fix(agent): stabilize stream token output

tests/routes/chatStream.test.ts

@@ -19,14 +19,9 @@ const createEventStream = (events: unknown[]) => ({
 const sleep = (ms: number) => new Promise((resolve) => setTimeout(resolve, ms));
 
 describe("streamPromptResponse", () => {
-  it("detects Chinese punctuation chunks for smooth token output", () => {
-    expect(detectTokenSmoothingChunk("管网压力异常，需要继续分析", "zh")).toBe(
-      "管网压力异常，",
-    );
-
-    expect(detectTokenSmoothingChunk("管网压力异常需要继续分析。后续排查阀门", "zh")).toBe(
-      "管网压力异常需要继续分析。",
-    );
+  it("does not split short Chinese text only because it has punctuation", () => {
+    expect(detectTokenSmoothingChunk("管网压力异常，需要继续分析", "zh")).toBeNull();
+    expect(detectTokenSmoothingChunk("管网压力异常需要继续分析。", "zh")).toBeNull();
   });
 
   it("does not immediately release a single Chinese character", () => {
@@ -67,7 +62,7 @@ describe("streamPromptResponse", () => {
     expect(events.map((item) => item.data.content)).toEqual(["第一段", "第二段"]);
   });
 
-  it("buffers Chinese single-character deltas until punctuation", async () => {
+  it("buffers Chinese single-character deltas until a bounded chunk is available", async () => {
     const events: Array<{ event: string; data: Record<string, unknown> }> = [];
     const smoother = createTokenSmoother({
       enabled: true,
@@ -81,13 +76,15 @@ describe("streamPromptResponse", () => {
     await sleep(8);
     expect(events).toHaveLength(0);
 
-    for (const char of Array.from("网压力异常，")) {
+    const content = "网压力异常需要继续分析东部主干供水走廊和相关阀门状态";
+    for (const char of Array.from(content)) {
       smoother.writeToken(char);
     }
     await sleep(20);
 
-    expect(events.map((item) => item.data.content).join("")).toBe("管网压力异常，");
+    expect(events.length).toBeGreaterThan(0);
     await smoother.flush();
+    expect(events.map((item) => item.data.content).join("")).toBe(`管${content}`);
   });
 
   it("smooths long Chinese text without waiting for paragraph-sized buffers", async () => {