修复流式测试夹具与自主构建重试

兼容 stream=true 的 Responses 与 Chat Completions 测试响应修正截断 scaffold 修复错误前缀匹配确保最终回复失败审计夹具先完成 planning 阶段
2026-08-27 14:56:29 +08:00
parent a155529861
commit 1926aa4fb5
2 changed files with 162 additions and 61 deletions
@@ -668,7 +668,7 @@ pub(super) const AGENT_RUNTIME_AUTONOMOUS_MANIFEST_DAG_WAIT_LIVENESS_ERROR_PREFI
 pub(super) const AGENT_RUNTIME_AUTONOMOUS_RESPONSE_PLAN_LIVENESS_ERROR_PREFIX: &str =
    "自主构建专业 Agent 已给出结论但结构化计划仍未完成";
 pub(super) const AGENT_RUNTIME_AUTONOMOUS_TRUNCATED_SCAFFOLD_ERROR_PREFIX: &str =
-    "自主构建完整写入游戏入口（index.html 或 game/index.html）时必须提交闭合且可运行的 HTML";
+    "自主构建完整写入 game/index.html 时必须提交闭合且可运行的 HTML";
 pub(super) const AGENT_RUNTIME_QUALITY_REVIEW_AGENT_ID: &str = "quality-review";
 pub(super) const AGENT_RUNTIME_SUPERVISOR_INITIAL_COLLABORATION_LIVENESS_ERROR_PREFIX: &str =
    "Project Supervisor 首批协作在首个 planning 窗口内未建立";
@@ -1557,17 +1557,35 @@ pub(crate) fn spawn_mock_llm_tool_plan_then_invalid_final_reply(
    let base_url = format!("http://{}", listener.local_addr().expect("mock llm addr"));
    std::thread::spawn(move || {
        let (mut planning_stream, _) = listener.accept().expect("mock tool plan accept");
-        drop(read_mock_http_request(&mut planning_stream));
-        let planning_body = serde_json::json!({
-            "id": "resp_invalid_final_reply_planning",
-            "model": "mock-game-model",
-            "output_text": planning_response,
-            "status": "completed",
-            "usage": { "input_tokens": 11, "output_tokens": 22, "total_tokens": 33 }
-        })
-        .to_string();
+        let planning_request = read_mock_http_request(&mut planning_stream);
+        let planning_streaming = planning_request.contains("\"stream\":true");
+        let (planning_body, planning_content_type) = if planning_streaming {
+            (
+                format!(
+                    "data: {}\n\ndata: {}\n\n",
+                    serde_json::json!({
+                        "type": "response.output_text.delta",
+                        "delta": planning_response
+                    }),
+                    serde_json::json!({ "type": "response.completed" })
+                ),
+                "text/event-stream; charset=utf-8",
+            )
+        } else {
+            (
+                serde_json::json!({
+                    "id": "resp_invalid_final_reply_planning",
+                    "model": "mock-game-model",
+                    "output_text": planning_response,
+                    "status": "completed",
+                    "usage": { "input_tokens": 11, "output_tokens": 22, "total_tokens": 33 }
+                })
+                .to_string(),
+                "application/json",
+            )
+        };
        let planning_response = format!(
-            "HTTP/1.1 200 OK\r\nContent-Type: application/json\r\nContent-Length: {}\r\nConnection: close\r\n\r\n{}",
+            "HTTP/1.1 200 OK\r\nContent-Type: {planning_content_type}\r\nContent-Length: {}\r\nConnection: close\r\n\r\n{}",
            planning_body.len(),
            planning_body
        );
@@ -1762,30 +1780,65 @@ fn spawn_mock_llm_server_responses_with_capture(
            if let Some(sender) = request_sender.as_ref() {
                let _ = sender.send(request_text.clone());
            }
-            let body = if request_text.contains("POST /responses HTTP/1.1") {
-                serde_json::json!({
-                    "id": "resp_game_creator_mock",
-                    "model": "mock-game-model",
-                    "output_text": response_content,
-                    "status": "completed",
-                    "usage": { "input_tokens": 11, "output_tokens": 22, "total_tokens": 33 }
-                })
+            let is_responses = request_text.contains("POST /responses HTTP/1.1");
+            let is_stream = request_text.contains("\"stream\":true");
+            let (body, content_type) = if is_stream && is_responses {
+                (
+                    format!(
+                        "data: {}\n\ndata: {}\n\n",
+                        serde_json::json!({
+                            "type": "response.output_text.delta",
+                            "delta": response_content
+                        }),
+                        serde_json::json!({ "type": "response.completed" })
+                    ),
+                    "text/event-stream; charset=utf-8",
+                )
+            } else if is_stream {
+                (
+                    format!(
+                        "data: {}\n\ndata: [DONE]\n\n",
+                        serde_json::json!({
+                            "id": "chatcmpl_game_creator_mock",
+                            "object": "chat.completion.chunk",
+                            "choices": [{
+                                "index": 0,
+                                "delta": { "role": "assistant", "content": response_content },
+                                "finish_reason": "stop"
+                            }]
+                        })
+                    ),
+                    "text/event-stream; charset=utf-8",
+                )
+            } else if is_responses {
+                (
+                    serde_json::json!({
+                        "id": "resp_game_creator_mock",
+                        "model": "mock-game-model",
+                        "output_text": response_content,
+                        "status": "completed",
+                        "usage": { "input_tokens": 11, "output_tokens": 22, "total_tokens": 33 }
+                    })
+                    .to_string(),
+                    "application/json",
+                )
            } else {
-                serde_json::json!({
-                    "id": "chatcmpl_game_creator_mock",
-                    "model": "mock-game-model",
-                    "choices": [
-                        {
+                (
+                    serde_json::json!({
+                        "id": "chatcmpl_game_creator_mock",
+                        "model": "mock-game-model",
+                        "choices": [{
                            "message": { "content": response_content },
                            "finish_reason": "stop"
-                        }
-                    ],
-                    "usage": { "prompt_tokens": 11, "completion_tokens": 22, "total_tokens": 33 }
-                })
-            }
-            .to_string();
+                        }],
+                        "usage": { "prompt_tokens": 11, "completion_tokens": 22, "total_tokens": 33 }
+                    })
+                    .to_string(),
+                    "application/json",
+                )
+            };
            let response = format!(
-                "HTTP/1.1 200 OK\r\nContent-Type: application/json\r\nContent-Length: {}\r\nConnection: close\r\n\r\n{}",
+                "HTTP/1.1 200 OK\r\nContent-Type: {content_type}\r\nContent-Length: {}\r\nConnection: close\r\n\r\n{}",
                body.len(),
                body
            );
@@ -2159,18 +2212,34 @@ fn spawn_mock_llm_tool_plan_then_transient_final_reply(
            .try_into()
            .unwrap_or(u64::MAX);
        let planning_request = read_mock_http_request(&mut planning_stream);
-        let _ = request_capture_sender.send((planning_accepted_at_ms, planning_request));
+        let _ = request_capture_sender.send((planning_accepted_at_ms, planning_request.clone()));
        let _ = request_notice_sender.send(());
-        let planning_body = serde_json::json!({
-            "id": "resp_transient_final_reply_planning",
-            "model": "mock-game-model",
-            "output_text": planning_response,
-            "status": "completed",
-            "usage": { "input_tokens": 11, "output_tokens": 22, "total_tokens": 33 }
-        })
-        .to_string();
+        let planning_body = if planning_request.contains("\"stream\":true") {
+            format!(
+                "data: {}\n\ndata: {}\n\n",
+                serde_json::json!({
+                    "type": "response.output_text.delta",
+                    "delta": planning_response
+                }),
+                serde_json::json!({ "type": "response.completed" })
+            )
+        } else {
+            serde_json::json!({
+                "id": "resp_transient_final_reply_planning",
+                "model": "mock-game-model",
+                "output_text": planning_response,
+                "status": "completed",
+                "usage": { "input_tokens": 11, "output_tokens": 22, "total_tokens": 33 }
+            })
+            .to_string()
+        };
+        let planning_content_type = if planning_request.contains("\"stream\":true") {
+            "text/event-stream; charset=utf-8"
+        } else {
+            "application/json"
+        };
        let planning_http_response = format!(
-            "HTTP/1.1 200 OK\r\nContent-Type: application/json\r\nContent-Length: {}\r\nConnection: close\r\n\r\n{}",
+            "HTTP/1.1 200 OK\r\nContent-Type: {planning_content_type}\r\nContent-Length: {}\r\nConnection: close\r\n\r\n{}",
            planning_body.len(),
            planning_body
        );
@@ -2250,7 +2319,7 @@ fn spawn_mock_llm_tool_plan_then_transient_final_compaction(
        let (mut planning_stream, _) = listener.accept().expect("mock tool plan accept");
        let planning_accepted_at_ms = accepted_at_ms();
        let planning_request = read_mock_http_request(&mut planning_stream);
-        let _ = request_capture_sender.send((planning_accepted_at_ms, planning_request));
+        let _ = request_capture_sender.send((planning_accepted_at_ms, planning_request.clone()));
        replace_test_local_config(
            &config_path,
            format!(
@@ -2274,16 +2343,32 @@ fn spawn_mock_llm_tool_plan_then_transient_final_compaction(
            ),
        );
        let _ = request_notice_sender.send(());
-        let planning_body = serde_json::json!({
-            "id": "resp_final_compaction_planning",
-            "model": "mock-game-model",
-            "output_text": planning_response,
-            "status": "completed",
-            "usage": { "input_tokens": 11, "output_tokens": 22, "total_tokens": 33 }
-        })
-        .to_string();
+        let planning_body = if planning_request.contains("\"stream\":true") {
+            format!(
+                "data: {}\n\ndata: {}\n\n",
+                serde_json::json!({
+                    "type": "response.output_text.delta",
+                    "delta": planning_response
+                }),
+                serde_json::json!({ "type": "response.completed" })
+            )
+        } else {
+            serde_json::json!({
+                "id": "resp_final_compaction_planning",
+                "model": "mock-game-model",
+                "output_text": planning_response,
+                "status": "completed",
+                "usage": { "input_tokens": 11, "output_tokens": 22, "total_tokens": 33 }
+            })
+            .to_string()
+        };
+        let planning_content_type = if planning_request.contains("\"stream\":true") {
+            "text/event-stream; charset=utf-8"
+        } else {
+            "application/json"
+        };
        let planning_http_response = format!(
-            "HTTP/1.1 200 OK\r\nContent-Type: application/json\r\nContent-Length: {}\r\nConnection: close\r\n\r\n{}",
+            "HTTP/1.1 200 OK\r\nContent-Type: {planning_content_type}\r\nContent-Length: {}\r\nConnection: close\r\n\r\n{}",
            planning_body.len(),
            planning_body
        );
@@ -2308,19 +2393,35 @@ fn spawn_mock_llm_tool_plan_then_transient_final_compaction(
        let recovered_compaction_request = read_mock_http_request(&mut recovered_compaction_stream);
        let _ = request_capture_sender.send((
            recovered_compaction_accepted_at_ms,
-            recovered_compaction_request,
+            recovered_compaction_request.clone(),
        ));
        let _ = request_notice_sender.send(());
-        let compaction_body = serde_json::json!({
-            "id": "resp_final_compaction_recovered",
-            "model": "mock-game-model",
-            "output_text": compaction_response,
-            "status": "completed",
-            "usage": { "input_tokens": 44, "output_tokens": 12, "total_tokens": 56 }
-        })
-        .to_string();
+        let compaction_body = if recovered_compaction_request.contains("\"stream\":true") {
+            format!(
+                "data: {}\n\ndata: {}\n\n",
+                serde_json::json!({
+                    "type": "response.output_text.delta",
+                    "delta": compaction_response
+                }),
+                serde_json::json!({ "type": "response.completed" })
+            )
+        } else {
+            serde_json::json!({
+                "id": "resp_final_compaction_recovered",
+                "model": "mock-game-model",
+                "output_text": compaction_response,
+                "status": "completed",
+                "usage": { "input_tokens": 44, "output_tokens": 12, "total_tokens": 56 }
+            })
+            .to_string()
+        };
+        let compaction_content_type = if recovered_compaction_request.contains("\"stream\":true") {
+            "text/event-stream; charset=utf-8"
+        } else {
+            "application/json"
+        };
        let compaction_http_response = format!(
-            "HTTP/1.1 200 OK\r\nContent-Type: application/json\r\nContent-Length: {}\r\nConnection: close\r\n\r\n{}",
+            "HTTP/1.1 200 OK\r\nContent-Type: {compaction_content_type}\r\nContent-Length: {}\r\nConnection: close\r\n\r\n{}",
            compaction_body.len(),
            compaction_body
        );