add-coder 0.3.11 → 0.3.13

package/README.md

@@ -54,6 +54,42 @@ dist-tags:
 published 2 weeks ago by wujixmm <wujixmm@gmail.com>
 ```
 
+
+---
+
+> ## ⚠️ 数据共享知情清单（Data-Sharing Disclosure）
+>
+> add-coder 无意于搞各家厂商的对抗叙事。以下内容仅为**基于各家官方公开条款原文的事实摘录**，目的是让开发者在选择 AI 编程 IDE 前知情。条款可能随时变更，请以官方最新版本为准。
+>
+> | 产品 | 运营主体 | 官方条款要点（原文摘录） | 产品实测（用户反馈） | 出处 | 核实日期 |
+> |------|---------|------------------------|---------------------|------|----------|
+> | **Qoder** | BRIGHT ZENITH PTE. LTD.（新加坡） | 用户内容（含聊天、编程及智能体会话的输入/输出，即含代码）默认用于"研究、开发和改进我们的服务"，保留期最长 5 年；条款文本称可关闭"共享与改进"（Share & Improve）退出，**但关闭前已为上述目的处理的数据仍保留和使用**；数据跨境存储于美国/新加坡/日本 | 实测为强制隐私共享模式，界面无有效切换入口 | [Qoder 隐私政策](https://qoder.com/zh/privacy-policy) | 2026-05-12 / 2026-08-04 |
+> | **Qoder CN** | 阿里云（通义灵码系列） | 官方 FAQ 称：仅在用户点赞/点踩后，针对聊天记录（不包含代码）脱敏去标识化处理，用于算法升级与迭代 | **实测（v1.10.0）：强制隐私共享模式，无切换功能**；官方 CLI 配置（`~/.qoder-cn/settings.json`）中亦无共享/改进开关项 | [阿里云帮助中心 FAQ](https://help.aliyun.com/zh/lingma/qoder-cn/support/faq) | 2026-07-15 / 2026-08-04 |
+> | **Trae CN** | 北京引力弹弓科技有限公司（字节跳动） | 服务所收集的数据经加密、严格去标识化且无法重新识别个人后，**可能用于模型训练**以优化模型效果与产品体验；可通过隐私模式"限制"该数据使用；代码文件不提供给其他第三方 | — | [Trae 隐私协议](https://www.trae.cn/privacy-policy) | 现行版 / 2026-08-04 |
+>
+> **要点**：三家官方条款均为"**默认开启共享、需用户主动退出**"（opt-out）模式。**但产品实测层面**：Qoder / Qoder CN 的退出开关在实际产品中不可用（Qoder CN v1.10.0 无切换功能，CLI 配置亦无对应开关），条款上的退出承诺在 CN 产品中无落地入口，实际效果即强制共享。Qoder 明确将用户内容（含代码）用于改进服务并保留 5 年；Trae 明确将数据用于模型训练（可通过隐私模式限制）；Qoder CN 官方条款范围最窄（仅反馈后的聊天记录、不含代码）。
+>
+> **建议**：涉及商业机密或个人数据的项目，使用前请逐项核对官方条款，并主动检查产品设置：Qoder（国际版）尝试关闭"共享与改进"；Qoder CN（v1.10.0 实测无开关）与 Trae CN 请结合隐私模式与最小化输入策略使用；必要时改用数据完全本地化的替代方案。
+
+---
+> ### 🔮 下一步：add-coder IDE 插件（实施中）
+
+> add-coder 当前是 **脚手架 + 治理层**：把 ADD 范式（裁决集中、审计即基础设施、门禁驱动）部署进你的项目。
+> 下一阶段会以 **IDE 插件** 形式落地，让治理能力直接嵌在编辑与 Agent 生命周期里。
+>
+> **为什么值得期待：**
+>
+> | 维度 | 说明 |
+> |------|------|
+> | **范式优势** | 不是「更好用的补全」，而是把规则、证据、审批、回溯做成一等公民。改规则不改代码，跨轮不靠聊天记忆。 |
+> | **与 LLM 低耦合** | 核心是结构与门禁，不是绑死某一家模型。换模型不换治理逻辑；模型只负责生成，裁决与审计由范式层承担。 |
+> | **与 IDE 低耦合** | 治理逻辑统一，适配层分离。今天支持 Claude Code / Qoder / VS Code / Trae / Codex，明天换宿主只需加适配，不必重写规则与审计链。 |
+> | **不收集用户隐私** | 不上传代码、对话与项目数据。审计与裁决默认留在本地 / 你自己的基础设施里，不做「默认共享、事后退出」。 |
+>
+> 目标不是再做一个「AI 编程 IDE」，而是让 **任意 IDE + 任意 LLM** 都能跑在同一套可审计、可收敛、且不拿你数据的 ADD 之上。
+
+---
+
 **AI 代码治理的落地方案** — [codein2027](https://github.com/xiaomingming92/codein2027) 快速构建 ADD 编程范式的完整脚手架。以「审计即基础设施」为核心，彻底打破编程过程黑盒与跨轮失忆，让编程范式进化为可审计、可追溯、可收敛的新时代。 [NPM](https://www.npmjs.com/package/add-coder) · [GitHub](https://github.com/xiaomingming92/add-coder)
 
 
@@ -124,7 +160,7 @@ ADD 范式 + Qoder:     cache 命中率 99.31%, 每次请求 MISS 仅 2,426 toke
 
 ```
 DPS (Documentation Precision Score) — TF-IDF/Jaccard 语义 + 香农/Deng 熵 + CPM 关键路径 + 结构完整度
-  → 四维复合评分 + FFT 自适应权重，≥ 85 进入 Step 1
+  → 四维复合评分 + FFT 自适应权重，≥ 阈值（以 dps-scoring-rules.toml 为准，当前 PASS=80）进入 Step 1
 RAHS (Runtime Architecture Health Score) — 运行时架构健康度
   → 五维判定：范围保真 + 类型安全 + 审计完整 + Spec 合规 + 阶段对称，≥ 90 通过
 ```
@@ -446,7 +482,7 @@ Traditional AI coding is "you say, I do" — quality depends entirely on the LLM
 
 ```
 DPS (Documentation Precision Score) — TF-IDF/Jaccard semantics + Shannon/Deng entropy + CPM critical path + structural completeness
-  → 4D composite scoring + FFT adaptive weights, ≥ 85 to enter Step 1
+  → 4D composite scoring + FFT adaptive weights, ≥ threshold (per dps-scoring-rules.toml, currently PASS=80) to enter Step 1
 RAHS (Runtime Architecture Health Score) — runtime architecture health
   → 5D assessment: scope fidelity + type safety + audit completeness + spec compliance + phase symmetry, ≥ 90 to pass
 ```

package/dist/caijuehub/dps-scoring-rules.toml

@@ -0,0 +1,66 @@
+# DPS 评分全参数 —— caijuehub 驱动，后续调优只需改此 TOML
+
+# ═══════════════ 维度一：语义相关性 ═══════════════
+[semantic]
+# 子权重：Jaccard Plan↔Specs / Cosine Plan↔Specs / Jaccard Plan↔Review / Cosine Plan↔Review
+weights = [0.35, 0.45, 0.08, 0.12]
+# Review 缺失时的惩罚系数（0.0=不惩罚，0.4=打六折，即得分×0.6）
+missing_review_penalty = 0.0
+
+# ═══════════════ 维度二：信息熵匹配 ═══════════════
+[entropy]
+# 熵差距乘数（entropyGap × multiplier，越小越不敏感）
+gap_multiplier = 10
+# Deng 不确定性惩罚：每个标记扣分，上限
+deng_per_marker = 0.05
+deng_max_penalty = 0.3
+# Specs 缺失时的熵差距（无 Specs 时 gap 设为此值）
+missing_specs_gap = 10
+
+# ═══════════════ 维度三：CPM 关键路径 ═══════════════
+[cpm]
+# 子权重：原子化 / 注意力匹配 / 依赖完整性
+sub_weights = [0.4, 0.35, 0.25]
+# 业务原子化：Jaccard 重叠乘数（overlap × multiplier → penalty）
+overlap_multiplier = 2
+# Task 数超过此阈值时随机采样配对（避免 O(N²) 爆炸，0 = 不限制）
+max_task_pairs = 200
+# 注意力匹配：文件/Task 阈值
+file_per_task_low = 3    # ≤3 → 100
+file_per_task_mid = 5    # 3~5 → 70，>5 → 40
+# 注意力匹配：描述熵阈值（bits）
+entropy_low = 6          # <6 → 100
+entropy_mid = 8          # 6~8 → 75，>8 → 50
+# 注意力匹配：子权重（文件耦合 / 描述熵）
+attention_weights = [0.5, 0.5]
+# 依赖完整性：子权重（覆盖率 / Task Dependencies 章节存在）
+dep_weights = [0.6, 0.4]
+
+# ═══════════════ 维度四：结构完整度 ═══════════════
+[structure]
+# 扣分项
+placeholder_penalty = 15
+missing_specs_penalty = 30
+missing_tasks_penalty = 15
+missing_checklist_penalty = 15
+# 子权重：结构分 / 回流分
+sub_weights = [0.6, 0.4]
+
+# ═══════════════ FFT 自适应权重 ═══════════════
+[fft]
+cold_start_threshold = 5
+history_limit = 50
+default_weights = [0.25, 0.25, 0.25, 0.25]
+
+# ═══════════════ DPS 阈值 ═══════════════
+[thresholds]
+pass = 80
+warn = 65
+
+# ═══════════════ Embedding 模型（@xenova/transformers ONNX）═══════════════
+[embedding]
+# 首次 DPS 调用自动下载并缓存到本地 ~/.cache/huggingface/hub/
+# 可选值（注释掉的前两个是备选方案）：
+#   "Xenova/all-MiniLM-L6-v2"      英文为主，80MB，速度快
+#   "Xenova/multilingual-e5-small"  多语言，118MB，平衡选择
+model = "Xenova/bge-small-zh-v1.5"

package/dist/index.js

@@ -400,9 +400,10 @@ function diffPrisma(basePath, targetPath) {
 }
 
 // src/core/renderer.ts
-import { readFileSync as readFileSync4, readdirSync, statSync } from "fs";
+import { readFileSync as readFileSync4, readdirSync, statSync, existsSync as existsSync5 } from "fs";
 import { join as join4, relative, dirname as dirname2 } from "path";
 import { fileURLToPath } from "url";
+import { parse } from "smol-toml";
 var __filename = fileURLToPath(import.meta.url);
 var __dirname = dirname2(__filename);
 var PLACEHOLDERS = {
@@ -417,12 +418,43 @@ var PLACEHOLDERS = {
   "{{mcpServerCommand}}": "mcpServerCommand",
   "{{agentAuditImport}}": "agentAuditImport"
 };
+function loadDpsThresholds() {
+  try {
+    const candidates = [
+      join4(__dirname, "../caijuehub/dps-scoring-rules.toml"),
+      join4(__dirname, "caijuehub/dps-scoring-rules.toml"),
+      join4(__dirname, "../../caijuehub/dps-scoring-rules.toml")
+    ];
+    const tomlPath = candidates.find(existsSync5);
+    if (!tomlPath) {
+      throw new Error(`\u5019\u9009\u8DEF\u5F84\u5747\u4E0D\u5B58\u5728: ${candidates.join(" | ")}`);
+    }
+    const doc = parse(readFileSync4(tomlPath, "utf-8"));
+    if (doc.thresholds && typeof doc.thresholds.pass === "number") {
+      return {
+        pass: String(doc.thresholds.pass),
+        warn: String(doc.thresholds.warn ?? doc.thresholds.pass)
+      };
+    }
+  } catch (e) {
+    const errorMessage = e instanceof Error ? e.message : String(e);
+    console.warn(
+      `[renderer] DPS \u9608\u503C TOML \u8BFB\u53D6\u5931\u8D25\uFF0C\u4FDD\u7559 dpsPass/dpsWarn \u5360\u4F4D\u7B26: ${errorMessage}`
+    );
+  }
+  return null;
+}
 function render(content, config) {
   let result = content;
   for (const [placeholder, key] of Object.entries(PLACEHOLDERS)) {
     const value = config[key];
     result = result.replaceAll(placeholder, value);
   }
+  const thresholds = loadDpsThresholds();
+  if (thresholds) {
+    result = result.replaceAll("{{dpsPass}}", thresholds.pass);
+    result = result.replaceAll("{{dpsWarn}}", thresholds.warn);
+  }
   return result;
 }
 var TEMPLATES_ROOT = join4(__dirname, "../templates");
@@ -512,7 +544,7 @@ function renderAdapter5(config, targetDir, dryRun, magicDir) {
 
 // src/lib/utils.ts
 import { createInterface as createInterface2 } from "readline";
-import { existsSync as existsSync5 } from "fs";
+import { existsSync as existsSync6 } from "fs";
 import { resolve as resolve2 } from "path";
 function ask(q) {
   const rl = createInterface2({ input: process.stdin, output: process.stdout });
@@ -524,7 +556,7 @@ function ask(q) {
   });
 }
 function detectPm(projectRoot) {
-  return existsSync5(resolve2(projectRoot, "pnpm-lock.yaml")) ? "pnpm" : "npm";
+  return existsSync6(resolve2(projectRoot, "pnpm-lock.yaml")) ? "pnpm" : "npm";
 }
 
 // src/cli/prisma-injector.ts
@@ -533,7 +565,7 @@ import { fileURLToPath as fileURLToPath2 } from "url";
 
 // src/caijuehub/strategies/prisma.strategy.ts
 import { spawnSync } from "child_process";
-import { copyFileSync, existsSync as existsSync6, readFileSync as readFileSync5, unlinkSync, writeFileSync as writeFileSync2, mkdirSync as mkdirSync2 } from "fs";
+import { copyFileSync, existsSync as existsSync7, readFileSync as readFileSync5, unlinkSync, writeFileSync as writeFileSync2, mkdirSync as mkdirSync2 } from "fs";
 import { resolve as resolve3 } from "path";
 var PRISMA_CONFIG = {
   onMissing: "ask",
@@ -586,10 +618,10 @@ function runPrismaInit(projectRoot, provider, schemaPath) {
     stdio: "inherit",
     shell: false
   });
-  if (initResult.status !== 0 || !existsSync6(schemaPath)) {
+  if (initResult.status !== 0 || !existsSync7(schemaPath)) {
     console.log("prisma init \u5931\u8D25\uFF0C\u624B\u52A8\u521B\u5EFA schema.prisma ...");
     const prismaDir = resolve3(projectRoot, "prisma");
-    if (!existsSync6(prismaDir)) mkdirSync2(prismaDir, { recursive: true });
+    if (!existsSync7(prismaDir)) mkdirSync2(prismaDir, { recursive: true });
     const content = `generator client {
   provider = "prisma-client-js"
 }
@@ -600,7 +632,7 @@ datasource db {
 `;
     writeFileSync2(schemaPath, content, "utf-8");
     const devEnvPath = resolve3(projectRoot, ".env.development");
-    if (!existsSync6(devEnvPath)) {
+    if (!existsSync7(devEnvPath)) {
       const defaultUrl = provider === "sqlite" ? 'DATABASE_URL="file:./data/dev.db"' : '# \u8BF7\u7F16\u8F91\u4E3A\u4F60\u7684\u6570\u636E\u5E93\u8FDE\u63A5\u4FE1\u606F\nDATABASE_URL="postgresql://USER:PASSWORD@HOST:PORT/DB?schema=public"';
       writeFileSync2(devEnvPath, defaultUrl + "\n", "utf-8");
       console.log("\u5DF2\u521B\u5EFA .env.development");
@@ -612,16 +644,16 @@ datasource db {
 function postInitSetup(projectRoot, schemaPath, addPrismaTemplate, destPath) {
   const envPath = resolve3(projectRoot, ".env");
   const devEnvPath = resolve3(projectRoot, ".env.development");
-  if (existsSync6(envPath)) {
+  if (existsSync7(envPath)) {
     const envContent = readFileSync5(envPath, "utf-8");
     const dbUrl = envContent.match(/DATABASE_URL=.*/);
     if (dbUrl) {
-      const existing = existsSync6(devEnvPath) ? readFileSync5(devEnvPath, "utf-8") : "";
+      const existing = existsSync7(devEnvPath) ? readFileSync5(devEnvPath, "utf-8") : "";
       if (!existing.includes("DATABASE_URL=")) {
         writeFileSync2(devEnvPath, `${existing}${existing ? "\n" : ""}${dbUrl[0]}
 `, "utf-8");
       }
-      if (existsSync6(envPath)) unlinkSync(envPath);
+      if (existsSync7(envPath)) unlinkSync(envPath);
       console.log("\u5DF2\u5C06 DATABASE_URL \u8FC1\u79FB\u5230 .env.development");
     }
   }
@@ -634,7 +666,7 @@ async function injectPrisma(projectRoot, addPrismaTemplate, options = {}) {
   const schemaPath = resolve3(prismaDir, "schema.prisma");
   const destPath = resolve3(prismaDir, "add.prisma");
   let justInited = false;
-  if (!existsSync6(prismaDir) || !existsSync6(schemaPath)) {
+  if (!existsSync7(prismaDir) || !existsSync7(schemaPath)) {
     if (C.onMissing === "skip") {
       console.log("\u8DF3\u8FC7\uFF1A\u7F3A\u5C11 Prisma");
       return true;
@@ -653,7 +685,7 @@ async function injectPrisma(projectRoot, addPrismaTemplate, options = {}) {
     postInitSetup(projectRoot, schemaPath, addPrismaTemplate, destPath);
     justInited = true;
   }
-  if (existsSync6(destPath) && !justInited) {
+  if (existsSync7(destPath) && !justInited) {
     if (options.dryRun) {
       console.log("[dry-run] \u5DF2\u6709 add.prisma");
       return true;
@@ -726,7 +758,7 @@ async function injectPrisma2(projectRoot, options = {}) {
 }
 
 // src/cli/commands/init.ts
-import { readFileSync as readFileSync6, writeFileSync as writeFileSync3, existsSync as existsSync7, mkdirSync as mkdirSync3, copyFileSync as copyFileSync2, readdirSync as readdirSync2 } from "fs";
+import { readFileSync as readFileSync6, writeFileSync as writeFileSync3, existsSync as existsSync8, mkdirSync as mkdirSync3, copyFileSync as copyFileSync2, readdirSync as readdirSync2 } from "fs";
 import { createHash } from "crypto";
 import { resolve as resolve5 } from "path";
 import { spawnSync as spawnSync2 } from "child_process";
@@ -906,13 +938,13 @@ main().catch((e) => { console.error(e); process.exit(1); });
     console.log(`[dry-run] \u5C06\u5199\u5165 ${scriptPath}`);
     return;
   }
-  if (!existsSync7(scriptsDir)) mkdirSync3(scriptsDir, { recursive: true });
+  if (!existsSync8(scriptsDir)) mkdirSync3(scriptsDir, { recursive: true });
   writeFileSync3(scriptPath, content, "utf-8");
   console.log("\u5DF2\u751F\u6210 scripts/export-db.ts");
 }
 function injectDbExportScript(projectRoot, dryRun) {
   const pkgPath = resolve5(projectRoot, "package.json");
-  if (!existsSync7(pkgPath)) return;
+  if (!existsSync8(pkgPath)) return;
   if (dryRun) {
     console.log("[dry-run] \u5C06\u6CE8\u5165 db:export");
     return;
@@ -947,13 +979,13 @@ function writeComposeEnv(ctx) {
   if (!db.reuseExisting) {
     const composeName = db.container === "podman" ? "podman-compose.add.yml" : "docker-compose.add.yml";
     const composePath = resolve5(projectRoot, composeName);
-    if (!options.dryRun && (!existsSync7(composePath) || options.force)) {
+    if (!options.dryRun && (!existsSync8(composePath) || options.force)) {
       writeFileSync3(composePath, composeContent(config.projectName || "add-project"), "utf-8");
       console.log(`\u5DF2\u521B\u5EFA ${composeName}`);
     }
   }
   const devEnvPath = resolve5(projectRoot, ".env.development");
-  if (!options.dryRun && existsSync7(devEnvPath)) {
+  if (!options.dryRun && existsSync8(devEnvPath)) {
     const existing = readFileSync6(devEnvPath, "utf-8");
     if (!/^DATABASE_USER=/m.test(existing)) {
       writeFileSync3(devEnvPath, existing + `
@@ -994,7 +1026,7 @@ async function renderAndWrite(ctx) {
   }
   for (const d of [".add", magicDir]) {
     const reviewsDir = resolve5(projectRoot, d, "reviews");
-    if (!existsSync7(reviewsDir)) {
+    if (!existsSync8(reviewsDir)) {
       if (dry) {
         console.log(`[dry-run] \u5C06\u521B\u5EFA ${reviewsDir}/`);
       } else {
@@ -1038,7 +1070,7 @@ async function deployDatabase(ctx) {
   }
   if (db.engine === "postgresql" && db.container === "manual") {
     const dbScript = resolve5(projectRoot, magicDir, "scripts", "db-ensure.sh");
-    if (existsSync7(dbScript)) spawnSync2("bash", [dbScript, "postgresql", "manual"], { cwd: projectRoot, stdio: "inherit" });
+    if (existsSync8(dbScript)) spawnSync2("bash", [dbScript, "postgresql", "manual"], { cwd: projectRoot, stdio: "inherit" });
     console.log([
       "",
       "\u2501".repeat(30),
@@ -1079,12 +1111,12 @@ function deployDocs(ctx) {
   for (const d of ["00-\u9700\u6C42", "01-\u67B6\u6784", "02-\u89C4\u8303"]) {
     const srcDir = resolve5(groundingSrc, d);
     const destDir = resolve5(docsBase, d);
-    if (!existsSync7(destDir)) mkdirSync3(destDir, { recursive: true });
-    if (!existsSync7(srcDir)) continue;
+    if (!existsSync8(destDir)) mkdirSync3(destDir, { recursive: true });
+    if (!existsSync8(srcDir)) continue;
     for (const f of readdirSync2(srcDir)) {
       const src = resolve5(srcDir, f);
       const dest = resolve5(destDir, f);
-      if (existsSync7(dest)) continue;
+      if (existsSync8(dest)) continue;
       try {
         copyFileSync2(src, dest);
       } catch {
@@ -1112,7 +1144,7 @@ function finalize(ctx, result) {
 }
 
 // src/cli/commands/sync.ts
-import { existsSync as existsSync8, readFileSync as readFileSync7, writeFileSync as writeFileSync4, mkdirSync as mkdirSync4 } from "fs";
+import { existsSync as existsSync9, readFileSync as readFileSync7, writeFileSync as writeFileSync4, mkdirSync as mkdirSync4 } from "fs";
 import { resolve as resolve6, dirname as dirname4 } from "path";
 import { createHash as createHash2 } from "crypto";
 
@@ -1250,7 +1282,7 @@ async function syncCommand(options = {}) {
     let sameCount = 0;
     for (const [relPath, content] of candidates) {
       const absPath = resolve6(projectRoot, relPath);
-      if (!existsSync8(absPath)) {
+      if (!existsSync9(absPath)) {
         missingFiles.set(relPath, content);
       } else if (establishBaseline) {
         missingFiles.set(relPath, content);
@@ -1288,7 +1320,7 @@ async function syncCommand(options = {}) {
   }
   const missing = /* @__PURE__ */ new Map();
   for (const [relPath, content] of allFiles) {
-    if (!existsSync8(resolve6(projectRoot, relPath))) {
+    if (!existsSync9(resolve6(projectRoot, relPath))) {
       missing.set(relPath, content);
     }
   }
@@ -1310,7 +1342,7 @@ async function syncCommand(options = {}) {
 }
 function printMigrateGuidance(targetPath, changed) {
   const g = SYNC_PRISMA_CONFIG.POST_SYNC;
-  const isManaged = existsSync8(resolve6(dirname4(targetPath), "migrations"));
+  const isManaged = existsSync9(resolve6(dirname4(targetPath), "migrations"));
   const actions = isManaged ? g.MANAGED_ACTIONS : g.UNMANAGED_ACTIONS;
   console.log(`  \u25B6 \u5DF2\u5199\u5165 ${changed} \u5904\u53D8\u66F4\uFF0C${g.HEADER}`);
   for (const a of actions) {
@@ -1334,7 +1366,7 @@ async function checkPrismaDiff(projectRoot, options) {
   if (!options.patch) return;
   const basePath = resolve6(projectRoot, SYNC_PRISMA_CONFIG.BASE_SCHEMA);
   const targetPath = resolve6(projectRoot, SYNC_PRISMA_CONFIG.TARGET_PATTERN);
-  if (!existsSync8(basePath)) {
+  if (!existsSync9(basePath)) {
     console.log(`
 \u26A0\uFE0F  \u57FA\u51C6 schema \u4E0D\u5B58\u5728: ${basePath}`);
     console.log(`  \u8BF7\u786E\u4FDD add-coder \u5DF2\u6B63\u786E\u5B89\u88C5\u3002`);
@@ -1346,7 +1378,7 @@ async function checkPrismaDiff(projectRoot, options) {
 \u2705 Prisma schema \u4E0E add-coder \u6807\u51C6\u4E00\u81F4\uFF0C\u65E0\u9700\u540C\u6B65\u3002`);
     return;
   }
-  const targetExists = existsSync8(targetPath);
+  const targetExists = existsSync9(targetPath);
   let modifiedCount = 0;
   console.log(`
 \u26A0\uFE0F  Prisma schema \u5DEE\u5F02\u68C0\u6D4B:`);
@@ -1533,13 +1565,12 @@ function injectFieldLines(targetPath, basePath, modelName, fieldKeys) {
     const fieldName = key.split(":")[0];
     const fieldLine = baseFields[fieldName];
     if (!fieldLine) continue;
-    if (new RegExp(`^\\s*${fieldName}\\s+`, "m").test(content)) continue;
-    const modelRegex = new RegExp(`(model\\s+${modelName}\\s*\\{[^}]*?)\\n(\\s*\\})`, "m");
+    const modelRegex = new RegExp(`(model\\s+${modelName}\\s*\\{)([^}]*?)(\\n\\s*\\})`, "m");
     const m = content.match(modelRegex);
     if (!m) continue;
-    content = content.replace(modelRegex, `$1
-  ${fieldLine}
-$2`);
+    if (new RegExp(`^\\s*${fieldName}\\s+`, "m").test(m[2])) continue;
+    content = content.replace(modelRegex, `$1$2
+  ${fieldLine}$3`);
     count++;
   }
   if (count > 0) writeFileSync4(targetPath, content, "utf-8");
@@ -1564,7 +1595,7 @@ function overwriteFieldLines(targetPath, basePath, modelName, conflicts) {
 }
 
 // src/cli/commands/status.ts
-import { existsSync as existsSync9 } from "fs";
+import { existsSync as existsSync10 } from "fs";
 import { resolve as resolve7 } from "path";
 async function statusCommand() {
   const projectRoot = process.cwd();
@@ -1574,7 +1605,7 @@ async function statusCommand() {
   const missing = [];
   const present = [];
   for (const [relPath] of coreFiles) {
-    if (existsSync9(resolve7(projectRoot, relPath))) {
+    if (existsSync10(resolve7(projectRoot, relPath))) {
       present.push(relPath);
     } else {
       missing.push(relPath);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "add-coder",
-  "version": "0.3.11",
+  "version": "0.3.13",
   "private": false,
   "description": "[codein2027](https://github.com/xiaomingming92/codein2027) - A complete scaffolding for building the ADD programming paradigm, the implementation layer for AI code governance. Core principle: Audit as Infrastructure. Breaks the black-box programming process and cross-session amnesia, evolving programming paradigms into an auditable, traceable, and convergent new era. npx-ready.\n\n[codein2027](https://github.com/xiaomingming92/codein2027) 快速构建 ADD 编程范式的完整脚手架——AI 代码治理的落地方案。以「审计即基础设施」为核心，彻底打破编程过程黑盒与跨轮失忆，让编程范式进化为可审计、可追溯、可收敛的新时代。npx 即用，人人可体验。",
   "repository": {

package/templates/.add-coder-src-hash.json

@@ -1,8 +1,8 @@
 {
-  "_version": "0.3.11",
+  "_version": "0.3.13",
   "adapters/claude/hooks/doc-format-guard.sh": "ec59ad3f",
-  "adapters/claude/hooks/lib/common.sh": "db321ef6",
-  "adapters/claude/hooks/lib/context-inject.sh": "4132f203",
+  "adapters/claude/hooks/lib/common.sh": "4f3e64ad",
+  "adapters/claude/hooks/lib/context-inject.sh": "cfc8c65e",
   "adapters/claude/hooks/lib/notify.sh": "e0033e53",
   "adapters/claude/hooks/lib/preload-templates.sh": "96dd67fe",
   "adapters/claude/hooks/lib/session-end.sh": "29813cbc",
@@ -16,7 +16,7 @@
   "adapters/claude/hooks/post-tool-use.sh": "f54dd009",
   "adapters/claude/hooks/pre-compact.sh": "4248f84f",
   "adapters/claude/hooks/pre-tool-use.sh": "ecbdc628",
-  "adapters/claude/hooks/prompt-submit.sh": "680b41f4",
+  "adapters/claude/hooks/prompt-submit.sh": "2fe9e42c",
   "adapters/claude/hooks/review-checklist.sh": "b383a983",
   "adapters/claude/hooks/session-end.sh": "777ae808",
   "adapters/claude/hooks/session-start.sh": "fd1d5e4f",
@@ -27,8 +27,8 @@
   "adapters/claude/mcp.json": "4ddc5120",
   "adapters/claude/settings.json": "067ccc45",
   "adapters/codex/hooks/doc-format-guard.sh": "8e134e78",
-  "adapters/codex/hooks/lib/common.sh": "db321ef6",
-  "adapters/codex/hooks/lib/context-inject.sh": "4132f203",
+  "adapters/codex/hooks/lib/common.sh": "4f3e64ad",
+  "adapters/codex/hooks/lib/context-inject.sh": "cfc8c65e",
   "adapters/codex/hooks/lib/notify.sh": "e0033e53",
   "adapters/codex/hooks/lib/preload-templates.sh": "96dd67fe",
   "adapters/codex/hooks/lib/session-end.sh": "29813cbc",
@@ -41,7 +41,7 @@
   "adapters/codex/hooks/post-tool-use.sh": "27123aee",
   "adapters/codex/hooks/pre-compact.sh": "6164c007",
   "adapters/codex/hooks/pre-tool-use.sh": "ffa7936b",
-  "adapters/codex/hooks/prompt-submit.sh": "bea6f381",
+  "adapters/codex/hooks/prompt-submit.sh": "a41371ea",
   "adapters/codex/hooks/review-checklist.sh": "b383a983",
   "adapters/codex/hooks/session-end.sh": "3cc8c346",
   "adapters/codex/hooks/session-start.sh": "aa8faa0f",
@@ -51,8 +51,8 @@
   "adapters/codex/hooks.json": "d9c99f6b",
   "adapters/codex/settings.json": "44136fa3",
   "adapters/qoder/hooks/doc-format-guard.sh": "bd4ffffb",
-  "adapters/qoder/hooks/lib/common.sh": "db321ef6",
-  "adapters/qoder/hooks/lib/context-inject.sh": "4132f203",
+  "adapters/qoder/hooks/lib/common.sh": "4f3e64ad",
+  "adapters/qoder/hooks/lib/context-inject.sh": "cfc8c65e",
   "adapters/qoder/hooks/lib/notify.sh": "e0033e53",
   "adapters/qoder/hooks/lib/preload-templates.sh": "96dd67fe",
   "adapters/qoder/hooks/lib/session-end.sh": "29813cbc",
@@ -65,7 +65,7 @@
   "adapters/qoder/hooks/post-tool-use.sh": "051cb6ed",
   "adapters/qoder/hooks/pre-compact.sh": "092fa6cf",
   "adapters/qoder/hooks/pre-tool-use.sh": "5ba6d9ae",
-  "adapters/qoder/hooks/prompt-submit.sh": "f36e6ae8",
+  "adapters/qoder/hooks/prompt-submit.sh": "336f262e",
   "adapters/qoder/hooks/review-checklist.sh": "666831c8",
   "adapters/qoder/hooks/session-end.sh": "c50339df",
   "adapters/qoder/hooks/session-start.sh": "a4307a78",
@@ -76,8 +76,8 @@
   "adapters/qoder/settings.json": "4bae3c9a",
   "adapters/qoder/sync-policy.json": "a4bcbbaf",
   "adapters/trae/hooks/doc-format-guard.sh": "9b99f253",
-  "adapters/trae/hooks/lib/common.sh": "db321ef6",
-  "adapters/trae/hooks/lib/context-inject.sh": "4132f203",
+  "adapters/trae/hooks/lib/common.sh": "4f3e64ad",
+  "adapters/trae/hooks/lib/context-inject.sh": "cfc8c65e",
   "adapters/trae/hooks/lib/notify.sh": "e0033e53",
   "adapters/trae/hooks/lib/preload-templates.sh": "96dd67fe",
   "adapters/trae/hooks/lib/session-end.sh": "29813cbc",
@@ -90,7 +90,7 @@
   "adapters/trae/hooks/post-tool-use.sh": "27123aee",
   "adapters/trae/hooks/pre-compact.sh": "6164c007",
   "adapters/trae/hooks/pre-tool-use.sh": "8e57d1fa",
-  "adapters/trae/hooks/prompt-submit.sh": "bea6f381",
+  "adapters/trae/hooks/prompt-submit.sh": "a41371ea",
   "adapters/trae/hooks/review-checklist.sh": "b383a983",
   "adapters/trae/hooks/session-end.sh": "3cc8c346",
   "adapters/trae/hooks/session-start.sh": "aa8faa0f",
@@ -111,8 +111,8 @@
   "adapters/vscode/.github/hooks/user-prompt-submit.json": "3a608d62",
   "adapters/vscode/extensions.json": "0f4024c1",
   "adapters/vscode/hooks/doc-format-guard.sh": "0db27261",
-  "adapters/vscode/hooks/lib/common.sh": "db321ef6",
-  "adapters/vscode/hooks/lib/context-inject.sh": "4132f203",
+  "adapters/vscode/hooks/lib/common.sh": "4f3e64ad",
+  "adapters/vscode/hooks/lib/context-inject.sh": "cfc8c65e",
   "adapters/vscode/hooks/lib/notify.sh": "e0033e53",
   "adapters/vscode/hooks/lib/preload-templates.sh": "96dd67fe",
   "adapters/vscode/hooks/lib/session-end.sh": "29813cbc",
@@ -125,7 +125,7 @@
   "adapters/vscode/hooks/post-tool-use.sh": "5f188718",
   "adapters/vscode/hooks/pre-compact.sh": "a0908f0e",
   "adapters/vscode/hooks/pre-tool-use.sh": "6232624b",
-  "adapters/vscode/hooks/prompt-submit.sh": "4529a394",
+  "adapters/vscode/hooks/prompt-submit.sh": "3e56d1a9",
   "adapters/vscode/hooks/review-checklist.sh": "b383a983",
   "adapters/vscode/hooks/session-end.sh": "7b407b9f",
   "adapters/vscode/hooks/session-start.sh": "ad724a3e",
@@ -135,15 +135,15 @@
   "adapters/vscode/launch.json": "f2b79855",
   "adapters/vscode/settings.json": "7cabd00e",
   "adapters/vscode/tasks.json": "654d7fbd",
-  "core/agents/add-flow-guardian.md": "a23f50cf",
+  "core/agents/add-flow-guardian.md": "c3e650d2",
   "core/docs/ADD-governance-claude-code.md": "0f582a7e",
   "core/docs/ADD-governance-codex.md": "289e8b71",
   "core/docs/ADD-governance-qoder-cn.md": "332a7253",
   "core/docs/ADD-governance-trae.md": "ffb29a0e",
   "core/docs/ADD-governance-vscode-copilot.md": "74062e88",
   "core/hooks/doc-format-guard.sh": "5cd6d1f2",
-  "core/hooks/lib/common.sh": "db321ef6",
-  "core/hooks/lib/context-inject.sh": "4132f203",
+  "core/hooks/lib/common.sh": "4f3e64ad",
+  "core/hooks/lib/context-inject.sh": "cfc8c65e",
   "core/hooks/lib/notify.sh": "e0033e53",
   "core/hooks/lib/preload-templates.sh": "96dd67fe",
   "core/hooks/lib/session-end.sh": "29813cbc",
@@ -156,7 +156,7 @@
   "core/hooks/post-tool-use.sh": "27123aee",
   "core/hooks/pre-compact.sh": "6164c007",
   "core/hooks/pre-tool-use.sh": "fa73df98",
-  "core/hooks/prompt-submit.sh": "bea6f381",
+  "core/hooks/prompt-submit.sh": "a41371ea",
   "core/hooks/review-checklist.sh": "b383a983",
   "core/hooks/session-end.sh": "3cc8c346",
   "core/hooks/session-start.sh": "aa8faa0f",
@@ -175,7 +175,7 @@
   "core/reports/index.md": "6348c374",
   "core/reports/runtime-report/gateway.md": "a5751665",
   "core/reviews/.gitkeep": "e3b0c442",
-  "core/rules/project_rules.md": "b2680762",
+  "core/rules/project_rules.md": "fea8bfaa",
   "core/rules/theory-practice-map.toml": "b1db99fd",
   "core/scripts/db-ensure.sh": "faeb6b92",
   "core/scripts/gen-plan-index.sh": "bfc6ad3e",
@@ -204,11 +204,11 @@
   "core/scripts/mcp-server/tasks/runner.ts": "d84bd35d",
   "core/scripts/mcp-server/tasks/store.ts": "2fce1a92",
   "core/scripts/mcp-server/tools/audit.ts": "3c23458b",
-  "core/scripts/mcp-server/tools/context.ts": "463df4f4",
+  "core/scripts/mcp-server/tools/context.ts": "5a00457b",
   "core/scripts/mcp-server/tools/docs.ts": "a3937aec",
   "core/scripts/mcp-server/tools/gateway/check_add_route_completeness.ts": "c205989d",
   "core/scripts/mcp-server/tools/gateway/check_add_route_status.ts": "3662b6dc",
-  "core/scripts/mcp-server/tools/gateway/check_dps.ts": "12f3b375",
+  "core/scripts/mcp-server/tools/gateway/check_dps.ts": "cea4cb8b",
   "core/scripts/mcp-server/tools/gateway/check_rahs.ts": "c37f6810",
   "core/scripts/mcp-server/tools/gateway/check_spec_sync.ts": "a683697f",
   "core/scripts/mcp-server/tools/gateway/helpers.ts": "5ed11a3b",
@@ -222,12 +222,12 @@
   "core/scripts/mcp-server/tools/review.ts": "c2d464ed",
   "core/scripts/mcp-server/types.ts": "6c5e60bc",
   "core/scripts/mcp-server.ts": "99cc6833",
-  "core/skills/add-paradigm/SKILL.md": "b74ef053",
+  "core/skills/add-paradigm/SKILL.md": "5d4fa389",
   "core/skills/session-init/SKILL.md": "bb1205b0",
   "core/specs/add-coder-npm-package/checklist.md": "ed92032c",
   "core/specs/add-coder-npm-package/spec.md": "c7d0f235",
   "core/specs/add-coder-npm-package/tasks.md": "c34301e5",
-  "core/templates/01-架构/《ADD开发工作路径与文档协同规范》.md": "4a566775",
+  "core/templates/01-架构/《ADD开发工作路径与文档协同规范》.md": "c86814d0",
   "core/templates/TERMINOLOGY.md": "e9f4f781",
   "core/templates/add-route-template-heavyweight.md": "0e107ac0",
   "core/templates/add-route-template.md": "da7022df",
@@ -267,6 +267,6 @@
   "core/templates/tasks-template.md": "2b9d0d52",
   "core/templates/tasks-template.schema.json": "fff6eb47",
   "core/tools/README.md": "26a63460",
-  "core/vocabulary/add-governance-vocabulary.md": "0bd9f725",
+  "core/vocabulary/add-governance-vocabulary.md": "62d71481",
   "shared/hooks-lib/common.sh": "441eee52"
 }

package/templates/adapters/claude/hooks/lib/common.sh

@@ -214,7 +214,7 @@ Plan 不是"文档开销"——它是代码治理的基础设施。跳过 Plan
 
 方案 A — 补 ADD 流程（招安）:
   Step 0: 读 .qoder/templates/plan-template.md → 生成 Plan
-          生成 add-route → check_dps ≥ 85
+          生成 add-route → check_dps ≥ {{dpsPass}}
   Step 1: 扩展 AgentAuditPhase（如需要）
   Step 2: 确认 agentAudit() 通道
   Step 3: 将已写代码关联到 tasks.md

package/templates/adapters/claude/hooks/lib/context-inject.sh

@@ -36,7 +36,7 @@ Plan 不是"文档开销"——它是代码治理的基础设斻。跳过 Plan
 
 方案 A — 补 ADD 流程（招安）:
   Step 0: 读 .qoder/templates/plan-template.md → 生成 Plan → .qoder/plans/{today}/{keyword}-plan-v1.md
-          生成 add-route → check_dps ≥ 85
+          生成 add-route → check_dps ≥ {{dpsPass}}
   Step 1: 扩展 AgentAuditPhase（如需要）
   Step 2: 确认 agentAudit() 通道
   Step 3: 将已写代码关联到 tasks.md

package/templates/adapters/claude/hooks/prompt-submit.sh

@@ -9,6 +9,7 @@ export CURRENT_MAGIC=$(basename "$(dirname "$HOOK_DIR")")
 COMMON_LIB="$HOOK_DIR/lib/common.sh"
 [ -f "$COMMON_LIB" ] && source "$COMMON_LIB"
 [ -f "$HOOK_DIR/lib/notify.sh" ] && source "$HOOK_DIR/lib/notify.sh" 2>/dev/null || true
+[ -f "$HOOK_DIR/lib/vocabulary.sh" ] && source "$HOOK_DIR/lib/vocabulary.sh" 2>/dev/null || true
 
 export PROJECT_DIR="${CLAUDE_PROJECT_DIR:-$PWD}"
 
@@ -40,25 +41,29 @@ EOF
   exit 0
 fi
 
-# ─── 开发关键词检测 ───
-if ! echo "$prompt" | grep -qiE '开发|改功能|修.?bug|加需求|新增|重构|实施|验收|继续|生成plan|生成计划'; then
+# ─── 开发关键词检测（从词汇表动态加载） ───
+dev_kw=$(load_dev_keywords 2>/dev/null || true)
+if [ -z "$dev_kw" ]; then
+  exit 0
+fi
+
+if ! echo "$prompt" | grep -qiE "$dev_kw"; then
   exit 0
 fi
 
 echo "[ADD PromptSubmit] 检测到开发关键词" >&2
 
-# ─── Layer 2: 无活跃 ADD → 阻断 ───
+# ─── Layer 2: 无活跃 ADD → 提示（不阻断） ───
 state=$(detect_active_add 2>/dev/null || true)
 if [ -z "$state" ]; then
   cat <<'EOF'
-[ADD 强制规则] 检测到开发任务。你必须先执行 add-paradigm SKILL 完成 ADD 工作流:
+[ADD 提示] 检测到开发任务，但无活跃 ADD Plan。建议先执行 add-paradigm SKILL:
   Step 0: 文档先行 (Plan → Review → Specs)
   Step 3: 代码实现 + 审计植入
   Step 8: 收敛判断
-如果 add-paradigm SKILL 尚未激活，请先调用它。
 EOF
-  write_hook_event "prompt-submit" "deny" "$prompt" "无活跃 ADD Plan 下检测到开发任务" "no-active-plan" "none" 2>/dev/null || true
-  exit 2
+  write_hook_event "prompt-submit" "info" "$prompt" "无活跃 ADD Plan 下检测到开发任务" "no-active-plan" "none" 2>/dev/null || true
+  exit 0
 fi
 
 # ─── Layer 3: 有活跃 ADD → 注入状态 + 模板全文 ───
@@ -78,9 +83,9 @@ if [ -f "$HOOK_JSONL" ]; then
   TOTAL=$(grep -c "\"ts\":\"${TODAY}" "$HOOK_JSONL" 2>/dev/null || echo 0)
   NO_PLAN=$(grep "\"ts\":\"${TODAY}" "$HOOK_JSONL" 2>/dev/null | grep -c '"planKeyword":"no-active-plan"' || echo 0)
   if [ "$TOTAL" -gt 0 ] 2>/dev/null; then
-    echo "[Hook 治理] 今日拦截: ${TOTAL} 次 | 无 Plan 违规: ${NO_PLAN} 次"
+    echo "[Hook 治理] 今日提示: ${TOTAL} 次 | 无 Plan 提示: ${NO_PLAN} 次"
     if [ "$NO_PLAN" -ge 10 ] 2>/dev/null; then
-      echo "[Hook ⚠️] 无 Plan 违规已达 ${NO_PLAN} 次（≥10），建议创建 Plan"
+      echo "[Hook ⚠️] 无 Plan 提示已达 ${NO_PLAN} 次（≥10），建议创建 Plan"
     fi
   fi
 fi

package/templates/adapters/codex/hooks/lib/common.sh

@@ -214,7 +214,7 @@ Plan 不是"文档开销"——它是代码治理的基础设施。跳过 Plan
 
 方案 A — 补 ADD 流程（招安）:
   Step 0: 读 .qoder/templates/plan-template.md → 生成 Plan
-          生成 add-route → check_dps ≥ 85
+          生成 add-route → check_dps ≥ {{dpsPass}}
   Step 1: 扩展 AgentAuditPhase（如需要）
   Step 2: 确认 agentAudit() 通道
   Step 3: 将已写代码关联到 tasks.md

package/templates/adapters/codex/hooks/lib/context-inject.sh

@@ -36,7 +36,7 @@ Plan 不是"文档开销"——它是代码治理的基础设斻。跳过 Plan
 
 方案 A — 补 ADD 流程（招安）:
   Step 0: 读 .qoder/templates/plan-template.md → 生成 Plan → .qoder/plans/{today}/{keyword}-plan-v1.md
-          生成 add-route → check_dps ≥ 85
+          生成 add-route → check_dps ≥ {{dpsPass}}
   Step 1: 扩展 AgentAuditPhase（如需要）
   Step 2: 确认 agentAudit() 通道
   Step 3: 将已写代码关联到 tasks.md

package/templates/adapters/codex/hooks/prompt-submit.sh

@@ -53,19 +53,15 @@ fi
 # ─── Layer 2/3: 按活跃 ADD 分流 ───
 state=$(detect_active_add 2>/dev/null || true)
 if [ -z "$state" ]; then
-  # Layer 2: 无活跃 ADD → 强制启动
+  # Layer 2: 无活跃 ADD → 提示启动（不阻断）
   cat >&2 <<'EOF'
-[ADD 强制规则] 检测到开发任务。你必须先执行 add-paradigm SKILL 完成 ADD 工作流:
+[ADD 提示] 检测到开发任务，但无活跃 ADD Plan。建议先执行 add-paradigm SKILL:
   Step 0: 文档先行 (Plan → Review → Specs)
-  Step 1: 扩展 AgentAuditPhase
-  Step 2: 确认 agentAudit() 通道
   Step 3: 代码实现 + 审计植入
-  ...
   Step 8: 收敛判断
-如果 add-paradigm SKILL 尚未激活，请先调用它。
 EOF
-  write_hook_event "prompt-submit" "deny" "$prompt" "无活跃 ADD Plan 下检测到开发任务" "no-active-plan" "none" 2>/dev/null || true
-  exit 2
+  write_hook_event "prompt-submit" "info" "$prompt" "无活跃 ADD Plan 下检测到开发任务" "no-active-plan" "none" 2>/dev/null || true
+  exit 0
 fi
 
 # Layer 3: 有活跃 ADD → 注入状态
@@ -82,9 +78,9 @@ if [ -f "$HOOK_JSONL" ]; then
   TOTAL=$(grep -c "\"ts\":\"${TODAY}" "$HOOK_JSONL" 2>/dev/null || echo 0)
   NO_PLAN=$(grep "\"ts\":\"${TODAY}" "$HOOK_JSONL" 2>/dev/null | grep -c '"planKeyword":"no-active-plan"' || echo 0)
   if [ "$TOTAL" -gt 0 ] 2>/dev/null; then
-    echo "[Hook 治理] 今日拦截: ${TOTAL} 次 | 无 Plan 违规: ${NO_PLAN} 次"
+    echo "[Hook 治理] 今日提示: ${TOTAL} 次 | 无 Plan 提示: ${NO_PLAN} 次"
     if [ "$NO_PLAN" -ge 10 ] 2>/dev/null; then
-      echo "[Hook ⚠️] 无 Plan 违规已达 ${NO_PLAN} 次（≥10），建议创建 Plan 或检查 hooks 误报"
+      echo "[Hook ⚠️] 无 Plan 提示已达 ${NO_PLAN} 次（≥10），建议创建 Plan 或检查 hooks 误报"
     fi
   fi
 fi

package/templates/adapters/qoder/hooks/lib/common.sh

@@ -214,7 +214,7 @@ Plan 不是"文档开销"——它是代码治理的基础设施。跳过 Plan
 
 方案 A — 补 ADD 流程（招安）:
   Step 0: 读 .qoder/templates/plan-template.md → 生成 Plan
-          生成 add-route → check_dps ≥ 85
+          生成 add-route → check_dps ≥ {{dpsPass}}
   Step 1: 扩展 AgentAuditPhase（如需要）
   Step 2: 确认 agentAudit() 通道
   Step 3: 将已写代码关联到 tasks.md

package/templates/adapters/qoder/hooks/lib/context-inject.sh

@@ -36,7 +36,7 @@ Plan 不是"文档开销"——它是代码治理的基础设斻。跳过 Plan
 
 方案 A — 补 ADD 流程（招安）:
   Step 0: 读 .qoder/templates/plan-template.md → 生成 Plan → .qoder/plans/{today}/{keyword}-plan-v1.md
-          生成 add-route → check_dps ≥ 85
+          生成 add-route → check_dps ≥ {{dpsPass}}
   Step 1: 扩展 AgentAuditPhase（如需要）
   Step 2: 确认 agentAudit() 通道
   Step 3: 将已写代码关联到 tasks.md

package/templates/adapters/qoder/hooks/prompt-submit.sh

@@ -55,22 +55,18 @@ if ! echo "$prompt" | grep -qiE "$dev_kw"; then
   exit 0
 fi
 
-# ─── Layer 2/3: 按活跃 ADD 分流（原 Layer 2/3，不修改） ───
+# ─── Layer 2/3: 按活跃 ADD 分流 ───
 state=$(detect_active_add 2>/dev/null || true)
 if [ -z "$state" ]; then
-  # Layer 2: 无活跃 ADD → 强制启动
+  # Layer 2: 无活跃 ADD → 提示启动（不阻断）
   cat >&2 <<'EOF'
-[ADD 强制规则] 检测到开发任务。你必须先执行 add-paradigm SKILL 完成 ADD 工作流:
+[ADD 提示] 检测到开发任务，但无活跃 ADD Plan。建议先执行 add-paradigm SKILL:
   Step 0: 文档先行 (Plan → Review → Specs)
-  Step 1: 扩展 AgentAuditPhase
-  Step 2: 确认 agentAudit() 通道
   Step 3: 代码实现 + 审计植入
-  ...
   Step 8: 收敛判断
-如果 add-paradigm SKILL 尚未激活，请先调用它。
 EOF
-  write_hook_event "prompt-submit" "deny" "$prompt" "无活跃 ADD Plan 下检测到开发任务" "no-active-plan" "none" 2>/dev/null || true
-  exit 2
+  write_hook_event "prompt-submit" "info" "$prompt" "无活跃 ADD Plan 下检测到开发任务" "no-active-plan" "none" 2>/dev/null || true
+  exit 0
 fi
 
 # Layer 3: 有活跃 ADD → 注入状态
@@ -88,9 +84,9 @@ if [ -f "$HOOK_JSONL" ]; then
   TOTAL=$(grep -c "\"ts\":\"${TODAY}" "$HOOK_JSONL" 2>/dev/null || echo 0)
   NO_PLAN=$(grep "\"ts\":\"${TODAY}" "$HOOK_JSONL" 2>/dev/null | grep -c '"planKeyword":"no-active-plan"' || echo 0)
   if [ "$TOTAL" -gt 0 ] 2>/dev/null; then
-    add_ctx="${add_ctx}\n[Hook 治理] 今日拦截: ${TOTAL} 次 | 无 Plan 违规: ${NO_PLAN} 次"
+    add_ctx="${add_ctx}\n[Hook 治理] 今日提示: ${TOTAL} 次 | 无 Plan 提示: ${NO_PLAN} 次"
     if [ "$NO_PLAN" -ge 10 ] 2>/dev/null; then
-      add_ctx="${add_ctx}\n[Hook ⚠️] 无 Plan 违规已达 ${NO_PLAN} 次（≥10），建议创建 Plan"
+      add_ctx="${add_ctx}\n[Hook ⚠️] 无 Plan 提示已达 ${NO_PLAN} 次（≥10），建议创建 Plan"
     fi
   fi
 fi

package/templates/adapters/trae/hooks/lib/common.sh

@@ -214,7 +214,7 @@ Plan 不是"文档开销"——它是代码治理的基础设施。跳过 Plan
 
 方案 A — 补 ADD 流程（招安）:
   Step 0: 读 .qoder/templates/plan-template.md → 生成 Plan
-          生成 add-route → check_dps ≥ 85
+          生成 add-route → check_dps ≥ {{dpsPass}}
   Step 1: 扩展 AgentAuditPhase（如需要）
   Step 2: 确认 agentAudit() 通道
   Step 3: 将已写代码关联到 tasks.md

package/templates/adapters/trae/hooks/lib/context-inject.sh

@@ -36,7 +36,7 @@ Plan 不是"文档开销"——它是代码治理的基础设斻。跳过 Plan
 
 方案 A — 补 ADD 流程（招安）:
   Step 0: 读 .qoder/templates/plan-template.md → 生成 Plan → .qoder/plans/{today}/{keyword}-plan-v1.md
-          生成 add-route → check_dps ≥ 85
+          生成 add-route → check_dps ≥ {{dpsPass}}
   Step 1: 扩展 AgentAuditPhase（如需要）
   Step 2: 确认 agentAudit() 通道
   Step 3: 将已写代码关联到 tasks.md

package/templates/adapters/trae/hooks/prompt-submit.sh

@@ -53,19 +53,15 @@ fi
 # ─── Layer 2/3: 按活跃 ADD 分流 ───
 state=$(detect_active_add 2>/dev/null || true)
 if [ -z "$state" ]; then
-  # Layer 2: 无活跃 ADD → 强制启动
+  # Layer 2: 无活跃 ADD → 提示启动（不阻断）
   cat >&2 <<'EOF'
-[ADD 强制规则] 检测到开发任务。你必须先执行 add-paradigm SKILL 完成 ADD 工作流:
+[ADD 提示] 检测到开发任务，但无活跃 ADD Plan。建议先执行 add-paradigm SKILL:
   Step 0: 文档先行 (Plan → Review → Specs)
-  Step 1: 扩展 AgentAuditPhase
-  Step 2: 确认 agentAudit() 通道
   Step 3: 代码实现 + 审计植入
-  ...
   Step 8: 收敛判断
-如果 add-paradigm SKILL 尚未激活，请先调用它。
 EOF
-  write_hook_event "prompt-submit" "deny" "$prompt" "无活跃 ADD Plan 下检测到开发任务" "no-active-plan" "none" 2>/dev/null || true
-  exit 2
+  write_hook_event "prompt-submit" "info" "$prompt" "无活跃 ADD Plan 下检测到开发任务" "no-active-plan" "none" 2>/dev/null || true
+  exit 0
 fi
 
 # Layer 3: 有活跃 ADD → 注入状态
@@ -82,9 +78,9 @@ if [ -f "$HOOK_JSONL" ]; then
   TOTAL=$(grep -c "\"ts\":\"${TODAY}" "$HOOK_JSONL" 2>/dev/null || echo 0)
   NO_PLAN=$(grep "\"ts\":\"${TODAY}" "$HOOK_JSONL" 2>/dev/null | grep -c '"planKeyword":"no-active-plan"' || echo 0)
   if [ "$TOTAL" -gt 0 ] 2>/dev/null; then
-    echo "[Hook 治理] 今日拦截: ${TOTAL} 次 | 无 Plan 违规: ${NO_PLAN} 次"
+    echo "[Hook 治理] 今日提示: ${TOTAL} 次 | 无 Plan 提示: ${NO_PLAN} 次"
     if [ "$NO_PLAN" -ge 10 ] 2>/dev/null; then
-      echo "[Hook ⚠️] 无 Plan 违规已达 ${NO_PLAN} 次（≥10），建议创建 Plan 或检查 hooks 误报"
+      echo "[Hook ⚠️] 无 Plan 提示已达 ${NO_PLAN} 次（≥10），建议创建 Plan 或检查 hooks 误报"
     fi
   fi
 fi

package/templates/adapters/vscode/hooks/lib/common.sh

@@ -214,7 +214,7 @@ Plan 不是"文档开销"——它是代码治理的基础设施。跳过 Plan
 
 方案 A — 补 ADD 流程（招安）:
   Step 0: 读 .qoder/templates/plan-template.md → 生成 Plan
-          生成 add-route → check_dps ≥ 85
+          生成 add-route → check_dps ≥ {{dpsPass}}
   Step 1: 扩展 AgentAuditPhase（如需要）
   Step 2: 确认 agentAudit() 通道
   Step 3: 将已写代码关联到 tasks.md

package/templates/adapters/vscode/hooks/lib/context-inject.sh

@@ -36,7 +36,7 @@ Plan 不是"文档开销"——它是代码治理的基础设斻。跳过 Plan
 
 方案 A — 补 ADD 流程（招安）:
   Step 0: 读 .qoder/templates/plan-template.md → 生成 Plan → .qoder/plans/{today}/{keyword}-plan-v1.md
-          生成 add-route → check_dps ≥ 85
+          生成 add-route → check_dps ≥ {{dpsPass}}
   Step 1: 扩展 AgentAuditPhase（如需要）
   Step 2: 确认 agentAudit() 通道
   Step 3: 将已写代码关联到 tasks.md

package/templates/adapters/vscode/hooks/prompt-submit.sh

@@ -9,6 +9,7 @@ export CURRENT_MAGIC=$(basename "$(dirname "$HOOK_DIR")")
 COMMON_LIB="$HOOK_DIR/lib/common.sh"
 [ -f "$COMMON_LIB" ] && source "$COMMON_LIB"
 [ -f "$HOOK_DIR/lib/notify.sh" ] && source "$HOOK_DIR/lib/notify.sh" 2>/dev/null || true
+[ -f "$HOOK_DIR/lib/vocabulary.sh" ] && source "$HOOK_DIR/lib/vocabulary.sh" 2>/dev/null || true
 
 export PROJECT_DIR="$PWD"
 
@@ -40,25 +41,29 @@ EOF
   exit 0
 fi
 
-# ─── 开发关键词检测 ───
-if ! echo "$prompt" | grep -qiE '开发|改功能|修.?bug|加需求|新增|重构|实施|验收|继续|生成plan|生成计划'; then
+# ─── 开发关键词检测（从词汇表动态加载） ───
+dev_kw=$(load_dev_keywords 2>/dev/null || true)
+if [ -z "$dev_kw" ]; then
+  exit 0
+fi
+
+if ! echo "$prompt" | grep -qiE "$dev_kw"; then
   exit 0
 fi
 
 echo "[ADD PromptSubmit] 检测到开发关键词" >&2
 
-# ─── Layer 2: 无活跃 ADD → 阻断 ───
+# ─── Layer 2: 无活跃 ADD → 提示（不阻断） ───
 state=$(detect_active_add 2>/dev/null || true)
 if [ -z "$state" ]; then
   cat <<'EOF'
-[ADD 强制规则] 检测到开发任务。你必须先执行 add-paradigm SKILL 完成 ADD 工作流:
+[ADD 提示] 检测到开发任务，但无活跃 ADD Plan。建议先执行 add-paradigm SKILL:
   Step 0: 文档先行 (Plan → Review → Specs)
   Step 3: 代码实现 + 审计植入
   Step 8: 收敛判断
-如果 add-paradigm SKILL 尚未激活，请先调用它。
 EOF
-  write_hook_event "prompt-submit" "deny" "$prompt" "无活跃 ADD Plan 下检测到开发任务" "no-active-plan" "none" 2>/dev/null || true
-  exit 2
+  write_hook_event "prompt-submit" "info" "$prompt" "无活跃 ADD Plan 下检测到开发任务" "no-active-plan" "none" 2>/dev/null || true
+  exit 0
 fi
 
 # ─── Layer 3: 有活跃 ADD → 注入状态 + 模板全文 ───
@@ -78,9 +83,9 @@ if [ -f "$HOOK_JSONL" ]; then
   TOTAL=$(grep -c "\"ts\":\"${TODAY}" "$HOOK_JSONL" 2>/dev/null || echo 0)
   NO_PLAN=$(grep "\"ts\":\"${TODAY}" "$HOOK_JSONL" 2>/dev/null | grep -c '"planKeyword":"no-active-plan"' || echo 0)
   if [ "$TOTAL" -gt 0 ] 2>/dev/null; then
-    echo "[Hook 治理] 今日拦截: ${TOTAL} 次 | 无 Plan 违规: ${NO_PLAN} 次"
+    echo "[Hook 治理] 今日提示: ${TOTAL} 次 | 无 Plan 提示: ${NO_PLAN} 次"
     if [ "$NO_PLAN" -ge 10 ] 2>/dev/null; then
-      echo "[Hook ⚠️] 无 Plan 违规已达 ${NO_PLAN} 次（≥10），建议创建 Plan"
+      echo "[Hook ⚠️] 无 Plan 提示已达 ${NO_PLAN} 次（≥10），建议创建 Plan"
     fi
   fi
 fi

package/templates/core/agents/add-flow-guardian.md

@@ -67,7 +67,7 @@ Step N-1 出口     ──→  Guardian 出口检查  ──→  BLOCKED? 回退
 #### Step 1 入口（功能分析）
 | 检查项 | 方法 | 不通过 |
 |--------|------|--------|
-| DPS 闸门 ≥ 85 | 读取 add-route §0.8 的最新 `check_dps` 结果（如无记录则提醒主 agent 先调用） | **BLOCKED** — 回退 Step 0 补齐文档 |
+| DPS 闸门 ≥ {{dpsPass}} | 读取 add-route §0.8 的最新 `check_dps` 结果（如无记录则提醒主 agent 先调用） | **BLOCKED** — 回退 Step 0 补齐文档 |
 | 0.6.5 Review 回流 | 检查 add-route Step 0 产出中是否已勾选 Review 回流项；如 add-route 无此勾选项但 reviews/ 下有 review 文件 → 提醒主 agent 先执行回流 | **BLOCKED** |
 | 原子闭包判定 | 检查 add-route 是否包含"原子闭包三可性判定"章节 | **BLOCKED** — 回退 Step 0.7 |
 
@@ -126,9 +126,9 @@ Step N-1 出口     ──→  Guardian 出口检查  ──→  BLOCKED? 回退
 | Review 文件存在 | 确认 `{{magicDir}}/reviews/` 下有对应的 review 文件 | 缺失 → **BLOCKED** |
 | add-route 已生成 | 确认 `{{magicDir}}/plans/` 下（含日期子目录）有 `*add-route*.md` | 缺失 → **BLOCKED** |
 | 项目文档已更新 | 检查 add-route Step 0 产出中是否勾选了"项目文档已更新"或"无需更新声明已记录" | 未勾选 → **FAIL** |
-| 0.6.5 回流完成 | 检查 add-route §0.8 是否记录了 DPS ≥ 85（或有 DPS 待通过标记） | 无记录 → **FAIL** |
+| 0.6.5 回流完成 | 检查 add-route §0.8 是否记录了 DPS ≥ {{dpsPass}}（或有 DPS 待通过标记） | 无记录 → **FAIL** |
 | 原子闭包判定已执行 | add-route 含"原子闭包三可性判定"章节且有结论 | 缺失 → **BLOCKED** |
-| **DPS 闸门 ≥ 85** | **此为主 agent 在出口前必须调用的 MCP 闸门。Guardian 检查 add-route §0.8 中是否已记录 DPS 结果** | **DPS < 85 或无记录 → BLOCKED** |
+| **DPS 闸门 ≥ {{dpsPass}}** | **此为主 agent 在出口前必须调用的 MCP 闸门。Guardian 检查 add-route §0.8 中是否已记录 DPS 结果** | **DPS < {{dpsPass}} 或无记录 → BLOCKED** |
 | TypeScript 编译 | 检查 add-route 中是否记录了主 agent 的 tsc 编译结果（主 agent 应先执行 `npx tsc --noEmit`） | 无记录 → **FAIL** |
 | eslint 检查 | 检查 add-route 中是否记录了主 agent 的 eslint 结果（主 agent 应先执行 `npx eslint`） | 无记录 → **FAIL** |
 
@@ -224,7 +224,7 @@ Plan: {✅ 存在 | ❌ 缺失}
 add-route: {文件路径 | ❌ 缺失 → BLOCKED}
 Specs 三元组: {spec/tasks/checklist 各状态}
 Review: {文件路径 | ❌ 缺失}
-DPS 记录: {已通过 ≥85 | 未通过 | 无记录}
+DPS 记录: {已通过 ≥{{dpsPass}} | 未通过 | 无记录}
 RAHS 记录: {已通过 ≥90 | 未通过 | 无记录}
 
 【add-route 勾选状态】

package/templates/core/hooks/lib/common.sh

@@ -214,7 +214,7 @@ Plan 不是"文档开销"——它是代码治理的基础设施。跳过 Plan
 
 方案 A — 补 ADD 流程（招安）:
   Step 0: 读 .qoder/templates/plan-template.md → 生成 Plan
-          生成 add-route → check_dps ≥ 85
+          生成 add-route → check_dps ≥ {{dpsPass}}
   Step 1: 扩展 AgentAuditPhase（如需要）
   Step 2: 确认 agentAudit() 通道
   Step 3: 将已写代码关联到 tasks.md

package/templates/core/hooks/lib/context-inject.sh

@@ -36,7 +36,7 @@ Plan 不是"文档开销"——它是代码治理的基础设斻。跳过 Plan
 
 方案 A — 补 ADD 流程（招安）:
   Step 0: 读 .qoder/templates/plan-template.md → 生成 Plan → .qoder/plans/{today}/{keyword}-plan-v1.md
-          生成 add-route → check_dps ≥ 85
+          生成 add-route → check_dps ≥ {{dpsPass}}
   Step 1: 扩展 AgentAuditPhase（如需要）
   Step 2: 确认 agentAudit() 通道
   Step 3: 将已写代码关联到 tasks.md

package/templates/core/hooks/prompt-submit.sh

@@ -53,19 +53,15 @@ fi
 # ─── Layer 2/3: 按活跃 ADD 分流 ───
 state=$(detect_active_add 2>/dev/null || true)
 if [ -z "$state" ]; then
-  # Layer 2: 无活跃 ADD → 强制启动
+  # Layer 2: 无活跃 ADD → 提示启动（不阻断）
   cat >&2 <<'EOF'
-[ADD 强制规则] 检测到开发任务。你必须先执行 add-paradigm SKILL 完成 ADD 工作流:
+[ADD 提示] 检测到开发任务，但无活跃 ADD Plan。建议先执行 add-paradigm SKILL:
   Step 0: 文档先行 (Plan → Review → Specs)
-  Step 1: 扩展 AgentAuditPhase
-  Step 2: 确认 agentAudit() 通道
   Step 3: 代码实现 + 审计植入
-  ...
   Step 8: 收敛判断
-如果 add-paradigm SKILL 尚未激活，请先调用它。
 EOF
-  write_hook_event "prompt-submit" "deny" "$prompt" "无活跃 ADD Plan 下检测到开发任务" "no-active-plan" "none" 2>/dev/null || true
-  exit 2
+  write_hook_event "prompt-submit" "info" "$prompt" "无活跃 ADD Plan 下检测到开发任务" "no-active-plan" "none" 2>/dev/null || true
+  exit 0
 fi
 
 # Layer 3: 有活跃 ADD → 注入状态
@@ -82,9 +78,9 @@ if [ -f "$HOOK_JSONL" ]; then
   TOTAL=$(grep -c "\"ts\":\"${TODAY}" "$HOOK_JSONL" 2>/dev/null || echo 0)
   NO_PLAN=$(grep "\"ts\":\"${TODAY}" "$HOOK_JSONL" 2>/dev/null | grep -c '"planKeyword":"no-active-plan"' || echo 0)
   if [ "$TOTAL" -gt 0 ] 2>/dev/null; then
-    echo "[Hook 治理] 今日拦截: ${TOTAL} 次 | 无 Plan 违规: ${NO_PLAN} 次"
+    echo "[Hook 治理] 今日提示: ${TOTAL} 次 | 无 Plan 提示: ${NO_PLAN} 次"
     if [ "$NO_PLAN" -ge 10 ] 2>/dev/null; then
-      echo "[Hook ⚠️] 无 Plan 违规已达 ${NO_PLAN} 次（≥10），建议创建 Plan 或检查 hooks 误报"
+      echo "[Hook ⚠️] 无 Plan 提示已达 ${NO_PLAN} 次（≥10），建议创建 Plan 或检查 hooks 误报"
     fi
   fi
 fi

package/templates/core/rules/project_rules.md

@@ -23,7 +23,7 @@
 | P2 | ADD-10 意图与实现的语义鸿沟 | 意图与实现是两层抽象，对齐只能靠显式对照 | ✅ 已实现 | |
 | P2 | ADD-11 证据的不可再生性 | 运行时上下文不可再生，证据必须优先持久化 | 🔶 待实现 | 需 runtime hook + predev 脚本 |
 | P2 | ADD-12 双源头漂移的必然性 | 代码与文档无同步必漂移，漂移代价由下一个接手者承担 | ✅ 已实现 | |
-| P2 | ADD-13 DPS 上游文档质量闸门 | Plan 概括度 → Review 注意力稀释 → Specs 遗漏 → 实现偏差。`check_dps`（DPS ≥ 85）在 Step 0 末尾量化阻断 | ✅ 已实现 | 2026-06-11 引入 |
+| P2 | ADD-13 DPS 上游文档质量闸门 | Plan 概括度 → Review 注意力稀释 → Specs 遗漏 → 实现偏差。`check_dps`（DPS ≥ {{dpsPass}}）在 Step 0 末尾量化阻断 | ✅ 已实现 | 2026-06-11 引入 |
 | P2 | ADD-14 RAHS 下游执行健康度闸门 | 范围保真度 + 类型安全 + 审计完整度 + Spec 合规 + 阶段对称性。`check_rahs`（RAHS ≥ 90）在 Step 4/8 量化阻断 | ✅ 已实现 | 2026-06-11 引入 |
 | P2 | ADD-15 add-route 闭环自检 | Step 3 代码完成后必须调用 `check_add_route_completeness` 扫描 add-route Step 完成度，防止执行遗漏。返回 complete 方可进入 Step 3.5 | ✅ 已实现 | 2026-06-11 引入 |
 | P2 | ADD-17 HITL 磋商临时文件机制 | doc-format-guard 要求 Plan/Review 文件包含完整章节才放行，HITL 第一步只写总览表会被 guard 阻断。用 `{name}.temporary.md` 做磋商（不受 guard 检查），人类拍板后再写正式文件并删除 temporary.md | ✅ 已实现 | 2026-07-23 引入 |

package/templates/core/scripts/mcp-server/tools/context.ts

@@ -161,7 +161,7 @@ export function registerContextTools(server: McpServer) {
             const planKw = activePlan.replace(/-plan-v\d+\.md$/, "")
             const sd = existsSync(specsDir) ? await readdir(specsDir) : []
             if (!sd.some(d => d.toLowerCase().includes(planKw.toLowerCase())) && reviewBackflowRate >= 70) { todoItems.push("3. [Step 0] 生成 Specs 三元组") }
-            if (todoItems.length === 0) { todoItems.push("✅ ADD 就绪 — check_dps ≥ 85 后可进入 Step 1") }
+            if (todoItems.length === 0) { todoItems.push("✅ ADD 就绪 — check_dps ≥ {{dpsPass}} 后可进入 Step 1") }
           }
           for (const item of todoItems) parts.push(item)
           parts.push(""); parts.push("快速指令: 说「执行 add-paradigm Step 0」进入文档先行流程")

package/templates/core/scripts/mcp-server/tools/gateway/check_dps.ts

@@ -34,7 +34,7 @@ export function registerCheckDps(server: McpServer) {
     "check_dps",
     {
       description:
-        "DPS 闸门。四维复合评分: 语义相关性(TF-IDF/Jaccard) + 信息熵匹配(香农/Deng) + CPM关键路径 + 结构完整度 + FFT自适应权重。DPS >= 85 可进入 Step 1。",
+        `DPS 闸门。四维复合评分: 语义相关性(TF-IDF/Jaccard) + 信息熵匹配(香农/Deng) + CPM关键路径 + 结构完整度 + FFT自适应权重。DPS >= ${CFG.THRESHOLD_PASS} 可进入 Step 1。`,
       inputSchema: z.object({
         planKeyword: z.string().describe("Plan 文件的关键词"),
       }),

package/templates/core/skills/add-paradigm/SKILL.md

@@ -280,7 +280,7 @@ find_related_docs({ query: "功能关键词" })
 
 5. **验证回流完整性**：
    - 对 Review 中每个 P0/P1 问题，确认能在 Plan/Specs 中找到对应的"已修正后的文字"
-   - 调用 `check_dps({ planKeyword: "..." })` 确认 DPS ≥ 85——如果回流后 DPS 仍然不达标，说明 Review 本身覆盖度不足或回流遗漏
+   - 调用 `check_dps({ planKeyword: "..." })` 确认 DPS ≥ {{dpsPass}}——如果回流后 DPS 仍然不达标，说明 Review 本身覆盖度不足或回流遗漏
 
 6. **记录审计**：调用 `record_dev_operation`（`targetType: "PLAN"`，`action: "DOC_UPDATED"`，`beforeState`: Review 问题编号列表，`afterState`: 每个问题的回流位置）
 
@@ -399,7 +399,7 @@ Plan 级闭包: {业务功能描述}
 - [ ] 文档合约一致性已确认
 - [ ] **[0.6.5] Review 结论已回流至 Plan 与 Specs**——Review 中每个 P0/P1 问题在 Plan/Specs 中有对应的修正后文字
 - [ ] 原子闭包判定已执行（Plan 级：确认单一业务功能；轮次级：确认文件边界独立、无跨轮修改、每轮可独立验证），轮次拆分方案已写入 add-route
-- [ ] **DPS 上游文档质量闸门已通过**——调用 `check_dps({ planKeyword: "<Plan 核心关键词>" })`，DPS ≥ 85 方可进入 Step 1。未通过时回退补齐短板（Plan 粒度不足 / Review 覆盖维度缺失 / Specs Requirements 遗漏）。详见 [ADD 协同规范 §八]({{projectRoot}}/{{docsDir}}/knowledge/01-架构/《ADD开发工作路径与文档协同规范》.md#八双质量闸门dps-与-rahs)
+- [ ] **DPS 上游文档质量闸门已通过**——调用 `check_dps({ planKeyword: "<Plan 核心关键词>" })`，DPS ≥ {{dpsPass}} 方可进入 Step 1。未通过时回退补齐短板（Plan 粒度不足 / Review 覆盖维度缺失 / Specs Requirements 遗漏）。详见 [ADD 协同规范 §八]({{projectRoot}}/{{docsDir}}/knowledge/01-架构/《ADD开发工作路径与文档协同规范》.md#八双质量闸门dps-与-rahs)
 
 ---
 

package/templates/core/templates/01-/346/236/266/346/236/204//343/200/212ADD/345/274/200/345/217/221/345/267/245/344/275/234/350/267/257/345/276/204/344/270/216/346/226/207/346/241/243/345/215/217/345/220/214/350/247/204/350/214/203/343/200/213.md

@@ -24,7 +24,7 @@ ADD 不是"写代码时顺便打日志"，而是一套覆盖全开发周期的
 │            │  产物：{{magicDir}}/plans/{需求域名}-plan-v{n}.md                        │
 │            │        {{magicDir}}/plans/{需求域名}-add-route-v{n}.md                   │
 │            │        {{magicDir}}/reviews/{需求域名}-review-v{n}.md                    │
-│            │  阈值：check_dps（DPS ≥ 85 方可进入 Step 1）                        │
+│            │  阈值：check_dps（DPS ≥ {{dpsPass}} 方可进入 Step 1）                        │
 ├────────────┼─────────────────────────────────────────────────────────────────┤
 │  Step 1    │  功能分析与审计阶段定义                                            │
 │  审计定义   │  分析业务阶段 → 扩展 AgentAuditPhase 联合类型 → 确认审计通道        │

package/templates/core/vocabulary/add-governance-vocabulary.md

@@ -56,7 +56,7 @@
 
 | 触发词 | LLM 默认操作 | 优先级 |
 |--------|-------------|--------|
-| `DPS` / `DPS闸门` | 调 `check_dps({ planKeyword: "..." })`，DPS ≥ 85 通过 | 🔴 P0 |
+| `DPS` / `DPS闸门` | 调 `check_dps({ planKeyword: "..." })`，DPS ≥ {{dpsPass}} 通过 | 🔴 P0 |
 | `RAHS` / `RAHS闸门` | 调 `check_rahs({ planKeyword: "..." })`，RAHS ≥ 90 通过 | 🔴 P0 |
 | `Guardian` / `门禁` / `add-flow-guardian` | 调 `add-flow-guardian` Subagent（入口或出口模式） | 🔴 P0 |
 | `add-route 闭环` / `闭环自检` | 调 `check_add_route_completeness({ planKeyword: "..." })` | 🟡 P1 |
@@ -197,7 +197,7 @@ LLM: "DPS 偏低，但不影响代码，继续 Step 1..." [忽略阻断]
 **正确输出**：
 ```
 LLM: "⛔ BLOCKED → 不得继续。修正 P0 阻断项后重新跑 DPS 门禁。"
-[回退到 Step 0，修复问题，重新 DPS，直到 ≥ 85]
+[回退到 Step 0，修复问题，重新 DPS，直到 ≥ {{dpsPass}}]
 ```
 
 ### 场景 6: "继续" 被误解为跳到代码
@@ -278,7 +278,7 @@ LLM: "验收通过 → 自动写 devlog日志(走mcp)（无需用户提醒）→
 
 | 优先级 | 触发词 | LLM 默认操作 |
 |:--:|------|-------------|
-| P0 | `开发` / `改功能` / `修.?bug` / `fix.?bug` / `加需求` / `新增` / `重构` / `实现` / `接入` / `改造` / `升级` / `加个` / `添加功能` / `新建` / `改一下` / `修改.*逻辑` / `优化.*代码` / `对接` / `迁移` / `重写` / `implement` / `refactor` / `feature` | 开发任务检测——无活跃 ADD 时强制退出并提示启动 add-paradigm SKILL；有活跃 Plan 时注入当前 Step/轮次/handoff 上下文 |
+| P0 | `开发` / `改功能` / `修.?bug` / `fix.?bug` / `加需求` / `新增` / `重构` / `实现` / `接入` / `改造` / `升级` / `加个` / `添加功能` / `新建` / `改一下` / `修改.*逻辑` / `优化.*代码` / `对接` / `迁移` / `重写` / `implement` / `refactor` / `feature` | 开发任务检测——无活跃 ADD 时提示启动 add-paradigm SKILL |
 | P0 | `实施` / `开始实施` / `进入实施` | Plan 就绪后 → 进入 add-paradigm SKILL → 从当前 Step 开始执行代码实现。如无活跃 add-route 先回 Step 0.5 生成 |
 | P0 | `继续` | **上下文锁定**：沿用上一条 AI 消息中明确提出的 Step/操作（如 "继续 Step 0.5 生成 add-route？" → "继续" = 执行 Step 0.5）。**禁止**将 "继续" 理解为"跳到代码编写（Step 3）"。**禁止**在 AI 刚提出 Step N 但用户说"继续"时跳过 Step N 做别的事 |
 | P1 | `Step 0` / `文档先行` | 进入 add-paradigm Step 0：分析变更 → 更新文档 → 生成 add-route → DPS 门禁 |
@@ -306,7 +306,7 @@ LLM: "验收通过 → 自动写 devlog日志(走mcp)（无需用户提醒）→
 
 | 优先级 | 触发词 | LLM 默认操作 |
 |:--:|------|-------------|
-| P0 | `DPS` / `DPS闸门` | 调 `check_dps({ planKeyword: "..." })`，DPS ≥ 85 通过 |
+| P0 | `DPS` / `DPS闸门` | 调 `check_dps({ planKeyword: "..." })`，DPS ≥ {{dpsPass}} 通过 |
 | P0 | `RAHS` / `RAHS闸门` | 调 `check_rahs({ planKeyword: "..." })`，RAHS ≥ 90 通过 |
 | P0 | `Guardian` / `门禁` / `add-flow-guardian` | 调 `add-flow-guardian` Subagent（入口或出口模式） |
 | P1 | `add-route 闭环` / `闭环自检` | 调 `check_add_route_completeness({ planKeyword: "..." })` |