diff --git a/apps/ai-game-creator-shell/src-tauri/src/agent/direct_tool_bridge.rs b/apps/ai-game-creator-shell/src-tauri/src/agent/direct_tool_bridge.rs
index 00bf6675d..30a7ad3bb 100644
--- a/apps/ai-game-creator-shell/src-tauri/src/agent/direct_tool_bridge.rs
+++ b/apps/ai-game-creator-shell/src-tauri/src/agent/direct_tool_bridge.rs
@@ -4422,67 +4422,100 @@ pub(super) fn drain_analytics_test_writer(
 #[cfg(test)]
 mod tests {
     #[test]
-    fn browser_gameplay_failure_reaches_model_with_both_viewports_and_screenshots() {
-        let temp = tempfile::tempdir().unwrap();
-        let evidence =
-            super::super::direct_validation::tests::failed_browser_evidence_fixture(temp.path());
-        for viewport in &evidence.viewport_results {
-            image::RgbaImage::new(2, 2)
-                .save(&viewport.screenshot_path)
-                .unwrap();
-        }
-        let receipt = super::super::direct_validation::browser_tool_result(
-            temp.path(),
-            &evidence,
-            Some(crate::browser::BrowserPlaytestScenario::GenericV1),
-            1,
-        );
-        assert!(evidence.viewport_results.iter().all(|result| result.passed));
-        assert_eq!(receipt["passed"], false);
-        let failure =
-            bridge_validation_result(temp.path(), Ok(receipt.clone()), |report, screenshots| {
-                BrowserPlaytestError::PlaytestNotPassed {
+    fn browser_gameplay_results_reach_model_with_matching_viewports_and_screenshots() {
+        for passed in [false, true] {
+            let temp = tempfile::tempdir().unwrap();
+            let evidence = if passed {
+                super::super::direct_validation::tests::browser_evidence_fixture(temp.path())
+            } else {
+                super::super::direct_validation::tests::failed_browser_evidence_fixture(temp.path())
+            };
+            for viewport in &evidence.viewport_results {
+                image::RgbaImage::new(2, 2)
+                    .save(&viewport.screenshot_path)
+                    .unwrap();
+            }
+            let receipt = super::super::direct_validation::browser_tool_result(
+                temp.path(),
+                &evidence,
+                Some(crate::browser::BrowserPlaytestScenario::GenericV1),
+                1,
+            );
+            assert!(evidence.viewport_results.iter().all(|result| result.passed));
+            assert_eq!(receipt["passed"], passed);
+            let outcome = bridge_validation_result(
+                temp.path(),
+                Ok(receipt.clone()),
+                |report, screenshots| BrowserPlaytestError::PlaytestNotPassed {
                     report,
                     screenshots,
-                }
-            })
-            .unwrap_err();
-        let error = ToolCallError::from(&failure);
-        assert_eq!(error.images.len(), 2);
-        assert_eq!(error.error["PlaytestNotPassed"]["report"], receipt);
-        assert!(error.error["PlaytestNotPassed"]
-            .get("screenshots")
-            .is_none());
-        let state = direct_tool_bridge_state(temp.path().to_path_buf());
-        let response =
-            compose_direct_tool_outcome(&state, "agc_browser_playtest", "{}", false, Err(error));
-        assert_eq!(response["isError"], true);
-        let text = response["content"][0]["text"].as_str().unwrap();
-        assert_eq!(
-            response["content"]
-                .as_array()
-                .unwrap()
-                .iter()
-                .filter(|block| block["type"] == "image")
-                .count(),
-            2
-        );
-        for (index, entry) in evidence.viewport_playtests.iter().enumerate() {
-            let report = &receipt["gameplayResults"][index];
+                },
+            )
+            .map_err(|failure| {
+                let error = ToolCallError::from(&failure);
+                assert_eq!(error.error["PlaytestNotPassed"]["report"], receipt);
+                assert!(error.error["PlaytestNotPassed"]
+                    .get("screenshots")
+                    .is_none());
+                error
+            });
+            let state = direct_tool_bridge_state(temp.path().to_path_buf());
+            let response =
+                compose_direct_tool_outcome(&state, "agc_browser_playtest", "{}", false, outcome);
+            assert_eq!(response["isError"], !passed);
+            let text = response["content"][0]["text"].as_str().unwrap();
+            let model_receipt: Value =
+                serde_json::from_str(&text[text.find('{').unwrap()..]).unwrap();
+            assert_eq!(model_receipt, receipt);
             assert_eq!(
-                report["assertions"],
-                serde_json::to_value(&entry.result.assertions).unwrap()
+                response["content"]
+                    .as_array()
+                    .unwrap()
+                    .iter()
+                    .filter(|block| block["type"] == "image")
+                    .count(),
+                2
             );
-            assert_eq!(report["finalPhase"], "ready");
-            assert_eq!(report["finalSequence"], index as u64);
-            assert!(text.contains(&entry.result.diagnostics[0]));
+            let summary = model_receipt["summary"].as_str().unwrap();
+            for (index, entry) in evidence.viewport_playtests.iter().enumerate() {
+                let report = &model_receipt["gameplayResults"][index];
+                assert_eq!(
+                    report["viewport"],
+                    serde_json::to_value(entry.viewport).unwrap()
+                );
+                assert_eq!(report["passed"], entry.result.passed);
+                assert_eq!(
+                    report["assertions"],
+                    serde_json::to_value(&entry.result.assertions).unwrap()
+                );
+                assert_eq!(report["finalPhase"], "ready");
+                assert_eq!(
+                    report["finalSequence"],
+                    entry.result.final_sequence.unwrap()
+                );
+                let viewport = report["viewport"].as_str().unwrap();
+                let line = summary
+                    .lines()
+                    .find(|line| line.starts_with(&format!("{viewport}: gameplayPassed=")))
+                    .unwrap();
+                assert!(line.starts_with(&format!(
+                    "{viewport}: gameplayPassed={}",
+                    entry.result.passed
+                )));
+                if let Some(reason) = entry.result.diagnostics.first() {
+                    assert_eq!(report["diagnostics"][0], reason.as_str());
+                    assert!(line.contains(reason));
+                }
+                if let Some(assertion) = entry
+                    .result
+                    .assertions
+                    .iter()
+                    .find(|assertion| !assertion.passed)
+                {
+                    assert!(line.contains(&assertion.name));
+                }
+            }
         }
-        let summary = receipt["summary"].as_str().unwrap();
-        assert!(summary.contains("desktop: gameplayPassed=false"));
-        assert!(summary.contains("mobile: gameplayPassed=false"));
-        assert!(summary.contains("start-control-clicked"));
-        assert!(summary.contains("start-phase-playing"));
-        assert!(summary.contains("未执行"));
     }
 
     #[tokio::test]
diff --git a/apps/ai-game-creator-shell/src-tauri/src/agent/direct_validation.rs b/apps/ai-game-creator-shell/src-tauri/src/agent/direct_validation.rs
index ecee74473..a42aed1b0 100644
--- a/apps/ai-game-creator-shell/src-tauri/src/agent/direct_validation.rs
+++ b/apps/ai-game-creator-shell/src-tauri/src/agent/direct_validation.rs
@@ -786,7 +786,9 @@ pub(super) mod tests {
             assert_eq!(receipt["passed"], mode != "missing-mobile");
             if mode == "visual" {
                 assert_eq!(receipt["gameplayResults"], json!([]));
-                assert!(summary.contains("未执行"));
+                assert!(summary
+                    .lines()
+                    .any(|line| line.starts_with("玩法检查：未执行")));
                 assert_eq!(receipt["coverage"], "visual-only");
                 continue;
             }
diff --git a/apps/ai-game-creator-shell/src-tauri/src/browser/tests.rs b/apps/ai-game-creator-shell/src-tauri/src/browser/tests.rs
index adcc64f1b..cb5d1cf8b 100644
--- a/apps/ai-game-creator-shell/src-tauri/src/browser/tests.rs
+++ b/apps/ai-game-creator-shell/src-tauri/src/browser/tests.rs
@@ -1514,6 +1514,41 @@ async fn real_chrome_generic_playtest_pointer_events_keeps_unavailable_controls_
     }
 }
 
+#[tokio::test]
+#[ignore = "requires an installed Chrome/Chromium/Edge and explicit local browser execution"]
+async fn real_chrome_generic_playtest_reports_start_click_without_playing_phase() {
+    let result = run_stable_generic_viewport_fixture(false, Some(
+        r#"<button type="button" data-playtest-id="start" data-start-phase="ready">Start</button>"#,
+    )).await;
+    assert!(!result.passed);
+    assert!(result
+        .viewport_results
+        .iter()
+        .all(|viewport| viewport.passed));
+    for entry in &result.viewport_playtests {
+        let playtest = &entry.result;
+        assert!(!playtest.passed);
+        assert_eq!(playtest.final_phase, Some(BrowserPlaytestPhase::Ready));
+        assert_eq!(playtest.final_sequence, Some(1));
+        for (name, expected) in [
+            ("start-control-clicked", true),
+            ("start-phase-playing", false),
+        ] {
+            assert_eq!(
+                playtest
+                    .assertions
+                    .iter()
+                    .find(|assertion| assertion.name == name)
+                    .unwrap()
+                    .passed,
+                expected
+            );
+        }
+        let reason = &playtest.diagnostics[0];
+        assert!(reason.contains("必须进入 playing"), "{reason}");
+    }
+}
+
 async fn run_stable_generic_viewport_fixture(
     mobile_action_broken: bool,
     start_markup: Option<&str>,
@@ -1556,7 +1591,7 @@ async fn run_stable_generic_viewport_fixture(
   };
   document.querySelector('[data-playtest-id="start"]').addEventListener('click', (event) => {
     if (event.isTrusted && state.phase === 'ready') {
-      advance(() => { state.phase = 'playing'; });
+      advance(() => { state.phase = event.currentTarget.dataset.startPhase || 'playing'; });
     }
   });
   const primary = document.querySelector('[data-playtest-id="primary-action"]');
@@ -1656,6 +1691,17 @@ async fn run_stable_generic_viewport_fixture(
     {
         let projected = &receipt["gameplayResults"][index];
         assert_eq!(projected["viewport"], entry["viewport"]);
+        let viewport = entry["viewport"].as_str().unwrap();
+        let line = receipt["summary"]
+            .as_str()
+            .unwrap()
+            .lines()
+            .find(|line| line.starts_with(&format!("{viewport}: gameplayPassed=")))
+            .unwrap();
+        assert!(line.starts_with(&format!(
+            "{viewport}: gameplayPassed={}",
+            entry["result"]["passed"]
+        )));
         for field in [
             "passed",
             "assertions",
@@ -1670,7 +1716,15 @@ async fn run_stable_generic_viewport_fixture(
         }
         if let Some(reason) = entry["result"]["diagnostics"][0].as_str() {
             assert_eq!(projected["diagnostics"][0], reason);
-            assert!(receipt["summary"].as_str().unwrap().contains(reason));
+            assert!(line.contains(reason));
+        }
+        if let Some(assertion) = entry["result"]["assertions"]
+            .as_array()
+            .unwrap()
+            .iter()
+            .find(|assertion| assertion["passed"] == false)
+        {
+            assert!(line.contains(assertion["name"].as_str().unwrap()));
         }
     }
     result
diff --git a/docs/technical/【技术方案】AI游戏创作智能体App实施计划-2026-06-24.md b/docs/technical/【技术方案】AI游戏创作智能体App实施计划-2026-06-24.md
index c3e60dde4..39ea7a151 100644
--- a/docs/technical/【技术方案】AI游戏创作智能体App实施计划-2026-06-24.md
+++ b/docs/technical/【技术方案】AI游戏创作智能体App实施计划-2026-06-24.md
@@ -196,6 +196,7 @@ UI 编辑器的“分析参考图”步骤、Rust 命令 `suggest_ui_design_sema
 
 - 复用客户端浏览器与现有固定玩法场景。视觉检查采集双端画面/布局/资源/诊断；玩法检查分别在 desktop/mobile 执行明确的固定场景和真实输入，按视口保存结果。旧报告缺少移动端玩法结果时保持未知，不补通过。报告必须声明检查层级；视觉通过不能宣称玩法通过，固定场景通过也不能宣称覆盖未执行的完整关卡。
 - 浏览器工具回执在 `mode=gameplay` 时新增 `gameplayResults` 双端有界投影，每个视口包含 `viewport`、`passed`、`diagnostics`、`assertions` 及 `initialPhase` / `initialSequence` / `initialLevel`、`finalPhase` / `finalSequence` / `finalLevel`。`assertions` 仅列当前固定场景的断言名称和通过状态；每端最多公开 4 条诊断，每条最多 512 字符。缺少某端结果时仍返回该视口 `passed=false`，诊断说明证据缺失，断言为空且六个状态字段为 `null`；断言的 `passed=false` 可包含未执行，不能据此断言该断言已独立执行且失败。`mode=visual` 的 `gameplayResults` 为空，并在摘要中明确玩法未执行。摘要从同一投影给出每端首条诊断和首个未通过断言，不能与结构化结果相矛盾。工具回执不得展开完整宿主报告；`reportPath` 只作宿主证据引用。宿主报告结构、验收门禁、持久化预算和私有路径保护保持原合同，旧回执不回填新字段。
+- 回执验收覆盖 start 禁用、点击后 phase 不符合和玩法成功；逐视口核对持久报告与摘要中的状态、首条诊断和首个未通过断言，并检查成功、失败两条最终 MCP 回包路径。
 - 输入或碰撞改变先做定点玩法检查；纯图像/颜色变化做视觉检查；首次交付和影响闭环的修改做所需玩法验证。新增失败或相关代码变化才重跑对应层，不因改说明文字重复完整验证。
 - 内置试玩和客户端托管的外部 Node/npm 验证共用当前 clientTurnId 的持久化预算。客户端分配递增执行序号，模型提供的旧 attempt 仅作兼容输入，不能减少计数或重置预算；同一轮错误反馈、工具切换和进程重启均不能刷新已消费次数。
 - 新增本地 validation.maxRuns（默认 3，正整数）独立于 llm.maxRetries；显式配置原样使用，不按角色或运行模式改写。超限直接返回已用/上限和最近证据，停止新的验证。预检与正常构建不计作重复试玩。