@haaaiawd/anws 2.2.6 → 2.4.0

package/templates/.agents/workflows/upgrade.md

@@ -1,182 +1,145 @@
----
-description: "在执行 anws update 后，读取 .anws/changelog/vX.Y.Z.md，判断升级等级（Minor / Major），生成人类可审核的升级计划，并路由到 /change 或 /genesis 执行后续修改。"
----
-
-# /upgrade
-
-<phase_context>
-你是 **UPGRADE ORCHESTRATOR (升级编排师)**。
-
-**核心使命**：
-在 `anws update` 执行完成后，读取 `.anws/changelog/` 中最新的升级记录，分析框架层变化对业务文档的影响，判断应走 `/change` 还是 `/genesis`，并在获得人类批准后**路由到对应工作流继续执行**。
-
-**核心原则**：
-- **changelog 是升级依据** - 不凭印象升级，必须读取 `.anws/changelog/vX.Y.Z.md`
-- **先定级，后路由** - 先判断 Minor / Major，再决定调用 `/change` 还是 `/genesis`
-- **保护业务常量** - 禁止覆盖业务域名词、业务规则、产品约束
-- **upgrade 只负责编排** - `/upgrade` 不自行绕过规范直接改文档，实际写操作必须遵循被路由工作流
-- **人类审批** - 写操作前必须先展示升级计划，等待用户批准
-
-**Output Goal**:
-- 输出升级等级：`Minor` / `Major`
-- 输出升级计划：影响范围、受影响文档、风险提示、推荐路由
-- 获得人类批准后，明确切换到 `/change` 或 `/genesis`
-</phase_context>
-
----
-
-## CRITICAL 执行顺序
-
-> [!IMPORTANT]
-> 必须严格按 Step 0 → Step 1 → Step 2 → Step 3 → Step 4 执行。
-> 禁止跳过 changelog 读取，禁止未定级先决定路由，禁止绕过人类检查点，禁止不读 `/change` 或 `/genesis` 就直接落笔。
-
----
-
-## Step 0: 定位升级输入
-
-1. 扫描 `.anws/changelog/`
-2. 找到最新的 `vX.Y.Z.md`
-3. 读取最新升级记录，提取：
-   - 文件级变更
-   - 内容级变更详情
-   - 可能受影响的 workflow / skill / template
-4. 扫描 `.anws/` 目录，找到最新的架构版本 `v{N}`
-5. 设定上下文变量：
-   - `LATEST_CHANGELOG = .anws/changelog/vX.Y.Z.md`
-   - `CURRENT_ARCH = .anws/v{N}`
-
-**若缺失任一目录**：停止并提示用户先运行 `anws update` 或 `/genesis`。
-
----
-
-## Step 1: 升级定级 (Minor / Major)
-
-> [!IMPORTANT]
-> 升级类型由 AI 判断，不从 changelog 静态读取。
-> **不再使用 Patch 级别**，只保留 `Minor` 与 `Major`，因为 `/upgrade` 的目标是决定跳转逻辑，而不是表达实现细粒度。
-
-使用以下判定规则：
-
-| 级别 | 判定标准 |
-|------|---------|
-| Minor | 变更可在当前版本内通过 `/change` 完成，不需要创建新的架构版本 |
-| Major | 版本目录规则变化、核心工作流协议变化、架构边界变化、需要新版本承载 |
-
-### 强制评估问题
-
-1. 是否改变版本目录或核心路径约定？
-2. 是否改变多个工作流的执行协议？
-3. 是否影响 `01_PRD.md`、`02_ARCHITECTURE_OVERVIEW.md`、`03_ADR/` 的结构语义？
-4. 是否需要保留旧版架构文档作为兼容参考？
-
-**判定逻辑**：
-- 影响局部文档、无需新版本 → `Minor`
-- 需要新版本承载、会改变架构语义或目录协议 → `Major`
-
----
-
-## Step 2: 影响分析与路由建议
-
-1. 读取 `CURRENT_ARCH` 下的以下文件（按需）:
-   - `01_PRD.md`
-   - `02_ARCHITECTURE_OVERVIEW.md`
-   - `03_ADR/*`
-   - `04_SYSTEM_DESIGN/*`
-   - `05A_TASKS.md` 与 `05B_VERIFICATION_PLAN.md`（若存在）
-2. 建立“框架变更 → 业务文档节点”的映射
-3. 识别以下三类影响：
-   - **路径迁移**：如 `.agent/` → `.agents/` 或工作流目录位置变化
-   - **流程迁移**：如新增 `/upgrade`、`anws update --check`
-   - **协议迁移**：如工作流优先原则、changelog 依赖
-4. 对每个影响点标注：
-   - 受影响文件
-   - 受影响章节
-   - 修改原因
-   - 是否涉及 AI 推断填充
-5. 生成**推荐路由**：
-   - `Minor` → 推荐调用 `/change`
-   - `Major` → 推荐调用 `/genesis`
-
-> [!IMPORTANT]
-> 此处只产出“升级计划”和“路由建议”，**不执行实际文档写入**。
-
----
-
-## Step 3: 人类检查点 
-
-> [!IMPORTANT]
-> 未经用户明确批准，禁止写任何文件。
-
-必须向用户展示以下内容：
-
-```markdown
- 人类检查点 — 升级计划确认
-
-**最新 changelog**: `.anws/changelog/vX.Y.Z.md`
-**当前架构版本**: `.anws/v{N}`
-**升级定级**: Minor / Major
-**推荐路由**: `/change` 或 `/genesis`
-
-## 受影响文件
-- `.anws/v{N}/01_PRD.md` — 原因: 路径约定变更
-- `.anws/v{N}/02_ARCHITECTURE_OVERVIEW.md` — 原因: 新增 update --check 流程
-
-## 执行策略
-- Minor: 进入 `/change`，按 `/change` 的权限边界和检查点修改
-- Major: 进入 `/genesis`，按 `/genesis` 的版本化规则创建/演进新版本
-
-## 风险提示
-- 哪些段落需要 AI 推断
-- 哪些业务常量将被保护不改
-
-请确认:  批准并路由 /  拒绝 /  调整
-```
-
----
-
-## Step 4: 路由到目标工作流
-
-### Case A: Minor → `/change`
-
-1. 接下来**必须读取**当前 target 对应的 `/change` 原生投影文件
-2. 将 Step 2 的影响分析结果带入 `/change`
-3. 后续所有修改动作必须遵守 `/change` 的权限边界、人类检查点和 CHANGELOG 记录规则
-4. 若在 `/change` 评估中发现超出其权限边界，立即终止并改走 `/genesis`
-
-### Case B: Major → `/genesis`
-
-1. 接下来**必须读取**当前 target 对应的 `/genesis` 原生投影文件
-2. 将 Step 2 的影响分析结果作为新版本演进输入带入 `/genesis`
-3. 后续版本复制、文档演进、Manifest/ADR 变更必须遵守 `/genesis` 的版本管理逻辑
-
-### AI 推断填充规则
-
-若某段内容需要 AI 基于上下文补全，必须在段前加入：
-
-```markdown
-> [!WARNING]
-> AI 推断填充，请人类复核。
-```
-
-### 业务常量保护规则
-
-以下内容禁止被框架升级覆盖：
-- 业务领域术语
-- 产品目标
-- 用户故事中的业务意图
-- 团队特定约束
-- 自定义系统边界
-
----
-
-## 完成报告
-
-完成路由后，向用户输出：
-- 升级级别
-- 推荐路由 (`/change` 或 `/genesis`)
-- 计划影响的文件列表
-- 是否预计创建新版本
-- 是否存在 AI 推断填充风险
-- 下一步必须读取的工作流文件
-
+---
+description: "【ALPHA】/upgrade：anws update 后读 changelog、定级 Minor/Major、产人类可审计划并路由到 /change 或 /genesis；宿主不替下游落盘长模板。"
+---
+
+# /upgrade (ALPHA)
+
+<phase_context>
+你是 **UPGRADE ORCHESTRATOR（升级编排师）— ALPHA 线**。
+
+**使命**：在 **`anws update` 已完成**的前提下，读取 `.anws/changelog/` 最新记录，判断 **Minor / Major**，生成可审升级计划，并在人类批准后**路由**到 `/change` 或 `/genesis`（由目标工作流执行写盘）。  
+**能力**：定位 changelog 与当前 `v{N}`、定级、框架→业务文档影响映射、路由建议、推断段 WARNING 标记规范。  
+**限制**：`/upgrade` **只做编排与计划**；**禁止**跳过 changelog、未定级先选路由、未经批准写业务文档；**不**在本文件嵌入完整「人类检查点」长 Markdown 样例。  
+**与用户的关系**：先展示计划再行动；批准后显式声明下一工作流。  
+**Output Goal**：定级（Minor/Major）+ 影响列表 + 推荐路由 + 推断风险提示；**不**自带完成业务文档修改。
+</phase_context>
+
+---
+
+## CRITICAL 凝练与版式（/craft + /challenge 思想）
+
+> [!IMPORTANT]
+> **craft**：改稿前 Read shipped `.agents/skills/craft-authoring/SKILL.md` 与 `.agents/workflows/craft.md`；各 `## Step …` 使用 **`### 做什么` / `### 为什么` / `### 怎么验收`**；`<completion_criteria>` 必填。  
+> **凝练**：计划与汇报 **一句一事**；定级规则与执行序 **不得削弱** shipped `templates/.../upgrade.md` 的硬约束。  
+> **不注入**：人类检查点展示**须含职能**（changelog 路径、当前 `v{N}`、定级、推荐路由、受影响文件+原因、推断风险提示、批准/拒绝/调整）——**不**粘贴整页 fenced 模板。
+
+---
+
+## CRITICAL 执行顺序（不可削弱）
+
+> [!IMPORTANT]
+> 严格 **Step 0 → 4**；**禁止**跳过 changelog 读取、**禁止**未定级先定路由、**禁止**绕过人类批准直接按记忆改 `.anws/v{N}`、**禁止**不读目标工作流就开写。
+
+---
+
+## Step 0: 定位升级输入
+
+### 做什么
+
+1. 列出 `.anws/changelog/` 下**最新** `vX.Y.Z.md`，设 `LATEST_CHANGELOG`；**须真实读入**（会话留一行路径或摘要，禁止口头假设）。  
+2. 扫描 `.anws/` 最大 `v{N}` → `CURRENT_ARCH = .anws/v{N}`。  
+3. 缺 `changelog` 或无可读版本：停止，提示先 `anws update` 或 `/genesis`。
+
+### 为什么
+
+无 changelog 则无升级事实源。
+
+### 怎么验收
+
+- 会话含 `LATEST_CHANGELOG` 与 `CURRENT_ARCH`；能引用 changelog 中至少一类变更（文件级/内容级/影响面）。
+
+---
+
+## Step 1: 升级定级（Minor / Major）
+
+### 做什么
+
+按 shipped `upgrade` 规则（**仅** Minor/Major）：评估是否需**新架构版本**、是否动目录/多工作流协议、`01`/`02`/`03` **结构语义**、是否需保留旧版兼容叙事。逐条记录**是/否+一句理由**。
+
+### 为什么
+
+定级决定路由，Patch 级不在本工作流展开。
+
+### 怎么验收
+
+- 输出明确 `Minor` 或 `Major` 与判定依据；无「未定级就推荐 forge」类跳跃。
+
+---
+
+## Step 2: 影响分析与路由建议
+
+### 做什么
+
+1. 按需读 `CURRENT_ARCH` 下 `01`、`02`、`03`、`04`、`05A`、`05B`。  
+2. 建「框架变更 → 业务节点」映射；标注路径迁移 / 流程迁移 / 协议迁移三类之一或多类。  
+3. 产出**推荐路由**：Minor → **`/change`**；Major → **`/genesis`**。  
+4. **本步不写盘**业务文档——只出计划与文件级意图列表。
+
+### 为什么
+
+upgrade 与执行工作流解耦，避免双写。
+
+### 怎么验收
+
+- 影响列表每条含：文件、章节/意图、是否可能 AI 推断填充。
+
+---
+
+## Step 3: 人类检查点
+
+### 做什么
+
+向用户展示 **CRITICAL 不注入** 所列职能（changelog、`v{N}`、定级、路由、受影响文件、推断风险、批准/拒绝/调整）。**未经明确批准禁止写任何文件。**
+
+### 为什么
+
+人类是升级风险的最后闸门。
+
+### 怎么验收
+
+- 用户可见完整决策包；拒绝时零写盘。
+
+---
+
+## Step 4: 路由到目标工作流
+
+### 做什么
+
+- **Minor**：读取宿主挂载的 **`/change`** 工作流（若使用 `templates_alpha` overlay 则读同树 `change.md`），将 Step 2 映射作为输入；后续**全部**遵守 `/change` 权限与签名；若执行中发现超出 `/change` → 停止并改 `/genesis`。  
+- **Major**：读取 **`/genesis`**（同上 overlay 规则），将 Step 2 作为新版本演进输入；遵守 Copy & Evolve 与版本规则。  
+- 需 AI 补全且非纯机械替换的段落：段前加 shipped 规定的 **`> [!WARNING] AI 推断填充，请人类复核。`**  
+- **业务常量**（领域术语、产品目标、用户故事业务意图、团队约束、自定义边界）**禁止**被框架升级覆盖。
+
+### 为什么
+
+写盘语义归属目标工作流，upgrade 只交接。
+
+### 怎么验收
+
+- 会话声明下一工作流名；若已批准，目标工作流的读盘与门禁已被引用。
+
+---
+
+## Step 5: 完成报告
+
+### 做什么
+
+向用户汇总：定级、路由、计划影响文件、是否预计新 `v{N}`、推断填充风险、**下一步须先读**的工作流路径。
+
+### 为什么
+
+闭环可审计。
+
+### 怎么验收
+
+- 报告含上述六项；无静默跳转。
+
+---
+
+<completion_criteria>
+- **凝练与版式**：各 Step 三小节齐备；执行序与定级闸门未削弱。  
+- changelog 已真实读取；`Minor`/`Major` 与路由一致。  
+- Step 3 人类批准或拒绝路径正确；批准前零业务写盘。  
+- Step 4 显式绑定 `/change` 或 `/genesis` 并引用其权限；WARNING/业务常量规则已声明。  
+- Step 5 完成报告已交付。  
+</completion_criteria>

package/templates_en/.agents/skills/anws-system/SKILL.md

@@ -0,0 +1,108 @@
+---
+name: anws-system
+description: Use when users in a skills-only environment need to decide which anws workflow to start from, or need routing among forge / change / genesis / probe / blueprint / challenge / upgrade. It is the navigation entry for the anws workflow set.
+---
+
+# ANWS System Router Manual
+
+You are the **ANWS Router**.
+
+Your responsibility is not to replace all workflows directly, but to serve as the unified navigation entry in **skills-only mode**:
+
+- Decide which workflow the current request should route to
+- Tell the model which `references/*.md` still need to be read
+- Clarify permission boundaries before execution to avoid mixing `/forge`, `/change`, and `/genesis`
+
+## Activation Rules
+
+Use this skill when any of the following applies:
+
+1. The user does not know which workflow to start with
+2. The user explicitly mentions workflows like `/quickstart`, `/forge`, `/change`, `/genesis`, but the current environment only has skills
+3. The user asks "which process should we follow next"
+4. The request spans multiple phases (design, tasking, implementation) and phase boundaries must be identified first
+
+## Mandatory Steps on First Activation
+
+1. Read `references/quickstart.md` first
+2. Judge which scenario the current request is closest to
+3. Then read corresponding workflow references as needed
+4. Before finishing required references, do not directly execute write operations of that workflow
+
+## Workflow Map
+
+- `references/quickstart.md`
+  - Purpose: global entry. Determine current project stage and which workflow to call first
+- `references/probe.md`
+  - Purpose: system risk probing before taking over legacy projects or major changes
+- `references/genesis.md`
+  - Purpose: new project, major refactor, architecture upgrade, or when a new version is needed
+- `references/design-system.md`
+  - Purpose: add detailed design docs for a single system
+- `references/blueprint.md`
+  - Purpose: decompose architecture design into executable `05A_TASKS.md` and `05B_VERIFICATION_PLAN.md`
+- `references/challenge.md`
+  - Purpose: systematically challenge design or task list before coding
+- `references/forge.md`
+  - Purpose: execute coding tasks according to `05A_TASKS.md` and verification commitments in `05B_VERIFICATION_PLAN.md`, and maintain Wave progress
+- `references/change.md`
+  - Purpose: only fine-tune existing task definitions; do not create new tasks; do not advance implementation status
+- `references/explore.md`
+  - Purpose: research, explore, diverge and converge solution options
+- `references/craft.md`
+  - Purpose: create workflows, skills, or prompts; scaffolds in **`craft-authoring`** (`skills/craft-authoring/SKILL.md`); shared persisted spec, delegation, and single-writer rules in **`output-contract`** (`skills/output-contract/SKILL.md`).
+- `references/upgrade.md`
+  - Purpose: handle upgrade orchestration after `anws update`
+
+## Routing Rules
+
+### Route 1: Uncertain starting point
+
+If the user does not know where to start, or you are unsure about the current stage:
+
+1. Read `references/quickstart.md`
+2. Determine the entry point based on project status
+3. Recommend a workflow and explain why
+4. Wait for user confirmation
+
+### Route 2: Request is "start coding / continue implementation / do current wave"
+
+1. Read `references/forge.md`
+2. Check whether `.anws/v{N}/05A_TASKS.md` and `.anws/v{N}/05B_VERIFICATION_PLAN.md` exist and are defined
+3. If task list is missing, do not implement directly; return to `blueprint` or `genesis`
+4. If tasks include **E2E / browser-assisted manual validation**: read the **`e2e-testing-guide`** skill under `skills/e2e-testing-guide/SKILL.md` (paths follow the target IDE projection).
+5. Before commit when **static contract fidelity** is required: read **`code-reviewer`**. **If the host provides Agent / Task / subagent delegation tools**, you **must prefer** running that skill **via delegation** (same checklist and outputs; orchestrator retrieves the body and persists). **If none exist**, run **`code-reviewer` in this session**—requirements do not shrink.
+
+### Route 3: Request is "fine-tune existing tasks / fix wording / adjust acceptance criteria"
+
+1. Read `references/change.md`
+2. Confirm only existing tasks are modified, no new tasks added
+3. If new tasks are needed or scope exceeds PRD, route to `genesis`
+
+### Route 4: Request is "new project / major refactor / new version / architecture upgrade"
+
+1. Read `references/genesis.md`
+2. Enter versioned architecture workflow
+
+### Route 5: Request is "research first / probe risk first"
+
+- Legacy project or before major change -> `references/probe.md`
+- Technical research or solution exploration -> `references/explore.md`
+
+## Boundary Rules
+
+1. You are the navigation layer, not a substitute for all workflows
+2. Route to only one primary workflow at a time; if chaining is needed, explain the sequence
+3. Before reading target workflow references, do not pretend the process is already followed
+4. If the user request spans both design and implementation, converge boundaries first, then choose `/change` or `/genesis`
+5. In skills-only mode, workflow details are all under `references/`
+
+## Output Format
+
+When routing decision is complete, output should include:
+
+- Current stage judgment
+- Recommended references to read
+- Recommended workflow to enter
+- Why other workflows are not chosen
+- Whether user confirmation is needed

package/templates_en/.agents/skills/code-reviewer/SKILL.md

@@ -0,0 +1,170 @@
+---
+name: code-reviewer
+description: 【ALPHA】Pure static "contract fidelity / implementation-side evidence" review: Against PRD, ADR, system design, 05A_TASKS, and 05B_VERIFICATION_PLAN, produce traceable conclusions on contract closure, task fulfillment, architecture health, security boundaries, verification evidence, and backflow consistency; shared by /challenge (CODE/FULL) and /forge (Step 3 §3.6 end-of-wave).
+---
+
+# Code Reviewer — Implementation-side evidence layer【ALPHA】
+
+You are **CODE REVIEWER**. Your job is not a generic PR review or style grading, but to answer with purely static evidence: **whether the implementation faithfully fulfills commitments in PRD / ADR / System Design / 05A_TASKS / 05B_VERIFICATION_PLAN; if not, what the risks are and where the evidence is.**
+
+## CRITICAL methodological anchors
+
+- **Static-only is the boundary**: Only readable artifacts and code form are admitted; anything that depends on processes, networks, browsers, or real runtime ordering must be labeled **cannot be confirmed via static review** or **requires manual verification**, and must not be written as proven.
+- **Contract over intuition**: Ordering and wording follow PRD / ADR / System Design / `05A_TASKS.md` / `05B_VERIFICATION_PLAN.md` / this round’s task description; preference-driven criticism without an anchor must not appear in strong conclusions.
+- **Evidence tiers**: Assertions such as Critical / High / Fail / Pass must cite `**path:line**`; without a location, downgrade to “suspected” or “cannot confirm”; do not overstate certainty.
+- **Root cause over stacking**: Merge similar issues into a fixable root cause; do not inflate severity by duplicate entries.
+
+## Hard boundaries (must follow)
+
+- **Pure static**: Do not start the project, run Docker, auto-run tests, modify code, or connect to external services.
+- **No exaggeration**: For runtime, network, browser, and external-integration conclusions, only write **cannot be confirmed via static review** or **requires manual verification**.
+- **Evidence**: Strong conclusions such as Critical / High / Pass / Fail must include `**path:line**`. Without evidence, downgrade to “suspected” or “cannot confirm”.
+- **Anchors**: Judgments must trace back to PRD / ADR / System Design / `05A_TASKS.md` / `05B_VERIFICATION_PLAN.md` / this round’s task description.
+
+## Severity levels (aligned with challenge reports)
+
+**Critical / High / Medium / Low** (consistent with `/challenge`). **Critical** = blocking severity where merge or delivery should not proceed without fixing (other processes’ Blocker bar can align with this).
+
+## Activation moments
+
+- `**/challenge`**: When `REVIEW_MODE` is `CODE` / `FULL`, or when design/task review **adaptively escalates** to the implementation side.
+- `**/forge`**: Step 3 **§3.6 end-of-wave gate** (mandatory after this wave’s last task’s §3.5 submission completes; **once per wave** by default). After §3.6, `/forge` also has **§3.8 delivery index table** (workflow rule; **not** the body of this skill’s report, **must not** substitute that table for the review narrative).
+
+## Execution shape and sub-agent orchestration (delegate to AGENT when available)
+
+### What to do
+
+- **Preferred**: If the host exposes delegatable **Agent / Task / sub-agent** (collectively **AGENT tools**), this skill **must** be executed by AGENT dedicated to it; the orchestration side prepares input, sends full skill constraints and output shape, invokes AGENT, and places the review body **verbatim** at the `/forge`-mandated path (or authorizes AGENT to write there directly). Output format, Lenses, and evidence rules come from this file; AGENT **must not** trim them on its own.
+- **Sub-agent orchestration (consistent with AGENT-first)**: Orchestration delivers in one shot: "required-reading input list + hard boundaries + six-section output template + on-disk path and first-line format + Issues field contract"; after handoff only do **structured acceptance** (see handoff checklist / completion_criteria), and do not replace Lens walk-through inside the orchestration session.
+- **Fallback**: Only when the host has **no** delegatable AGENT capability does the **current session** fully execute Lenses 1–6 and produce the full text per the six-section "Output structure (concise)" below.
+
+### Why
+
+Isolate fine-grained implementation reading from orchestration context so evidence rules and Lens coverage are not diluted by “running it informally”.
+
+### How to accept
+
+**No excuses**: When AGENT tools exist, the current session **must not** substitute to save steps; **must not** lower evidence requirements, skip Lenses, or skip execution because of "insufficient context", "small change", or "time pressure". Session execution without AGENT is normal fallback. `/forge` exemption **only** when the user states it explicitly at wave sign-off.
+
+## On-disk requirements (`/forge` path mandatory)
+
+When `/forge` §3.6 fires, the full review **must** be written to a physical file:
+
+- **Path**: `{TARGET_DIR}/wave-reviews/wave-{N}-review.md` (`{N}` = current wave number).
+- **First line**: `# Wave {N} Code Review — {YYYY-MM-DD}`.
+- **Not on disk = §4.0 hard block**, no exceptions (including AUTO). Self-filled tables or verbal “reviewed” counts as not executed.
+- On user exemption, do not write the review body; instead create `{TARGET_DIR}/wave-reviews/wave-{N}-WAIVED.md` (see `/forge` §3.6 waiver protocol).
+
+When `/challenge` triggers, follow that workflow’s report path (default `07_CHALLENGE_REPORT.md` section); **do not** force `wave-reviews/`.
+
+## Handoff checklist (orchestration ↔ AGENT ↔ on disk)
+
+- [ ] Required inputs are ready, or an acceptable narrowed scope is stated.
+- [ ] AGENT was given the full constraints of this skill (or equivalent summary + explicit “must not omit sections”).
+- [ ] AGENT output includes the six-section structure; each Issue satisfies **CRITICAL: Issues output contract** below.
+- [ ] `/forge`: `wave-{N}-review.md` path and first-line format are correct, or `wave-{N}-WAIVED.md` was generated per rules.
+- [ ] Strong conclusions all have `path:line`, or were downgraded to suspected / cannot confirm.
+
+## completion_criteria (this review counts as complete if and only if)
+
+- Lenses 1–6 are all covered, or each has a one-line “not applicable”.
+- Summary is one of Pass / Partial Pass / Fail / Cannot Confirm (static sense), and matches Issue severities.
+- "Review scope and static boundaries" honestly lists unread items and items needing manual verification.
+- `/forge` on-disk requirements are met, or the user explicitly waived and completed the WAIVED flow.
+
+## Required reading
+
+1. `src/` (or the repo’s agreed implementation root)
+2. `{TARGET_DIR}/01_PRD.md`, `02_ARCHITECTURE_OVERVIEW.md`, `03_ADR/`, `04_SYSTEM_DESIGN/`
+3. `{TARGET_DIR}/05A_TASKS.md`
+4. `{TARGET_DIR}/05B_VERIFICATION_PLAN.md`
+5. If present: `{TARGET_DIR}/07_CHALLENGE_REPORT.md`
+
+When inputs are missing, narrow scope and state it in the output.
+
+## Thinking order (risk first, conclusions back to contract)
+
+README/config → entry / routing / CLI → authn & authz → core business / data model → admin / debug endpoints → tests / logs → UI (if applicable).
+
+## Review surfaces (Lenses 1–6)
+
+### What to do
+
+Review each of the six lenses in turn; the final report may organize by findings, but each lens needs a conclusion or one line “not applicable”.
+
+### Why
+
+Keep contract, delivery, architecture, static safety, verification, and backflow under one evidence standard.
+
+### How to accept
+
+When applicable, give conclusion + evidence; when not, one sentence. Strong conclusions carry `**path:line**`.
+
+### Lens 1: Contract fidelity
+
+Do public API / CLI / config keys / layout / error semantics match PRD, ADR, System Design? Any public contract introduced without backflow? Typical: `Contract Drift`, `Undocumented Contract`, `Static Verifiability Gap`.
+
+### Lens 2: Task fulfillment and delivery closure
+
+Do `05A_TASKS.md` outputs, acceptance, and boundaries, and `05B_VERIFICATION_PLAN.md` plans/evidence requirements, have real implementation/test/docs coverage? Are Mock/Stub/Hardcode boundaries clear; could they leak into production paths? Typical: `Task Drift`, `Acceptance Gap`, `Mock Boundary Risk`.
+
+### Lens 3: Architecture fit and complexity health
+
+Module boundaries, dependency direction, data model, state flow vs Architecture / System Design? Single-file pile-up, over-abstraction, high coupling, or hard-to-test structure? For UI, only structure and interaction that affect usability; no pure aesthetic demerits. Typical: `Architecture Drift`, `Complexity Risk`, `Maintainability Gap`.
+
+### Lens 4: Runtime risk from static evidence and security boundaries
+
+From static evidence: input validation, error paths, edge cases, duplication/concurrency, cleanup/rollback; auth entry and route/object/function-level authz, tenant isolation, admin protection, secrets and PII leakage. Security first; without direct evidence use “suspected” or “cannot confirm”. Typical: `Safety Gap`, `Auth Boundary Gap`, `Input/Error Path Risk`, `Sensitive Data Exposure`.
+
+### Lens 5: Verification evidence and observability
+
+Do test/verification entry points exist? Minimal coverage mapping for core requirements and high risk? Assertions too weak or prone to false positives? Logs diagnosable without leaking secrets? Coverage wording: `sufficient` / `basically covered` / `insufficient` / `missing` / `not applicable` / `cannot confirm`. Typical: `Test Drift`, `Foundational Test Gap`, `Observability Gap`.
+
+### Lens 6: Backflow consistency and handoff evidence
+
+Are new public behaviors synced to README / CLI help / changelog / ADR / System Design / task status? Export entry points, manifest, registry, install/update paths consistent? Enough handoff information? Typical: `Missing Change Backflow`, `Documentation Drift`, `Handoff Gap`.
+
+## Output structure (concise)
+
+1. **Summary conclusion**: Pass / Partial Pass / Fail / Cannot Confirm (static sense).
+2. **Review scope and static boundaries**: What was read, what was not, what was deliberately not run, what needs manual verification.
+3. **Contract → code mapping summary**: Core promises → corresponding implementation areas (short).
+4. **Lens result summary**: One line per applicable Lens + evidence.
+5. **Issues**: Critical → High → Medium → Low; each satisfies **CRITICAL: Issues output contract**.
+6. **Security / test coverage addendum**: Only high-risk gaps and boundaries that cannot be confirmed statically.
+
+## CRITICAL: Issues output contract (per-item fields; one-line fill preferred)
+
+Each Issue **must** be parseable into the following fields (one line with `|` separators or fixed subheadings; information must not be missing):
+
+`Severity` | `Lens` | `Title` | `Evidence` (`**path:line**` or multiple points) | `Impact` | `Minimum fix` | `Anchor` (corresponding PRD/ADR/Task/Verify snippet or section pointer)
+
+- **Severity**: Critical / High / Medium / Low.
+- **Lens**: L1–L6 or combination (e.g. L2+L4).
+- **Title**: Short label; may include typical type names (e.g. `Contract Drift`).
+- **Evidence**: If not locatable, write "no static location: suspected" and downgrade severity or move to a “cannot confirm” style note; do not fake Critical/High.
+- **Concision (challenge-aligned)**: `Title` / `Impact` / `Minimum fix`—**one sentence each** (tiny compound allowed); **merge** same root cause; no synonym spam for count.
+
+## Discipline (self-check before strong conclusions)
+
+### What to do
+
+Before issuing Critical/High, Pass, Fail, or equivalent strong assertions, check each item below.
+
+### Why
+
+Prevent packaging speculation, repeated symptoms, or personal preference as blocking conclusions.
+
+### How to accept
+
+If any check fails, reword or downgrade.
+
+- Is there a direct `**path:line**`?
+- Is this a static fact, or implying unproven runtime behavior?
+- Is this the root cause or a repeated symptom?
+- Is the judgment anchored to PRD/tasks/ADR, not personal preference?
+- When uncertain, is it written as **cannot be confirmed via static review**?
+
+## Skip protocol
+
+If there is no `src/` or scope does not apply: output `Code review skipped` + one line of reason; do not fabricate review results.