fix(sse): treat "length" stop_reason as legitimate in detectMalformedNonStream for Claude messages (#12935)

* fix(sse): treat "length" stop_reason as legitimate in detectMalformedNonStream for Claude messages

* docs(changelog): add fragment for length stop_reason fix

Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>

---------

Co-authored-by: diegosouzapw <8016841+diegosouzapw@users.noreply.github.com>
Co-authored-by: jasminsehic <jasminsehic@users.noreply.github.com>

changelog.d/fixes/ollama-qwen3-length-stop-reason.md

@@ -0,0 +1 @@
+- **fix(sse):** `detectMalformedNonStream` no longer flags a Claude-format message as malformed when the content array only contains empty text blocks or the `(empty response)` sentinel and `stop_reason` is `"length"` — a legitimate truncation (e.g. `ollama/qwen3:1.7b` exhausting its reasoning budget before producing visible text), not a real 502-worthy empty response.

open-sse/utils/diagnostics.ts

@@ -291,7 +291,11 @@ export function detectMalformedNonStream(
     // text:""}] — one block, just with no visible text — which is the exact
     // same legitimate truncated-completion shape, so the exemption must apply
     // whenever there is no visible output, not only when content is [].
-    if (stopReason === "max_tokens" || stopReason === "tool_use") return null;
+    // "length" is the OpenAI-style spelling some Claude-compatible shims
+    // (ollama qwen3 with the reasoning budget exhausted) emit for the same
+    // truncated-completion case — sentinel content + stop_reason "length".
+    if (stopReason === "max_tokens" || stopReason === "tool_use" || stopReason === "length")
+      return null;
     // content:[] with no stop_reason at all is non-terminal, not empty (#9971).
     if (content.length === 0 && stopReason.length === 0) return null;
     return "empty_choices";

tests/unit/diagnostics.test.ts

@@ -56,6 +56,19 @@ test("synthOpenAIErrorChunk references provider in message", () => {
   );
 });
 
+test("detectMalformedNonStream allows Claude message with (empty response) + stop_reason=length (ollama qwen3)", () => {
+  const resp = {
+    type: "message",
+    content: [{ type: "text", text: "(empty response)" }],
+    stop_reason: "length",
+  };
+  assert.strictEqual(
+    detectMalformedNonStream(resp),
+    null,
+    "ollama reasoning truncation should not be empty_choices"
+  );
+});
+
 // ── (b) synthResponsesFailure matches a response.failed event ────────────────
 
 test("synthResponsesFailure produces a response.failed SSE event", () => {