@gordon.gan/specflow 1.4.0-beta → 1.4.1

package/README.md

@@ -126,7 +126,7 @@ npm install -g @gordon.gan/specflow
 npm install -g github:Gordon-Gan-Jiang/specflow
 
 # 验证
-specflow --version   # 以 npm / package.json 为准（当前 1.4.0-beta）
+specflow --version   # 以 npm / package.json 为准（当前 1.4.1）
 specflow --help
 ```
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gordon.gan/specflow",
-  "version": "1.4.0-beta",
+  "version": "1.4.1",
   "type": "module",
   "description": "SpecFlow — unified spec-driven development: OpenSpec planning + Superpowers execution in one CLI and cross-IDE workflow",
   "keywords": [

package/prompts/approval/database-guidance.md

@@ -0,0 +1,79 @@
+# Approval · Database Guidance Router
+
+> Used by `/specflow:approval` when generating §4.4 (数据结构 / 数据模型).  
+> **本地技能**: `skills/database/`（本仓库长期维护）。  
+> **禁止**依赖远程 skill 仓库（禁止 `npx skills add` / clone / 运行时 fetch）。  
+> **Not** an MCP tool — on hit, `Read` local matched skill files; on miss, LLM-only §4.4.
+
+---
+
+## 1. Detect project database stack
+
+Scan the **project under review** (change's repo root), not SpecFlow itself. Collect signals
+(non-recursive first; optional one-level into `docker-compose*.yml` / `migrations/`).
+
+| Signal | Suggested stack |
+|--------|-----------------|
+| `go.mod` / deps: `go-sql-driver/mysql`, `gorm.io` + mysql DSN, `ENGINE=InnoDB` in `*.sql` | `mysql` |
+| `package.json`: `mysql2`, `mysql`, `@prisma/client` + datasource mysql, drizzle mysql | `mysql` |
+| `pyproject` / requirements: `PyMySQL`, `mysqlclient`, `aiomysql` | `mysql` |
+| `go.mod`: `jackc/pgx`, `lib/pq`; `package.json`: `pg`, `postgres`; Prisma/drizzle postgres | `postgresql` |
+| `docker-compose`: image `postgres` / `mysql` / `mariadb` / `redis` / `elasticsearch` | matching image |
+| SQL files with `ENGINE=InnoDB` / `CHARSET=utf8mb4` | `mysql` |
+| SQL with `JSONB`, `SERIAL`, `gen_random_uuid()` | `postgresql` |
+| Oracle JDBC / `oracledb` / `go-ora` | `oracle` |
+| `ioredis` / `redis` client / `go-redis` | `redis` (cache/KV — usually **not** §4.4 relational DDL) |
+| `@elastic/elasticsearch` / OpenSearch client | `elasticsearch` (search — usually not relational DDL) |
+
+**Multi-hit**: pick primary OLTP store for §4.4 (prefer `mysql` / `postgresql` / `oracle` over
+`redis` / `elasticsearch`). Mention secondary stores in §4.7/架构 if relevant.
+
+**Announce** after detection:
+
+```text
+DB stack for approval: <mysql|postgresql|oracle|redis|elasticsearch|none>
+Guidance: skills/database/<stack>/SKILL.md | LLM-fallback
+```
+
+---
+
+## 2. On hit — load **local** guidance (Read, do not invent, do not fetch)
+
+Base: `skills/database/<stack>/`（相对 SpecFlow 仓库根；离线可读）
+
+| Stack | Must Read | Also Read when §4.4 needs it |
+|-------|-----------|------------------------------|
+| `mysql` | `SKILL.md` | `references/05-sql-ddl-types.md`, `references/06-index-optimization.md`; JSON → `references/04-functions-json.md` |
+| `postgresql` | `SKILL.md` | `references/06-index-types.md`; JSONB → `references/04-functions-jsonb.md` |
+| `oracle` | `SKILL.md` | `references/09-sql-syntax.md`, `references/10-features.md` |
+| `redis` | `SKILL.md` | Only if change is cache/key design — **not** a substitute for relational DDL |
+| `elasticsearch` | `SKILL.md` | Only if change is index/mapping — **not** a substitute for relational DDL |
+
+Apply guidance to §4.4 output:
+
+1. Prefer engine/charset/collation idioms from the local skill (MySQL: InnoDB + utf8mb4…).
+2. Prefer type/index gotchas from the skill (e.g. DECIMAL for money, no FLOAT amounts).
+3. Keep SpecFlow hard rules (ER, full CREATE TABLE, 本迭代用法, G3/G4) — skill **supplements**, does not replace.
+
+Cite in §4.4 总则, e.g. `DB 技能: skills/database/mysql (本地)`.
+
+---
+
+## 3. On miss — LLM fallback
+
+If `DB stack = none` (no durable store signals, or only unknown/custom DB):
+
+1. **Do not** force-load mysql/postgresql skills.
+2. Generate §4.4 with SpecFlow rules only (`generate.md` §4.4 + Quality Gates G3/G4).
+3. If design/tasks claim a DB but signals are absent → `WARNING` in §8:
+   `design 声称落库但未检测到 DB 栈信号 — 已用 LLM 通用 DDL;建议 refine 标明引擎`.
+4. Pure CLI/config changes → keep `不涉及数据库变更(...)`.
+
+---
+
+## 4. Why not remote skills / MCP tools?
+
+- **Remote**: approval must work offline and must not drift with upstream installs; maintain
+  `skills/database/` in this repo.
+- **MCP tools**: these packs are markdown guidance, not APIs; routing chooses which local
+  files to `Read`.