fix: suppress tool-call finish_reason planning content

src/proxy.tool-forwarding.test.ts

@@ -214,7 +214,9 @@ describe("tool forwarding", () => {
       body: JSON.stringify({
         model: "moonshot/kimi-k2.6",
         stream: true,
-        messages: [{ role: "user", content: "What time is it in Chicago right now? Use the tool." }],
+        messages: [
+          { role: "user", content: "What time is it in Chicago right now? Use the tool." },
+        ],
         tools: [
           {
             type: "function",
@@ -286,4 +288,57 @@ describe("tool forwarding", () => {
     );
     expect(finishReasons).toContain("tool_calls");
   });
+
+  it("suppresses assistant content when upstream marks finish_reason as tool_calls", async () => {
+    upstreamResponse = {
+      id: "chatcmpl-tool-finish-reason",
+      object: "chat.completion",
+      created: Math.floor(Date.now() / 1000),
+      model: "moonshot/kimi-k2.6",
+      choices: [
+        {
+          index: 0,
+          message: {
+            role: "assistant",
+            content: "I should look up Barcelona's next match before replying.",
+          },
+          finish_reason: "tool_calls",
+        },
+      ],
+      usage: { prompt_tokens: 10, completion_tokens: 20, total_tokens: 30 },
+    };
+
+    const res = await fetch(`${proxy.baseUrl}/v1/chat/completions`, {
+      method: "POST",
+      headers: { "Content-Type": "application/json" },
+      body: JSON.stringify({
+        model: "moonshot/kimi-k2.6",
+        stream: false,
+        messages: [{ role: "user", content: "When is Barcelona's next match?" }],
+        tools: [
+          {
+            type: "function",
+            function: {
+              name: "web_search",
+              description: "Search the web",
+              parameters: { type: "object" },
+            },
+          },
+        ],
+      }),
+    });
+
+    expect(res.status).toBe(200);
+    const json = (await res.json()) as {
+      choices?: Array<{
+        finish_reason?: string | null;
+        message?: {
+          content?: string;
+          tool_calls?: unknown[];
+        };
+      }>;
+    };
+    expect(json.choices?.[0]?.finish_reason).toBe("tool_calls");
+    expect(json.choices?.[0]?.message?.content).toBe("");
+  });
 });

src/proxy.ts

@@ -5069,14 +5069,19 @@ async function proxyRequest(
           // Process each choice (usually just one)
           if (rsp.choices && Array.isArray(rsp.choices)) {
             for (const choice of rsp.choices) {
+              const endsWithToolCalls = choice.finish_reason === "tool_calls";
               // Some OpenAI-compatible providers include planning prose in content
-              // alongside tool_calls. Tool execution only needs tool_calls, so do
-              // not forward that prose to chat channels.
+              // alongside tool_calls, or mark the turn as a tool-call turn via
+              // finish_reason before exposing the tool_calls array at the same
+              // object shape. Tool execution only needs tool_calls, so do not
+              // forward that prose to chat channels.
               const toolCalls = choice.message?.tool_calls ?? choice.delta?.tool_calls;
               // Strip thinking tokens (Kimi <｜...｜> and standard <think> tags)
               const rawContent = choice.message?.content ?? choice.delta?.content ?? "";
               const content =
-                toolCalls && toolCalls.length > 0 ? "" : stripThinkingTokens(rawContent);
+                endsWithToolCalls || (toolCalls && toolCalls.length > 0)
+                  ? ""
+                  : stripThinkingTokens(rawContent);
               const role = choice.message?.role ?? choice.delta?.role ?? "assistant";
               const index = choice.index ?? 0;
 
@@ -5170,7 +5175,7 @@ async function proxyRequest(
                     delta: {},
                     logprobs: null,
                     finish_reason:
-                      toolCalls && toolCalls.length > 0
+                      endsWithToolCalls || (toolCalls && toolCalls.length > 0)
                         ? "tool_calls"
                         : (choice.finish_reason ?? "stop"),
                   },
@@ -5300,6 +5305,7 @@ async function proxyRequest(
         try {
           const parsed = JSON.parse(responseBody.toString()) as {
             choices?: Array<{
+              finish_reason?: string | null;
               message?: {
                 content?: string;
                 tool_calls?: unknown[];
@@ -5311,7 +5317,10 @@ async function proxyRequest(
             const message = choice.message;
             if (!message || typeof message.content !== "string") continue;
 
-            if (Array.isArray(message.tool_calls) && message.tool_calls.length > 0) {
+            if (
+              choice.finish_reason === "tool_calls" ||
+              (Array.isArray(message.tool_calls) && message.tool_calls.length > 0)
+            ) {
               if (message.content !== "") {
                 message.content = "";
                 changed = true;