Fix Kimi Coding tool-call replay (#82550)

amknight · web-flow · commit f8b7008f7c53 · 2026-05-16T11:54:46.000Z
Summary: - The PR preserves Kimi Coding reasoning_content replay for OpenAI-compatible tool-call follow-up turns, extends replay model-id matching, adds Kimi wrapper/tests, and updates the changelog. - Reproducibility: yes. at source level: current main drops or fails to synthesize reasoning_content for kimi- ... es a concrete Kimi 400 after tool-call history. I did not run a live Kimi request in this read-only review. Automerge notes: - No ClawSweeper repair was needed after automerge opt-in. Validation: - ClawSweeper review passed for head 9a4605e. - Required merge gates passed before the squash merge. Prepared head SHA: 9a4605e Review: #82550 (comment) Co-authored-by: Alex Knight <15041791+amknight@users.noreply.github.com>
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -82,6 +82,7 @@ Docs: https://docs.openclaw.ai
 - Providers/polling: reject array, null, or scalar successful operation status responses with provider-owned malformed JSON errors instead of waiting until timeout.
 - ACPX/Codex: reap plugin-local Codex ACP adapter orphans on startup after wrapper crashes while keeping direct adapter commands out of launch-lease injection. Fixes #82364. (#82459) Thanks @joshavant.
 - Telegram: send presentation-only payloads by rendering fallback text and inline buttons instead of treating them as empty. Fixes #82404. (#82449) Thanks @joshavant.
+- Providers/Kimi: preserve Kimi Coding `reasoning_content` replay and backfill assistant tool-call placeholders when thinking is enabled, so `kimi-for-coding` follow-up tool turns no longer fail after prior tool use. Fixes #82161. Thanks @amknight.
 - Providers/search tools: reject malformed successful xAI, Gemini, and Kimi web/code search responses with provider-owned errors instead of silent `No response` payloads or ungrounded fallback state.
 - Trajectory export: skip and report malformed session/runtime JSONL rows in `manifest.json` instead of letting wrong-shaped session rows crash support bundle export.
 - Voice calls: persist rejected inbound-call replay keys so duplicate carrier webhook retries stay ignored after a Gateway restart.
diff --git a/extensions/kimi-coding/stream.test.ts b/extensions/kimi-coding/stream.test.ts
@@ -307,6 +307,126 @@ describe("kimi tool-call markup wrapper", () => {
     });
   });
 
+  it("backfills Kimi OpenAI-compatible tool-call reasoning_content when thinking is enabled", () => {
+    const { streamFn: baseStreamFn, getCapturedPayload } = createPayloadCapturingStream({
+      messages: [
+        { role: "user", content: "run pwd" },
+        {
+          role: "assistant",
+          content: null,
+          tool_calls: [
+            {
+              id: "call_1",
+              type: "function",
+              function: { name: "exec", arguments: "{\"command\":\"pwd\"}" },
+            },
+          ],
+        },
+        {
+          role: "assistant",
+          content: "kept",
+          reasoning_content: "native reasoning",
+          tool_calls: [
+            {
+              id: "call_2",
+              type: "function",
+              function: { name: "read", arguments: "{}" },
+            },
+          ],
+        },
+      ],
+    });
+
+    const wrapped = createKimiThinkingWrapper(baseStreamFn, "enabled");
+    void wrapped(
+      {
+        api: "openai-completions",
+        provider: "kimi",
+        id: "kimi-for-coding",
+      } as Model<"openai-completions">,
+      { messages: [] } as Context,
+      {},
+    );
+
+    expect(getCapturedPayload()).toEqual({
+      messages: [
+        { role: "user", content: "run pwd" },
+        {
+          role: "assistant",
+          content: null,
+          reasoning_content: "",
+          tool_calls: [
+            {
+              id: "call_1",
+              type: "function",
+              function: { name: "exec", arguments: "{\"command\":\"pwd\"}" },
+            },
+          ],
+        },
+        {
+          role: "assistant",
+          content: "kept",
+          reasoning_content: "native reasoning",
+          tool_calls: [
+            {
+              id: "call_2",
+              type: "function",
+              function: { name: "read", arguments: "{}" },
+            },
+          ],
+        },
+      ],
+      thinking: { type: "enabled" },
+    });
+  });
+
+  it("strips Kimi OpenAI-compatible replay reasoning_content when thinking is disabled", () => {
+    const { streamFn: baseStreamFn, getCapturedPayload } = createPayloadCapturingStream({
+      messages: [
+        {
+          role: "assistant",
+          content: null,
+          reasoning_content: "old reasoning",
+          tool_calls: [
+            {
+              id: "call_1",
+              type: "function",
+              function: { name: "exec", arguments: "{\"command\":\"pwd\"}" },
+            },
+          ],
+        },
+      ],
+    });
+
+    const wrapped = createKimiThinkingWrapper(baseStreamFn, "disabled");
+    void wrapped(
+      {
+        api: "openai-completions",
+        provider: "kimi",
+        id: "kimi-for-coding",
+      } as Model<"openai-completions">,
+      { messages: [] } as Context,
+      {},
+    );
+
+    expect(getCapturedPayload()).toEqual({
+      messages: [
+        {
+          role: "assistant",
+          content: null,
+          tool_calls: [
+            {
+              id: "call_1",
+              type: "function",
+              function: { name: "exec", arguments: "{\"command\":\"pwd\"}" },
+            },
+          ],
+        },
+      ],
+      thinking: { type: "disabled" },
+    });
+  });
+
   it("enables Kimi Anthropic thinking with a high budget and enough output room", () => {
     const { streamFn: baseStreamFn, getCapturedPayload } = createPayloadCapturingStream();
 
diff --git a/extensions/kimi-coding/stream.ts b/extensions/kimi-coding/stream.ts
@@ -75,6 +75,39 @@ function ensureKimiAnthropicMaxTokens(
   payloadObj.max_tokens = current === undefined ? required : Math.max(current, required);
 }
 
+function messageHasOpenAIToolCalls(message: Record<string, unknown>): boolean {
+  return Array.isArray(message.tool_calls) && message.tool_calls.length > 0;
+}
+
+function ensureKimiOpenAIReasoningContent(payloadObj: Record<string, unknown>): void {
+  if (!Array.isArray(payloadObj.messages)) {
+    return;
+  }
+  for (const message of payloadObj.messages) {
+    if (!message || typeof message !== "object") {
+      continue;
+    }
+    const record = message as Record<string, unknown>;
+    if (record.role !== "assistant" || !messageHasOpenAIToolCalls(record)) {
+      continue;
+    }
+    if (!("reasoning_content" in record)) {
+      record.reasoning_content = "";
+    }
+  }
+}
+
+function stripKimiOpenAIReasoningContent(payloadObj: Record<string, unknown>): void {
+  if (!Array.isArray(payloadObj.messages)) {
+    return;
+  }
+  for (const message of payloadObj.messages) {
+    if (message && typeof message === "object") {
+      delete (message as Record<string, unknown>).reasoning_content;
+    }
+  }
+}
+
 function normalizeKimiThinkingType(value: unknown): KimiThinkingType | undefined {
   if (typeof value === "boolean") {
     return value ? "enabled" : "disabled";
@@ -331,6 +364,10 @@ export function createKimiThinkingWrapper(
         model.api === "anthropic-messages" ? { ...normalized } : { type: normalized.type };
       if (model.api === "anthropic-messages") {
         ensureKimiAnthropicMaxTokens(payloadObj, normalized);
+      } else if (normalized.type === "enabled") {
+        ensureKimiOpenAIReasoningContent(payloadObj);
+      } else {
+        stripKimiOpenAIReasoningContent(payloadObj);
       }
       delete payloadObj.reasoning;
       delete payloadObj.reasoning_effort;
diff --git a/src/agents/openai-transport-stream.test.ts b/src/agents/openai-transport-stream.test.ts
@@ -5725,6 +5725,14 @@ describe("buildOpenAICompletionsParams sanitizes reasoning replay fields", () =>
     maxTokens: 32_000,
   } satisfies Model<"openai-completions">;
 
+  const kimiCodingProxyModel = {
+    ...customKimiProxyModel,
+    id: "kimi-for-coding",
+    name: "Kimi for Coding",
+    provider: "kimi",
+    baseUrl: "https://api.kimi.com/coding/v1",
+  } satisfies Model<"openai-completions">;
+
   function getAssistantMessage(params: { messages: unknown }) {
     expect(Array.isArray(params.messages)).toBe(true);
     const list = params.messages as Array<Record<string, unknown>>;
@@ -5916,6 +5924,17 @@ describe("buildOpenAICompletionsParams sanitizes reasoning replay fields", () =>
     expect(assistant).not.toHaveProperty("reasoning_text");
   });
 
+  it("preserves reasoning_content replay for Kimi Coding OpenAI-compatible routes", () => {
+    const assistant = getAssistantMessage(
+      buildReplayParams(kimiCodingProxyModel, "reasoning_content"),
+    );
+
+    expect(assistant.reasoning_content).toBe("Need to answer politely.");
+    expect(assistant).not.toHaveProperty("reasoning_details");
+    expect(assistant).not.toHaveProperty("reasoning");
+    expect(assistant).not.toHaveProperty("reasoning_text");
+  });
+
   it("preserves reasoning_content replay for suffixed reasoning model ids", () => {
     const assistant = getAssistantMessage(
       buildReplayParams(
@@ -5930,6 +5949,20 @@ describe("buildOpenAICompletionsParams sanitizes reasoning replay fields", () =>
     expect(assistant.reasoning_content).toBe("Need to answer politely.");
   });
 
+  it("preserves reasoning_content replay for prefixed reasoning model ids", () => {
+    const assistant = getAssistantMessage(
+      buildReplayParams(
+        {
+          ...customKimiProxyModel,
+          id: "hf:moonshotai/kimi-k2-thinking",
+        },
+        "reasoning_content",
+      ),
+    );
+
+    expect(assistant.reasoning_content).toBe("Need to answer politely.");
+  });
+
   it("preserves OpenRouter array reasoning_details from tool-call signatures", () => {
     const reasoningDetail = { type: "reasoning.encrypted", id: "rs_1", data: "ciphertext" };
     const params = buildOpenAICompletionsParams(
diff --git a/src/agents/openai-transport-stream.ts b/src/agents/openai-transport-stream.ts
@@ -2552,6 +2552,7 @@ function sanitizeReasoningContentReplayFields(record: Record<string, unknown>):
 const REASONING_CONTENT_REPLAY_MODEL_IDS = new Set([
   "deepseek-v4-flash",
   "deepseek-v4-pro",
+  "kimi-for-coding",
   "kimi-k2.5",
   "kimi-k2.6",
   "kimi-k2-thinking",
@@ -2563,16 +2564,22 @@ const REASONING_CONTENT_REPLAY_MODEL_IDS = new Set([
   "mimo-v2.6-pro",
 ]);
 
-function normalizeReasoningContentReplayModelId(modelId: unknown): string | undefined {
+function getReasoningContentReplayModelIdCandidates(modelId: unknown): string[] {
   if (typeof modelId !== "string") {
-    return undefined;
+    return [];
   }
-  const normalized = modelId.trim().toLowerCase().split(":", 1)[0];
+  const normalized = modelId.trim().toLowerCase();
   if (!normalized) {
-    return undefined;
+    return [];
   }
   const parts = normalized.split("/").filter(Boolean);
-  return parts[parts.length - 1] ?? normalized;
+  const finalPart = parts[parts.length - 1] ?? normalized;
+  const candidates = [finalPart];
+  const colonParts = finalPart.split(":").filter(Boolean);
+  if (colonParts.length > 1) {
+    candidates.push(colonParts[0] ?? "", colonParts[colonParts.length - 1] ?? "");
+  }
+  return [...new Set(candidates.filter(Boolean))];
 }
 
 function shouldPreserveReasoningContentReplay(
@@ -2586,9 +2593,8 @@ function shouldPreserveReasoningContentReplay(
   ) {
     return true;
   }
-  const normalizedModelId = normalizeReasoningContentReplayModelId(model.id);
-  return (
-    normalizedModelId !== undefined && REASONING_CONTENT_REPLAY_MODEL_IDS.has(normalizedModelId)
+  return getReasoningContentReplayModelIdCandidates(model.id).some((modelId) =>
+    REASONING_CONTENT_REPLAY_MODEL_IDS.has(modelId),
   );
 }
 
diff --git a/src/agents/transcript-policy.test.ts b/src/agents/transcript-policy.test.ts
@@ -362,7 +362,14 @@ describe("resolveTranscriptPolicy", () => {
     expect(responsesPolicy.dropReasoningFromHistory).toBe(false);
   });
 
-  it.each(["moonshotai/kimi-k2.6", "kimi-k2-thinking", "xiaomi/mimo-v2.6-pro"])(
+  it.each([
+    "kimi-for-coding",
+    "moonshotai/kimi-k2.6",
+    "kimi-k2-thinking",
+    "hf:moonshotai/kimi-k2-thinking",
+    "xiaomi/mimo-v2.6-pro",
+    "xiaomi/mimo-v2.6-pro:cloud",
+  ])(
     "preserves historical reasoning for %s replay-required OpenAI-compatible models",
     (modelId) => {
       const policy = resolveTranscriptPolicy({
diff --git a/src/agents/transcript-policy.ts b/src/agents/transcript-policy.ts
@@ -158,6 +158,7 @@ function buildUnownedProviderTransportReplayFallback(params: {
 }
 
 const REASONING_CONTENT_REPLAY_MODEL_IDS = new Set([
+  "kimi-for-coding",
   "kimi-k2.5",
   "kimi-k2.6",
   "kimi-k2-thinking",
@@ -170,13 +171,18 @@ const REASONING_CONTENT_REPLAY_MODEL_IDS = new Set([
 ]);
 
 function requiresReasoningContentReplay(modelId: string | null | undefined): boolean {
-  const normalized = normalizeLowercaseStringOrEmpty(modelId).split(":", 1)[0];
+  const normalized = normalizeLowercaseStringOrEmpty(modelId);
   if (!normalized) {
     return false;
   }
   const parts = normalized.split("/").filter(Boolean);
   const finalPart = parts[parts.length - 1] ?? normalized;
-  return REASONING_CONTENT_REPLAY_MODEL_IDS.has(finalPart);
+  const candidates = [finalPart];
+  const colonParts = finalPart.split(":").filter(Boolean);
+  if (colonParts.length > 1) {
+    candidates.push(colonParts[0] ?? "", colonParts[colonParts.length - 1] ?? "");
+  }
+  return candidates.some((candidate) => REASONING_CONTENT_REPLAY_MODEL_IDS.has(candidate));
 }
 
 function mergeTranscriptPolicy(
