fix(gemini-cli): raise output floor for thinking and add validated toolConfig (#2486)

Gemini CLI requests with small max_tokens spend the whole output budget on
thoughts after reasoning_effort maps to thinkingConfig, returning blank
content or finish=length. Raise maxOutputTokens floors per thinking level/
budget (clamped to caps.maxOutput). Also emit toolConfig
functionCallingConfig.mode=VALIDATED for Gemini CLI tool requests to avoid
MALFORMED_FUNCTION_CALL.

Co-authored-by: Cursor <cursoragent@cursor.com>

tests/translator/golden-request.test.js

@@ -54,6 +54,45 @@ describe("GOLDEN request: OpenAI → Gemini", () => {
     const out = translateRequest(FORMATS.OPENAI, FORMATS.GEMINI, "gemini-3-pro", baseBody(), true, { apiKey: "k" }, "gemini");
     expect(clean(out)).toMatchSnapshot();
   });
+
+  it("Gemini CLI tool requests include validated toolConfig and enough output for high thinking", () => {
+    const body = {
+      messages: [{ role: "user", content: "Call add with 7 and 35." }],
+      tools: [
+        {
+          type: "function",
+          function: {
+            name: "add",
+            description: "Add two numbers",
+            parameters: {
+              type: "object",
+              properties: {
+                a: { type: "number" },
+                b: { type: "number" },
+              },
+              required: ["a", "b"],
+            },
+          },
+        },
+      ],
+      reasoning_effort: "high",
+      max_tokens: 128,
+    };
+    const out = translateRequest(
+      FORMATS.OPENAI,
+      FORMATS.GEMINI_CLI,
+      "gemini-3.1-pro-preview",
+      body,
+      true,
+      { accessToken: "t", projectId: "p" },
+      "gemini-cli"
+    );
+
+    expect(out.request.toolConfig).toEqual({ functionCallingConfig: { mode: "VALIDATED" } });
+    expect(out.request.safetySettings).toBeDefined();
+    expect(out.request.generationConfig.thinkingConfig).toEqual({ thinkingLevel: "high", includeThoughts: true });
+    expect(out.request.generationConfig.maxOutputTokens).toBe(65535);
+  });
 });
 
 describe("GOLDEN request: OpenAI → Kiro", () => {

tests/translator/thinking-unified.test.js

@@ -78,11 +78,27 @@ describe("applyThinking per provider format", () => {
     const out = apply("gemini", "gemini-3-pro", { reasoning_effort: "auto" }, "gemini");
     expect(out.generationConfig.thinkingConfig.thinkingLevel).toBe("high");
   });
+  it("gemini-3 high thinking raises too-small maxOutputTokens", () => {
+    const out = apply("gemini-cli", "gemini-3.1-pro-preview", {
+      request: { generationConfig: { maxOutputTokens: 128 } },
+      reasoning_effort: "high",
+    }, "gemini-cli");
+    expect(out.request.generationConfig.thinkingConfig).toEqual({ thinkingLevel: "high", includeThoughts: true });
+    expect(out.request.generationConfig.maxOutputTokens).toBe(65535);
+  });
   it("gemini-2.5 → thinkingBudget", () => {
     const out = apply("gemini", "gemini-2.5-flash", { reasoning_effort: "high" }, "gemini");
     expect(out.generationConfig.thinkingConfig.thinkingBudget).toBe(24576);
     expect(out.generationConfig.thinkingConfig.thinkingLevel).toBeUndefined();
   });
+  it("gemini-2.5 budget thinking keeps enough room for answer tokens", () => {
+    const out = apply("gemini-cli", "gemini-2.5-pro", {
+      request: { generationConfig: { maxOutputTokens: 1024 } },
+      reasoning_effort: "high",
+    }, "gemini-cli");
+    expect(out.request.generationConfig.thinkingConfig).toEqual({ thinkingBudget: 24576, includeThoughts: true });
+    expect(out.request.generationConfig.maxOutputTokens).toBe(32768);
+  });
   it("GLM off → enable_thinking:false (not thinking.disabled)", () => {
     const out = apply("openai", "glm-4.6", { reasoning_effort: "none" }, "glm");
     expect(out.enable_thinking).toBe(false);