fix(ollama): avoid double sanitizing kimi stream output

osolmaz · osolmaz · commit 0559f4d1e4e8 · 2026-05-25T21:06:10.000+08:00
diff --git a/extensions/ollama/src/sanitizers/kimi-inline-reasoning.ts b/extensions/ollama/src/sanitizers/kimi-inline-reasoning.ts
@@ -70,8 +70,5 @@ export function createKimiInlineReasoningSanitizer(): OllamaVisibleContentSaniti
       const resolution = resolveInlineReasoningVisibleText({ text, final: true });
       return resolution.kind === "visible" ? resolution.text : text;
     },
-    shouldSanitizeFinalMessage() {
-      return !bypassInlineReasoning;
-    },
   };
 }
diff --git a/extensions/ollama/src/sanitizers/visible-content-contract.ts b/extensions/ollama/src/sanitizers/visible-content-contract.ts
@@ -5,5 +5,4 @@ export type OllamaVisibleContentStreamResolution =
 export type OllamaVisibleContentSanitizer = {
   resolveStreamText(params: { text: string; final: boolean }): OllamaVisibleContentStreamResolution;
   sanitizeFinalText(text: string): string;
-  shouldSanitizeFinalMessage(): boolean;
 };
diff --git a/extensions/ollama/src/sanitizers/visible-content.ts b/extensions/ollama/src/sanitizers/visible-content.ts
@@ -11,9 +11,6 @@ const noopVisibleContentSanitizer: OllamaVisibleContentSanitizer = {
   sanitizeFinalText(text) {
     return text;
   },
-  shouldSanitizeFinalMessage() {
-    return true;
-  },
 };
 
 export function createOllamaVisibleContentSanitizer(
diff --git a/extensions/ollama/src/stream-runtime.test.ts b/extensions/ollama/src/stream-runtime.test.ts
@@ -2139,6 +2139,47 @@ describe("createOllamaStreamFn streaming events", () => {
     );
   });
 
+  it("does not re-sanitize visible Kimi stream output before done", async () => {
+    const visibleAnswer =
+      "This visible answer is intentionally long enough to look like a reasoning prefix if it is sanitized a second time. ️ keep this marker visible.";
+    await withMockNdjsonFetch(
+      [
+        JSON.stringify({
+          model: "kimi-k2.6:cloud",
+          created_at: "t",
+          message: {
+            role: "assistant",
+            content:
+              "I should think privately and not leak this planning text in the answer. I need to keep deciding what to say next. ️",
+          },
+          done: false,
+        }),
+        JSON.stringify({
+          model: "kimi-k2.6:cloud",
+          created_at: "t",
+          message: { role: "assistant", content: visibleAnswer },
+          done: false,
+        }),
+        '{"model":"kimi-k2.6:cloud","created_at":"t","message":{"role":"assistant","content":""},"done":true,"prompt_eval_count":3,"eval_count":4}',
+      ],
+      async () => {
+        const stream = await createOllamaTestStream({
+          baseUrl: "http://ollama-host:11434",
+          model: { id: "kimi-k2.6:cloud", provider: "ollama" },
+        });
+        const events = await collectStreamEvents(stream);
+        const textEnd = events.find((event) => event.type === "text_end");
+        const doneEvent = events.at(-1);
+
+        expect(textEnd?.content).toBe(visibleAnswer);
+        expect(doneEvent?.type).toBe("done");
+        if (doneEvent?.type === "done") {
+          expect(doneEvent.message.content).toEqual([{ type: "text", text: visibleAnswer }]);
+        }
+      },
+    );
+  });
+
   it("does not leak Kimi inline reasoning when a boundary is followed by tool calls only", async () => {
     const hiddenPrefix =
       "I should think privately and not leak this planning text in the answer. " +
diff --git a/extensions/ollama/src/stream.ts b/extensions/ollama/src/stream.ts
@@ -1367,7 +1367,7 @@ export function createOllamaStreamFn(
           };
           const assistantMessage = buildAssistantMessage(finalResponse, modelInfo, usageFallback, {
             ...toolCallNameOptions,
-            sanitizeVisibleContent: visibleContentSanitizer.shouldSanitizeFinalMessage(),
+            sanitizeVisibleContent: false,
           });
           closeThinkingBlock();
           closeTextBlock();

Original file line number	Diff line number	Diff line change
`@@ -70,8 +70,5 @@ export function createKimiInlineReasoningSanitizer(): OllamaVisibleContentSaniti`
`70`	`70`	`const resolution = resolveInlineReasoningVisibleText({ text, final: true });`
`71`	`71`	`return resolution.kind === "visible" ? resolution.text : text;`
`72`	`72`	`},`
`73`		`- shouldSanitizeFinalMessage() {`
`74`		`- return !bypassInlineReasoning;`
`75`		`- },`
`76`	`73`	`};`
`77`	`74`	`}`