@chenmk/superflow 0.5.12 → 0.5.14

package/assets/skills/superflow-implement/SKILL.md

@@ -83,6 +83,14 @@ Before implementation phase can exit:
 ## Prompt Splitting Rules
 
 - Split by independently verifiable feature closure, not by a giant backend module.
+- Read `tasks.md` predecessor links before splitting or dispatching. A batch is
+  executable only when its real predecessors are complete; independent batches
+  may run in parallel. Keep the existing checkbox and machine category as the
+  progress authority. An implementation discovery that changes a confirmed
+  owner decision or task acceptance requires canonical document repair,
+  handoff refresh, document review, and a new Coding Ready receipt before the
+  affected batch resumes. A free-text `Blocked by` note cannot downgrade a
+  locally executable task to an external category.
 - Use P0 for baseline only: worktree, branch, compile, SQL/entity/mapper/DTO consistency.
 - Keep business prompts small and explicit.
 - Every implementation prompt must begin with the mandatory superpower/team

package/assets/skills/superflow-pipeline/SKILL.md

@@ -7,6 +7,16 @@ description: SDD 总入口和路由技能。Use when the user says sdd, SDD 流
 
 This is the lightweight SDD router. Do not load the whole SDD process when a smaller phase skill is enough.
 
+## 角色 owner 与互斥边界
+
+| 角色 | 唯一 owner | 可以做 | 不可以做 |
+|---|---|---|---|
+| Host | 目标、澄清、监督、语义评审 | 选择路由、判断代码/架构风险、决定是否接受反馈 | 代替 Executor 写业务代码；启动嵌套 Supervisor |
+| Executor | 源码实现、测试、启动、整改 | 在冻结合同内修改工作区并提交证据 | 改合同/状态；自行降低验收等级；自动提交发布 |
+| Runner | 状态、哈希、退出码、证据持久化 | 调度、等待、恢复、确定性门禁 | 猜业务语义；重新设计 API/DB；代替 Host 评审 |
+
+菜单或 persona 只改变入口表达，不创建第二套状态机；发现 owner 冲突时回到 Host 澄清。
+
 维护 Superflow CLI 自身时，必须先读取仓库根目录
 `docs/superflow-cli-design-principles.md` 和
 `docs/superflow-cli-evaluation-framework.md`；涉及托管时再读取托管专项宪章与协议。
@@ -31,6 +41,9 @@ CLI 命令：
 停止，不创建托管 Task；已有用户评审通过的 change/Prompt 时原路径冻结；只有边界清晰的
 口头小需求才直接进入 minimal/standard。若口头需求存在会改变实现方向的 API、数据、
 并发、跨仓或 owner 决策，先进入澄清/文档流程，不能把这一判断交给 Runner 正则或 Executor。
+多页面、跨前后端、数据库字段横向改造要求闭环交付时，按完整 SDD 冻结真实入口、
+`tests.md` 和逐入口验收合同。仅有 OpenSpec `spec/design/tasks` 或通用冒烟报告，
+只能声明相应范围已实现/验证，不得称为完整 SDD 验收。
 
 ```bash
 superflow pipeline "<implementation-prompt 路径、change 目录或简单任务>" --managed --project "<项目根目录>" \
@@ -86,6 +99,12 @@ JSON，再用 `--resume-task <task-id> --submit-host-review <json>` 恢复。监
 
 ## Choose The Right SDD Skill
 
+- Use `$superflow-quick-dev` for one clear, low-blast-radius code change along
+  an existing test seam. Run its narrow gate before editing; API, DB, security,
+  transactions, concurrency, payments, cross-module work, a related active SDD
+  change, or uncertain ownership goes to `$superflow-clarify` instead. Reuse
+  the current quick managed profile after the Quick Spec is approved; do not
+  create a second managed-task state machine.
 - Use `$openspec-explore` when the user is still thinking through a requirement,
   wants to compare options, or needs codebase investigation before a change.
 - Route to `$superflow-clarify` in embedded deep-clarification mode when the
@@ -163,6 +182,12 @@ When the router selects embedded deep clarification, follow these rules inside
 - Exit deep clarification once only repository-verifiable facts remain and the
   user confirms the current feature understanding. Continue through the existing
   clarify workflow without invoking an external skill dependency.
+- After all relevant features are frozen, show the change-level Consensus
+  Summary from `superflow-clarify` before OpenSpec artifact generation. Carry
+  confirmed choices into canonical `proposal.md`; do not create a parallel
+  alignment fact source or add approval prompts for decisions already explicit
+  in the user's request. A changed owner choice requires affected documents,
+  handoff, review, and Coding Ready to be refreshed before execution.
 
 ## Non-Negotiable Rules
 
@@ -326,6 +351,11 @@ context_compression | created_at | verified_at | updated_at`
 
 - 设计范围、API/DB 合同、执行模式、验证边界、archive closeout 需要用户取舍时，
   使用 `references/decision-point.md` 的格式记录，不用聊天记忆代替。
+
+仓库级术语或持久架构决定可运行 `superflow context-audit --path <project>`
+做只读检查。业务术语放入 `CONTEXT.md`；只有难逆转且存在有意义备选方案时才写 ADR。
+审计是 advisory：缺少 ADR 只提醒，ADR 字段不全或术语混入实现细节才报告修复项，
+不新增审批门禁，也不替代 `design.md`。
 - 脏工作区、并行 worktree 或跨会话接力前，使用 `references/dirty-worktree.md`
   的检查项，避免覆盖用户改动或拿旧 handoff 继续。
 - `build_mode: subagent-driven-development` 时，维护
@@ -431,7 +461,7 @@ Mapper/XML、数据库样例、接口契约和真实消费入口验证。
 **影响面发现方式（优先但可降级）：**
 
 - 优先检查当前仓库是否存在 understand-anything 相关索引/产物，例如
-  `.understand-anything/`、`understand-anything.md`、`understand anything.md`、
+  `.ua/`、`.understand-anything/`、`understand-anything.md`、`understand anything.md`、
   或项目约定的 understand-anything 图谱文件。
 - 如果索引存在，可以用它快速定位整个平台相关代码，包括当前仓库、同平台
   sibling repo、共享 SDK、公共 DTO、数据库表消费方、定时任务、回调入口、外部平台

package/assets/skills/superflow-pipeline/references/document-review-coverage.md

@@ -24,6 +24,14 @@
    同页面不同角色的分支用不同入口 ID；不得只测试开发者选中的新接口。
 5. 验收同时断言返回、最终状态和禁止的副作用。只读/纯函数场景也要给出明确
    无持久化变更的理由，不能留空。browser 入口必须有 browser 验收，API 测试只能补充。
+6. 横向页面/字段改造先从前端路由、列表列和可点击动作建立页面入口清单，再从后端
+   Controller、Mapper statement、全部写入方反向建立清单。逐项对账后才生成 tasks/tests；
+   相似方法名不得作为调用关系证据。每个列表记录实际页面请求、Controller、Service、
+   精确 Mapper statement、返回 DTO 与表；每个管理按钮记录实际 mutation API 和写入点。
+   未定位或两份清单不一致的入口标记待调查，不得以总数相等替代逐项核验。
+7. 数据库字段展示用例必须分别覆盖一个非空记录与允许的历史 NULL 记录。非空记录
+   用同一业务 ID 比较 API 字段与数据库真源值，并在浏览器核对展示；仅有字段键、HTTP
+   200、空列表或历史 NULL 不能证明字段链路正确。修改入口还要验证写后读及禁止的副作用。
 
 ## 凭证扩展
 
@@ -80,3 +88,13 @@ coding-ready 复检失败会写 BLOCKED 撤销旧 READY。
 验收阶段沿同一 R/E/C ID 在 `test-report.md` 回填实际角色、动作、请求路由、状态/
 副作用断言、原始日志路径及结果。主 Agent 对照原始来源和实际证据，不只看测试汇总。
 计划用例不是执行证据；API-only 不能报页面 E2E 成功。脚本未验证业务语义的部分不得宣称自动保证。
+完整 SDD 报告使用唯一的逐用例执行证据表，供现有 report lint 核对冻结索引：
+
+| 用例 ID | 入口 ID | 验收级别 | 结果 | 证据路径 |
+|---|---|---|---|---|
+| C1 | E1 | browser | PASS | logs/C1-trace.zip |
+
+结果只用 `PASS/FAIL/BLOCKED/PARTIAL`；证据路径指向实际命令输出、浏览器 trace、
+响应或数据库对账记录。PASS 的本地证据文件必须保留到验收并可读取；远端证据可用
+HTTP(S) 链接。阻塞项在报告正文写明原因。表格仅证明用例逐项回填，不能证明
+证据内容正确。任何非 PASS 用例存在时，不得声明整体验证 PASS。

package/assets/skills/superflow-pipeline/references/superpower-technical-design-template.md

@@ -20,6 +20,23 @@ canonical_sources:
 
 # Superpowers Technical Design: <change title>
 
+## Design-It-Twice：接口与 seam 选择
+
+仅在新增或改变模块接口时填写；纯实现内重构可写“无接口变化”。
+
+| 项目 | 方案 A | 方案 B |
+|---|---|---|
+| 接口与调用方约束 |  |  |
+| seam 与 adapter owner |  |  |
+| 真实测试面 |  |  |
+| 复用证据（文件/符号） |  |  |
+| 爆炸半径与后果 |  |  |
+
+- 选择：A / B
+- 选择理由：
+- 被拒绝方案的具体代价：
+- 两种方案不能只是改名或换参数顺序。
+
 ## Boundary
 
 - This document owns source-level HOW only.

package/assets/skills/superflow-pipeline/references/test-execution-template.md

@@ -71,6 +71,17 @@ mvn test -Dtest=UserServiceTest
 
 ## 4.1 真实测试数据与证据
 
+完整 SDD 变更还需按 `document-review.json` 的 R/E/C 索引逐项回填，不得用总通过数
+代替单个入口的结果。下表供 `superflow-test-report-lint.py --review` 校验：
+
+| 用例 ID | 入口 ID | 验收级别 | 结果 | 证据路径 |
+|---|---|---|---|---|
+| C1 | E1 | browser | PASS | logs/C1-trace.zip |
+
+`PASS/FAIL/BLOCKED/PARTIAL` 之外的值无效；非 PASS 项在正文说明原因。
+数据库字段展示要分别记录非空数据按同一 ID 的 API/DB 值对账、历史 NULL 结果和
+真实浏览器展示。API 键存在、静态前端配置与浏览器显示分别记录证据等级。
+
 | 用例 ID | 真实参数 | 参数来源 | 自动化命令 | 证据等级 | 执行证据 | 结论 |
 |---------|----------|----------|------------|----------|----------|------|
 | TC-001 | parkCode=____ | 前端请求 / 用户提供 / DB查询 / Mock | curl/Newman/pytest/RestAssured | Mock only / 测试端点 / 真实入口 | 响应 + DB + 日志 | Passed / Blocked / Partial |

package/assets/skills/superflow-pipeline/scripts/superflow-guard.sh

@@ -87,15 +87,21 @@ project_root_for_change() {
 
 require_test_report_lint() {
   local report="$CHANGE_DIR/test-report.md"
-  local root lint output
+  local root lint output review
+  local -a lint_args
   root="$(project_root_for_change)"
+  review="$CHANGE_DIR/.sdd/reviews/document-review.json"
+  lint_args=(--tests "$CHANGE_DIR/tests.md")
+  if [[ -f "$review" ]]; then
+    lint_args+=(--review "$review")
+  fi
 
   for lint in \
     "$root/.codex/hooks/superflow-test-report-lint.py" \
     "$HOME/.codex/hooks/superflow-test-report-lint.py"; do
     if [[ -x "$lint" ]]; then
       output="$(mktemp)"
-      if ! "$lint" --tests "$CHANGE_DIR/tests.md" "$report" >"$output" 2>&1; then
+      if ! "$lint" "${lint_args[@]}" "$report" >"$output" 2>&1; then
         issues+=("superflow-test-report-lint failed: $(tr '\n' ' ' < "$output" | sed -E 's/[[:space:]]+/ /g')")
       fi
       rm -f "$output"

package/assets/skills/superflow-quick-dev/SKILL.md

@@ -0,0 +1,44 @@
+---
+name: superflow-quick-dev
+description: 小型代码工作的快速准入与 Quick Spec。目标、seam 或爆炸半径不清时停止并转正式 SDD。
+license: MIT
+metadata:
+  author: superflow
+  version: '1.0'
+---
+
+# Superflow Quick Dev
+
+只接单一、低爆炸半径、沿已有 seam 的代码工作。
+
+## 窄门
+
+先查真实源码、调用方和测试，再运行
+`superflow quick "<request>" --json --path <changed-files...> --seam <existing-file>`。
+只有确认**同一行为**由在途 SDD 变更拥有时才追加 `--active-sdd`；仅复用其代码或依赖
+其产出，不等于同一变更。只有返回
+`QUICK` 才能继续。公共 API、数据库/迁移、事务、并发、锁、权限、安全、支付、跨模块、
+在途 SDD 变更或非独占工作树必须返回 `STOP`，转 `superflow-clarify`。若项目还安装了
+ake-harness 的 `ssd-propose`，不得因为它可见就自动改派；只有用户明确选择 ake SSD
+流程时才切换。
+
+## Quick Spec
+
+在 `.sdd/tasks/<YYYY-MM-DD>-<slug>/spec.md` 写入并展示当前磁盘版本，内容必须包含：
+
+- Intent / Non-goals；
+- Current behavior / Target behavior；
+- 已确认的测试 seam；
+- 按依赖排序的实施步骤；
+- Given / When / Then 验收条件；
+- 风险、升级条件和 deferred；
+- baseline commit 与精确文件范围。
+
+用户批准前不得修改运行时代码；收到当前磁盘版本的明确批准后，才运行
+`superflow quick "<request>" --path <files...> --seam <existing-file> --spec <spec> --approve`。
+CLI 将当前内容摘要绑定到批准状态。Intent、Non-goals 或 Acceptance Criteria 变化时回到
+`draft`，重新批准后才实施。按 TDD 和 verification-loop 逐项跑 RED/GREEN 信号，完成后
+复用 `superflow review-coverage` 和 `superflow eval` 的证据格式收口。
+
+Quick 路径不创建 OpenSpec checkpoint，也不自动提交、推送或发布；一旦范围扩大，保留
+spec 作为线索并转正式 SDD。

package/assets/skills/superflow-verify/SKILL.md

@@ -5,13 +5,56 @@ description: Use when an SDD/OpenSpec change is ready for verification, test-rep
 
 # SDD Verify
 
+Before semantic code review, run `superflow review-coverage --json` (or pass
+`--base <ref>` for a branch review). Use its included/excluded inventory and
+path-specific rules to account for every changed file. Review related Java
+and Mapper XML files together when they share a bundle. Do not treat this
+mechanical inventory as a correctness verdict: findings still need source
+anchors, a fresh reading of the affected call path, and independent judgment.
+
+Use the reported risk tier to choose review depth: low may be reviewed inline;
+medium requires an independent view; high requires focused coverage of the
+named high-impact paths in addition to the ordinary Host review. Every
+reviewer report starts with `DONE`, `DONE_WITH_CONCERNS`, `BLOCKED`, or
+`NEEDS_CONTEXT`. Run `superflow review-coverage --report <file> --json` to
+validate that marker and every `file:line` finding anchor. `BLOCKED` and
+`NEEDS_CONTEXT` require more evidence before triage; `DONE_WITH_CONCERNS`
+requires Host investigation of each concern before deciding. Re-read each
+anchor and attempt to disprove the finding before assigning severity. Keep
+unverifiable claims separate rather than giving them a guessed severity.
+
+## 评审回应纪律
+
+收到 finding 后先通读全部反馈，复述要求，再对照当前源码、项目规则、调用方和测试验证。
+不得只依据评审者措辞附和或直接改。确认有效时说明代码锚点和修复方向；认为有误时给出
+具体源码或合同证据及被拒绝方案。一次处理一个 finding，跑受影响信号后才能声称已修复。
+遇到“做成通用版本”等宽泛建议，先检查真实调用点；没有第二个调用方或合同要求时保持最小实现。
+项目规则只有用户确认后才回写，评审反馈不会静默改变仓库政策。
+
 完整 workflow 验收必须读取[需求×真实入口评审](../superflow-pipeline/references/document-review-coverage.md)。
 沿 R/E/C ID 在现有 test-report 回填实际角色、页面动作、请求路由、最终状态与禁止副作用
 的原始证据。独立对照原始来源及实际调用方；API-only 不得冒充 browser E2E，计划不等于执行。
+横向字段改造还要复核页面请求命中的精确 Mapper statement，并查看非空记录的同 ID
+API/DB 值对照与浏览器显示。历史 NULL、字段键存在和静态列配置各自只能支持较弱结论。
 
 Verification is a phase, not a final sentence. Use this skill after
 implementation work claims completion and before archive or delivery.
 
+## Behavior-preserving simplification
+
+After the implementation signals are green, inspect only the files changed by
+this change for missed reuse, duplicated logic, redundant defensive branches,
+and needless abstraction. Before removing a helper or guard, identify its
+callers, error behavior, side effects, and reason for existing. If that reason
+is unclear, leave it in place. Simplification is optional; clean code needs no
+ceremonial edit.
+
+Apply one small simplification at a time. Run the same tests and build signals
+without changing their assertions; if behavior, errors, or side effects change,
+revert that simplification. Keep simplification in a separate commit from the
+feature, and stop if it reaches files outside the original change. This is a
+readability pass, not a second code review or a license for unrelated cleanup.
+
 ## Entry Check
 
 Run:
@@ -99,7 +142,8 @@ Before pass:
    DB/log/API evidence, handoff hash, and phase state.
 2. `~/.codex/hooks/superflow-verify-integration.sh <test-report.md>` passes when the
    report involves integration evidence.
-3. `~/.codex/hooks/superflow-test-report-lint.py --tests <tests.md> <test-report.md>`
+3. `~/.codex/hooks/superflow-test-report-lint.py --tests <tests.md> --review
+   <.sdd/reviews/document-review.json> <test-report.md>`
    passes. If `tests.md` declares L3/L4, real-entry, third-party, device,
    database, log, or dev-tool evidence, mock-only/unit-only reports are hard
    failures unless the report is explicitly `Blocked` or `Partially verified`

package/assets/skills-en/openspec-propose/SKILL.md

@@ -22,6 +22,14 @@ When ready to implement, run /opsx:apply
 
 **Input**: The user's request should include a change name (kebab-case) OR a description of what they want to build.
 
+When this skill follows `superflow-clarify`, read the confirmed change-level
+Consensus Summary and its source evidence before creating artifacts. Put the
+goal, scope, material owner decisions, rejected alternatives, migration risks,
+and non-blocking reopen triggers in canonical `proposal.md`; cite the existing
+source audit rather than copying it. A pending owner decision that changes the
+solution blocks downstream spec/design/tasks. An explicit decision in the
+user's original request needs no second confirmation.
+
 ### Canonical OpenSpec Change Directory (Blocking)
 
 - A change must have exactly one physical directory:
@@ -111,6 +119,12 @@ After completing all artifacts, summarize:
 - Follow the `instruction` field from `openspec instructions` for each artifact type
 - The schema defines what each artifact should contain - follow it
 - Read dependency artifacts for context before creating new ones
+- For a Superflow change, keep ordinary OpenSpec checkbox task syntax. For each
+  independently verifiable task, state its observable delivery and acceptance
+  signal; record real predecessor task IDs as `Blocked by` and link the relevant
+  requirement/scenario or traceability row. A free-text blocker does not change
+  Superflow's `[local_required]` / `[environment_required]` /
+  `[release_required]` machine classification. Do not invent dependencies.
 - Use `template` as the structure for your output file - fill in its sections
 - **IMPORTANT**: `context` and `rules` are constraints for YOU, not content for the file
   - Do NOT copy `<context>`, `<rules>`, `<project_context>` blocks into the artifact

package/assets/skills-en/superflow-clarify/SKILL.md

@@ -13,7 +13,7 @@ any full SDD document or implementation prompt is generated.
 
 - Before freezing any new requirement or bug fix, perform platform-level impact
   discovery. Prefer understand-anything artifacts when present, such as
-  `.understand-anything/`, `understand-anything.md`, `understand anything.md`,
+  `.ua/`, `.understand-anything/`, `understand-anything.md`, `understand anything.md`,
   or the project-specific graph output, but treat them only as locator evidence.
   If artifacts are missing, stale, incomplete, or unavailable, degrade to `rg`,
   source reading, Mapper/XML, configuration, database table names, log
@@ -174,6 +174,28 @@ Use the detailed workflow in:
     indexed feature. Do not batch-confirm multiple features from memory or a
     compressed chat summary.
 
+### Change-level alignment before OpenSpec artifacts
+
+After all in-scope features are frozen, present one concise change-level
+`Consensus Summary` covering the goal, in/out scope, verified source facts,
+owner decisions with rejected alternatives, production and migration risks,
+and remaining non-blocking questions with owner and reopen trigger. Derive it
+from `source-code-audit.md`, `feature-gates.md`, and the confirmed decisions in
+`.sdd/handoff/brainstorm-summary.md`; the summary is a handoff view, not a
+second source of requirements. Record unresolved decisions as blockers and
+ask only the next owner decision. Do not ask the user to reconfirm facts or
+choices already explicit in the original request.
+
+When a material owner choice was made during clarification, show the current
+summary before generating OpenSpec artifacts. Reuse an explicit confirmation
+already given for that exact choice and summary; ask once only if the summary
+introduces a new choice or changes its consequence. The later `proposal.md`
+must carry the confirmed change-level
+decisions and cite their evidence. If a later finding changes a confirmed
+choice, return to that decision, update affected canonical documents, refresh
+the handoff hash and repeat document review and Coding Ready. Task checkbox
+progress alone does not change the agreed contract.
+
 ## Handoff
 
 When the feature is frozen, tell the user to continue with `$openspec-propose`

package/assets/skills-en/superflow-design/SKILL.md

@@ -58,7 +58,12 @@ Superpowers own source-level HOW without weakening OpenSpec/SDD contracts.
 5. Write:
    `docs/superpowers/specs/YYYY-MM-DD-<change-id>-technical-design.md`
    using `../superflow-pipeline/references/superpower-technical-design-template.md`.
-6. Complete `Minimal Design Review` before expanding source-level HOW. For
+6. For each new or changed module interface, add a `Design-It-Twice` card to
+   the technical design: two materially different interfaces, seam and adapter
+   ownership, the real test surface, reuse evidence, caller consequences, and
+   an explicit choice. Renaming the same interface does not count as two
+   options; revisit it. This card is advisory and does not add a state gate.
+7. Complete `Minimal Design Review` before expanding source-level HOW. For
    every new table, field, API, service/component, abstraction, cache,
    async/MQ/event flow, scheduled job, or compatibility layer, prove existing
    reuse options, necessity, and why the simplest direct implementation is

package/assets/skills-en/superflow-docs/SKILL.md

@@ -38,6 +38,12 @@ Do not invent new fields or requirements.
 
 For complex requirements, verify these before writing docs:
 
+- `proposal.md` carries the confirmed change-level Consensus Summary from
+  clarification: goal, scope, owner choices and rejected alternatives,
+  production/migration risks, and non-blocking reopen triggers. Check it
+  against original sources and `source-code-audit.md`; do not treat the
+  handoff summary as a second authority. If a material choice remains open,
+  return to `$superflow-clarify` before finalizing design or tasks.
 - `source-ingestion.md`, `feature-inventory.md`, and `feature-gates.md` prove
   that the original PRD/Lark/Feishu/screenshots were processed section by
   section, not summarized into tasks in one whole-document pass. Each feature
@@ -236,6 +242,12 @@ applicable:
   `[local_required]`, `[environment_required]`, or `[release_required]`.
   Locally executable code, tests, startup, and browser checks are local_required;
   free-form Blocker/owner prose cannot replace the category.
+- For each implementation task, review observable delivery, acceptance signal,
+  requirement/scenario coverage, and actual predecessors. Show the currently
+  executable tasks whose predecessors are complete. Reuse
+  `traceability-matrix.md` for detailed coverage; do not create another task
+  state file. `Blocked by` expresses execution order and never changes the
+  machine task classification above.
 - `tests.md`
 - `test-report.md` skeleton
 - `sdd-quality-gate.md`
@@ -364,6 +376,18 @@ Use references only as needed:
   `Case ID | Requirement/scenario | Level L1/L2/L3/L4 | Precondition data | Steps |
   Automation command | Response assertion | DB assertion | Log assertion | RED expected failure | GREEN expected pass |
   test-report evidence location`.
+- For cross-page management or database-backed display changes, reconcile the
+  frontend action/list inventory with real Controller, Service, exact Mapper
+  statement and table anchors before freezing tasks. Put the confirmed chain
+  and entry ID in `source-code-audit.md` and reuse it in `design.md`, `tasks.md`
+  and `tests.md`. A similarly named query is not evidence for the page route.
+  Include every visible mutation button, such as enable/disable, as an entry
+  even when the original task list omitted it.
+- For a returned persisted field, `tests.md` must freeze at least one non-null
+  fixture and an allowed historical-NULL fixture separately. Assert API and DB
+  values for the same record ID, then browser rendering where a page is in
+  scope. Key-presence, HTTP 200, empty-page and NULL-only assertions cannot
+  close the non-null data path. Name the exact query/command and evidence file.
 - For L3/L4 input validation, protocol rejection, and exceptional branches,
   `tests.md` also freezes the real injected payload and evidence that it reaches
   the intended parser or validation branch; matching the same HTTP status alone

package/assets/skills-en/superflow-implement/SKILL.md

@@ -96,6 +96,15 @@ Use references/batch-split-guide.md and references/batch-prompt-template.md.
 Split by business closure, not by file type. Each batch must have independent
 entry, implementation scope, verification, and report obligations.
 
+Read `tasks.md` predecessor links before splitting or dispatching. A batch is
+executable only when its real predecessors are complete; independent batches
+may run in parallel. Keep the existing checkbox and machine category as the
+progress authority. An implementation discovery that changes a confirmed
+owner decision or task acceptance requires canonical document repair, handoff
+refresh, document review, and a new Coding Ready receipt before the affected
+batch resumes. A free-text `Blocked by` note cannot downgrade a locally
+executable task to an external category.
+
 Workers may update only their own embedded-changes/pXX-* docs. Root aggregate
 docs are updated by the leader during closeout after related worktrees merge.
 

package/assets/skills-en/superflow-pipeline/SKILL.md

@@ -15,6 +15,17 @@ the unified evaluation framework.
 ## Role
 
 This is the main router and stateful workflow orchestrator for SuperBridge Flow.
+
+## Role ownership and mutual exclusions
+
+| Role | Sole owner | May do | Must not do |
+|---|---|---|---|
+| Host | goal, clarification, supervision, semantic review | choose routing and judge code/architecture risk | write business code or start a nested Supervisor |
+| Executor | source implementation, tests, startup, repair | change the workspace within the frozen contract and return evidence | edit the contract/state, lower verification, or release automatically |
+| Runner | state, hashes, exit codes, evidence persistence | schedule, wait, recover, and enforce deterministic facts | guess business meaning or replace Host review |
+
+Menus and personas change entry wording only; they never create a second state
+machine. Return to Host clarification when ownership conflicts.
 It combines OpenSpec/SDD and Superpowers without letting them overwrite each
 other:
 
@@ -50,6 +61,10 @@ user-approved change/prompt is frozen from its original path. Only a bounded
 verbal request enters minimal or standard directly. Direction-changing API,
 data, concurrency, cross-repository, or owner choices return to
 clarification/documents; Runner regexes and Executor improvisation do not decide them.
+For cross-page, cross-stack or database-field changes requesting closed delivery,
+use full SDD to freeze real entries, `tests.md` and per-entry acceptance. A raw
+OpenSpec `spec/design/tasks` set or generic smoke report supports only its
+actually verified scope, not a full SDD acceptance claim.
 
 ```bash
 superflow pipeline "<implementation prompt, change directory, or direct task>" --managed --project "<project root>" \
@@ -138,6 +153,11 @@ hotfix:
 
 ## Routing
 
+- One clear, low-blast-radius code change with an existing test seam: use
+  superflow-quick-dev before editing, then reuse the existing quick managed
+  profile after Quick Spec approval. API, DB, security, transaction,
+  concurrency, payment, cross-module, active-SDD, or ownership uncertainty
+  routes to superflow-clarify instead; do not add a second managed state machine.
 - When the user explicitly asks for rigorous questioning, or a bounded feature
   has unresolved owner decisions, mutually exclusive approaches, an unclear
   source of truth, unclear acceptance behavior, or cross-system responsibility,
@@ -179,6 +199,12 @@ When the router selects embedded deep clarification, follow these rules inside
 - Exit deep clarification once only repository-verifiable facts remain and the
   user confirms the current feature understanding. Continue through the existing
   clarify workflow without invoking an external skill dependency.
+- After all relevant features are frozen, show the change-level Consensus
+  Summary from `superflow-clarify` before OpenSpec artifact generation. Carry
+  confirmed choices into canonical `proposal.md`; do not create a parallel
+  alignment fact source or add approval prompts for decisions already explicit
+  in the user's request. A changed owner choice requires affected documents,
+  handoff, review, and Coding Ready to be refreshed before execution.
 
 ## Required Artifacts
 
@@ -316,6 +342,14 @@ questions in one final batch.
 Record irreversible or ambiguous decisions in references/decision-point.md
 format. Do not store decisions only in chat memory.
 
+For repository-wide terminology, run `superflow context-audit --path <project>`
+when a change introduces a domain term or a durable architectural decision. Keep
+business vocabulary in `CONTEXT.md`; use an ADR only when the decision is hard
+to reverse and has meaningful alternatives. The audit is advisory: a missing
+ADR is a warning, while incomplete ADR fields or implementation details in the
+terminology list are reported for repair. It does not create approval gates or
+replace the canonical `design.md` contract.
+
 Decision points include:
 
 - API compatibility choices

package/assets/skills-en/superflow-pipeline/references/document-review-coverage.md

@@ -26,6 +26,16 @@ investigate source-resolvable questions autonomously.
 5. Assert response, final state and forbidden effects. For read-only/pure-function work explain
    why no persistent change applies; do not leave blanks. Browser paths require browser acceptance;
    API checks can supplement, not replace them.
+6. For cross-page or field changes, inventory frontend routes, list columns and clickable actions
+   independently from backend Controllers, exact Mapper statements and all writers. Reconcile both
+   inventories before writing tasks/tests. Similar method names are not call-chain evidence. For
+   every list record page request, Controller, Service, exact Mapper statement, response DTO and
+   table; for every management button record its mutation API and write point. Investigate missing
+   or conflicting entries instead of treating matching totals as proof.
+7. A database-backed display case needs both a non-null fixture and a separate allowed historical
+   NULL case. Compare API fields with authoritative DB values for the same business ID, then check
+   browser rendering. Field keys, HTTP 200, empty lists and historical NULL alone are insufficient.
+   Mutation cases also check write-then-read behavior and forbidden side effects.
 
 ## Receipt extension
 
@@ -87,3 +97,13 @@ During verification, reuse R/E/C IDs in `test-report.md`, recording actual actor
 request route, state/forbidden-effect assertions, raw evidence paths and outcomes. The Host
 compares them to original sources, not only a test summary. Planned cases are not execution
 evidence; API-only results cannot claim browser E2E. Do not claim automatic semantic assurance.
+Full SDD reports use one per-case execution evidence table for the existing report linter:
+
+| Case ID | Entry ID | Acceptance level | Result | Evidence path |
+|---|---|---|---|---|
+| C1 | E1 | browser | PASS | logs/C1-trace.zip |
+
+Results are `PASS/FAIL/BLOCKED/PARTIAL`. Evidence points to actual command output, browser trace,
+response or DB reconciliation. Retain readable local evidence files for PASS; HTTP(S) links may
+identify remote evidence. Explain blockers in the report body. This table proves case-level
+accounting only; the Host must inspect evidence contents. Any non-PASS case prevents an overall PASS.

package/assets/skills-en/superflow-pipeline/references/test-execution-template.md

@@ -55,6 +55,18 @@ Archive Readiness: PASS / BLOCKED
 
 Required for `verify_mode: full`:
 
+For full SDD changes, fill every R/E/C case from `document-review.json` in one
+execution table checked by `superflow-test-report-lint.py --review`:
+
+| Case ID | Entry ID | Acceptance level | Result | Evidence path |
+|---|---|---|---|---|
+| C1 | E1 | browser | PASS | logs/C1-trace.zip |
+
+Use only `PASS/FAIL/BLOCKED/PARTIAL`; explain non-PASS results in the report.
+For database-backed display fields, record same-ID API/DB values on non-null
+data, historical NULL separately, and real browser rendering. Distinguish
+field-key checks, static frontend inspection and browser evidence.
+
 - Test environment:
 - Branch/commit/build number or image:
 - Base URL:

package/assets/skills-en/superflow-pipeline/scripts/superflow-guard.sh

@@ -87,15 +87,21 @@ project_root_for_change() {
 
 require_test_report_lint() {
   local report="$CHANGE_DIR/test-report.md"
-  local root lint output
+  local root lint output review
+  local -a lint_args
   root="$(project_root_for_change)"
+  review="$CHANGE_DIR/.sdd/reviews/document-review.json"
+  lint_args=(--tests "$CHANGE_DIR/tests.md")
+  if [[ -f "$review" ]]; then
+    lint_args+=(--review "$review")
+  fi
 
   for lint in \
     "$root/.codex/hooks/superflow-test-report-lint.py" \
     "$HOME/.codex/hooks/superflow-test-report-lint.py"; do
     if [[ -x "$lint" ]]; then
       output="$(mktemp)"
-      if ! "$lint" --tests "$CHANGE_DIR/tests.md" "$report" >"$output" 2>&1; then
+      if ! "$lint" "${lint_args[@]}" "$report" >"$output" 2>&1; then
         issues+=("superflow-test-report-lint failed: $(tr '\n' ' ' < "$output" | sed -E 's/[[:space:]]+/ /g')")
       fi
       rm -f "$output"

package/assets/skills-en/superflow-quick-dev/SKILL.md

@@ -0,0 +1,43 @@
+---
+name: superflow-quick-dev
+description: Fast admission and Quick Spec for small code tasks. Stop and switch to formal SDD when intent, seam, or blast radius is unclear.
+license: MIT
+metadata:
+  author: superflow
+  version: '1.0'
+---
+
+# Superflow Quick Dev
+
+Accept only one low-blast-radius code task that follows an existing seam.
+
+## Narrow gate
+
+Inspect current source, callers, and tests, then run
+`superflow quick "<request>" --json --path <changed-files...> --seam <existing-file>`.
+Pass `--active-sdd` only after confirming an active SDD change owns the **same
+behavior**; reusing its code or depending on its output is insufficient. Continue only
+when it returns `QUICK`. Public APIs, databases/migrations, transactions,
+concurrency, locks, permissions, security, payments, cross-module work, active
+SDD changes, or a non-exclusive worktree must return `STOP` and switch to
+`superflow-clarify`. If ake-harness's `ssd-propose` is also visible, do not
+route there merely because it is installed. Switch to ake SSD only when the
+user explicitly chooses that workflow.
+
+## Quick Spec
+
+Write and show the current disk version at
+`.sdd/tasks/<YYYY-MM-DD>-<slug>/spec.md` with Intent / Non-goals, current and
+target behavior, confirmed test seam, ordered implementation steps, Given /
+When / Then acceptance, risks, escalation conditions, deferred work, baseline
+commit, and exact file scope.
+
+Do not edit runtime code before approval. After explicit approval of the current
+disk version, run `superflow quick "<request>" --path <files...> --seam <existing-file> --spec <spec> --approve`.
+The CLI binds the approval to the current content digest. If Intent, Non-goals, or Acceptance
+Criteria change, return to `draft` and obtain approval again. Use TDD and
+verification-loop RED/GREEN signals, then close with `superflow review-coverage`
+and `superflow eval` evidence.
+
+Quick work creates no OpenSpec checkpoint and never commits, pushes, or releases.
+When scope expands, keep the spec as a lead and switch to formal SDD.