补齐试玩回包审查要求的关键验收（#548）

绑定逐视口摘要中的状态、阻断原因与首个未通过断言验证成功和失败结果均完整进入最终 MCP 回包新增真实点击后未进入 playing 的浏览器场景并对照持久报告收紧视觉未执行断言并补充长期验收要求
2026-10-03 16:59:53 +08:00
parent 6c0f504c12
commit 1a27e40758
4 changed files with 149 additions and 59 deletions
@@ -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]
@@ -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;
            }
@@ -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
@@ -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。