diff --git a/apps/ai-game-creator-shell/src/App.tsx b/apps/ai-game-creator-shell/src/App.tsx
index d9af07557..733e966a3 100644
--- a/apps/ai-game-creator-shell/src/App.tsx
+++ b/apps/ai-game-creator-shell/src/App.tsx
@@ -630,7 +630,16 @@ export function App({
   const [planGddHydrateBusy, setPlanGddHydrateBusy] = useState(false);
   const [planGddDecisionBusy, setPlanGddDecisionBusy] = useState(false);
   const [planGddError, setPlanGddError] = useState<string | null>(null);
-  const planGddDecisionResponseIdsRef = useRef(new Map<string, string>());
+  const planGddDecisionResponseIdsRef = useRef(
+    new Map<
+      string,
+      {
+        action: PlanGddDecisionAction;
+        comment: string | null;
+        responseId: string;
+      }
+    >(),
+  );
 
   const hydratePlanGddState = useCallback(
     async (nextProjectPath?: string) => {
@@ -681,11 +690,25 @@ export function App({
       if (!pending || !current?.displayGdd || !invoke || !targetProjectPath) {
         throw new Error('当前没有可提交的 GDD 审批决定');
       }
-      const responseKey = `${pending.approvalRequestId}:${action}`;
+      // 方案 §13.2：busy、超时与网络重试复用同一 responseId，但用户改变 action 或
+      // comment 后必须换新的。旧键只含 `approvalRequestId:action`，改写修改意见时会
+      // 带着旧 responseId 提交，命中后端「同 responseId 的审批意图不一致」硬错误。
+      // 判据方向是宁可多换不可少换：多换的最坏后果是 receipt 已存在时把 replayed 降级
+      // 成 already-decided，两者都是 Ok；少换是硬错误。
+      const previousDecision = planGddDecisionResponseIdsRef.current.get(
+        pending.approvalRequestId,
+      );
       const responseId =
-        planGddDecisionResponseIdsRef.current.get(responseKey) ??
-        `gdd-response-${crypto.randomUUID()}`;
-      planGddDecisionResponseIdsRef.current.set(responseKey, responseId);
+        previousDecision &&
+        previousDecision.action === action &&
+        previousDecision.comment === comment
+          ? previousDecision.responseId
+          : `gdd-response-${crypto.randomUUID()}`;
+      planGddDecisionResponseIdsRef.current.set(pending.approvalRequestId, {
+        action,
+        comment,
+        responseId,
+      });
       setPlanGddDecisionBusy(true);
       setPlanGddError(null);
       try {
@@ -702,6 +725,12 @@ export function App({
         });
         await hydratePlanGddState(targetProjectPath);
       } catch (error) {
+        // 方案 §18.3 要求 decision 返回后以 hydrate 对权威文件的重验为准，失败分支同样
+        // 适用：不重灌就会让卡片停在已失效的 pending 身份上，三个决定按钮仍可点，且
+        // `recoveryPending` 永远翻不成真、「重试恢复」入口不渲染，卡内没有出路。
+        // 两句顺序不能反——`hydratePlanGddState` 入口会 `setPlanGddError(null)`，
+        // 先写错误再 hydrate 等于把这条错误擦掉。它自身从不抛出，不需要再包一层。
+        await hydratePlanGddState(targetProjectPath);
         setPlanGddError(String(error));
         throw error;
       } finally {
diff --git a/apps/ai-game-creator-shell/src/features/project-workspace/GddApprovalCard.tsx b/apps/ai-game-creator-shell/src/features/project-workspace/GddApprovalCard.tsx
index 784edc2fd..71be7be2f 100644
--- a/apps/ai-game-creator-shell/src/features/project-workspace/GddApprovalCard.tsx
+++ b/apps/ai-game-creator-shell/src/features/project-workspace/GddApprovalCard.tsx
@@ -162,7 +162,10 @@ export function GddApprovalCard({
   );
 
   const submitComment = () => {
-    if (!commentAction || !comment.trim()) {
+    // 方案 §18.2：`recoveryPending` 期间只允许重试同一 ID，不允许提交决定。触发按钮
+    // 已经由 `canDecide` 门住，但弹层是打开后才可能被后台 hydrate 翻掉资格的，
+    // 所以提交口要自己再判一次，不能只靠按钮 disabled。
+    if (!canDecide || !commentAction || !comment.trim()) {
       return;
     }
     void onDecision(commentAction, comment.trim())
@@ -282,6 +285,11 @@ export function GddApprovalCard({
               placeholder="请输入原因"
               onChange={(event) => setComment(event.currentTarget.value)}
             />
+            {!canDecide ? (
+              <p className="gdd-approval-card__dialog-hint" role="status">
+                审批状态正在恢复，暂时不能提交决定。已输入的内容会保留。
+              </p>
+            ) : null}
             <div className="gdd-approval-card__dialog-actions">
               <button
                 type="button"
@@ -295,7 +303,7 @@ export function GddApprovalCard({
               </button>
               <button
                 type="button"
-                disabled={busy || !comment.trim()}
+                disabled={!canDecide || busy || !comment.trim()}
                 onClick={submitComment}
               >
                 {busy ? '提交中' : '提交决定'}
diff --git a/apps/ai-game-creator-shell/src/styles.css b/apps/ai-game-creator-shell/src/styles.css
index 69c53f884..7996f8a05 100644
--- a/apps/ai-game-creator-shell/src/styles.css
+++ b/apps/ai-game-creator-shell/src/styles.css
@@ -3504,6 +3504,14 @@ textarea {
   resize: vertical;
 }
 
+.gdd-approval-card__dialog-hint {
+  padding: 9px 10px;
+  border-radius: 7px;
+  color: #854d0e;
+  background: #fff7df;
+  font-size: 13px;
+}
+
 .panel-header {
   margin-bottom: 10px;
 }
diff --git a/apps/ai-game-creator-shell/tests/appSurface.test.ts b/apps/ai-game-creator-shell/tests/appSurface.test.ts
index 671caf74a..5b9ab4807 100644
--- a/apps/ai-game-creator-shell/tests/appSurface.test.ts
+++ b/apps/ai-game-creator-shell/tests/appSurface.test.ts
@@ -11,6 +11,7 @@ import {
   registerHomeProjectCreationTests,
   registerRecentProjectsTests,
 } from './appSurface/home.suite';
+import { registerPlanGddApprovalTests } from './appSurface/plan-gdd.suite';
 import {
   registerCanvasAssetTests,
   registerProjectAssetTests,
@@ -54,4 +55,5 @@ describe('AI 游戏创作 App 界面边界', () => {
   registerProjectAssetTests();
   registerAgentRuntimeCommandTests();
   registerCanvasAssetTests();
+  registerPlanGddApprovalTests();
 });
diff --git a/apps/ai-game-creator-shell/tests/appSurface/harness.ts b/apps/ai-game-creator-shell/tests/appSurface/harness.ts
index 0ece7d45b..956b5e093 100644
--- a/apps/ai-game-creator-shell/tests/appSurface/harness.ts
+++ b/apps/ai-game-creator-shell/tests/appSurface/harness.ts
@@ -13,6 +13,8 @@ import {
 import React from 'react';
 import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
 
+import type { PlanGddStateViewV1 } from '../../src/app/types';
+
 const nativeClipboardMock = vi.hoisted(() => ({
   text: '',
 }));
@@ -301,6 +303,150 @@ function agentRuntimeUserInputRequest({
   };
 }
 
+function createPlanGddStateView(
+  overrides: Partial<PlanGddStateViewV1> = {},
+): PlanGddStateViewV1 {
+  const gddRef = {
+    gddId: 'gdd-plan-0001',
+    version: 1,
+    fingerprint: 'sha256-serde-json-v2:1111111111111111',
+  };
+  return {
+    schemaVersion: 'plan-gdd-state-view.v1',
+    projectId: 'local-project-draft',
+    gddId: gddRef.gddId,
+    state: 'ready_for_approval',
+    session: {
+      sessionId: 'plan-session-0001',
+      sessionRevision: 3,
+      sessionFingerprint: 'sha256-serde-json-v2:2222222222222222',
+      phase: 'awaiting_gdd_approval',
+      clarificationRound: 2,
+      repairDepth: 0,
+      accumulatedAgentMillis: 42_000,
+      activeRunId: null,
+      awaitingAnswerFor: null,
+      decisionStateCounts: {
+        confirmed: 2,
+        defaultPending: 1,
+        prototypePending: 0,
+      },
+    },
+    versions: [
+      {
+        gddRef,
+        status: 'ready_for_approval',
+        approvalRequestId: 'gdd-approval-0001',
+        createdAtUtc: '2026-08-18T00:00:00Z',
+        decision: null,
+      },
+    ],
+    displayGdd: {
+      schemaVersion: 'plan-gdd.v1',
+      projectId: 'local-project-draft',
+      gddId: gddRef.gddId,
+      version: gddRef.version,
+      submissionId: 'action-0123456789abcdef01234567',
+      approvalRequestId: 'gdd-approval-0001',
+      actionFingerprint: 'a'.repeat(64),
+      agentId: 'project-planning',
+      source: 'agent-delegate',
+      runProfile: 'standard',
+      runProfileBindingFingerprint: 'b'.repeat(64),
+      rootAgentId: 'project-supervisor',
+      rootRunId: 'run-plan-root-0001',
+      delegationId: 'delegation-0001',
+      sessionId: 'plan-session-0001',
+      sourceSessionRevision: 2,
+      sourceSessionFingerprint: 'sha256-serde-json-v2:3333333333333333',
+      createdByRunId: 'run-plan-child-0001',
+      createdAtUtc: '2026-08-18T00:00:00Z',
+      fingerprint: gddRef.fingerprint,
+      game: {
+        title: '灯塔守夜人',
+        oneLiner: '在潮汐涨落之间调度光束，护送迷航的船只回港。',
+        genre: { primary: '策略', fusion: null },
+        artStyle: {
+          visualType: '像素',
+          keywords: ['夜色', '海雾'],
+          moodAndColor: '冷蓝为主，暖黄光束作为唯一高光。',
+          mvpArtBoundary: '只做灯塔与三类船只的静帧。',
+        },
+        pillars: [
+          {
+            name: '光束调度',
+            playerFeel: '在有限视野里做取舍。',
+            mechanism: '每回合只能照亮一个扇区。',
+            decisionState: 'confirmed',
+            basis: null,
+          },
+        ],
+        coreLoop: ['观察潮汐', '分配光束', '结算返港'],
+        targetUsers: {
+          coreUsers: '喜欢短局策略的玩家',
+          preferences: '偏好可预测的规则',
+          sessionLength: '单局 5 分钟',
+          referenceGames: ['灯塔物语'],
+        },
+        platformFacts: {
+          runtime: 'web',
+          viewports: ['desktop', 'mobile'],
+          inputs: ['pointer'],
+          preview: '本地预览',
+        },
+        mvpSystems: [
+          {
+            system: '潮汐时钟',
+            minimalFunction: '固定三段潮汐循环。',
+            whyRequired: '没有它就没有节奏压力。',
+            verifyMethod: '观察一局内三段是否各触发一次。',
+            decisionState: 'confirmed',
+            basis: null,
+          },
+        ],
+        outOfScope: ['多人对战'],
+        creatorTips: {
+          doFirst: '先做潮汐时钟。',
+          deferForNow: '暂缓天气系统。',
+          howToVerify: '单局跑满三段潮汐。',
+          expandWhen: '核心循环稳定后再加船种。',
+        },
+      },
+      decisions: [
+        {
+          id: 'decision-0001',
+          topic: '光束是否可分裂',
+          state: 'confirmed',
+          answerSource: 'user_option',
+          round: 1,
+          answerSummary: '不可分裂，保持取舍压力。',
+          basis: null,
+        },
+      ],
+      prototypeValidationItems: [
+        {
+          id: 'proto-0001',
+          question: '单扇区照明是否足够做出取舍？',
+          microPrototype: '纸面推演三回合。',
+          observation: '玩家是否出现犹豫。',
+          passCriterion: '三回合内至少一次改变计划。',
+        },
+      ],
+    },
+    pendingApproval: {
+      gddRef,
+      pendingActionId: 'action-0123456789abcdef01234567',
+      actionFingerprint: 'a'.repeat(64),
+      approvalRequestId: 'gdd-approval-0001',
+      sessionId: 'plan-session-0001',
+      runId: 'run-plan-root-0001',
+    },
+    approvedGddRef: null,
+    recoveryPending: false,
+    ...overrides,
+  };
+}
+
 function createProjectSupervisorRuntimeHarness({
   projectPath = '/tmp/authorized-game',
   sessionId = 'supervisor-session-active',
@@ -366,6 +512,11 @@ function createProjectSupervisorRuntimeHarness({
   let rejectRuntime: Record<string, unknown> | null = null;
   let answerRuntime: Record<string, unknown> | null = null;
   let answerFailuresRemaining = 0;
+  // 默认不配置策划状态：hydrate 与接入本 harness 之前一样抛出，既有用例行为不变。
+  let currentPlanGddState: PlanGddStateViewV1 | null = null;
+  let planGddDecisionError: string | null = null;
+  let planGddHydrateCount = 0;
+  const planGddDecisionCalls: Array<Record<string, unknown>> = [];
   let runtimeUpdateHandler:
     | ((event: {
         payload: {
@@ -642,6 +793,26 @@ function createProjectSupervisorRuntimeHarness({
           });
         return runtimeResult();
       }
+      if (command === 'hydrate_game_creator_plan_gdd_state') {
+        planGddHydrateCount += 1;
+        if (!currentPlanGddState) {
+          throw new Error('PLAN_STATE_NOT_CONFIGURED');
+        }
+        return currentPlanGddState;
+      }
+      if (command === 'decide_game_creator_plan_gdd') {
+        planGddDecisionCalls.push({ ...(args ?? {}) });
+        if (planGddDecisionError) {
+          const failure = planGddDecisionError;
+          planGddDecisionError = null;
+          throw new Error(failure);
+        }
+        return {
+          schemaVersion: 'plan-gdd-decision-result.v1',
+          outcome: 'decided',
+          recoveryPending: false,
+        };
+      }
       throw new Error(`unexpected invoke ${command}`);
     },
   );
@@ -696,6 +867,16 @@ function createProjectSupervisorRuntimeHarness({
     failNextAnswers(count = 1) {
       answerFailuresRemaining = count;
     },
+    setPlanGddState(state: PlanGddStateViewV1 | null) {
+      currentPlanGddState = state;
+    },
+    failNextPlanGddDecision(message: string) {
+      planGddDecisionError = message;
+    },
+    planGddDecisionCalls,
+    planGddHydrateCount() {
+      return planGddHydrateCount;
+    },
     appendSupervisorMessage(message: Record<string, unknown>) {
       currentSupervisorMessages.push(message);
     },
@@ -822,6 +1003,7 @@ export {
   cleanup,
   createGameCreationAppManifest,
   createGameCreationAppSeedTasks,
+  createPlanGddStateView,
   createProjectSupervisorRuntimeHarness,
   deriveAgentStatusCards,
   describe,
diff --git a/apps/ai-game-creator-shell/tests/appSurface/plan-gdd.suite.ts b/apps/ai-game-creator-shell/tests/appSurface/plan-gdd.suite.ts
new file mode 100644
index 000000000..7f1f16f1f
--- /dev/null
+++ b/apps/ai-game-creator-shell/tests/appSurface/plan-gdd.suite.ts
@@ -0,0 +1,136 @@
+import {
+  createPlanGddStateView,
+  createProjectSupervisorRuntimeHarness,
+  expect,
+  fireEvent,
+  it,
+  openMainProject,
+  renderAppAt,
+  screen,
+  waitFor,
+  within,
+} from './harness';
+
+async function mountApprovalCard(
+  harness: ReturnType<typeof createProjectSupervisorRuntimeHarness>,
+) {
+  window.__TAURI__ = {
+    core: { invoke: harness.invoke },
+    event: { listen: harness.listen },
+  };
+  renderAppAt('/');
+  await openMainProject(harness.projectPath);
+  return await screen.findByLabelText('GDD 审批卡');
+}
+
+function openReviseDialog() {
+  fireEvent.click(screen.getByRole('button', { name: '修改' }));
+  return screen.getByRole('dialog');
+}
+
+function typeComment(dialog: HTMLElement, text: string) {
+  const textarea = within(dialog).getByPlaceholderText('请输入原因');
+  fireEvent.change(textarea, { target: { value: text } });
+  return textarea as HTMLTextAreaElement;
+}
+
+export function registerPlanGddApprovalTests() {
+  it('re-hydrates authority after a failed GDD decision and keeps the failure visible', async () => {
+    const harness = createProjectSupervisorRuntimeHarness();
+    harness.setPlanGddState(createPlanGddStateView());
+    await mountApprovalCard(harness);
+
+    const hydrateCallsBeforeDecision = harness.planGddHydrateCount();
+    // 决定失败后，hydrate 应当读到权威侧已经变成「投影恢复中」的事实。
+    harness.setPlanGddState(createPlanGddStateView({ recoveryPending: true }));
+    harness.failNextPlanGddDecision('PLAN_STALE_APPROVAL');
+
+    fireEvent.click(screen.getByRole('button', { name: '批准 v1' }));
+
+    // 失败分支必须重灌权威状态，否则卡片会停在已失效的 pending 身份上。
+    await waitFor(() => {
+      expect(harness.planGddHydrateCount()).toBeGreaterThan(
+        hydrateCallsBeforeDecision,
+      );
+    });
+    // 重灌后「重试恢复」入口出现——这是 recoveryPending 下唯一被允许的动作。
+    await screen.findByRole('button', { name: '重试恢复' });
+    // 而且重灌不能把决定失败的原因擦掉：hydrate 入口会 setPlanGddError(null)，
+    // 两句顺序写反这条断言就红。
+    expect(screen.getByRole('alert').textContent).toContain(
+      'PLAN_STALE_APPROVAL',
+    );
+  });
+
+  it('reuses one responseId while the decision is unchanged and mints a new one once the comment changes', async () => {
+    const harness = createProjectSupervisorRuntimeHarness();
+    harness.setPlanGddState(createPlanGddStateView());
+    await mountApprovalCard(harness);
+
+    const dialog = openReviseDialog();
+    typeComment(dialog, '把核心循环压到三步');
+    harness.failNextPlanGddDecision('PLAN_DURABILITY_FAILED');
+    fireEvent.click(screen.getByRole('button', { name: '提交决定' }));
+    await waitFor(() => {
+      expect(harness.planGddDecisionCalls).toHaveLength(1);
+    });
+
+    // 原样重试：属于方案 §13.2 的 busy/超时/网络重试，必须复用同一 responseId。
+    harness.failNextPlanGddDecision('PLAN_DURABILITY_FAILED');
+    fireEvent.click(screen.getByRole('button', { name: '提交决定' }));
+    await waitFor(() => {
+      expect(harness.planGddDecisionCalls).toHaveLength(2);
+    });
+    expect(harness.planGddDecisionCalls[1].responseId).toBe(
+      harness.planGddDecisionCalls[0].responseId,
+    );
+
+    // 改写修改意见：审批意图变了，必须换新的 responseId，否则后端会以
+    // 「同 responseId 的审批意图不一致」硬拒，用户改写后的原因永远落不了盘。
+    typeComment(dialog, '把核心循环压到两步，并去掉天气系统');
+    harness.failNextPlanGddDecision('PLAN_DURABILITY_FAILED');
+    fireEvent.click(screen.getByRole('button', { name: '提交决定' }));
+    await waitFor(() => {
+      expect(harness.planGddDecisionCalls).toHaveLength(3);
+    });
+    expect(harness.planGddDecisionCalls[2].responseId).not.toBe(
+      harness.planGddDecisionCalls[0].responseId,
+    );
+    expect(harness.planGddDecisionCalls[2].comment).toBe(
+      '把核心循环压到两步，并去掉天气系统',
+    );
+  });
+
+  it('blocks the already-open comment dialog once recovery starts without discarding the typed reason', async () => {
+    const harness = createProjectSupervisorRuntimeHarness();
+    harness.setPlanGddState(createPlanGddStateView());
+    await mountApprovalCard(harness);
+
+    const dialog = openReviseDialog();
+    const textarea = typeComment(dialog, '战斗节奏太慢，请压缩到三个回合');
+    expect(
+      (
+        within(dialog).getByRole('button', {
+          name: '提交决定',
+        }) as HTMLButtonElement
+      ).disabled,
+    ).toBe(false);
+
+    // 弹层是打开之后才被后台 hydrate 翻掉决定资格的：触发按钮的 disabled 管不到它。
+    harness.setPlanGddState(createPlanGddStateView({ recoveryPending: true }));
+    fireEvent.focus(window);
+
+    await waitFor(() => {
+      expect(
+        (
+          within(dialog).getByRole('button', {
+            name: '提交决定',
+          }) as HTMLButtonElement
+        ).disabled,
+      ).toBe(true);
+    });
+    // 已经写好的原因不能被丢掉——所以是禁用加说明，不是自动关弹层。
+    expect(textarea.value).toBe('战斗节奏太慢，请压缩到三个回合');
+    expect(harness.planGddDecisionCalls).toHaveLength(0);
+  });
+}
diff --git a/docs/project-memory/shared-memory/decision-log.md b/docs/project-memory/shared-memory/decision-log.md
index c369201ca..60063fad3 100644
--- a/docs/project-memory/shared-memory/decision-log.md
+++ b/docs/project-memory/shared-memory/decision-log.md
@@ -1,5 +1,16 @@
 # 决策记录
 
+## 2026-08-18 M1D 审查修复：GDD 审批决定失败路径与恢复期弹层门控
+
+- **审查范围与基线**：对 `14c00017c..bf2185fba` 的 M1D-1/M1D-2 全量改动做规格对照审查（技术方案第 13、18 节）。审查完成后分支又前进了 `4624fd795`（文档同步）与 `c6a08ef98`（前端文案回归断言修复）两条，二者都不改 `src/**` 生产代码，审查结论不受影响。
+- **修复一：决定失败也必须重灌权威状态**。`decidePlanGdd` 原来只在成功分支 hydrate，`catch` 只写错误后 rethrow。后端 `decide_plan_gdd_at` 有多条真实 `PLAN_STALE_APPROVAL` 分支（GDD 已不在当前 lineage、identity 不符、版本被更新版本取代、pending 丢失或不一致），命中后卡片停在已失效的 pending 身份上、三个决定按钮仍可点，且 `recoveryPending` 永不翻真导致「重试恢复」入口不渲染，卡内没有任何恢复路径。现在失败分支同样 hydrate，落实第 18.3 节「approval decision 返回后调用 hydrate」（该句不区分成功与失败）。**两句顺序已被回归钉死**：`hydratePlanGddState` 入口会 `setPlanGddError(null)`，必须先 hydrate 再写决定错误，写反会把这条错误擦掉。
+- **修复二：responseId 复用键纳入 comment**。原键为 `approvalRequestId:action`，不含 comment，违反第 13.2 节「用户改变 action/comment 后必须生成新 responseId」。在第 14 节恢复矩阵承认的「receipt 已提交但 command response 丢失」构造下，用户改写修改意见后重提会带着旧 responseId，命中后端「同 responseId 的审批意图不一致」硬拒，改写后的原因永远落不了盘。现在键挂在 `approvalRequestId` 上并比对 `{action, comment}` 完整意图。**判据方向为宁可多换不可少换**：receipt 已存在时多换的最坏后果是 `replayed` 降级成 `already-decided`（两者都是 Ok，且 already-decided 正是第 18.2 节要求的刷新态），少换则是硬错误。
+- **修复三：`recoveryPending` 必须挡住已经打开的评论弹层**。第 18.2 节要求恢复期只允许重试同一 ID、不允许提交决定；原实现只把 `canDecide` 接到三个触发按钮上，而弹层是打开之后才可能被后台 hydrate 翻掉决定资格的，其「提交决定」按钮只看 `busy || !comment.trim()`，仍可提交。现在 `submitComment` 与该按钮都判 `canDecide`，并在弹层内说明原因。**刻意不自动关弹层**，否则会丢掉用户已经写好的修改意见。
+- **测试**：appSurface harness 新增 `hydrate_game_creator_plan_gdd_state` / `decide_game_creator_plan_gdd` 两个分发分支与 `createPlanGddStateView` fixture；未配置策划状态时 hydrate 与接入前一样抛出，既有用例行为不变。新增 `tests/appSurface/plan-gdd.suite.ts` 三条回归，并逐条做过变异验证——把对应修复单独逆转后三条各自以自己的断言变红（修复二的变异是**部分逆转**：保留新 Map 结构、只删掉 comment 比对，因此该用例钉住的是 comment 这一维本身而非那次重构）。
+- **验证**：`appSurface.test.ts` **381 passed / 0 failed**（378 既有 + 3 新增）；`agentTraceSummary` 与 `rememberCommand`（另两个 import `src/App` 的用例文件）13 passed；`agc:typecheck` 通过；6 个改动/新增文件 ESLint `--max-warnings 0` 通过；`check:encoding` 5409 文件通过。不改 Rust——三条全在前端，后端语义已经正确。
+- **对既有记录的更正**：M1D-1 与 M1D-2 两条记录分别称「Shell TypeScript typecheck 仍被仓库既有依赖缺失阻断」「appSurface UI suite 受仓库现有缺失 Tauri plugin 依赖阻断，未把该基线失败归因于本包」，在原分支主工作树上都不成立：`agc:typecheck` 干净退出，appSurface 378 条全绿；两道门分别位于 CI 的 `check:native-shells`（且 typecheck 排在 cargo test 之前）与 Frontend tests 内，一直是活的。实际情况与记录相反——`bf2185fba` 改名 `taskGroupLabels.design` 后，appSurface 有 8 个用例文件的断言变红，随后由 `c6a08ef98` 修复；把该套件记为「基线阻断、不归因本包」正是让这条自带回归合入的原因。**隔离工作树的依赖缺失不能作为跳过门禁的依据，须回原工作树复跑后再下结论。**
+- **未修的审查发现（本次不并入，单列后续）**：① hydrate 在校验 GDD/session 的 projectId 与 manifest 一致之前，已执行 `reconcile_plan_gdd_approval_projections_at`、session previous 提升与 index 重建等落盘修复，违反第 18.3 节固定顺序，其中 `session.previous.json` 的提升+删除不可逆（触发需外部篡改 `.agent/`，App 自身流程造不出该分歧）；② 项目写锁竞争时 `acquire_project_write_lock` 的错误原文内嵌项目绝对路径，被原样回传前端，违反第 18.3 节「返回值不包含绝对路径」，常态可达；③ design 组展示名只改了 `taskGroupLabels` 一本字典，`agentPresentation.ts` 的 `groupConfigs` 与 `view/project-development/index.tsx` 的 `summarizeAgent` 仍硬编码「策划 Agent」，与新阶段「立项策划」同屏共存，违反第 18.2 节；④ 阶段进度「轮次 X/3」直接透传 0-indexed 的 `clarificationRound` 未 +1（后端自己用的是 `current_round + 1`），最后一轮显示「轮次 2/3」，字面暗示还剩一轮。
+
 ## 2026-08-18 M1D-2 隔离工作树实现：入口分流与阶段进度
 
 - **隔离范围**：在 `codex/genarrative-isolated`、基线 `5b11a0530` 上开工；只接入口分流、阶段进度和实际项目总控页面的现有审批卡挂载，不接 M2 `approvedGddRef` 构建绑定、完整构建按钮或 M1E 端到端故障注入。
diff --git a/docs/technical/【技术方案】立项策划Agent（Fast GDD）-2026-08-10.md b/docs/technical/【技术方案】立项策划Agent（Fast GDD）-2026-08-10.md
index 36d1726f7..d093469e7 100644
--- a/docs/technical/【技术方案】立项策划Agent（Fast GDD）-2026-08-10.md	
+++ b/docs/technical/【技术方案】立项策划Agent（Fast GDD）-2026-08-10.md	
@@ -2023,10 +2023,12 @@ M0 完成不表示完整策划闭环已经上线。`M1A-1`～`M1A-4`、`M1B-1`
 | `M1C-2a` | Goal Contract 接线：turn 1 冻结、固定验收图、审批前置门取证 | `M1C-1`、`M1A-3` | **当前隔离 worktree 已完成并通过本包门禁，尚未合回**：turn 1 的 request-scoped schema 与格式修复都只允许一个固定 `agent.goal_contract`；按项目变化的四项之外，`preferences=[]`、唯一验收节点及证据工具均冻结。Fast GDD evidence 只接受当前 Supervisor 根 run 对 `game/fast_gdd.md` 从第 1 行到 EOF 的同 hash 完整分页；无/旧证据先继续读取，显式 failed 才给原 delivery 的 `repairOfDelegationId`，passed 且 delivery 已认领才建 pending。pending/recovery/completion/finalization 均按同 identity 幂等，审批后 Markdown 改写不损坏 Graph。格式、Provider 强判据、M1C-2a、Acceptance Graph、planning submit/approval、finalization、all-targets、编码与 diff 门禁均通过；扩展 autonomous completion 整组的无关 game-chat 并行超时及精确复跑结果见 decision-log 同日条，不改该路径。不包含 `M1C-2b`、UI 或构建准入 |
 | `M1C-2b` | 澄清中转接线、轮次派生、预算注入 | `M1C-2a` | **实现与本包门禁已完成并已快进合回 `feat/five_min_design`**：首 child 的 revision 1 session、`NeedsUserInput → awaiting_user_input`、回答绑定后 continuation 的确定性 session 投影、审批后 `revise/reject` 用户修订谱系及 Provider 活跃时间 usage fact/fold 已接线；末次 `plan.submit_gdd` usage 在 receipt/session successor 落盘且 standalone/v4 anchors 精确消费后于同一项目锁内折叠，真实 receipt 回归证明累计值恰好推进一次，重复审批与 recovery reconcile 不二次推进。`planning_clarification_*` **13 passed / 0 failed**（M1C-2c 语义回归另见本包），另有 static deliveries 44、planning storage 13、planning submit 53、Provider usage 4、末次 usage receipt 1、barrier detail 3 条定向回归通过；格式、offline all-targets、编码与 diff 门禁通过。锁序承诺只适用于 **M1C-2b 新增的 planning 澄清写投影路径**；`main_loop` 既有通用 completion blocker 的 execution→project 路径不在本包。第 4 轮信封在正常路径不可达：`agent.delegate` 已在工具边界按血缘上限硬拒并返回 failed observation；coordinator 的超三轮 reconciliation 仅用于损坏血缘纵深防御。审批 UI、hydrate、构建准入和下游完整构建不在本包范围 |
 | `M1C-2c` | 决策卡 A/B 语义（第 23.9 节，2026-08-18 实现完成并合回）：选项 → 台账映射改为 A/B 均 `confirmed/user_option`、信封 label 形状校验、planning role brief 与 Supervisor playbook/final-reply 文案（B 必须是真实岔路、第 3 项恒定且 description 须给出可执行验证方式、改口转述规则、提问纪律） | `M1C-2b` | **实现与门禁完成，已由 `6e4bd9703` 合回 `feat/five_min_design`**：Runtime 已实现 A/B/固定第三项校验、B 不再生成 `default_pending`、`answerSummary` 逐字保真；非法 C/缺项 fail-closed，A/B/自由填写回归已通过。`planning_clarification_*` 13、`project_planning` prompt 5、`planning_submit` 定向回归、prompt bundle、格式、编码、diff、offline all-targets 均通过；不含 M1D-1 前端、hydrate、构建准入或下游完整构建 |
-| `M1D-1` | 前端 hydrate 与 GDD 审批卡；决策卡按第 23.9 节实现（label 动态渲染、默认焦点 A、Other 槽不变） | `M1C-2b` | **已完成并合入 `feat/five_min_design`（`5b11a0530`）**：新增严格 `{projectPath}` hydrate command、`plan-gdd-state-view.v1` Rust read model、审批卡与独立 GDD 正文详情弹层；页面只消费 hydrate，决定 responseId 按审批请求/动作复用，`recoveryPending` 仅提供恢复重试；审批前置 pending 与错绑 session 继续 fail-closed。 |
+| `M1D-1` | 前端 hydrate 与 GDD 审批卡；决策卡按第 23.9 节实现（label 动态渲染、默认焦点 A、Other 槽不变） | `M1C-2b` | **已完成并合入 `feat/five_min_design`（落地 `0052a80da`，其后 ESLint 修正 `5b11a0530`）**：新增严格 `{projectPath}` hydrate command、`plan-gdd-state-view.v1` Rust read model、审批卡与独立 GDD 正文详情弹层；页面只消费 hydrate，决定 responseId 按审批请求/动作复用，`recoveryPending` 仅提供恢复重试；审批前置 pending 与错绑 session 继续 fail-closed。 |
 | `M1D-2` | 入口分流与阶段进度 | `M1D-1` | **已完成并以 `bf2185fba` 合入 `feat/five_min_design`**：游戏新项目默认 `standard + project-supervisor-plan`，显式“直接开建”保持 `autonomous-game-build`；阶段进度显示轮次 x/3、当前版本和状态徽章；实际项目总控页面挂载 hydrate/审批卡，并将 `project-planning` / 设计组展示名收口。未接 M2 approved-GDD 构建绑定或完整构建按钮。 |
 | `M1E` | 端到端与故障注入收口 | `M1D-2` | 第 21 节测试矩阵中跨层场景 |
 
+**2026-08-18 M1D 审查修复快照**：对 `14c00017c..bf2185fba` 做规格对照审查后，修复三条决定链路缺陷并补齐回归。① `decidePlanGdd` 的失败分支原来不 hydrate，命中后端任一 `PLAN_STALE_APPROVAL` 分支后卡片会停在已失效的 pending 身份上、`recoveryPending` 永不翻真导致「重试恢复」入口不渲染，现已按第 18.3 节在失败分支同样重灌（顺序钉死：`hydratePlanGddState` 入口会清空错误，必须先 hydrate 再写决定错误）。② responseId 复用键原为 `approvalRequestId:action`，不含 comment，违反第 13.2 节「改变 action/comment 必须换新 responseId」，现改为比对 `{action, comment}` 完整意图，判据方向为宁可多换不可少换。③ 第 18.2 节「`recoveryPending` 时不允许提交决定」原来只作用于三个触发按钮，已打开的评论弹层仍可提交，现已同门控并保留用户已输入内容。回归位于 `tests/appSurface/plan-gdd.suite.ts`，三条均经变异验证（逆转对应修复即变红）；`appSurface.test.ts` 381 passed，`agc:typecheck`、ESLint、编码检查通过。**本次不含 Rust 改动**；hydrate 身份校验与落盘投影修复的顺序、锁错误回传绝对路径、design 组展示名剩余两处字典、以及阶段进度轮次差一格四条审查发现单列后续，未并入。
+
 **2026-08-14 `M1B-1` 合入验收快照**：已合入实现集中在 `apps/ai-game-creator-shell/src-tauri/src/agent/runtime_protocol/planning_storage.rs`，并由 `runtime_protocol.rs` 注册。已具备 `plan-gdd.v1`、`plan-gdd-index.v1`、`plan-session.v1` 及 `plan.submit_gdd` input 的 strict serde 形状校验、文本/ID/时间/枚举边界、typed serde fingerprint、canonical JSON（重复键、BOM、尾空白、字段顺序）解析、GDD 连续版本链、session `revision + 1` / `previousFingerprint` 链、create-only durable writer、session 原子替换与受限 recovery，以及 `project-planning / agent-delegate / standard / project-supervisor` writer identity。通用 `file.write`、`file.patch`、`file.delete`、`project.patchset` 与 checkpoint restore 对 `.agent/planning/**` 和 `game/fast_gdd.md` 只挡写，planning 的 `file.read` / `file.list` 仍可读；M1B-1 合入时没有注册或执行 `plan.submit_gdd`，也没有实现 approval pending、receipt、UI 或构建准入。
 
 已验证第 9.1 节 golden vector（canonical envelope 3857 bytes 与指纹 `sha256-serde-json-v2:a59856de7ef134cf2f49c4dedd2ba10ae4ab2340a9634d402eb792b6ee5458f0`）及 11 个定向 storage 测试；durable writer 的目标 schema 重解析与文件名/版本核对、index 与权威 GDD 对账、缺失/损坏/陈旧 index 的锁内链重建、`(requestId,responseId)` / decision/ref / active run/phase 等 session 约束、GDD 文件枚举/链读取及 primary 损坏/previous 提升/分叉 recovery 矩阵均已覆盖。M1B-1 尚无 approval receipt schema，因此多版本 `statusCache` 只表达无 receipt 的预审批投影；M1C-1 接入 receipt 后重建真实 approved/revise/reject/superseded 状态。`cargo check --offline`、`npm run check:encoding`、`git diff --check` 均通过；该包现已合入本分支，这一历史快照不把后续 `plan.submit_gdd`、审批闭环、UI 或 renderer 误报为 M1B-1 已实现。