + TestStrategist draft and finalize `10_Plan.md` (How-only plan).
-- Integrate: Orchestrator consolidates delegated outputs and presents them to the user for confirmation.
-- Gate: Reviewer is delegated independently and returns only `PASS` or `REVISE`.
+- Delegate: `requirements-analyst` drafts requirement-aligned spec content and open-question handling.
+- Delegate: `solution-architect` drafts structural, contract, and architecture-sensitive sections.
+- Delegate: `test-design-analyst` drafts traceability, examples, and test-design structure.
+- Delegate: `product-experience-architect` only when the target is UI-bearing.
+- Integrate: `orchestrator` consolidates delegated outputs and presents them to the user for confirmation.
+- Gate: `completion-reviewer` is delegated independently and returns only `PASS` or `REVISE`.
 - Orchestrator must not draft the primary artifact body and must not self-approve.
 
 ### No-argument batch delegation (MUST)
@@ -115,12 +134,8 @@ Every major artifact in this stage MUST include a `## Work Orders Summary` secti
 - If `/qfai-sdd` is invoked without `<spec-id-or-name>`, treat the run as an all-capability batch.
 - Enumerate targets from `.qfai/specs/_policies/03_Capabilities.md` and keep `spec-0001..N` mapping stable by Capability order.
 - In batch mode, run Contracts-first and Outline exactly once as shared outputs.
-- Delegate Slice in parallel per spec:
-  - `SpecWriter + TraceabilityBuilder` own `spec-XXXX/01..08`.
-- After each spec passes Slice gate, delegate Plan in parallel per spec:
-  - `Architect + TestStrategist` own `spec-XXXX/10_Plan.md`.
-- Delegate Delta in parallel per spec:
-  - `SpecWriter` (or `DocSteward`) owns `spec-XXXX/09_delta.md` with `Adopted/Rejected` and rejected guardrails (`DO NOT`, `Temptation`).
+- Delegate Slice in parallel per spec using the routing for `qfai-sdd` in `.qfai/assistant/steering/agent-routing.yml`.
+- Delegate Plan and Delta per spec after the design phase passes its blocking reviewers.
 - Validate gate and Review gate run once at batch tail after all target specs are integrated.
 - Evidence is mandatory per spec: `.qfai/evidence/sdd-spec-XXXX.md`.
 - Optional batch summary is allowed: `.qfai/evidence/sdd-batch-<timestamp>.md`.
@@ -128,7 +143,7 @@ Every major artifact in this stage MUST include a `## Work Orders Summary` secti
 
 ### Reviewer Gate (MUST)
 
-- Final completion gate MUST be delegated to an independent Reviewer sub-agent.
+- Final completion gate MUST be delegated to an independent `completion-reviewer`.
 - Reviewer checks (minimum):
   - Required roles were delegated (no orchestrator self-authoring).
   - DoD satisfied (validate gate, test-layer hard gate, evidence, DR-IDs).
@@ -150,15 +165,15 @@ Every major artifact in this stage MUST include a `## Work Orders Summary` secti
       - `tests/e2e/**` -> `QFAI:SPEC-XXXX:US-YYYY`
       - `tests/integration/**` -> `QFAI:SPEC-XXXX:TC-YYYY`
       - `tests/api/**` -> `QFAI:CON-API-XXXX` (and no TC annotations)
-- Do not declare DONE or handoff until Reviewer returns `PASS`.
-- **All reviewers: alternative proposal obligation**:
-  - Every reviewer MUST provide a concrete alternative or fix proposal when returning FAIL. Feedback without a concrete alternative is invalid and triggers re-judgment.
-- **devils-advocate gate**:
-  - devils-advocate FAIL must include a concrete alternative proposal. Bare negation FAIL triggers re-judgment.
-  - 3 consecutive FAILs trigger advisory demotion and allow progression to the next phase.
-- **pattern-doubler gate**:
-  - Each pattern proposed by pattern-doubler must include rationale.
-  - Artifacts with no ID-bearing items (US/AC/BR/EX/TC) are marked N/A.
+- Route specialist reviewers from `.qfai/assistant/steering/agent-routing.yml` and `.qfai/assistant/steering/review-profiles.yml`.
+- Default SDD review set:
+  - `completion-reviewer`
+- Conditional SDD reviewers:
+  - `architecture-reviewer` for structural / contract / CLI changes
+  - `product-surface-reviewer` for UI-bearing specs
+  - `qa-gatekeeper` when validate, coverage, runtime, or prototyping evidence is affected
+- Do not declare DONE or handoff until all routed blocking reviewers return `PASS`.
+- Every reviewer MUST provide a concrete alternative or fix proposal when returning `FAIL`.
 
 ### Work order template (copy/paste)
 
@@ -196,21 +211,13 @@ Evidence checked:
 
 ## Review Cycle Protocol (RCP)
 
-- Roster SSOT: `.qfai/assistant/steering/review-roster.yml`.
+- Routing SSOT: `.qfai/assistant/steering/agent-routing.yml`.
+- Review profile SSOT: `.qfai/assistant/steering/review-profiles.yml`.
 - Footer SSOT: `.qfai/assistant/skills/qfai-sdd/references/rcp_footer.md`.
-- Every discuss/require/sdd review cycle must execute the full roster.
-- Allowed reviewer verdicts: `PASS`, `FAIL`, `N/A` (`N/A` requires `na_rule` reason).
-- Any `FAIL` triggers return/fix/full-rerun from the first reviewer.
-- `fixed` is forbidden until all reviewers are `PASS` or valid `N/A`.
-- Execution order: existing 10 reviewers (1-10) → devils-advocate (11) → pattern-doubler (12).
-- devils-advocate (11th):
-  - `can_be_na: false` — N/A is not allowed.
-  - FAIL must include a concrete alternative. Bare negation FAIL is invalid.
-  - 3 consecutive FAILs trigger advisory demotion (current review cycle only). Blocking power is lost after demotion.
-- pattern-doubler (12th):
-  - `can_be_na: true` — N/A only when the target artifact has no ID-bearing items.
-  - Sets a 2x target for current ID-bearing items (US, AC, BR, EX, TC) and identifies missing patterns.
-  - Rationale for each proposed addition is required.
+- Every review cycle executes only the reviewers selected for the current phase and conditions.
+- Allowed reviewer verdicts: `PASS`, `FAIL`.
+- Any `FAIL` triggers targeted rerun for the failed reviewer and reviewers affected by the changed scope.
+- `fixed` is forbidden until all routed blocking reviewers are `PASS`.
 
 ## Stage 0 - Steering completion refresh (mandatory)
 
@@ -241,11 +248,25 @@ Rules:
 - **Plan finalize MUST happen after at least one user-story slice is grounded.**
 - **Unresolved items MUST be moved to `08_Open-questions.md` (spec scope) or `_policies/09_Open-questions.md` (shared scope).**
 
+## Slice Policy Protocol (Mandatory)
+
+Before any spec creation, update, or deletion, the agent MUST:
+
+1. Treat `_policies/11_Slice-Policy.md` as the SSOT for spec slicing decisions.
+2. If `_policies/11_Slice-Policy.md` is missing, stale, or no longer matches the repository structure, create or update it before continuing.
+3. Classify each target into a category (structural / CLI / skill / agent).
+4. Enforce 1:1 mapping per category rule (1 command = 1 spec, 1 skill = 1 spec, etc.).
+5. Use `_policies/11_Slice-Policy.md` to decide whether each detected change is CREATE, UPDATE, or DELETE.
+6. If a new CLI command or skill is detected that has no corresponding spec, propose CREATE (requires AskUserQuestion).
+7. If a CLI command or skill has been removed but its spec still exists, propose DELETE (requires AskUserQuestion).
+8. If the action violates slice rules, STOP and report the violation — do NOT proceed without user override.
+9. Do not start Phase 2 (Slice) until `_policies/11_Slice-Policy.md` exists and reflects the current slicing model.
+
 ## Arguments and Target Selection (Mandatory)
 
 - With argument (`/qfai-sdd <spec-id-or-name> [--auto]`): update only the matched single spec target.
 - Without argument (`/qfai-sdd`): target all capabilities listed in `_policies/03_Capabilities.md`.
-- If `_policies/03_Capabilities.md` does not exist, bootstrap shared templates first, then enumerate capabilities.
+- If `_policies/03_Capabilities.md` or `_policies/11_Slice-Policy.md` does not exist, bootstrap shared templates first, then enumerate capabilities.
 - Capability order in `_policies/03_Capabilities.md` is SSOT for `spec-0001..N` assignment and ID stability.
 - Reordering capability-to-spec mapping is a Change Request decision and must not be done implicitly.
 - Batch policy (no argument):
@@ -272,13 +293,14 @@ Rules:
   - never duplicate existing declared IDs; update in-place when the contract already exists
   - contract stubs must be syntactically valid (OpenAPI YAML / UI YAML / executable SQL skeleton)
 - `/qfai-sdd` must stop when discussion-pack is missing/incomplete or has blocking OQ (guide to `/qfai-discussion` first).
-- Review roster is fixed by `.qfai/assistant/steering/review-roster.yml` and must be executed in full.
+- Reviewer routing is fixed by `.qfai/assistant/steering/agent-routing.yml` and `.qfai/assistant/steering/review-profiles.yml`.
 - RCP wording must be sourced from `.qfai/assistant/skills/qfai-sdd/references/rcp_footer.md`.
 - `_policies/04_Business-Flow.md` must be Markdown and include at least one Mermaid `flowchart` or `sequenceDiagram`.
 - Business Flow must not be authored as Gherkin (`*Business-flow*.feature` is deprecated).
-- If diagrams are written in discuss/require/spec/evidence artifacts, Mermaid syntax must be inside ` ```mermaid ` fences only.
+- If diagrams are written in discussion/spec/evidence artifacts, Mermaid syntax must be inside ` ```mermaid ` fences only.
 - `05_Examples.md` must include `EX-ID` and `BR-Ref` mappings.
-- `06_Test-Cases.md` must include `TC-ID`, `EX-Ref`, and `AC-Refs`.
+- `06_Test-Cases.md` must include `TC-ID`, `EX-Ref`, `AC-Refs`, and `Type` (normal/error/boundary/edge).
+- `06_Test-Cases.md` quality depth: each AC must have at minimum one normal-path and one error/boundary test case. Normal-path-only coverage is incomplete.
 - Do not complete this stage until:
   - `qfai validate --fail-on error --format github | tee .qfai/report/validate.log` exits successfully.
   - `.qfai/report/specs-coverage/spec-*.md` has been read for density review.
@@ -325,12 +347,19 @@ Create/update:
 - `_policies/08_Decisions.md`
 - `_policies/09_Open-questions.md`
 - `_policies/10_delta.md`
+- `_policies/11_Slice-Policy.md`
 
 Rules:
 
 - Temporary `TBD` is allowed, but each `TBD` must be mirrored into `_policies/09_Open-questions.md`.
 - `_policies/04_Business-Flow.md` must include Mermaid and keep diagram syntax inside ` ```mermaid ` fences.
 - `_policies/08_Decisions.md` and `_policies/10_delta.md` must exist even when empty, and must explicitly state `0 items`.
+- `_policies/11_Slice-Policy.md` must be created or refreshed in this phase and must explain:
+  - slice categories
+  - create/update/delete decision rules
+  - ID stability rules
+  - gap / renumbering policy
+- If the current repository shape and `_policies/11_Slice-Policy.md` disagree, update the policy file first and record the decision in delta.
 
 ### Phase 2 - Slice (slice-first)
 
@@ -413,6 +442,7 @@ Create or update layered SDD artifacts in one run so downstream execution phases
 - `.qfai/specs/_policies/08_Decisions.md`
 - `.qfai/specs/_policies/09_Open-questions.md`
 - `.qfai/specs/_policies/10_delta.md`
+- `.qfai/specs/_policies/11_Slice-Policy.md`
 - `.qfai/specs/spec-XXXX/01_Spec.md`
 - `.qfai/specs/spec-XXXX/02_User-stories.md`
 - `.qfai/specs/spec-XXXX/03_Acceptance-Criteria.md`
@@ -435,24 +465,27 @@ Create or update layered SDD artifacts in one run so downstream execution phases
 4. Write `.qfai/report/preflight_summary.md` from `templates/report/preflight_summary.md`.
 5. Execute Phase 0 (Contracts-first) and ensure `_policies/05_Contracts.md` index and `.qfai/contracts/**` are aligned.
 6. Execute Phase 1 (Outline) in layer-first order.
-7. Execute Phase 2 (Slice) and pass slice gate for each target spec (single target: at least one user-story slice; no-argument batch: all enumerated specs).
-8. Execute Phase 3 (Plan finalize) and make every target `10_Plan.md` actionable as How-only.
-9. Execute Phase 4 (Delta update) and record adoption/rejection rationale for every target spec.
-10. Run `qfai validate --fail-on error --format github | tee .qfai/report/validate.log`.
-11. Run Density Review Pass using `.qfai/report/specs-coverage/spec-*.md` and `QFAI-COV-207` warnings.
-12. If any validate error exists, fix the source layer table(s) and repeat steps 10-11 until `error=0`.
-13. Record static checks, validate evidence, and density review outcomes in evidence.
+7. Ensure `_policies/11_Slice-Policy.md` exists and matches the current repository slicing model before any CREATE / UPDATE / DELETE spec decision.
+8. Execute Phase 2 (Slice) and pass slice gate for each target spec (single target: at least one user-story slice; no-argument batch: all enumerated specs).
+9. Execute Phase 3 (Plan finalize) and make every target `10_Plan.md` actionable as How-only.
+10. Execute Phase 4 (Delta update) and record adoption/rejection rationale for every target spec.
+11. Run `qfai validate --fail-on error --format github | tee .qfai/report/validate.log`.
+12. Run Density Review Pass using `.qfai/report/specs-coverage/spec-*.md` and `QFAI-COV-207` warnings.
+13. If any validate error exists, fix the source layer table(s) and repeat steps 11-12 until `error=0`.
+14. Record static checks, validate evidence, and density review outcomes in evidence.
 
 ## Unified SDD Quality Gate
 
 Run static checks:
 
 - Confirm required `_policies` and `spec-XXXX` layered files exist.
+- Confirm `_policies/11_Slice-Policy.md` exists and matches the repository's current slice model.
 - Confirm `_policies/04_Business-Flow.md` includes Mermaid and at least one `flowchart` or `sequenceDiagram`.
 - Confirm `01_Spec.md` includes copy-down context and Escalation Hook to `_policies`.
 - Confirm Mermaid syntax is not written in ` ```text ` or language-less fences.
 - Confirm `05_Examples.md` provides `EX-ID` + `BR-Ref` mappings.
-- Confirm `06_Test-Cases.md` provides `TC-ID` + `EX-Ref` + `AC-Refs`.
+- Confirm `06_Test-Cases.md` provides `TC-ID` + `EX-Ref` + `AC-Refs` + `Type`.
+- Confirm `06_Test-Cases.md` includes error/boundary/edge test cases, not only normal paths.
 - Confirm reference direction follows lower-to-upper only.
 - Confirm required edges `US -> AC -> BR -> EX -> TC`.
 - Confirm BR/Examples/Test-cases contain non-empty IDs and coverage mapping.
@@ -499,6 +532,7 @@ When declaring DONE, include:
 - [ ] Upper-to-lower references were not introduced.
 - [ ] At least one user-story slice passed gate before plan finalization.
 - [ ] Required `_policies` + `spec-XXXX` outputs exist and are internally consistent.
+- [ ] `_policies/11_Slice-Policy.md` exists and was created or refreshed before spec slice decisions.
 - [ ] `_policies/04_Business-Flow.md` is Markdown + Mermaid (`flowchart` or `sequenceDiagram`).
 - [ ] Mermaid syntax was not written in ` ```text ` or language-less fences.
 - [ ] `10_Plan.md` is finalized with implementation/test strategy (How-only).

package/assets/init/.qfai/assistant/skills/qfai-sdd/references/rcp_footer.md

@@ -1,27 +1,31 @@
 # RCP Footer (qfai-sdd / SSOT)
 
-この文書は `/qfai-sdd` のレビュー周回（Review Cycle）を「spec/contract 前提」で固定するための SSOT です。
-discussion-pack 等の共通規約ではありません。
+この文書は `/qfai-sdd` の Review Cycle を spec / contract 前提で固定する SSOT です。
 
 ---
 
 ## Review Target（固定）
 
 - Scope: `sdd`
-- 主要成果物（レビュー対象）:
-  - `.qfai/specs/spec-*/**`（spec pack）
-  - `.qfai/contracts/**`（API/DB/UI 契約）
-  - `.qfai/evidence/**`（意思決定根拠・実験ログ）
-  - `.qfai/report/**`（validate / coverage / preflight 出力）
+- レビュー対象:
+  - `.qfai/specs/spec-*/**`
+  - `.qfai/contracts/**`
+  - `.qfai/evidence/**`
+  - `.qfai/report/**`
 
 ---
 
-## Roster Execution Rule（固定）
+## Routing Rule（固定）
 
-- Roster は `.qfai/assistant/steering/review-roster.yml` を読む
-- 各レビューは `PASS` / `FAIL` / `N/A` を返す
-- `FAIL` が1つでも出たら **即修正へ戻る**
-- 修正後は **review cycle を新規作成し** roster を先頭から再実行する（スキップ禁止）
+- reviewer の選定は `.qfai/assistant/steering/agent-routing.yml` と `.qfai/assistant/steering/review-profiles.yml` を読む
+- 常設 reviewer:
+  - `completion-reviewer`
+- 条件付き reviewer:
+  - `architecture-reviewer`
+  - `product-surface-reviewer`
+  - `qa-gatekeeper`
+- 各レビューは `PASS` / `FAIL` を返す
+- `FAIL` が出たら、失敗 reviewer と修正影響を受ける reviewer のみ再実行する
 
 ---
 
@@ -40,74 +44,11 @@ discussion-pack 等の共通規約ではありません。
 - `.qfai/review/review-<timestamp>/R01_<reviewer>.md`, `R02_<reviewer>.md`, ...
 - `.qfai/review/review-<timestamp>/summary.json`
 
-`summary.json` の最低要件:
-
-- `version`
-- `created_at`
-- `target`
-- `roster`
-- `overall_status`
-
----
-
-## spec-pack 固有のレビュー観点（sdd 特化）
-
-1. 仕様の一貫性
-
-- spec の「目的/スコープ/非スコープ」が、後続の user story / acceptance criteria / examples と矛盾しない
-- “例（Examples）” が acceptance criteria を **具体ケースとして裏付け**ている（単なる繰り返しではない）
-
-2. 意思決定の可観測性（Decision Log）
-
-- `delta` / decisions / rejected が「なぜ採用/不採用か」を保持している
-- “Temptation（再発しがちな誤り）” が明文化され、再採用防止になっている
-
-3. Contracts の妥当性
-
-- API / UI / DB 契約が spec の用語と一致している（同一概念に別名を付けない）
-- 禁止参照（contracts 特例ルール等）がある場合は、契約側の README のルールに従っている
-
-4. Traceability（必要なら）
-
-- spec → tests（ATDD/TDD）への紐付けが破綻していない
-- “数を増やす” のではなく “境界/負例/権限/状態遷移” の観点が埋まっている
-
----
-
-## 代表的な FAIL と復旧（sdd 特化）
-
-- FAIL: acceptance criteria が抽象的で、例に落ちていない
-  - 復旧: 例（Examples）に「入力→状態→出力」形式のケースを追加し、境界/異常系を1つは入れる
-- FAIL: 契約が先行し、spec の用語・概念とズレる
-  - 復旧: 先に Glossary/Capabilities を補強し、契約を spec に合わせて修正する
-- FAIL: decision/rejected が薄く、なぜそうしたかが追えない
-  - 復旧: 代替案A/B/Cと採用基準を書き、Rejected に DO NOT/Temptation を残す
-
 ---
 
-## 拡張レビュアー
-
-### devils-advocate（11番目）
-
-- 役割: 全否定エージェント — 「現状すべてが間違っている」前提でレビューし、あるべき姿を提示する
-- `can_be_na: false` — N/A は許可されない
-- FAIL 判定時は必ず具体的代替案を提示すること。代替案なしの FAIL は無効とし、再判定を要求する
-- 3 回連続 FAIL → アドバイザリー降格（当該レビューサイクル限定）。降格後はブロッキング力が消失し、フィードバックのみ記録する
-- レビュー記録: `R11_devils-advocate.md`
-
-### pattern-doubler（12番目）
-
-- 役割: パターン倍増エージェント — ID 付き項目（US, AC, BR, EX, TC）の数を現状の 2 倍にする提案を行う
-- `can_be_na: true` — ID 付き項目のない成果物の場合のみ N/A 可
-- 追加パターンの根拠提示が必須。根拠なしの追加要求は無効
-- カウント対象: US/AC/BR/EX/TC プレフィックスの連番形式 ID を持つ項目のみ
-- レビュー記録: `R12_pattern-doubler.md`
-
-### 代表的な FAIL と復旧（拡張レビュアー特化）
+## sdd 特化レビュー観点
 
-- FAIL (devils-advocate): 代替案なしの否定のみ
-  - 復旧: 具体的な「あるべき姿」と移行パスを記述させ、再判定する
-- FAIL (devils-advocate, 3回連続): 無限ループ検知
-  - 復旧: アドバイザリー降格を記録し、devils-advocate のフィードバックを advisory として保存。次フェーズに進行する
-- FAIL (pattern-doubler): 根拠なしの倍増要求
-  - 復旧: 各追加パターンに「なぜ必要か」（境界値/負例/権限/状態遷移/冪等性）の観点を付与させる
+- spec の目的 / スコープ / 非スコープが下位レイヤと矛盾しない
+- Contracts が spec 用語と一致している
+- delta / decisions / rejected が再発防止に十分か
+- examples / test-cases / traceability が境界・負例・権限・状態遷移を落としていないか

package/assets/init/.qfai/assistant/skills/qfai-sdd/templates/evidence/import-lite.md

@@ -25,4 +25,4 @@
 ## Notes
 
 - This file is a pointer artifact for preflight, not requirement/spec SSOT.
-- Reflect unresolved items into require/spec Open Questions.
+- Reflect unresolved items into discussion/spec Open Questions.

package/assets/init/.qfai/assistant/skills/qfai-sdd/templates/specs/_policies/11_Slice-Policy.md

@@ -0,0 +1,50 @@
+# 11 Slice Policy
+
+> **SSOT notice**: The runtime slice policy is `.qfai/specs/_policies/11_Slice-Policy.md`. This template seeds the initial file used by that runtime policy.
+
+Define the slicing model for `.qfai/specs/`. This template is used by `/qfai-sdd` when creating or updating the runtime slice policy file.
+
+## Slice categories
+
+| Category   | Slice Rule                     | ID Range        |
+| ---------- | ------------------------------ | --------------- |
+| structural | 1 pack-type = 1 spec           | spec-0001..0002 |
+| cli        | 1 command = 1 spec             | spec-0003..0007 |
+| skill      | 1 skill = 1 spec               | spec-0008..0014 |
+| agent      | all agents = 1 collective spec | spec-0015       |
+
+## Category definitions
+
+- `structural`: framework-level pack definitions such as spec-pack and discussion-pack.
+- `cli`: commands implemented under `packages/qfai/src/cli/commands/`. One command maps to one spec.
+- `skill`: skills defined under `packages/qfai/assets/init/.qfai/assistant/skills/`. One skill maps to one spec.
+- `agent`: sub-agents defined under `packages/qfai/assets/init/.qfai/assistant/agents/`. All agents are grouped into one shared spec unless a newer approved slice policy says otherwise.
+
+## Create / Update / Delete decision rules
+
+| Operation | Trigger                                                               | AskUserQuestion |
+| --------- | --------------------------------------------------------------------- | --------------- |
+| CREATE    | New CLI command, skill, or structural pack-type with no matching spec | Required        |
+| UPDATE    | Requirements or implementation scope changed for an existing slice    | Not required    |
+| DELETE    | CLI command, skill, or structural slice was removed from the product  | Required        |
+
+### Decision procedure
+
+1. Discover the current repository subjects for each category.
+2. Match each subject to existing specs and `_policies/03_Capabilities.md`.
+3. Apply CREATE when a subject exists without a corresponding spec.
+4. Apply UPDATE when the subject exists and the spec already owns it.
+5. Apply DELETE when the owned subject no longer exists in the repository.
+
+## ID stability rules
+
+1. Keep existing IDs stable whenever the subject still exists.
+2. Follow `_policies/03_Capabilities.md` order as the SSOT for capability-to-spec mapping.
+3. Do not renumber surviving specs only to close gaps.
+4. Record any reorder or category-boundary change as an explicit Change Request plus delta entry.
+
+## Gap policy
+
+- Leave gaps after deletions unless an approved migration explicitly renumbers them.
+- Append new specs at the end of the relevant category block.
+- Do not merge or split categories implicitly; update this file first, then apply the approved structural change.

package/assets/init/.qfai/assistant/skills/qfai-sdd/templates/specs/spec/06_Test-Cases.md

@@ -4,9 +4,29 @@
 
 - Verify examples and acceptance criteria with explicit refs.
 - Include both `AC-Refs` and `EX-Ref` whenever possible.
+- Ensure test cases cover not only normal paths but also error paths, boundary values, and edge cases.
 
 ## Test Case Table (required)
 
-| TC-ID   | Level | AC-Refs | EX-Ref  | Steps   | Expected   | Notes   |
-| ------- | ----- | ------- | ------- | ------- | ---------- | ------- |
-| TC-0001 | L2    | AC-0001 | EX-0001 | <steps> | <expected> | <notes> |
+| TC-ID   | Level | AC-Refs | EX-Ref  | Type     | Steps   | Expected   | Notes   |
+| ------- | ----- | ------- | ------- | -------- | ------- | ---------- | ------- |
+| TC-0001 | L2    | AC-0001 | EX-0001 | normal   | <steps> | <expected> | <notes> |
+| TC-0002 | L2    | AC-0001 | —       | error    | <steps> | <expected> | <notes> |
+| TC-0003 | L2    | AC-0001 | —       | boundary | <steps> | <expected> | <notes> |
+
+### Type column values
+
+- `normal` — Happy path / expected successful behavior.
+- `error` — Error, failure, or invalid input path.
+- `boundary` — Boundary value (min, max, just-outside-range).
+- `edge` — Edge case (null, empty, concurrent, timing, max payload, etc.).
+
+## Quality depth guideline
+
+Each AC should have at minimum:
+
+- One `normal` test case.
+- One `error` or `boundary` test case.
+
+If an AC has only `normal` type test cases, the test case set is considered incomplete.
+Refer to `.qfai/assistant/skills/qfai-atdd/references/test-case-depth-checklist.md` for the full depth checklist.

package/assets/init/.qfai/assistant/skills/qfai-verify/SKILL.md

@@ -4,7 +4,17 @@ title: QFAI Verify (Quality Gates + Evidence)
 description: "Run and document quality gates (repo + qfai validate/report), fix until PASS."
 argument-hint: "[--auto]"
 allowed-tools: [Read, Glob, Bash, Write, TodoWrite, Task]
-roles: [DevOpsCIEngineer, QAEngineer, CodeReviewer, Planner]
+roles:
+  [
+    orchestrator,
+    delivery-planner,
+    qa-strategist,
+    devops-ci-engineer,
+    completion-reviewer,
+    qa-gatekeeper,
+    implementation-reviewer,
+  ]
+routing-profile: runtime-heavy
 mode: evidence-focused
 ---
 
@@ -32,7 +42,7 @@ QFAI Skill Body (SSOT)
 ## FORMAT SSOT (Mandatory)
 
 - **Before writing or editing any `.qfai/**` artifact\*\*, read and follow the relevant directory README template and sample:
-  - `.qfai/require/README.md`
+  - `.qfai/discussion/README.md`
   - `.qfai/specs/README.md`
   - `.qfai/contracts/**/README.md`
   - `.qfai/evidence/README.md`
@@ -97,7 +107,7 @@ Every major artifact in this stage MUST include a `## Work Orders Summary` secti
 
 ### Reviewer Gate (MUST)
 
-- Final completion gate MUST be delegated to an independent Reviewer sub-agent.
+- Final completion gate MUST be delegated to an independent `completion-reviewer`.
 - Reviewer checks (minimum):
   - Required roles were delegated (no orchestrator self-authoring).
   - DoD satisfied (validate gate, test-layer hard gate, evidence, DR-IDs).
@@ -108,15 +118,13 @@ Every major artifact in this stage MUST include a `## Work Orders Summary` secti
   - **Test-layer policy enforced**:
     - E2E/API/Integration coverage aligns with `steering/test-layers.md` and the project’s plan.
     - Do not use pyramid ratios as a gate; use floors/ratios only as signals. Coverage obligations are the gate.
-- Do not declare DONE or handoff until Reviewer returns `PASS`.
-- **All reviewers: alternative proposal obligation**:
-  - Every reviewer MUST provide a concrete alternative or fix proposal when returning FAIL. Feedback without a concrete alternative is invalid and triggers re-judgment.
-- **devils-advocate gate**:
-  - devils-advocate FAIL must include a concrete alternative proposal. Bare negation FAIL triggers re-judgment.
-  - 3 consecutive FAILs trigger advisory demotion and allow progression to the next phase.
-- **pattern-doubler gate**:
-  - Each pattern proposed by pattern-doubler must include rationale.
-  - Artifacts with no ID-bearing items (US/AC/BR/EX/TC) are marked N/A.
+- Route specialist reviewers from `.qfai/assistant/steering/agent-routing.yml`.
+- Default verify review set:
+  - `qa-gatekeeper`
+  - `completion-reviewer`
+- Add `implementation-reviewer` only when code fixes are in scope.
+- Do not declare DONE or handoff until all routed blocking reviewers return `PASS`.
+- Every reviewer MUST provide a concrete alternative or fix proposal when returning FAIL.
 
 ### Work order template (copy/paste)
 
@@ -278,7 +286,7 @@ This workflow assumes the environment _may_ support subagents (e.g., Claude Code
 
 Delegate to multiple roles and then merge the results. Use a “real‑world workflow” order:
 
-- Facilitator → Interviewer → Requirements Analyst → Planner → Architect → (Contract Designer) → Test Engineer → QA Engineer → Code Reviewer → DevOps/CI Engineer
+- discovery-analyst -> requirements-analyst -> delivery-planner -> solution-architect -> test-design-analyst -> qa-strategist -> devops-ci-engineer -> completion-reviewer / qa-gatekeeper
 
 **Pseudo‑invocation pattern** (adjust to your tool):
 
@@ -298,8 +306,8 @@ Only with explicit user approval (`Simulation mode allowed`), simulate roles by
 
 ## Completion Separation (mandatory)
 
-- Gate execution (DevOpsCIEngineer) and completion approval (CodeReviewer) must be separate.
-- QAEngineer must confirm gate coverage before approval.
+- Gate execution (`devops-ci-engineer`) and completion approval (`completion-reviewer`) must be separate.
+- `qa-gatekeeper` must confirm gate coverage before approval.
 
 ## Context Refresh (mandatory for long tasks)
 
@@ -322,7 +330,7 @@ Every 5 major actions, pause and restate:
    - `.qfai/assistant/instructions/workflow.md` (or equivalent)
 
 3. Read existing artifacts for the current work item (if present):
-   - `.qfai/require/`
+   - `.qfai/discussion/`
    - `.qfai/specs/spec-*/`
    - `.qfai/contracts/`
 
@@ -427,7 +435,7 @@ If anything fails:
 - Identify whether it’s spec mismatch, test issue, or implementation defect.
 - Fix the root cause (do not silence tests without reason).
 
-## Step 5 — Produce Evidence Summary (Planner)
+## Step 5 — Produce Evidence Summary (Delivery Planner)
 
 Output this format: