From 29c8cae9ba5c1ad2064dbc7aa964e92055dcfea6 Mon Sep 17 00:00:00 2001
From: roboomp <omp@can.ac>
Date: Tue, 14 Jul 2026 16:39:48 +0000
Subject: [PATCH] fix(catalog): extend stream idle floor to kimi k2.7 code

Native Kimi K2.7 Code (kimi-k2.7-code / kimi-k2.7-code-highspeed) reasons
for minutes before the first stream event like K2.6, but the
streamIdleTimeoutMs branch in buildOpenAICompat gated only on
isKimiK26ModelId, so K2.7 Code fell through to the 120s default and
aborted on long reasoning turns. Match matchesKimiK27CodeFamily in the
same branch and rename the constant to KIMI_REASONING_STREAM_IDLE_TIMEOUT_MS.

Fixes #4836
---
 packages/catalog/CHANGELOG.md         |  4 ++++
 packages/catalog/src/compat/openai.ts |  8 ++++----
 packages/catalog/test/build.test.ts   | 21 +++++++++++++++++++++
 3 files changed, 29 insertions(+), 4 deletions(-)

diff --git a/packages/catalog/CHANGELOG.md b/packages/catalog/CHANGELOG.md
index 42e39218a..3ae8f9833 100644
--- a/packages/catalog/CHANGELOG.md
+++ b/packages/catalog/CHANGELOG.md
@@ -2,6 +2,10 @@
 
 ## [Unreleased]
 
+### Fixed
+
+- Extended the reasoning `streamIdleTimeoutMs` floor (300s) to native Kimi K2.7 Code (`kimi-k2.7-code` / `kimi-k2.7-code-highspeed`), which previously fell through to the 120s default and aborted on long reasoning turns ([#4836](https://github.com/can1357/oh-my-pi/issues/4836)).
+
 ## [16.3.11] - 2026-07-06
 
 ### Added
diff --git a/packages/catalog/src/compat/openai.ts b/packages/catalog/src/compat/openai.ts
index d5627ccd5..8198f93b8 100644
--- a/packages/catalog/src/compat/openai.ts
+++ b/packages/catalog/src/compat/openai.ts
@@ -37,8 +37,8 @@ const GLM_CODING_PLAN_MODEL_PATTERN = /(^|\/)glm-5(?:[.-]|$)/i;
 const GLM_CODING_PLAN_STREAM_IDLE_TIMEOUT_MS = 600_000;
 /** Direct DeepSeek reasoning models stall between thinking and answer phases. */
 const DEEPSEEK_REASONING_STREAM_IDLE_TIMEOUT_MS = 300_000;
-/** Kimi K2.6 can spend several minutes reasoning before the first visible token. */
-const KIMI_K26_REASONING_STREAM_IDLE_TIMEOUT_MS = 300_000;
+/** Kimi K2.6 and native K2.7 Code can spend several minutes reasoning before the first visible token. */
+const KIMI_REASONING_STREAM_IDLE_TIMEOUT_MS = 300_000;
 /**
  * Native Kimi K2.7 Code requires `thinking.type: "enabled"` and rejects
  * disabled thinking. Match the public id, its Fast variant, and the
@@ -383,8 +383,8 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 				? ALIBABA_CODING_PLAN_STREAM_IDLE_TIMEOUT_MS
 				: isXiaomiMimo
 					? XIAOMI_MIMO_STREAM_IDLE_TIMEOUT_MS
-					: spec.reasoning && isKimiK26ModelId(spec.id)
-						? KIMI_K26_REASONING_STREAM_IDLE_TIMEOUT_MS
+					: spec.reasoning && (isKimiK26ModelId(spec.id) || matchesKimiK27CodeFamily(spec))
+						? KIMI_REASONING_STREAM_IDLE_TIMEOUT_MS
 						: spec.reasoning && isDirectDeepseekApi
 							? DEEPSEEK_REASONING_STREAM_IDLE_TIMEOUT_MS
 							: isLocalOpenAICompatBackend
diff --git a/packages/catalog/test/build.test.ts b/packages/catalog/test/build.test.ts
index 706321869..628d01cd3 100644
--- a/packages/catalog/test/build.test.ts
+++ b/packages/catalog/test/build.test.ts
@@ -282,6 +282,27 @@ describe("openai-completions wire-quirk compat detection", () => {
 		expect(buildOpenAICompat(completionsSpec()).reasoningDeltasMayBeCumulative).toBe(false);
 	});
 
+	it("extends the reasoning stream idle floor to Kimi K2.6 and K2.7 Code, not other reasoning models", () => {
+		const kimiOverrides = {
+			provider: "moonshot",
+			baseUrl: "https://api.moonshot.ai/v1",
+			reasoning: true,
+		} as const;
+		expect(buildOpenAICompat(completionsSpec({ ...kimiOverrides, id: "kimi-k2.6" })).streamIdleTimeoutMs).toBe(
+			300_000,
+		);
+		expect(buildOpenAICompat(completionsSpec({ ...kimiOverrides, id: "kimi-k2.7-code" })).streamIdleTimeoutMs).toBe(
+			300_000,
+		);
+		expect(
+			buildOpenAICompat(completionsSpec({ ...kimiOverrides, id: "kimi-k2.7-code-highspeed" })).streamIdleTimeoutMs,
+		).toBe(300_000);
+		// A non-Kimi reasoning model on a generic host keeps the runtime default.
+		expect(
+			buildOpenAICompat(completionsSpec({ id: "some-reasoner", reasoning: true })).streamIdleTimeoutMs,
+		).toBeUndefined();
+	});
+
 	it("maps the remaining provider-keyed wire quirks", () => {
 		expect(buildOpenAICompat(completionsSpec({ provider: "ollama" })).emptyLengthFinishIsContextError).toBe(true);
 		expect(buildOpenAICompat(completionsSpec()).emptyLengthFinishIsContextError).toBe(false);