@shareai-lab/kode 2.0.1 → 2.0.3

package/dist/{llm-ZUQC4WYM.js → chunks/llm-62N6T5ZT.js}

@@ -1,47 +1,35 @@
-import { createRequire as __kodeCreateRequire } from "node:module";
-const require = __kodeCreateRequire(import.meta.url);
 import {
-  getToolDescription
-} from "./chunk-WVHORZQ5.js";
+  models_default
+} from "./chunk-2WEXPKHH.js";
+import {
+  addToTotalCost
+} from "./chunk-6BAS4WY6.js";
 import {
   processResponsesStream
-} from "./chunk-HN4E4UUQ.js";
+} from "./chunk-TI2CTTMA.js";
+import {
+  getCompletionWithProfile,
+  getGPT5CompletionWithProfile
+} from "./chunk-IFCIADS3.js";
+import {
+  getReasoningEffort
+} from "./chunk-53M46S5I.js";
 import {
   formatSystemPromptWithContext,
   generateKodeContext,
   getCLISyspromptPrefix,
-  getReasoningEffort,
-  models_default,
   refreshKodeContext
-} from "./chunk-FC5ZCKBI.js";
-import "./chunk-K2CWOTI2.js";
-import "./chunk-CM3EGTG6.js";
-import "./chunk-UHYRLID6.js";
-import "./chunk-QVLYOPO5.js";
-import "./chunk-ZVDRDPII.js";
-import "./chunk-MWRSY4X6.js";
-import "./chunk-KAA5BGMQ.js";
+} from "./chunk-HPYNW6TT.js";
+import "./chunk-LOD5ZHCI.js";
+import "./chunk-GELCZWMB.js";
+import "./chunk-3BYE3ME6.js";
+import "./chunk-M7P3QNRU.js";
+import "./chunk-SWQV4KSY.js";
 import {
   setRequestStatus
-} from "./chunk-3RUXVV4S.js";
-import "./chunk-ARZSBOAO.js";
-import "./chunk-NPFOMITO.js";
-import "./chunk-YC6LJCDE.js";
-import {
-  getCompletionWithProfile,
-  getGPT5CompletionWithProfile
-} from "./chunk-DZE5YA7L.js";
-import "./chunk-ND3XWFO6.js";
-import "./chunk-4FX3IVPT.js";
-import "./chunk-W7GRKO7Q.js";
-import "./chunk-AFFSCMYS.js";
-import "./chunk-HCBELH4J.js";
-import "./chunk-S6HRABTA.js";
-import "./chunk-7M2YN6TU.js";
-import "./chunk-E6YNABER.js";
-import "./chunk-73WGVYLQ.js";
-import "./chunk-UKHTVRJM.js";
-import "./chunk-S3J2TLV6.js";
+} from "./chunk-LOCXPQNJ.js";
+import "./chunk-ZCLTZIVP.js";
+import "./chunk-4CRUCZR4.js";
 import {
   API_ERROR_MESSAGE_PREFIX,
   CREDIT_BALANCE_TOO_LOW_ERROR_MESSAGE,
@@ -51,20 +39,13 @@ import {
   PROMPT_TOO_LONG_ERROR_MESSAGE,
   createAssistantAPIErrorMessage,
   normalizeContentFromAPI
-} from "./chunk-RZWOUA25.js";
+} from "./chunk-HRJ3ICQK.js";
 import {
   USE_BEDROCK,
   USE_VERTEX,
   getModelManager,
   getVertexRegionForModel
-} from "./chunk-IZVMU4S2.js";
-import "./chunk-5PDP7R6N.js";
-import "./chunk-7U7L4NMD.js";
-import {
-  getAnthropicApiKey,
-  getGlobalConfig
-} from "./chunk-LC4TVOCZ.js";
-import "./chunk-UYXEDKOZ.js";
+} from "./chunk-6ZWEOSEI.js";
 import {
   debug,
   getCurrentRequest,
@@ -72,166 +53,157 @@ import {
   logLLMInteraction,
   logSystemPromptConstruction,
   markPhase
-} from "./chunk-STSX7GIX.js";
+} from "./chunk-YIO5EBMQ.js";
 import {
   PRODUCT_COMMAND,
   env,
-  getCwd,
   logError
-} from "./chunk-YXYYDIMI.js";
+} from "./chunk-3OEJVB5A.js";
+import {
+  getCwd
+} from "./chunk-BBJFHTBC.js";
+import "./chunk-QHQOBUF6.js";
 import {
   MACRO
-} from "./chunk-F2SJXUDI.js";
+} from "./chunk-W5EGGA44.js";
+import "./chunk-DXD76CMV.js";
+import "./chunk-LB6TCPDI.js";
 import {
-  addToTotalCost
-} from "./chunk-EZXMVTDU.js";
-import "./chunk-JC6NCUG5.js";
+  getAnthropicApiKey,
+  getGlobalConfig
+} from "./chunk-XR2W3MAM.js";
+import "./chunk-B3MW3YGY.js";
+import "./chunk-54KOYG5C.js";
 
-// src/services/ai/llm.ts
+// packages/core/src/ai/llm.ts
 import "@anthropic-ai/sdk/shims/node";
-import Anthropic, { APIConnectionError, APIError } from "@anthropic-ai/sdk";
-import { AnthropicBedrock } from "@anthropic-ai/bedrock-sdk";
-import { AnthropicVertex } from "@anthropic-ai/vertex-sdk";
-import chalk from "chalk";
-import { randomUUID } from "crypto";
+import { randomUUID as randomUUID4 } from "crypto";
 import "dotenv/config";
 
-// src/utils/system/http.ts
-var USER_AGENT = `${PRODUCT_COMMAND}/${MACRO.VERSION} (${process.env.USER_TYPE})`;
+// packages/core/src/ai/llm/openai/queryOpenAI.ts
+import { randomUUID as randomUUID3 } from "crypto";
+import { zodToJsonSchema as zodToJsonSchema4 } from "zod-to-json-schema";
 
-// src/services/system/vcr.ts
-import { createHash } from "crypto";
-import { mkdirSync, readFileSync, writeFileSync } from "fs";
-import { dirname } from "path";
-import { existsSync } from "fs";
-import * as path from "path";
-import { mapValues } from "lodash-es";
-async function withVCR(messages, f) {
-  if (process.env.NODE_ENV !== "test") {
-    return await f();
-  }
-  const dehydratedInput = mapMessages(
-    messages.map((_) => _.message.content),
-    dehydrateValue
-  );
-  const filename = `./fixtures/${dehydratedInput.map((_) => createHash("sha1").update(JSON.stringify(_)).digest("hex").slice(0, 6)).join("-")}.json`;
-  if (existsSync(filename)) {
-    const cached = JSON.parse(readFileSync(filename, "utf-8"));
-    return mapAssistantMessage(cached.output, hydrateValue);
-  }
-  if (env.isCI) {
-    process.stderr.write(
-      `Anthropic API fixture missing. Re-run bun test locally, then commit the result. ${JSON.stringify({ input: dehydratedInput }, null, 2)}
-`
-    );
-  }
-  const result = await f();
-  if (env.isCI) {
-    return result;
-  }
-  if (!existsSync(dirname(filename))) {
-    mkdirSync(dirname(filename), { recursive: true });
-  }
-  writeFileSync(
-    filename,
-    JSON.stringify(
-      {
-        input: dehydratedInput,
-        output: mapAssistantMessage(result, dehydrateValue)
-      },
-      null,
-      2
-    )
-  );
-  return result;
+// packages/core/src/ai/llm/systemPromptUtils.ts
+var PROMPT_CACHING_ENABLED = !process.env.DISABLE_PROMPT_CACHING;
+function splitSysPromptPrefix(systemPrompt) {
+  const systemPromptFirstBlock = systemPrompt[0] || "";
+  const systemPromptRest = systemPrompt.slice(1);
+  return [systemPromptFirstBlock, systemPromptRest.join("\n")].filter(Boolean);
 }
-function mapMessages(messages, f) {
-  return messages.map((_) => {
-    if (typeof _ === "string") {
-      return f(_);
+
+// packages/core/src/ai/llm/retry.ts
+import "@anthropic-ai/sdk/shims/node";
+import { APIConnectionError, APIError } from "@anthropic-ai/sdk";
+var MAX_RETRIES = process.env.USER_TYPE === "SWE_BENCH" ? 100 : 10;
+var BASE_DELAY_MS = 500;
+function abortableDelay(delayMs, signal) {
+  return new Promise((resolve, reject) => {
+    if (signal?.aborted) {
+      reject(new Error("Request was aborted"));
+      return;
+    }
+    const timeoutId = setTimeout(() => {
+      resolve();
+    }, delayMs);
+    if (signal) {
+      const abortHandler = () => {
+        clearTimeout(timeoutId);
+        reject(new Error("Request was aborted"));
+      };
+      signal.addEventListener("abort", abortHandler, { once: true });
     }
-    return _.map((_2) => {
-      switch (_2.type) {
-        case "tool_result":
-          if (typeof _2.content === "string") {
-            return { ..._2, content: f(_2.content) };
-          }
-          if (Array.isArray(_2.content)) {
-            return {
-              ..._2,
-              content: _2.content.map((_3) => {
-                switch (_3.type) {
-                  case "text":
-                    return { ..._3, text: f(_3.text) };
-                  case "image":
-                    return _3;
-                }
-              })
-            };
-          }
-          return _2;
-        case "text":
-          return { ..._2, text: f(_2.text) };
-        case "tool_use":
-          return {
-            ..._2,
-            input: mapValues(_2.input, f)
-          };
-        case "image":
-          return _2;
-      }
-    });
   });
 }
-function mapAssistantMessage(message, f) {
-  return {
-    durationMs: "DURATION",
-    costUSD: "COST",
-    uuid: "UUID",
-    message: {
-      ...message.message,
-      content: message.message.content.map((_) => {
-        switch (_.type) {
-          case "text":
-            return {
-              ..._,
-              text: f(_.text),
-              citations: _.citations || []
-            };
-          case "tool_use":
-            return {
-              ..._,
-              input: mapValues(_.input, f)
-            };
-          default:
-            return _;
-        }
-      }).filter(Boolean)
-    },
-    type: "assistant"
-  };
+function getRetryDelay(attempt, retryAfterHeader) {
+  if (retryAfterHeader) {
+    const seconds = parseInt(retryAfterHeader, 10);
+    if (!isNaN(seconds)) {
+      return seconds * 1e3;
+    }
+  }
+  return Math.min(BASE_DELAY_MS * Math.pow(2, attempt - 1), 32e3);
 }
-function dehydrateValue(s) {
-  if (typeof s !== "string") {
-    return s;
+function shouldRetry(error) {
+  if (error.message?.includes('"type":"overloaded_error"')) {
+    return process.env.USER_TYPE === "SWE_BENCH";
   }
-  const s1 = s.replace(/num_files="\d+"/g, 'num_files="[NUM]"').replace(/duration_ms="\d+"/g, 'duration_ms="[DURATION]"').replace(/cost_usd="\d+"/g, 'cost_usd="[COST]"').replace(/\//g, path.sep).replaceAll(getCwd(), "[CWD]");
-  if (s1.includes("Files modified by user:")) {
-    return "Files modified by user: [FILES]";
+  const shouldRetryHeader = error.headers?.["x-should-retry"];
+  if (shouldRetryHeader === "true") return true;
+  if (shouldRetryHeader === "false") return false;
+  if (error instanceof APIConnectionError) {
+    return true;
   }
-  return s1;
+  if (!error.status) return false;
+  if (error.status === 408) return true;
+  if (error.status === 409) return true;
+  if (error.status === 429) return true;
+  if (error.status && error.status >= 500) return true;
+  return false;
 }
-function hydrateValue(s) {
-  if (typeof s !== "string") {
-    return s;
+async function withRetry(operation, options = {}) {
+  const maxRetries = options.maxRetries ?? MAX_RETRIES;
+  let lastError;
+  for (let attempt = 1; attempt <= maxRetries + 1; attempt++) {
+    try {
+      return await operation(attempt);
+    } catch (error) {
+      lastError = error;
+      if (attempt > maxRetries || !(error instanceof APIError) || !shouldRetry(error)) {
+        throw error;
+      }
+      if (options.signal?.aborted) {
+        throw new Error("Request cancelled by user");
+      }
+      const retryAfter = error.headers?.["retry-after"] ?? null;
+      const delayMs = getRetryDelay(attempt, retryAfter);
+      debug.warn("LLM_API_RETRY", {
+        name: error.name,
+        message: error.message,
+        status: error.status,
+        attempt,
+        maxRetries,
+        delayMs
+      });
+      try {
+        await abortableDelay(delayMs, options.signal);
+      } catch (delayError) {
+        if (delayError.message === "Request was aborted") {
+          throw new Error("Request cancelled by user");
+        }
+        throw delayError;
+      }
+    }
   }
-  return s.replaceAll("[NUM]", "1").replaceAll("[DURATION]", "100").replaceAll("[CWD]", getCwd());
+  throw lastError;
 }
 
-// src/services/ai/llm.ts
-import { zodToJsonSchema as zodToJsonSchema4 } from "zod-to-json-schema";
+// packages/core/src/ai/llm/errors.ts
+function getAssistantMessageFromError(error) {
+  if (error instanceof Error && error.message.includes("prompt is too long")) {
+    return createAssistantAPIErrorMessage(PROMPT_TOO_LONG_ERROR_MESSAGE);
+  }
+  if (error instanceof Error && error.message.includes("Your credit balance is too low")) {
+    return createAssistantAPIErrorMessage(CREDIT_BALANCE_TOO_LOW_ERROR_MESSAGE);
+  }
+  if (error instanceof Error && error.message.toLowerCase().includes("x-api-key")) {
+    return createAssistantAPIErrorMessage(INVALID_API_KEY_ERROR_MESSAGE);
+  }
+  if (error instanceof Error) {
+    if (process.env.NODE_ENV === "development") {
+      debug.error("ANTHROPIC_API_ERROR", {
+        message: error.message,
+        stack: error.stack
+      });
+    }
+    return createAssistantAPIErrorMessage(
+      `${API_ERROR_MESSAGE_PREFIX}: ${error.message}`
+    );
+  }
+  return createAssistantAPIErrorMessage(API_ERROR_MESSAGE_PREFIX);
+}
 
-// src/services/ai/adapters/base.ts
+// packages/core/src/ai/adapters/base.ts
 function normalizeTokens(apiResponse) {
   if (!apiResponse || typeof apiResponse !== "object") {
     return { input: 0, output: 0 };
@@ -257,13 +229,16 @@ var ModelAPIAdapter = class {
     this.modelProfile = modelProfile;
   }
   cumulativeUsage = { input: 0, output: 0 };
+  // Optional: subclasses can implement streaming for real-time updates
+  // Default implementation yields no events (not supported)
   async *parseStreamingResponse(response, signal) {
     return;
-    yield;
   }
+  // Reset cumulative usage for new requests
   resetCumulativeUsage() {
     this.cumulativeUsage = { input: 0, output: 0 };
   }
+  // Safely update cumulative usage
   updateCumulativeUsage(usage) {
     this.cumulativeUsage.input += usage.input;
     this.cumulativeUsage.output += usage.output;
@@ -274,6 +249,7 @@ var ModelAPIAdapter = class {
       this.cumulativeUsage.reasoning = (this.cumulativeUsage.reasoning || 0) + usage.reasoning;
     }
   }
+  // Shared utility methods
   getMaxTokensParam() {
     return this.capabilities.parameters.maxTokensField;
   }
@@ -294,12 +270,26 @@ var ModelAPIAdapter = class {
   }
 };
 
-// src/services/ai/adapters/openaiAdapter.ts
+// packages/core/src/tooling/Tool.ts
+function getToolDescription(tool) {
+  if (tool.cachedDescription) {
+    return tool.cachedDescription;
+  }
+  if (typeof tool.description === "string") {
+    return tool.description;
+  }
+  return `Tool: ${tool.name}`;
+}
+
+// packages/core/src/ai/adapters/openaiAdapter.ts
 import { zodToJsonSchema } from "zod-to-json-schema";
 var OpenAIAdapter = class extends ModelAPIAdapter {
   constructor(capabilities, modelProfile) {
     super(capabilities, modelProfile);
   }
+  /**
+   * Unified parseResponse that handles both streaming and non-streaming responses
+   */
   async parseResponse(response) {
     if (response?.body instanceof ReadableStream) {
       const { assistantMessage } = await this.parseStreamingOpenAIResponse(response);
@@ -320,6 +310,9 @@ var OpenAIAdapter = class extends ModelAPIAdapter {
     }
     return this.parseNonStreamingResponse(response);
   }
+  /**
+   * Common streaming response parser for all OpenAI APIs
+   */
   async *parseStreamingResponse(response) {
     const reader = response.body.getReader();
     const decoder = new TextDecoder();
@@ -387,6 +380,9 @@ var OpenAIAdapter = class extends ModelAPIAdapter {
       }
     };
   }
+  /**
+   * Parse SSE chunk - common for all OpenAI APIs
+   */
   parseSSEChunk(line) {
     if (line.startsWith("data: ")) {
       const data = line.slice(6).trim();
@@ -407,6 +403,9 @@ var OpenAIAdapter = class extends ModelAPIAdapter {
     }
     return null;
   }
+  /**
+   * Common helper for processing text deltas
+   */
   handleTextDelta(delta, responseId, hasStarted) {
     const events = [];
     if (!hasStarted && delta) {
@@ -428,6 +427,9 @@ var OpenAIAdapter = class extends ModelAPIAdapter {
     }
     return events;
   }
+  /**
+   * Common usage normalization
+   */
   normalizeUsageForAdapter(usage) {
     if (!usage) {
       return {
@@ -451,6 +453,9 @@ var OpenAIAdapter = class extends ModelAPIAdapter {
       reasoningTokens: usage.reasoningTokens ?? 0
     };
   }
+  /**
+   * Common tool building logic
+   */
   buildTools(tools) {
     return tools.map((tool) => ({
       type: "function",
@@ -463,22 +468,202 @@ var OpenAIAdapter = class extends ModelAPIAdapter {
   }
 };
 
-// src/services/ai/adapters/responsesAPI.ts
+// packages/core/src/ai/adapters/responsesAPI.ts
 import { zodToJsonSchema as zodToJsonSchema2 } from "zod-to-json-schema";
-var ResponsesAPIAdapter = class extends OpenAIAdapter {
-  createRequest(params) {
-    const {
-      messages,
-      systemPrompt,
-      tools,
-      maxTokens,
-      reasoningEffort,
-      stopSequences
+
+// packages/core/src/ai/adapters/responsesAPI/messageInput.ts
+function convertMessagesToInput(messages) {
+  const inputItems = [];
+  for (const message of messages) {
+    const role = message.role;
+    if (role === "tool") {
+      const callId = message.tool_call_id || message.id;
+      if (typeof callId === "string" && callId) {
+        let content2 = message.content || "";
+        if (Array.isArray(content2)) {
+          const texts = [];
+          for (const part of content2) {
+            if (typeof part === "object" && part !== null) {
+              const t = part.text || part.content;
+              if (typeof t === "string" && t) {
+                texts.push(t);
+              }
+            }
+          }
+          content2 = texts.join("\n");
+        }
+        if (typeof content2 === "string") {
+          inputItems.push({
+            type: "function_call_output",
+            call_id: callId,
+            output: content2
+          });
+        }
+      }
+      continue;
+    }
+    if (role === "assistant" && Array.isArray(message.tool_calls)) {
+      for (const tc of message.tool_calls) {
+        if (typeof tc !== "object" || tc === null) {
+          continue;
+        }
+        const tcType = tc.type || "function";
+        if (tcType !== "function") {
+          continue;
+        }
+        const callId = tc.id || tc.call_id;
+        const fn = tc.function;
+        const name = typeof fn === "object" && fn !== null ? fn.name : null;
+        const args = typeof fn === "object" && fn !== null ? fn.arguments : null;
+        if (typeof callId === "string" && typeof name === "string" && typeof args === "string") {
+          inputItems.push({
+            type: "function_call",
+            name,
+            arguments: args,
+            call_id: callId
+          });
+        }
+      }
+      continue;
+    }
+    const content = message.content || "";
+    const contentItems = [];
+    if (Array.isArray(content)) {
+      for (const part of content) {
+        if (typeof part !== "object" || part === null) continue;
+        const ptype = part.type;
+        if (ptype === "text") {
+          const text = part.text || part.content || "";
+          if (typeof text === "string" && text) {
+            const kind = role === "assistant" ? "output_text" : "input_text";
+            contentItems.push({ type: kind, text });
+          }
+        } else if (ptype === "image_url") {
+          const image = part.image_url;
+          const url = typeof image === "object" && image !== null ? image.url : image;
+          if (typeof url === "string" && url) {
+            contentItems.push({ type: "input_image", image_url: url });
+          }
+        }
+      }
+    } else if (typeof content === "string" && content) {
+      const kind = role === "assistant" ? "output_text" : "input_text";
+      contentItems.push({ type: kind, text: content });
+    }
+    if (contentItems.length) {
+      const roleOut = role === "assistant" ? "assistant" : "user";
+      inputItems.push({
+        type: "message",
+        role: roleOut,
+        content: contentItems
+      });
+    }
+  }
+  return inputItems;
+}
+function buildInstructions(systemPrompt) {
+  const systemContent = systemPrompt.filter((content) => content.trim()).join("\n\n");
+  return systemContent;
+}
+
+// packages/core/src/ai/adapters/responsesAPI/nonStreaming.ts
+function parseToolCalls(response) {
+  if (!response.output || !Array.isArray(response.output)) {
+    return [];
+  }
+  const toolCalls = [];
+  for (const item of response.output) {
+    if (item.type === "function_call") {
+      const callId = item.call_id || item.id;
+      const name = item.name || "";
+      const args = item.arguments || "{}";
+      if (typeof callId === "string" && typeof name === "string" && typeof args === "string") {
+        toolCalls.push({
+          id: callId,
+          type: "function",
+          function: {
+            name,
+            arguments: args
+          }
+        });
+      }
+    } else if (item.type === "tool_call") {
+      const callId = item.id || `tool_${Math.random().toString(36).substring(2, 15)}`;
+      toolCalls.push({
+        id: callId,
+        type: "tool_call",
+        name: item.name,
+        arguments: item.arguments
+      });
+    }
+  }
+  return toolCalls;
+}
+function parseNonStreamingResponse(response) {
+  let content = response.output_text || "";
+  let reasoningContent = "";
+  if (response.output && Array.isArray(response.output)) {
+    const messageItems = response.output.filter(
+      (item) => item.type === "message"
+    );
+    if (messageItems.length > 0) {
+      content = messageItems.map((item) => {
+        if (item.content && Array.isArray(item.content)) {
+          return item.content.filter((c) => c.type === "text").map((c) => c.text).join("\n");
+        }
+        return item.content || "";
+      }).filter(Boolean).join("\n\n");
+    }
+    const reasoningItems = response.output.filter(
+      (item) => item.type === "reasoning"
+    );
+    if (reasoningItems.length > 0) {
+      reasoningContent = reasoningItems.map((item) => item.content || "").filter(Boolean).join("\n\n");
+    }
+  }
+  if (reasoningContent) {
+    const thinkBlock = `
+
+${reasoningContent}
+
+`;
+    content = thinkBlock + content;
+  }
+  const toolCalls = parseToolCalls(response);
+  const contentArray = content ? [{ type: "text", text: content, citations: [] }] : [{ type: "text", text: "", citations: [] }];
+  const promptTokens = response.usage?.input_tokens || 0;
+  const completionTokens = response.usage?.output_tokens || 0;
+  const totalTokens = response.usage?.total_tokens ?? promptTokens + completionTokens;
+  return {
+    id: response.id || `resp_${Date.now()}`,
+    content: contentArray,
+    // Return as array (Anthropic format)
+    toolCalls,
+    usage: {
+      promptTokens,
+      completionTokens,
+      reasoningTokens: response.usage?.output_tokens_details?.reasoning_tokens
+    },
+    responseId: response.id
+    // Save for state management
+  };
+}
+
+// packages/core/src/ai/adapters/responsesAPI.ts
+var ResponsesAPIAdapter = class extends OpenAIAdapter {
+  createRequest(params) {
+    const {
+      messages,
+      systemPrompt,
+      tools,
+      maxTokens,
+      reasoningEffort,
+      stopSequences
     } = params;
     const request = {
       model: this.modelProfile.modelName,
-      input: this.convertMessagesToInput(messages),
-      instructions: this.buildInstructions(systemPrompt)
+      input: convertMessagesToInput(messages),
+      instructions: buildInstructions(systemPrompt)
     };
     const maxTokensField = this.getMaxTokensParam();
     request[maxTokensField] = maxTokens;
@@ -530,17 +715,19 @@ var ResponsesAPIAdapter = class extends OpenAIAdapter {
     return request;
   }
   buildTools(tools) {
+    const isPlainObject = (obj) => {
+      return obj !== null && typeof obj === "object" && !Array.isArray(obj);
+    };
     return tools.map((tool) => {
       let parameters = tool.inputJSONSchema;
       if (!parameters && tool.inputSchema) {
-        const isPlainObject = (obj) => {
-          return obj !== null && typeof obj === "object" && !Array.isArray(obj);
-        };
-        if (isPlainObject(tool.inputSchema) && ("type" in tool.inputSchema || "properties" in tool.inputSchema)) {
-          parameters = tool.inputSchema;
+        const inputSchema = tool.inputSchema;
+        if (isPlainObject(inputSchema) && ("type" in inputSchema || "properties" in inputSchema)) {
+          parameters = inputSchema;
         } else {
           try {
-            parameters = zodToJsonSchema2(tool.inputSchema);
+            const converted = zodToJsonSchema2(tool.inputSchema);
+            parameters = isPlainObject(converted) && ("type" in converted || "properties" in converted) ? converted : { type: "object", properties: {} };
           } catch (error) {
             logError(error);
             debug.warn("RESPONSES_API_TOOL_SCHEMA_CONVERSION_FAILED", {
@@ -555,10 +742,11 @@ var ResponsesAPIAdapter = class extends OpenAIAdapter {
         type: "function",
         name: tool.name,
         description: getToolDescription(tool),
-        parameters: parameters || { type: "object", properties: {} }
+        parameters: parameters ?? { type: "object", properties: {} }
       };
     });
   }
+  // Override parseResponse to handle Response API directly without double conversion
   async parseResponse(response) {
     if (response?.body instanceof ReadableStream) {
       const { assistantMessage } = await processResponsesStream(
@@ -579,53 +767,11 @@ var ResponsesAPIAdapter = class extends OpenAIAdapter {
     }
     return this.parseNonStreamingResponse(response);
   }
+  // Implement abstract method from OpenAIAdapter
   parseNonStreamingResponse(response) {
-    let content = response.output_text || "";
-    let reasoningContent = "";
-    if (response.output && Array.isArray(response.output)) {
-      const messageItems = response.output.filter(
-        (item) => item.type === "message"
-      );
-      if (messageItems.length > 0) {
-        content = messageItems.map((item) => {
-          if (item.content && Array.isArray(item.content)) {
-            return item.content.filter((c) => c.type === "text").map((c) => c.text).join("\n");
-          }
-          return item.content || "";
-        }).filter(Boolean).join("\n\n");
-      }
-      const reasoningItems = response.output.filter(
-        (item) => item.type === "reasoning"
-      );
-      if (reasoningItems.length > 0) {
-        reasoningContent = reasoningItems.map((item) => item.content || "").filter(Boolean).join("\n\n");
-      }
-    }
-    if (reasoningContent) {
-      const thinkBlock = `
-
-${reasoningContent}
-
-`;
-      content = thinkBlock + content;
-    }
-    const toolCalls = this.parseToolCalls(response);
-    const contentArray = content ? [{ type: "text", text: content, citations: [] }] : [{ type: "text", text: "", citations: [] }];
-    const promptTokens = response.usage?.input_tokens || 0;
-    const completionTokens = response.usage?.output_tokens || 0;
-    const totalTokens = response.usage?.total_tokens ?? promptTokens + completionTokens;
-    return {
-      id: response.id || `resp_${Date.now()}`,
-      content: contentArray,
-      toolCalls,
-      usage: {
-        promptTokens,
-        completionTokens,
-        reasoningTokens: response.usage?.output_tokens_details?.reasoning_tokens
-      },
-      responseId: response.id
-    };
+    return parseNonStreamingResponse(response);
   }
+  // Implement abstract method from OpenAIAdapter - Responses API specific streaming logic
   async *processStreamingChunk(parsed, responseId, hasStarted, accumulatedContent, reasoningContext) {
     if (parsed.type === "response.reasoning_summary_part.added") {
       const partIndex = parsed.summary_index || 0;
@@ -714,14 +860,17 @@ ${reasoningContent}
     }
     return state;
   }
+  // parseStreamingResponse and parseSSEChunk are now handled by the base OpenAIAdapter class
+  // Implement abstract method for parsing streaming OpenAI responses
   async parseStreamingOpenAIResponse(response) {
-    const { processResponsesStream: processResponsesStream2 } = await import("./responsesStreaming-AW344PQO.js");
+    const { processResponsesStream: processResponsesStream2 } = await import("./responsesStreaming-JNGE2P3D.js");
     return await processResponsesStream2(
       this.parseStreamingResponse(response),
       Date.now(),
       response.id ?? `resp_${Date.now()}`
     );
   }
+  // Implement abstract method for usage normalization
   normalizeUsageForAdapter(usage) {
     const baseUsage = super.normalizeUsageForAdapter(usage);
     return {
@@ -729,153 +878,10 @@ ${reasoningContent}
       reasoningTokens: usage?.output_tokens_details?.reasoning_tokens ?? 0
     };
   }
-  convertMessagesToInput(messages) {
-    const inputItems = [];
-    for (const message of messages) {
-      const role = message.role;
-      if (role === "tool") {
-        const callId = message.tool_call_id || message.id;
-        if (typeof callId === "string" && callId) {
-          let content2 = message.content || "";
-          if (Array.isArray(content2)) {
-            const texts = [];
-            for (const part of content2) {
-              if (typeof part === "object" && part !== null) {
-                const t = part.text || part.content;
-                if (typeof t === "string" && t) {
-                  texts.push(t);
-                }
-              }
-            }
-            content2 = texts.join("\n");
-          }
-          if (typeof content2 === "string") {
-            inputItems.push({
-              type: "function_call_output",
-              call_id: callId,
-              output: content2
-            });
-          }
-        }
-        continue;
-      }
-      if (role === "assistant" && Array.isArray(message.tool_calls)) {
-        for (const tc of message.tool_calls) {
-          if (typeof tc !== "object" || tc === null) {
-            continue;
-          }
-          const tcType = tc.type || "function";
-          if (tcType !== "function") {
-            continue;
-          }
-          const callId = tc.id || tc.call_id;
-          const fn = tc.function;
-          const name = typeof fn === "object" && fn !== null ? fn.name : null;
-          const args = typeof fn === "object" && fn !== null ? fn.arguments : null;
-          if (typeof callId === "string" && typeof name === "string" && typeof args === "string") {
-            inputItems.push({
-              type: "function_call",
-              name,
-              arguments: args,
-              call_id: callId
-            });
-          }
-        }
-        continue;
-      }
-      const content = message.content || "";
-      const contentItems = [];
-      if (Array.isArray(content)) {
-        for (const part of content) {
-          if (typeof part !== "object" || part === null) continue;
-          const ptype = part.type;
-          if (ptype === "text") {
-            const text = part.text || part.content || "";
-            if (typeof text === "string" && text) {
-              const kind = role === "assistant" ? "output_text" : "input_text";
-              contentItems.push({ type: kind, text });
-            }
-          } else if (ptype === "image_url") {
-            const image = part.image_url;
-            const url = typeof image === "object" && image !== null ? image.url : image;
-            if (typeof url === "string" && url) {
-              contentItems.push({ type: "input_image", image_url: url });
-            }
-          }
-        }
-      } else if (typeof content === "string" && content) {
-        const kind = role === "assistant" ? "output_text" : "input_text";
-        contentItems.push({ type: kind, text: content });
-      }
-      if (contentItems.length) {
-        const roleOut = role === "assistant" ? "assistant" : "user";
-        inputItems.push({
-          type: "message",
-          role: roleOut,
-          content: contentItems
-        });
-      }
-    }
-    return inputItems;
-  }
-  buildInstructions(systemPrompt) {
-    const systemContent = systemPrompt.filter((content) => content.trim()).join("\n\n");
-    return systemContent;
-  }
-  parseToolCalls(response) {
-    if (!response.output || !Array.isArray(response.output)) {
-      return [];
-    }
-    const toolCalls = [];
-    for (const item of response.output) {
-      if (item.type === "function_call") {
-        const callId = item.call_id || item.id;
-        const name = item.name || "";
-        const args = item.arguments || "{}";
-        if (typeof callId === "string" && typeof name === "string" && typeof args === "string") {
-          toolCalls.push({
-            id: callId,
-            type: "function",
-            function: {
-              name,
-              arguments: args
-            }
-          });
-        }
-      } else if (item.type === "tool_call") {
-        const callId = item.id || `tool_${Math.random().toString(36).substring(2, 15)}`;
-        toolCalls.push({
-          id: callId,
-          type: "tool_call",
-          name: item.name,
-          arguments: item.arguments
-        });
-      }
-    }
-    return toolCalls;
-  }
-  applyReasoningToMessage(message, reasoningSummaryText, reasoningFullText) {
-    const rtxtParts = [];
-    if (typeof reasoningSummaryText === "string" && reasoningSummaryText.trim()) {
-      rtxtParts.push(reasoningSummaryText);
-    }
-    if (typeof reasoningFullText === "string" && reasoningFullText.trim()) {
-      rtxtParts.push(reasoningFullText);
-    }
-    const rtxt = rtxtParts.filter((p) => p).join("\n\n");
-    if (rtxt) {
-      const thinkBlock = `<think>
-${rtxt}
-</think>
-`;
-      const contentText = message.content || "";
-      message.content = thinkBlock + (typeof contentText === "string" ? contentText : "");
-    }
-    return message;
-  }
 };
 
-// src/services/ai/adapters/chatCompletions.ts
+// packages/core/src/ai/adapters/chatCompletions.ts
+import { randomUUID } from "crypto";
 import { zodToJsonSchema as zodToJsonSchema3 } from "zod-to-json-schema";
 var ChatCompletionsAdapter = class extends OpenAIAdapter {
   createRequest(params) {
@@ -924,6 +930,8 @@ var ChatCompletionsAdapter = class extends OpenAIAdapter {
       }
     }));
   }
+  // parseResponse is now handled by the base OpenAIAdapter class
+  // Implement abstract method from OpenAIAdapter - Chat Completions specific non-streaming
   parseNonStreamingResponse(response) {
     if (!response || typeof response !== "object") {
       throw new Error("Invalid response: response must be an object");
@@ -980,6 +988,7 @@ var ChatCompletionsAdapter = class extends OpenAIAdapter {
       return msg;
     });
   }
+  // Implement abstract method from OpenAIAdapter - Chat Completions specific streaming logic
   async *processStreamingChunk(parsed, responseId, hasStarted, accumulatedContent, reasoningContext) {
     if (!parsed || typeof parsed !== "object") {
       return;
@@ -1037,6 +1046,7 @@ var ChatCompletionsAdapter = class extends OpenAIAdapter {
     }
     return state;
   }
+  // Implement abstract method for parsing streaming OpenAI responses
   async parseStreamingOpenAIResponse(response, signal) {
     const contentBlocks = [];
     const usage = {
@@ -1104,7 +1114,7 @@ var ChatCompletionsAdapter = class extends OpenAIAdapter {
           },
           costUSD: 0,
           durationMs: Date.now() - Date.now(),
-          uuid: `${Date.now()}-${Math.random().toString(36).slice(2, 11)}`,
+          uuid: randomUUID(),
           responseId
         };
         return {
@@ -1147,7 +1157,8 @@ var ChatCompletionsAdapter = class extends OpenAIAdapter {
       },
       costUSD: 0,
       durationMs: Date.now() - Date.now(),
-      uuid: `${Date.now()}-${Math.random().toString(36).slice(2, 11)}`,
+      // Placeholder
+      uuid: randomUUID(),
       responseId
     };
     return {
@@ -1164,7 +1175,7 @@ var ChatCompletionsAdapter = class extends OpenAIAdapter {
   }
 };
 
-// src/constants/modelCapabilities.ts
+// packages/core/src/constants/modelCapabilities.ts
 var GPT5_CAPABILITIES = {
   apiArchitecture: {
     primary: "responses_api",
@@ -1172,6 +1183,7 @@ var GPT5_CAPABILITIES = {
   },
   parameters: {
     maxTokensField: "max_output_tokens",
+    // Responses API uses max_output_tokens
     supportsReasoningEffort: true,
     supportsVerbosity: true,
     temperatureMode: "fixed_one"
@@ -1189,6 +1201,7 @@ var GPT5_CAPABILITIES = {
   },
   streaming: {
     supported: true,
+    // Responses API supports streaming
     includesUsage: true
   }
 };
@@ -1219,18 +1232,22 @@ var CHAT_COMPLETIONS_CAPABILITIES = {
   }
 };
 var MODEL_CAPABILITIES_REGISTRY = {
+  // GPT-5 series
   "gpt-5": GPT5_CAPABILITIES,
   "gpt-5-mini": GPT5_CAPABILITIES,
   "gpt-5-nano": GPT5_CAPABILITIES,
   "gpt-5-chat-latest": GPT5_CAPABILITIES,
   "gpt-5-codex": GPT5_CAPABILITIES,
+  // GPT-4 series
   "gpt-4o": CHAT_COMPLETIONS_CAPABILITIES,
   "gpt-4o-mini": CHAT_COMPLETIONS_CAPABILITIES,
   "gpt-4-turbo": CHAT_COMPLETIONS_CAPABILITIES,
   "gpt-4": CHAT_COMPLETIONS_CAPABILITIES,
+  // Claude series (supported through conversion layer)
   "claude-3-5-sonnet-20241022": CHAT_COMPLETIONS_CAPABILITIES,
   "claude-3-5-haiku-20241022": CHAT_COMPLETIONS_CAPABILITIES,
   "claude-3-opus-20240229": CHAT_COMPLETIONS_CAPABILITIES,
+  // O1 series (special reasoning models)
   o1: {
     ...CHAT_COMPLETIONS_CAPABILITIES,
     parameters: {
@@ -1274,6 +1291,7 @@ function inferModelCapabilities(modelName) {
       toolCalling: {
         ...CHAT_COMPLETIONS_CAPABILITIES.toolCalling,
         supportsAllowedTools: false
+        // GLM might not support this
       }
     };
   }
@@ -1309,8 +1327,11 @@ function getModelCapabilities(modelName) {
   return defaultCapabilities;
 }
 
-// src/services/ai/modelAdapterFactory.ts
+// packages/core/src/ai/modelAdapterFactory.ts
 var ModelAdapterFactory = class {
+  /**
+   * Create appropriate adapter based on model configuration
+   */
   static createAdapter(modelProfile) {
     const capabilities = getModelCapabilities(modelProfile.modelName);
     const apiType = this.determineAPIType(modelProfile, capabilities);
@@ -1322,6 +1343,9 @@ var ModelAdapterFactory = class {
         return new ChatCompletionsAdapter(capabilities, modelProfile);
     }
   }
+  /**
+   * Determine which API should be used
+   */
   static determineAPIType(modelProfile, capabilities) {
     if (capabilities.apiArchitecture.primary !== "responses_api") {
       return "chat_completions";
@@ -1335,6 +1359,9 @@ var ModelAdapterFactory = class {
     }
     return capabilities.apiArchitecture.primary;
   }
+  /**
+   * Check if model should use Responses API
+   */
   static shouldUseResponsesAPI(modelProfile) {
     const capabilities = getModelCapabilities(modelProfile.modelName);
     const apiType = this.determineAPIType(modelProfile, capabilities);
@@ -1342,254 +1369,39 @@ var ModelAdapterFactory = class {
   }
 };
 
-// src/services/ai/responseStateManager.ts
-var ResponseStateManager = class {
-  conversationStates = /* @__PURE__ */ new Map();
-  CLEANUP_INTERVAL = 60 * 60 * 1e3;
-  constructor() {
-    setInterval(() => {
-      this.cleanup();
-    }, this.CLEANUP_INTERVAL);
-  }
-  setPreviousResponseId(conversationId, responseId) {
-    this.conversationStates.set(conversationId, {
-      previousResponseId: responseId,
-      lastUpdate: Date.now()
-    });
-  }
-  getPreviousResponseId(conversationId) {
-    const state = this.conversationStates.get(conversationId);
-    if (state) {
-      state.lastUpdate = Date.now();
-      return state.previousResponseId;
-    }
-    return void 0;
-  }
-  clearConversation(conversationId) {
-    this.conversationStates.delete(conversationId);
-  }
-  clearAll() {
-    this.conversationStates.clear();
-  }
-  cleanup() {
-    const now = Date.now();
-    for (const [conversationId, state] of this.conversationStates.entries()) {
-      if (now - state.lastUpdate > this.CLEANUP_INTERVAL) {
-        this.conversationStates.delete(conversationId);
-      }
-    }
-  }
-  getStateSize() {
-    return this.conversationStates.size;
-  }
-};
-var responseStateManager = new ResponseStateManager();
-function getConversationId(agentId, messageId) {
-  return agentId || messageId || `conv_${Date.now()}_${Math.random().toString(36).substr(2, 9)}`;
-}
-
-// src/services/ai/llm.ts
+// packages/core/src/ai/llm/openai/conversion.ts
 import { nanoid } from "nanoid";
 
-// src/utils/tooling/toolUsePartialJson.ts
-function tokenizePartialJson(input) {
-  let index = 0;
-  const tokens = [];
-  while (index < input.length) {
-    let ch = input[index];
-    if (ch === "\\") {
-      index++;
-      continue;
-    }
-    if (ch === "{") {
-      tokens.push({ type: "brace", value: "{" });
-      index++;
-      continue;
-    }
-    if (ch === "}") {
-      tokens.push({ type: "brace", value: "}" });
-      index++;
-      continue;
-    }
-    if (ch === "[") {
-      tokens.push({ type: "paren", value: "[" });
-      index++;
-      continue;
-    }
-    if (ch === "]") {
-      tokens.push({ type: "paren", value: "]" });
-      index++;
-      continue;
-    }
-    if (ch === ":") {
-      tokens.push({ type: "separator", value: ":" });
-      index++;
-      continue;
-    }
-    if (ch === ",") {
-      tokens.push({ type: "delimiter", value: "," });
-      index++;
-      continue;
+// packages/core/src/utils/openaiMessageConversion.ts
+function asRecord(value) {
+  if (!value || typeof value !== "object") return null;
+  return value;
+}
+function convertAnthropicMessagesToOpenAIMessages(messages) {
+  const openaiMessages = [];
+  const toolResults = {};
+  for (const message of messages) {
+    const blocks = [];
+    if (typeof message.message.content === "string") {
+      blocks.push({ type: "text", text: message.message.content });
+    } else if (Array.isArray(message.message.content)) {
+      blocks.push(...message.message.content);
+    } else if (message.message.content) {
+      blocks.push(message.message.content);
     }
-    if (ch === '"') {
-      let value = "";
-      let incomplete = false;
-      ch = input[++index];
-      while (ch !== '"') {
-        if (index === input.length) {
-          incomplete = true;
-          break;
-        }
-        if (ch === "\\") {
-          if (++index === input.length) {
-            incomplete = true;
-            break;
-          }
-          value += ch + input[index];
-          ch = input[++index];
-        } else {
-          value += ch;
-          ch = input[++index];
-        }
-      }
-      ch = input[++index];
-      if (!incomplete) tokens.push({ type: "string", value });
-      continue;
-    }
-    if (ch && /\s/.test(ch)) {
-      index++;
-      continue;
-    }
-    const digit = /[0-9]/;
-    if (ch && digit.test(ch) || ch === "-" || ch === ".") {
-      let value = "";
-      if (ch === "-") {
-        value += ch;
-        ch = input[++index];
-      }
-      while (ch && digit.test(ch) || ch === ".") {
-        value += ch;
-        ch = input[++index];
-      }
-      tokens.push({ type: "number", value });
-      continue;
-    }
-    const alpha = /[a-z]/i;
-    if (ch && alpha.test(ch)) {
-      let value = "";
-      while (ch && alpha.test(ch)) {
-        if (index === input.length) break;
-        value += ch;
-        ch = input[++index];
-      }
-      if (value === "true" || value === "false" || value === "null") {
-        tokens.push({ type: "name", value });
-      } else {
-        index++;
-      }
-      continue;
-    }
-    index++;
-  }
-  return tokens;
-}
-function trimTrailingIncompleteTokens(tokens) {
-  if (tokens.length === 0) return tokens;
-  const last = tokens[tokens.length - 1];
-  if (last.type === "separator") {
-    return trimTrailingIncompleteTokens(tokens.slice(0, -1));
-  }
-  if (last.type === "number") {
-    const lastChar = last.value[last.value.length - 1];
-    if (lastChar === "." || lastChar === "-") {
-      return trimTrailingIncompleteTokens(tokens.slice(0, -1));
-    }
-  }
-  if (last.type === "string" || last.type === "number") {
-    const previous = tokens[tokens.length - 2];
-    if (previous?.type === "delimiter") {
-      return trimTrailingIncompleteTokens(tokens.slice(0, -1));
-    }
-    if (previous?.type === "brace" && previous.value === "{") {
-      return trimTrailingIncompleteTokens(tokens.slice(0, -1));
-    }
-  }
-  if (last.type === "delimiter") {
-    return trimTrailingIncompleteTokens(tokens.slice(0, -1));
-  }
-  return tokens;
-}
-function closeOpenBrackets(tokens) {
-  const missingClosers = [];
-  for (const token of tokens) {
-    if (token.type === "brace") {
-      if (token.value === "{") missingClosers.push("}");
-      else missingClosers.splice(missingClosers.lastIndexOf("}"), 1);
-      continue;
-    }
-    if (token.type === "paren") {
-      if (token.value === "[") missingClosers.push("]");
-      else missingClosers.splice(missingClosers.lastIndexOf("]"), 1);
-    }
-  }
-  if (missingClosers.length > 0) {
-    missingClosers.reverse();
-    for (const closer of missingClosers) {
-      if (closer === "}") tokens.push({ type: "brace", value: "}" });
-      else tokens.push({ type: "paren", value: "]" });
-    }
-  }
-  return tokens;
-}
-function tokensToJson(tokens) {
-  let out = "";
-  for (const token of tokens) {
-    if (token.type === "string") out += `"${token.value}"`;
-    else out += token.value;
-  }
-  return out;
-}
-function parseToolUsePartialJson(input) {
-  const tokens = tokenizePartialJson(input);
-  const trimmed = trimTrailingIncompleteTokens(tokens);
-  const completed = closeOpenBrackets(trimmed);
-  return JSON.parse(tokensToJson(completed));
-}
-function parseToolUsePartialJsonOrThrow(input) {
-  try {
-    return parseToolUsePartialJson(input);
-  } catch (error) {
-    throw new Error(
-      `Unable to parse tool parameter JSON from model. Please retry your request or adjust your prompt. Error: ${String(error)}. JSON: ${input}`
-    );
-  }
-}
-
-// src/utils/model/openaiMessageConversion.ts
-function convertAnthropicMessagesToOpenAIMessages(messages) {
-  const openaiMessages = [];
-  const toolResults = {};
-  for (const message of messages) {
-    const blocks = [];
-    if (typeof message.message.content === "string") {
-      blocks.push({ type: "text", text: message.message.content });
-    } else if (Array.isArray(message.message.content)) {
-      blocks.push(...message.message.content);
-    } else if (message.message.content) {
-      blocks.push(message.message.content);
-    }
-    const role = message.message.role;
-    const userContentParts = [];
-    const assistantTextParts = [];
-    const assistantToolCalls = [];
-    for (const block of blocks) {
-      if (block.type === "text") {
-        const text = typeof block.text === "string" ? block.text : "";
-        if (!text) continue;
-        if (role === "user") {
-          userContentParts.push({ type: "text", text });
-        } else if (role === "assistant") {
-          assistantTextParts.push(text);
+    const role = message.message.role;
+    const userContentParts = [];
+    const assistantTextParts = [];
+    const assistantToolCalls = [];
+    for (const block of blocks) {
+      if (block.type === "text") {
+        const record = asRecord(block);
+        const text = record && typeof record.text === "string" ? record.text : "";
+        if (!text) continue;
+        if (role === "user") {
+          userContentParts.push({ type: "text", text });
+        } else if (role === "assistant") {
+          assistantTextParts.push(text);
         }
         continue;
       }
@@ -1640,7 +1452,10 @@ function convertAnthropicMessagesToOpenAIMessages(messages) {
           content: userContentParts[0].text
         });
       } else if (userContentParts.length > 0) {
-        openaiMessages.push({ role: "user", content: userContentParts });
+        openaiMessages.push({
+          role: "user",
+          content: userContentParts
+        });
       }
       continue;
     }
@@ -1655,14 +1470,17 @@ function convertAnthropicMessagesToOpenAIMessages(messages) {
         continue;
       }
       if (text) {
-        openaiMessages.push({ role: "assistant", content: text });
+        openaiMessages.push({
+          role: "assistant",
+          content: text
+        });
       }
     }
   }
   const finalMessages = [];
   for (const message of openaiMessages) {
     finalMessages.push(message);
-    if ("tool_calls" in message && message.tool_calls) {
+    if (message.role === "assistant" && Array.isArray(message.tool_calls)) {
       for (const toolCall of message.tool_calls) {
         if (toolResults[toolCall.id]) {
           finalMessages.push(toolResults[toolCall.id]);
@@ -1673,216 +1491,165 @@ function convertAnthropicMessagesToOpenAIMessages(messages) {
   return finalMessages;
 }
 
-// src/services/ai/llm.ts
-function isGPT5Model(modelName) {
-  return modelName.startsWith("gpt-5");
-}
-var PROMPT_CACHING_ENABLED = !process.env.DISABLE_PROMPT_CACHING;
-var SONNET_COST_PER_MILLION_INPUT_TOKENS = 3;
-var SONNET_COST_PER_MILLION_OUTPUT_TOKENS = 15;
-var SONNET_COST_PER_MILLION_PROMPT_CACHE_WRITE_TOKENS = 3.75;
-var SONNET_COST_PER_MILLION_PROMPT_CACHE_READ_TOKENS = 0.3;
-var MAX_RETRIES = process.env.USER_TYPE === "SWE_BENCH" ? 100 : 10;
-var BASE_DELAY_MS = 500;
-function abortableDelay(delayMs, signal) {
-  return new Promise((resolve, reject) => {
-    if (signal?.aborted) {
-      reject(new Error("Request was aborted"));
-      return;
-    }
-    const timeoutId = setTimeout(() => {
-      resolve();
-    }, delayMs);
-    if (signal) {
-      const abortHandler = () => {
-        clearTimeout(timeoutId);
-        reject(new Error("Request was aborted"));
-      };
-      signal.addEventListener("abort", abortHandler, { once: true });
-    }
-  });
+// packages/core/src/ai/llm/openai/usage.ts
+function getMaxTokensFromProfile(modelProfile) {
+  return modelProfile?.maxTokens || 8e3;
 }
-function getRetryDelay(attempt, retryAfterHeader) {
-  if (retryAfterHeader) {
-    const seconds = parseInt(retryAfterHeader, 10);
-    if (!isNaN(seconds)) {
-      return seconds * 1e3;
-    }
+function normalizeUsage(usage) {
+  if (!usage) {
+    return {
+      input_tokens: 0,
+      output_tokens: 0,
+      cache_read_input_tokens: 0,
+      cache_creation_input_tokens: 0
+    };
   }
-  return Math.min(BASE_DELAY_MS * Math.pow(2, attempt - 1), 32e3);
+  return {
+    input_tokens: usage.input_tokens ?? usage.prompt_tokens ?? usage.promptTokens ?? usage.inputTokens ?? 0,
+    output_tokens: usage.output_tokens ?? usage.completion_tokens ?? usage.completionTokens ?? usage.outputTokens ?? 0,
+    cache_read_input_tokens: usage.cache_read_input_tokens ?? 0,
+    cache_creation_input_tokens: usage.cache_creation_input_tokens ?? 0,
+    prompt_tokens: usage.prompt_tokens ?? usage.input_tokens ?? usage.promptTokens ?? 0,
+    completion_tokens: usage.completion_tokens ?? usage.output_tokens ?? usage.completionTokens ?? 0,
+    promptTokens: usage.promptTokens ?? usage.prompt_tokens ?? usage.input_tokens ?? 0,
+    completionTokens: usage.completionTokens ?? usage.completion_tokens ?? usage.output_tokens ?? 0,
+    totalTokens: usage.totalTokens ?? (usage.prompt_tokens ?? usage.input_tokens ?? usage.promptTokens ?? 0) + (usage.completion_tokens ?? usage.output_tokens ?? usage.completionTokens ?? 0),
+    reasoningTokens: usage.reasoningTokens
+  };
 }
-function shouldRetry(error) {
-  if (error.message?.includes('"type":"overloaded_error"')) {
-    return process.env.USER_TYPE === "SWE_BENCH";
-  }
-  const shouldRetryHeader = error.headers?.["x-should-retry"];
-  if (shouldRetryHeader === "true") return true;
-  if (shouldRetryHeader === "false") return false;
-  if (error instanceof APIConnectionError) {
-    return true;
+
+// packages/core/src/ai/llm/openai/conversion.ts
+function mapFinishReasonToStopReason(reason) {
+  switch (reason) {
+    case "stop":
+      return "end_turn";
+    case "length":
+      return "max_tokens";
+    case "tool_calls":
+    case "function_call":
+      return "tool_use";
+    default:
+      return null;
   }
-  if (!error.status) return false;
-  if (error.status === 408) return true;
-  if (error.status === 409) return true;
-  if (error.status === 429) return true;
-  if (error.status && error.status >= 500) return true;
-  return false;
 }
-async function withRetry(operation, options = {}) {
-  const maxRetries = options.maxRetries ?? MAX_RETRIES;
-  let lastError;
-  for (let attempt = 1; attempt <= maxRetries + 1; attempt++) {
-    try {
-      return await operation(attempt);
-    } catch (error) {
-      lastError = error;
-      if (attempt > maxRetries || !(error instanceof APIError) || !shouldRetry(error)) {
-        throw error;
-      }
-      if (options.signal?.aborted) {
-        throw new Error("Request cancelled by user");
+function convertAnthropicMessagesToOpenAIMessages2(messages) {
+  return convertAnthropicMessagesToOpenAIMessages(messages);
+}
+function convertOpenAIResponseToAnthropic(response, tools) {
+  const normalizedUsage = normalizeUsage(response.usage);
+  let contentBlocks = [];
+  const message = response.choices?.[0]?.message;
+  if (!message) {
+    return {
+      id: nanoid(),
+      model: response.model ?? "<openai>",
+      role: "assistant",
+      content: [],
+      stop_reason: mapFinishReasonToStopReason(
+        response.choices?.[0]?.finish_reason
+      ),
+      stop_sequence: null,
+      type: "message",
+      usage: {
+        input_tokens: normalizedUsage.input_tokens ?? 0,
+        output_tokens: normalizedUsage.output_tokens ?? 0,
+        cache_creation_input_tokens: normalizedUsage.cache_creation_input_tokens ?? 0,
+        cache_read_input_tokens: normalizedUsage.cache_read_input_tokens ?? 0
       }
-      const retryAfter = error.headers?.["retry-after"] ?? null;
-      const delayMs = getRetryDelay(attempt, retryAfter);
-      debug.warn("LLM_API_RETRY", {
-        name: error.name,
-        message: error.message,
-        status: error.status,
-        attempt,
-        maxRetries,
-        delayMs
-      });
+    };
+  }
+  if (message?.tool_calls) {
+    for (const toolCall of message.tool_calls) {
+      const tool = toolCall.function;
+      const toolName = tool.name;
+      let toolArgs = {};
       try {
-        await abortableDelay(delayMs, options.signal);
-      } catch (delayError) {
-        if (delayError.message === "Request was aborted") {
-          throw new Error("Request cancelled by user");
-        }
-        throw delayError;
+        toolArgs = tool.arguments ? JSON.parse(tool.arguments) : {};
+      } catch (e) {
       }
+      contentBlocks.push({
+        type: "tool_use",
+        input: toolArgs,
+        name: toolName,
+        id: toolCall.id?.length > 0 ? toolCall.id : nanoid()
+      });
     }
   }
-  throw lastError;
-}
-async function fetchAnthropicModels(baseURL, apiKey) {
-  try {
-    const modelsURL = baseURL ? `${baseURL.replace(/\/+$/, "")}/v1/models` : "https://api.anthropic.com/v1/models";
-    const response = await fetch(modelsURL, {
-      method: "GET",
-      headers: {
-        "x-api-key": apiKey,
-        "anthropic-version": "2023-06-01",
-        "User-Agent": USER_AGENT
-      }
-    });
-    if (!response.ok) {
-      if (response.status === 401) {
-        throw new Error(
-          "Invalid API key. Please check your Anthropic API key and try again."
-        );
-      } else if (response.status === 403) {
-        throw new Error(
-          "API key does not have permission to access models. Please check your API key permissions."
-        );
-      } else if (response.status === 429) {
-        throw new Error(
-          "Too many requests. Please wait a moment and try again."
-        );
-      } else if (response.status >= 500) {
-        throw new Error(
-          "Anthropic service is temporarily unavailable. Please try again later."
-        );
-      } else {
-        throw new Error(
-          `Unable to connect to Anthropic API (${response.status}). Please check your internet connection and API key.`
-        );
-      }
-    }
-    const data = await response.json();
-    return data.data || [];
-  } catch (error) {
-    if (error instanceof Error && error.message.includes("API key") || error instanceof Error && error.message.includes("Anthropic")) {
-      throw error;
-    }
-    logError(error);
-    debug.warn("ANTHROPIC_MODELS_FETCH_FAILED", {
-      error: error instanceof Error ? error.message : String(error)
+  const record = message;
+  if (typeof record.reasoning === "string" && record.reasoning) {
+    contentBlocks.push({
+      type: "thinking",
+      thinking: record.reasoning,
+      signature: ""
     });
-    throw new Error(
-      "Unable to connect to Anthropic API. Please check your internet connection and try again."
-    );
   }
-}
-async function verifyApiKey(apiKey, baseURL, provider) {
-  if (!apiKey) {
-    return false;
+  if (typeof record.reasoning_content === "string" && record.reasoning_content) {
+    contentBlocks.push({
+      type: "thinking",
+      thinking: record.reasoning_content,
+      signature: ""
+    });
   }
-  if (provider && provider !== "anthropic") {
-    try {
-      const headers = {
-        Authorization: `Bearer ${apiKey}`,
-        "Content-Type": "application/json"
-      };
-      if (!baseURL) {
-        debug.warn("API_VERIFICATION_MISSING_BASE_URL", { provider });
-        return false;
-      }
-      const modelsURL = `${baseURL.replace(/\/+$/, "")}/models`;
-      const response = await fetch(modelsURL, {
-        method: "GET",
-        headers
-      });
-      return response.ok;
-    } catch (error) {
-      logError(error);
-      debug.warn("API_VERIFICATION_FAILED", {
-        provider,
-        error: error instanceof Error ? error.message : String(error)
-      });
-      return false;
-    }
+  if (message.content) {
+    contentBlocks.push({
+      type: "text",
+      text: message.content,
+      citations: []
+    });
   }
-  const clientConfig = {
-    apiKey,
-    dangerouslyAllowBrowser: true,
-    maxRetries: 3,
-    defaultHeaders: {
-      "User-Agent": USER_AGENT
+  const finalMessage = {
+    id: nanoid(),
+    model: response.model ?? "<openai>",
+    role: "assistant",
+    content: contentBlocks,
+    stop_reason: mapFinishReasonToStopReason(
+      response.choices?.[0]?.finish_reason
+    ),
+    stop_sequence: null,
+    type: "message",
+    usage: {
+      input_tokens: normalizedUsage.input_tokens ?? 0,
+      output_tokens: normalizedUsage.output_tokens ?? 0,
+      cache_creation_input_tokens: normalizedUsage.cache_creation_input_tokens ?? 0,
+      cache_read_input_tokens: normalizedUsage.cache_read_input_tokens ?? 0
     }
   };
-  if (baseURL && (provider === "anthropic" || provider === "minimax-coding")) {
-    clientConfig.baseURL = baseURL;
+  return finalMessage;
+}
+
+// packages/core/src/ai/llm/openai/params.ts
+function isGPT5Model(modelName) {
+  return modelName.startsWith("gpt-5");
+}
+function buildOpenAIChatCompletionCreateParams(args) {
+  const isGPT5 = isGPT5Model(args.model);
+  const opts = {
+    model: args.model,
+    ...isGPT5 ? { max_completion_tokens: args.maxTokens } : { max_tokens: args.maxTokens },
+    messages: args.messages,
+    temperature: args.temperature
+  };
+  if (args.stopSequences && args.stopSequences.length > 0) {
+    opts.stop = args.stopSequences;
+  }
+  if (args.stream) {
+    ;
+    opts.stream = true;
+    opts.stream_options = {
+      include_usage: true
+    };
   }
-  const anthropic = new Anthropic(clientConfig);
-  try {
-    await withRetry(
-      async () => {
-        const model = "claude-sonnet-4-20250514";
-        const messages = [{ role: "user", content: "test" }];
-        await anthropic.messages.create({
-          model,
-          max_tokens: 1e3,
-          messages,
-          temperature: 0
-        });
-        return true;
-      },
-      { maxRetries: 2 }
-    );
-    return true;
-  } catch (error) {
-    logError(error);
-    if (error instanceof Error && error.message.includes(
-      '{"type":"error","error":{"type":"authentication_error","message":"invalid x-api-key"}}'
-    )) {
-      return false;
-    }
-    throw error;
+  if (args.toolSchemas.length > 0) {
+    opts.tools = args.toolSchemas;
+    opts.tool_choice = "auto";
   }
+  if (args.reasoningEffort) {
+    opts.reasoning_effort = args.reasoningEffort;
+  }
+  return opts;
 }
-function convertAnthropicMessagesToOpenAIMessages2(messages) {
-  return convertAnthropicMessagesToOpenAIMessages(messages);
-}
+
+// packages/core/src/ai/llm/openai/stream.ts
 function messageReducer(previous, item) {
   const reduce = (acc, delta) => {
     acc = { ...acc };
@@ -2016,20 +1783,14 @@ async function handleMessageStream(stream, signal) {
     usage
   };
 }
-function convertOpenAIResponseToAnthropic(response, tools) {
-  let contentBlocks = [];
-  const message = response.choices?.[0]?.message;
-  if (!message) {
-    return {
-      role: "assistant",
-      content: [],
-      stop_reason: response.choices?.[0]?.finish_reason,
-      type: "message",
-      usage: response.usage
-    };
-  }
-  if (message?.tool_calls) {
-    for (const toolCall of message.tool_calls) {
+
+// packages/core/src/ai/llm/openai/unifiedResponse.ts
+import { nanoid as nanoid2 } from "nanoid";
+import { randomUUID as randomUUID2 } from "crypto";
+function buildAssistantMessageFromUnifiedResponse(unifiedResponse, startTime) {
+  const contentBlocks = [...unifiedResponse.content || []];
+  if (unifiedResponse.toolCalls && unifiedResponse.toolCalls.length > 0) {
+    for (const toolCall of unifiedResponse.toolCalls) {
       const tool = toolCall.function;
       const toolName = tool?.name;
       let toolArgs = {};
@@ -2041,73 +1802,333 @@ function convertOpenAIResponseToAnthropic(response, tools) {
         type: "tool_use",
         input: toolArgs,
         name: toolName,
-        id: toolCall.id?.length > 0 ? toolCall.id : nanoid()
+        id: toolCall.id?.length > 0 ? toolCall.id : nanoid2()
       });
     }
   }
-  if (message.reasoning) {
-    contentBlocks.push({
-      type: "thinking",
-      thinking: message.reasoning,
-      signature: ""
-    });
-  }
-  if (message.reasoning_content) {
-    contentBlocks.push({
-      type: "thinking",
-      thinking: message.reasoning_content,
-      signature: ""
-    });
-  }
-  if (message.content) {
-    contentBlocks.push({
-      type: "text",
-      text: message?.content,
-      citations: []
-    });
-  }
-  const finalMessage = {
-    role: "assistant",
-    content: contentBlocks,
-    stop_reason: response.choices?.[0]?.finish_reason,
-    type: "message",
-    usage: response.usage
+  const inputTokens = unifiedResponse.usage?.promptTokens ?? unifiedResponse.usage?.input_tokens ?? 0;
+  const outputTokens = unifiedResponse.usage?.completionTokens ?? unifiedResponse.usage?.output_tokens ?? 0;
+  return {
+    type: "assistant",
+    message: {
+      id: unifiedResponse.responseId ?? nanoid2(),
+      model: unifiedResponse.model ?? "",
+      role: "assistant",
+      type: "message",
+      stop_reason: unifiedResponse.stopReason ?? null,
+      stop_sequence: null,
+      content: contentBlocks,
+      usage: {
+        input_tokens: inputTokens,
+        output_tokens: outputTokens,
+        cache_creation_input_tokens: 0,
+        cache_read_input_tokens: 0
+      }
+    },
+    costUSD: 0,
+    durationMs: Date.now() - startTime,
+    uuid: randomUUID2(),
+    responseId: unifiedResponse.responseId
   };
-  return finalMessage;
 }
-var anthropicClient = null;
-function getAnthropicClient(model) {
+
+// packages/core/src/ai/llm/openai/queryOpenAI.ts
+var SONNET_COST_PER_MILLION_INPUT_TOKENS = 3;
+var SONNET_COST_PER_MILLION_OUTPUT_TOKENS = 15;
+var SONNET_COST_PER_MILLION_PROMPT_CACHE_WRITE_TOKENS = 3.75;
+var SONNET_COST_PER_MILLION_PROMPT_CACHE_READ_TOKENS = 0.3;
+async function queryOpenAI(messages, systemPrompt, maxThinkingTokens, tools, signal, options) {
   const config = getGlobalConfig();
-  const provider = config.primaryProvider;
-  if (anthropicClient && provider) {
-    anthropicClient = null;
-  }
-  if (anthropicClient) {
-    return anthropicClient;
-  }
-  const region = getVertexRegionForModel(model);
   const modelManager = getModelManager();
-  const modelProfile = modelManager.getModel("main");
-  const defaultHeaders = {
-    "x-app": "cli",
-    "User-Agent": USER_AGENT
-  };
-  if (process.env.ANTHROPIC_AUTH_TOKEN) {
-    defaultHeaders["Authorization"] = `Bearer ${process.env.ANTHROPIC_AUTH_TOKEN}`;
+  const toolUseContext = options?.toolUseContext;
+  const modelProfile = options?.modelProfile || modelManager.getModel("main");
+  let model;
+  const currentRequest = getCurrentRequest();
+  debug.api("MODEL_CONFIG_OPENAI", {
+    modelProfileFound: !!modelProfile,
+    modelProfileId: modelProfile?.modelName,
+    modelProfileName: modelProfile?.name,
+    modelProfileModelName: modelProfile?.modelName,
+    modelProfileProvider: modelProfile?.provider,
+    modelProfileBaseURL: modelProfile?.baseURL,
+    modelProfileApiKeyExists: !!modelProfile?.apiKey,
+    optionsModel: options?.model,
+    requestId: getCurrentRequest()?.id
+  });
+  if (modelProfile) {
+    model = modelProfile.modelName;
+  } else {
+    model = options?.model || modelProfile?.modelName || "";
   }
-  const ARGS = {
-    defaultHeaders,
-    maxRetries: 0,
-    timeout: parseInt(process.env.API_TIMEOUT_MS || String(60 * 1e3), 10)
-  };
-  if (USE_BEDROCK) {
-    const client = new AnthropicBedrock(ARGS);
-    anthropicClient = client;
-    return client;
+  if (options?.prependCLISysprompt) {
+    const [firstSyspromptBlock] = splitSysPromptPrefix(systemPrompt);
+    systemPrompt = [getCLISyspromptPrefix() + systemPrompt];
   }
-  if (USE_VERTEX) {
-    const vertexArgs = {
-      ...ARGS,
+  const system = splitSysPromptPrefix(systemPrompt).map(
+    (_) => ({
+      ...PROMPT_CACHING_ENABLED ? { cache_control: { type: "ephemeral" } } : {},
+      text: _,
+      type: "text"
+    })
+  );
+  const toolSchemas = await Promise.all(
+    tools.map(
+      async (_) => ({
+        type: "function",
+        function: {
+          name: _.name,
+          description: await _.prompt({
+            safeMode: options?.safeMode
+          }),
+          // Use tool's JSON schema directly if provided, otherwise convert Zod schema
+          parameters: "inputJSONSchema" in _ && _.inputJSONSchema ? _.inputJSONSchema : zodToJsonSchema4(_.inputSchema)
+        }
+      })
+    )
+  );
+  const openaiSystem = system.map(
+    (s) => ({
+      role: "system",
+      content: s.text
+    })
+  );
+  const openaiMessages = convertAnthropicMessagesToOpenAIMessages2(messages);
+  logSystemPromptConstruction({
+    basePrompt: systemPrompt.join("\n"),
+    kodeContext: generateKodeContext() || "",
+    reminders: [],
+    // 这里可以从 generateSystemReminders 获取
+    finalPrompt: systemPrompt.join("\n")
+  });
+  let start = Date.now();
+  let adapterContext = null;
+  if (modelProfile && modelProfile.modelName) {
+    debug.api("CHECKING_ADAPTER_SYSTEM", {
+      modelProfileName: modelProfile.modelName,
+      modelName: modelProfile.modelName,
+      provider: modelProfile.provider,
+      requestId: getCurrentRequest()?.id
+    });
+    const USE_NEW_ADAPTER_SYSTEM = process.env.USE_NEW_ADAPTERS !== "false";
+    if (USE_NEW_ADAPTER_SYSTEM) {
+      const shouldUseResponses = ModelAdapterFactory.shouldUseResponsesAPI(modelProfile);
+      if (shouldUseResponses) {
+        const adapter = ModelAdapterFactory.createAdapter(modelProfile);
+        const reasoningEffort = await getReasoningEffort(modelProfile, messages);
+        let verbosity = "medium";
+        const modelNameLower = modelProfile.modelName.toLowerCase();
+        if (modelNameLower.includes("high")) {
+          verbosity = "high";
+        } else if (modelNameLower.includes("low")) {
+          verbosity = "low";
+        }
+        const unifiedParams = {
+          messages: openaiMessages,
+          systemPrompt: openaiSystem.map((s) => s.content),
+          tools,
+          maxTokens: options?.maxTokens ?? getMaxTokensFromProfile(modelProfile),
+          stream: config.stream,
+          reasoningEffort: reasoningEffort ?? void 0,
+          temperature: options?.temperature ?? (isGPT5Model(model) ? 1 : MAIN_QUERY_TEMPERATURE),
+          previousResponseId: toolUseContext?.responseState?.previousResponseId,
+          verbosity,
+          ...options?.stopSequences && options.stopSequences.length > 0 ? { stopSequences: options.stopSequences } : {}
+        };
+        adapterContext = {
+          adapter,
+          request: adapter.createRequest(unifiedParams),
+          shouldUseResponses: true
+        };
+      }
+    }
+  }
+  let queryResult;
+  let startIncludingRetries = Date.now();
+  try {
+    queryResult = await withRetry(
+      async () => {
+        start = Date.now();
+        if (adapterContext) {
+          if (adapterContext.shouldUseResponses) {
+            const { callGPT5ResponsesAPI } = await import("./openai-XXK3YZG4.js");
+            const response = await callGPT5ResponsesAPI(
+              modelProfile,
+              adapterContext.request,
+              signal
+            );
+            const unifiedResponse = await adapterContext.adapter.parseResponse(response);
+            const assistantMessage2 = buildAssistantMessageFromUnifiedResponse(
+              unifiedResponse,
+              start
+            );
+            assistantMessage2.message.usage = normalizeUsage(
+              assistantMessage2.message.usage
+            );
+            return {
+              assistantMessage: assistantMessage2,
+              rawResponse: unifiedResponse,
+              apiFormat: "openai"
+            };
+          }
+          const s2 = await getCompletionWithProfile(
+            modelProfile,
+            adapterContext.request,
+            0,
+            10,
+            signal
+          );
+          let finalResponse2;
+          if (config.stream) {
+            finalResponse2 = await handleMessageStream(
+              s2,
+              signal
+            );
+          } else {
+            finalResponse2 = s2;
+          }
+          const message2 = convertOpenAIResponseToAnthropic(finalResponse2, tools);
+          const assistantMsg2 = {
+            type: "assistant",
+            message: message2,
+            costUSD: 0,
+            durationMs: Date.now() - start,
+            uuid: randomUUID3()
+          };
+          return {
+            assistantMessage: assistantMsg2,
+            rawResponse: finalResponse2,
+            apiFormat: "openai"
+          };
+        }
+        const maxTokens = options?.maxTokens ?? getMaxTokensFromProfile(modelProfile);
+        const opts = buildOpenAIChatCompletionCreateParams({
+          model,
+          maxTokens,
+          messages: [...openaiSystem, ...openaiMessages],
+          temperature: options?.temperature ?? (isGPT5Model(model) ? 1 : MAIN_QUERY_TEMPERATURE),
+          stream: config.stream,
+          toolSchemas,
+          stopSequences: options?.stopSequences,
+          reasoningEffort: await getReasoningEffort(modelProfile, messages)
+        });
+        const completionFunction = isGPT5Model(modelProfile?.modelName || "") ? getGPT5CompletionWithProfile : getCompletionWithProfile;
+        const s = await completionFunction(modelProfile, opts, 0, 10, signal);
+        let finalResponse;
+        if (opts.stream) {
+          finalResponse = await handleMessageStream(
+            s,
+            signal
+          );
+        } else {
+          finalResponse = s;
+        }
+        const message = convertOpenAIResponseToAnthropic(finalResponse, tools);
+        const assistantMsg = {
+          type: "assistant",
+          message,
+          costUSD: 0,
+          durationMs: Date.now() - start,
+          uuid: randomUUID3()
+        };
+        return {
+          assistantMessage: assistantMsg,
+          rawResponse: finalResponse,
+          apiFormat: "openai"
+        };
+      },
+      { signal }
+    );
+  } catch (error) {
+    logError(error);
+    return getAssistantMessageFromError(error);
+  }
+  const durationMs = Date.now() - start;
+  const durationMsIncludingRetries = Date.now() - startIncludingRetries;
+  const assistantMessage = queryResult.assistantMessage;
+  assistantMessage.message.content = normalizeContentFromAPI(
+    assistantMessage.message.content || []
+  );
+  const normalizedUsage = normalizeUsage(assistantMessage.message.usage);
+  assistantMessage.message.usage = normalizedUsage;
+  const inputTokens = normalizedUsage.input_tokens ?? 0;
+  const outputTokens = normalizedUsage.output_tokens ?? 0;
+  const cacheReadInputTokens = normalizedUsage.cache_read_input_tokens ?? 0;
+  const cacheCreationInputTokens = normalizedUsage.cache_creation_input_tokens ?? 0;
+  const costUSD = inputTokens / 1e6 * SONNET_COST_PER_MILLION_INPUT_TOKENS + outputTokens / 1e6 * SONNET_COST_PER_MILLION_OUTPUT_TOKENS + cacheReadInputTokens / 1e6 * SONNET_COST_PER_MILLION_PROMPT_CACHE_READ_TOKENS + cacheCreationInputTokens / 1e6 * SONNET_COST_PER_MILLION_PROMPT_CACHE_WRITE_TOKENS;
+  addToTotalCost(costUSD, durationMsIncludingRetries);
+  logLLMInteraction({
+    systemPrompt: systemPrompt.join("\n"),
+    messages: [...openaiSystem, ...openaiMessages],
+    response: assistantMessage.message || queryResult.rawResponse,
+    usage: {
+      inputTokens,
+      outputTokens
+    },
+    timing: {
+      start,
+      end: Date.now()
+    },
+    apiFormat: queryResult.apiFormat
+  });
+  assistantMessage.costUSD = costUSD;
+  assistantMessage.durationMs = durationMs;
+  assistantMessage.uuid = assistantMessage.uuid || randomUUID3();
+  return assistantMessage;
+}
+
+// packages/core/src/ai/llm/anthropic/native.ts
+import Anthropic2 from "@anthropic-ai/sdk";
+import { nanoid as nanoid3 } from "nanoid";
+import { zodToJsonSchema as zodToJsonSchema5 } from "zod-to-json-schema";
+
+// packages/core/src/utils/http.ts
+var USER_AGENT = `${PRODUCT_COMMAND}/${MACRO.VERSION} (${process.env.USER_TYPE})`;
+
+// packages/core/src/ai/llm/maxTokens.ts
+function getMaxTokensFromProfile2(modelProfile) {
+  return modelProfile?.maxTokens || 8e3;
+}
+
+// packages/core/src/ai/llm/anthropic/client.ts
+import "@anthropic-ai/sdk/shims/node";
+import Anthropic from "@anthropic-ai/sdk";
+import { AnthropicBedrock } from "@anthropic-ai/bedrock-sdk";
+import { AnthropicVertex } from "@anthropic-ai/vertex-sdk";
+import chalk from "chalk";
+var anthropicClient = null;
+function getAnthropicClient(model) {
+  const config = getGlobalConfig();
+  const provider = config.primaryProvider;
+  if (anthropicClient && provider) {
+    anthropicClient = null;
+  }
+  if (anthropicClient) {
+    return anthropicClient;
+  }
+  const region = getVertexRegionForModel(model);
+  const modelManager = getModelManager();
+  const modelProfile = modelManager.getModel("main");
+  const defaultHeaders = {
+    "x-app": "cli",
+    "User-Agent": USER_AGENT
+  };
+  if (process.env.ANTHROPIC_AUTH_TOKEN) {
+    defaultHeaders["Authorization"] = `Bearer ${process.env.ANTHROPIC_AUTH_TOKEN}`;
+  }
+  const ARGS = {
+    defaultHeaders,
+    maxRetries: 0,
+    // Disabled auto-retry in favor of manual implementation
+    timeout: parseInt(process.env.API_TIMEOUT_MS || String(60 * 1e3), 10)
+  };
+  if (USE_BEDROCK) {
+    const client = new AnthropicBedrock(ARGS);
+    anthropicClient = client;
+    return client;
+  }
+  if (USE_VERTEX) {
+    const vertexArgs = {
+      ...ARGS,
       region: region || process.env.CLOUD_ML_REGION || "us-east5"
     };
     const client = new AnthropicVertex(vertexArgs);
@@ -2135,6 +2156,7 @@ function getAnthropicClient(model) {
     dangerouslyAllowBrowser: true,
     ...ARGS,
     ...baseURL && { baseURL }
+    // Use baseURL directly, SDK will handle API versioning
   };
   anthropicClient = new Anthropic(clientConfig);
   return anthropicClient;
@@ -2142,13 +2164,15 @@ function getAnthropicClient(model) {
 function resetAnthropicClient() {
   anthropicClient = null;
 }
+
+// packages/core/src/ai/llm/anthropic/cacheControl.ts
 function applyCacheControlWithLimits(systemBlocks, messageParams) {
   if (!PROMPT_CACHING_ENABLED) {
     return { systemBlocks, messageParams };
   }
   const maxCacheBlocks = 4;
   let usedCacheBlocks = 0;
-  const processedSystemBlocks = systemBlocks.map((block, index) => {
+  const processedSystemBlocks = systemBlocks.map((block) => {
     if (usedCacheBlocks < maxCacheBlocks && block.text.length > 1e3) {
       usedCacheBlocks++;
       return {
@@ -2163,7 +2187,9 @@ function applyCacheControlWithLimits(systemBlocks, messageParams) {
     if (Array.isArray(message.content)) {
       const processedContent = message.content.map(
         (contentBlock, blockIndex) => {
-          const shouldCache = usedCacheBlocks < maxCacheBlocks && contentBlock.type === "text" && typeof contentBlock.text === "string" && (contentBlock.text.length > 2e3 || messageIndex === messageParams.length - 1 && blockIndex === message.content.length - 1 && contentBlock.text.length > 500);
+          const shouldCache = usedCacheBlocks < maxCacheBlocks && contentBlock.type === "text" && typeof contentBlock.text === "string" && // Long documents (over 2000 characters)
+          (contentBlock.text.length > 2e3 || // Last content block of the last message (may be important context)
+          messageIndex === messageParams.length - 1 && blockIndex === message.content.length - 1 && contentBlock.text.length > 500);
           if (shouldCache) {
             usedCacheBlocks++;
             return {
@@ -2171,8 +2197,10 @@ function applyCacheControlWithLimits(systemBlocks, messageParams) {
               cache_control: { type: "ephemeral" }
             };
           }
-          const { cache_control, ...blockWithoutCache } = contentBlock;
-          return blockWithoutCache;
+          if ("cache_control" in contentBlock) {
+            return { ...contentBlock, cache_control: void 0 };
+          }
+          return contentBlock;
         }
       );
       return {
@@ -2187,6 +2215,8 @@ function applyCacheControlWithLimits(systemBlocks, messageParams) {
     messageParams: processedMessageParams
   };
 }
+
+// packages/core/src/ai/llm/anthropic/messageParams.ts
 function userMessageToMessageParam(message, addCache = false) {
   if (addCache) {
     if (typeof message.message.content === "string") {
@@ -2235,157 +2265,351 @@ function assistantMessageToMessageParam(message, addCache = false) {
     content: message.message.content
   };
 }
-function splitSysPromptPrefix(systemPrompt) {
-  const systemPromptFirstBlock = systemPrompt[0] || "";
-  const systemPromptRest = systemPrompt.slice(1);
-  return [systemPromptFirstBlock, systemPromptRest.join("\n")].filter(Boolean);
+function addCacheBreakpoints(messages) {
+  return messages.map((msg, index) => {
+    return msg.type === "user" ? userMessageToMessageParam(msg, index > messages.length - 3) : assistantMessageToMessageParam(msg, index > messages.length - 3);
+  });
 }
-async function queryLLM(messages, systemPrompt, maxThinkingTokens, tools, signal, options) {
-  const modelManager = options.__testModelManager ?? getModelManager();
-  const modelResolution = modelManager.resolveModelWithInfo(options.model);
-  if (!modelResolution.success || !modelResolution.profile) {
-    const fallbackProfile = modelManager.resolveModel(options.model);
-    if (!fallbackProfile) {
-      throw new Error(
-        modelResolution.error || `Failed to resolve model: ${options.model}`
-      );
+
+// packages/core/src/utils/toolUsePartialJson.ts
+function tokenizePartialJson(input) {
+  let index = 0;
+  const tokens = [];
+  while (index < input.length) {
+    let ch = input[index];
+    if (ch === "\\") {
+      index++;
+      continue;
     }
-    debug.warn("MODEL_RESOLUTION_FALLBACK", {
-      inputParam: options.model,
-      error: modelResolution.error,
-      fallbackModelName: fallbackProfile.modelName,
-      fallbackProvider: fallbackProfile.provider,
-      requestId: getCurrentRequest()?.id
-    });
-    modelResolution.success = true;
-    modelResolution.profile = fallbackProfile;
-  }
-  const modelProfile = modelResolution.profile;
-  const resolvedModel = modelProfile.modelName;
-  const toolUseContext = options.toolUseContext;
-  if (toolUseContext && !toolUseContext.responseState) {
-    const conversationId = getConversationId(
-      toolUseContext.agentId,
-      toolUseContext.messageId
-    );
-    const previousResponseId = responseStateManager.getPreviousResponseId(conversationId);
-    toolUseContext.responseState = {
-      previousResponseId,
-      conversationId
-    };
-  }
-  debug.api("MODEL_RESOLVED", {
-    inputParam: options.model,
-    resolvedModelName: resolvedModel,
-    provider: modelProfile.provider,
-    isPointer: ["main", "task", "compact", "quick"].includes(options.model),
-    hasResponseState: !!toolUseContext?.responseState,
-    conversationId: toolUseContext?.responseState?.conversationId,
-    requestId: getCurrentRequest()?.id
-  });
-  const currentRequest = getCurrentRequest();
-  debug.api("LLM_REQUEST_START", {
-    messageCount: messages.length,
-    systemPromptLength: systemPrompt.join(" ").length,
-    toolCount: tools.length,
-    model: resolvedModel,
-    originalModelParam: options.model,
-    requestId: getCurrentRequest()?.id
-  });
-  markPhase("LLM_CALL");
-  try {
-    const queryFn = options.__testQueryLLMWithPromptCaching ?? queryLLMWithPromptCaching;
-    const cleanOptions = { ...options };
-    delete cleanOptions.__testModelManager;
-    delete cleanOptions.__testQueryLLMWithPromptCaching;
-    const runQuery = () => queryFn(
-      messages,
-      systemPrompt,
-      maxThinkingTokens,
-      tools,
-      signal,
-      {
-        ...cleanOptions,
-        model: resolvedModel,
-        modelProfile,
-        toolUseContext
+    if (ch === "{") {
+      tokens.push({ type: "brace", value: "{" });
+      index++;
+      continue;
+    }
+    if (ch === "}") {
+      tokens.push({ type: "brace", value: "}" });
+      index++;
+      continue;
+    }
+    if (ch === "[") {
+      tokens.push({ type: "paren", value: "[" });
+      index++;
+      continue;
+    }
+    if (ch === "]") {
+      tokens.push({ type: "paren", value: "]" });
+      index++;
+      continue;
+    }
+    if (ch === ":") {
+      tokens.push({ type: "separator", value: ":" });
+      index++;
+      continue;
+    }
+    if (ch === ",") {
+      tokens.push({ type: "delimiter", value: "," });
+      index++;
+      continue;
+    }
+    if (ch === '"') {
+      let value = "";
+      let incomplete = false;
+      ch = input[++index];
+      while (ch !== '"') {
+        if (index === input.length) {
+          incomplete = true;
+          break;
+        }
+        if (ch === "\\") {
+          if (++index === input.length) {
+            incomplete = true;
+            break;
+          }
+          value += ch + input[index];
+          ch = input[++index];
+        } else {
+          value += ch;
+          ch = input[++index];
+        }
       }
-    );
-    const result = options.__testQueryLLMWithPromptCaching ? await runQuery() : await withVCR(messages, runQuery);
-    debug.api("LLM_REQUEST_SUCCESS", {
-      costUSD: result.costUSD,
-      durationMs: result.durationMs,
-      responseLength: result.message.content?.length || 0,
-      requestId: getCurrentRequest()?.id
-    });
-    if (toolUseContext?.responseState?.conversationId && result.responseId) {
-      responseStateManager.setPreviousResponseId(
-        toolUseContext.responseState.conversationId,
-        result.responseId
-      );
-      debug.api("RESPONSE_STATE_UPDATED", {
-        conversationId: toolUseContext.responseState.conversationId,
-        responseId: result.responseId,
-        requestId: getCurrentRequest()?.id
-      });
+      ch = input[++index];
+      if (!incomplete) tokens.push({ type: "string", value });
+      continue;
     }
-    return result;
-  } catch (error) {
-    logErrorWithDiagnosis(
-      error,
-      {
-        messageCount: messages.length,
-        systemPromptLength: systemPrompt.join(" ").length,
-        model: options.model,
-        toolCount: tools.length,
-        phase: "LLM_CALL"
-      },
-      currentRequest?.id
-    );
-    throw error;
+    if (ch && /\s/.test(ch)) {
+      index++;
+      continue;
+    }
+    const digit = /[0-9]/;
+    if (ch && digit.test(ch) || ch === "-" || ch === ".") {
+      let value = "";
+      if (ch === "-") {
+        value += ch;
+        ch = input[++index];
+      }
+      while (ch && digit.test(ch) || ch === ".") {
+        value += ch;
+        ch = input[++index];
+      }
+      tokens.push({ type: "number", value });
+      continue;
+    }
+    const alpha = /[a-z]/i;
+    if (ch && alpha.test(ch)) {
+      let value = "";
+      while (ch && alpha.test(ch)) {
+        if (index === input.length) break;
+        value += ch;
+        ch = input[++index];
+      }
+      if (value === "true" || value === "false" || value === "null") {
+        tokens.push({ type: "name", value });
+      } else {
+        index++;
+      }
+      continue;
+    }
+    index++;
   }
+  return tokens;
 }
-async function queryLLMWithPromptCaching(messages, systemPrompt, maxThinkingTokens, tools, signal, options) {
-  const config = getGlobalConfig();
-  const modelManager = getModelManager();
-  const toolUseContext = options.toolUseContext;
-  const modelProfile = options.modelProfile || modelManager.getModel("main");
-  let provider;
-  if (modelProfile) {
-    provider = modelProfile.provider || config.primaryProvider || "anthropic";
-  } else {
-    provider = config.primaryProvider || "anthropic";
+function trimTrailingIncompleteTokens(tokens) {
+  if (tokens.length === 0) return tokens;
+  const last = tokens[tokens.length - 1];
+  if (last.type === "separator") {
+    return trimTrailingIncompleteTokens(tokens.slice(0, -1));
   }
-  if (provider === "anthropic" || provider === "bigdream" || provider === "opendev" || provider === "minimax-coding") {
-    return queryAnthropicNative(
-      messages,
-      systemPrompt,
-      maxThinkingTokens,
-      tools,
-      signal,
-      { ...options, modelProfile, toolUseContext }
-    );
+  if (last.type === "number") {
+    const lastChar = last.value[last.value.length - 1];
+    if (lastChar === "." || lastChar === "-") {
+      return trimTrailingIncompleteTokens(tokens.slice(0, -1));
+    }
   }
-  return queryOpenAI(messages, systemPrompt, maxThinkingTokens, tools, signal, {
-    ...options,
-    modelProfile,
-    toolUseContext
-  });
+  if (last.type === "string" || last.type === "number") {
+    const previous = tokens[tokens.length - 2];
+    if (previous?.type === "delimiter") {
+      return trimTrailingIncompleteTokens(tokens.slice(0, -1));
+    }
+    if (previous?.type === "brace" && previous.value === "{") {
+      return trimTrailingIncompleteTokens(tokens.slice(0, -1));
+    }
+  }
+  if (last.type === "delimiter") {
+    return trimTrailingIncompleteTokens(tokens.slice(0, -1));
+  }
+  return tokens;
 }
-async function queryAnthropicNative(messages, systemPrompt, maxThinkingTokens, tools, signal, options) {
-  const config = getGlobalConfig();
-  const modelManager = getModelManager();
-  const toolUseContext = options?.toolUseContext;
-  const modelProfile = options?.modelProfile || modelManager.getModel("main");
-  let anthropic;
-  let model;
-  let provider;
-  debug.api("MODEL_CONFIG_ANTHROPIC", {
-    modelProfileFound: !!modelProfile,
-    modelProfileId: modelProfile?.modelName,
-    modelProfileName: modelProfile?.name,
-    modelProfileModelName: modelProfile?.modelName,
-    modelProfileProvider: modelProfile?.provider,
+function closeOpenBrackets(tokens) {
+  const missingClosers = [];
+  for (const token of tokens) {
+    if (token.type === "brace") {
+      if (token.value === "{") missingClosers.push("}");
+      else missingClosers.splice(missingClosers.lastIndexOf("}"), 1);
+      continue;
+    }
+    if (token.type === "paren") {
+      if (token.value === "[") missingClosers.push("]");
+      else missingClosers.splice(missingClosers.lastIndexOf("]"), 1);
+    }
+  }
+  if (missingClosers.length > 0) {
+    missingClosers.reverse();
+    for (const closer of missingClosers) {
+      if (closer === "}") tokens.push({ type: "brace", value: "}" });
+      else tokens.push({ type: "paren", value: "]" });
+    }
+  }
+  return tokens;
+}
+function tokensToJson(tokens) {
+  let out = "";
+  for (const token of tokens) {
+    if (token.type === "string") out += `"${token.value}"`;
+    else out += token.value;
+  }
+  return out;
+}
+function parseToolUsePartialJson(input) {
+  const tokens = tokenizePartialJson(input);
+  const trimmed = trimTrailingIncompleteTokens(tokens);
+  const completed = closeOpenBrackets(trimmed);
+  return JSON.parse(tokensToJson(completed));
+}
+function parseToolUsePartialJsonOrThrow(input) {
+  try {
+    return parseToolUsePartialJson(input);
+  } catch (error) {
+    throw new Error(
+      `Unable to parse tool parameter JSON from model. Please retry your request or adjust your prompt. Error: ${String(error)}. JSON: ${input}`
+    );
+  }
+}
+
+// packages/core/src/ai/llm/anthropic/streaming.ts
+function asRecord2(value) {
+  if (!value || typeof value !== "object") return null;
+  if (Array.isArray(value)) return null;
+  return value;
+}
+async function createAnthropicStreamingMessage(anthropic, params, signal) {
+  const stream = await anthropic.beta.messages.create(
+    {
+      ...params,
+      stream: true
+    },
+    {
+      signal
+      // ← CRITICAL: Connect the AbortSignal to API call
+    }
+  );
+  let finalResponse = null;
+  let messageStartEvent = null;
+  const contentBlocks = [];
+  const inputJSONBuffers = /* @__PURE__ */ new Map();
+  let usage = null;
+  let stopReason = null;
+  let stopSequence = null;
+  let hasMarkedStreaming = false;
+  for await (const event of stream) {
+    if (signal.aborted) {
+      debug.flow("STREAM_ABORTED", {
+        eventType: event.type,
+        timestamp: Date.now()
+      });
+      throw new Error("Request was cancelled");
+    }
+    switch (event.type) {
+      case "message_start":
+        messageStartEvent = event;
+        finalResponse = {
+          ...event.message,
+          content: []
+          // Will be populated from content blocks
+        };
+        break;
+      case "content_block_start":
+        contentBlocks[event.index] = { ...event.content_block };
+        {
+          const contentBlock = asRecord2(event.content_block);
+          const blockType = contentBlock?.type;
+          if (blockType === "tool_use" || blockType === "server_tool_use" || blockType === "mcp_tool_use") {
+            setRequestStatus({
+              kind: "tool",
+              detail: typeof contentBlock?.name === "string" ? contentBlock.name : void 0
+            });
+            inputJSONBuffers.set(event.index, "");
+          }
+        }
+        break;
+      case "content_block_delta":
+        const blockIndex = event.index;
+        if (!contentBlocks[blockIndex]) {
+          contentBlocks[blockIndex] = {
+            type: event.delta.type === "text_delta" ? "text" : "tool_use",
+            text: event.delta.type === "text_delta" ? "" : void 0
+          };
+          if (event.delta.type === "input_json_delta") {
+            inputJSONBuffers.set(blockIndex, "");
+          }
+        }
+        if (event.delta.type === "text_delta") {
+          if (!hasMarkedStreaming) {
+            setRequestStatus({ kind: "streaming" });
+            hasMarkedStreaming = true;
+          }
+          contentBlocks[blockIndex].text += event.delta.text;
+        } else if (event.delta.type === "input_json_delta") {
+          const currentBuffer = inputJSONBuffers.get(blockIndex) || "";
+          const nextBuffer = currentBuffer + event.delta.partial_json;
+          inputJSONBuffers.set(blockIndex, nextBuffer);
+          const trimmed = nextBuffer.trim();
+          if (trimmed.length === 0) {
+            contentBlocks[blockIndex].input = {};
+            break;
+          }
+          contentBlocks[blockIndex].input = parseToolUsePartialJsonOrThrow(nextBuffer) ?? {};
+        }
+        break;
+      case "message_delta":
+        if (event.delta.stop_reason) stopReason = event.delta.stop_reason;
+        if (event.delta.stop_sequence) stopSequence = event.delta.stop_sequence;
+        if (event.usage) usage = { ...usage, ...event.usage };
+        break;
+      case "content_block_stop":
+        const stopIndex = event.index;
+        const block = contentBlocks[stopIndex];
+        if ((block?.type === "tool_use" || block?.type === "server_tool_use" || block?.type === "mcp_tool_use") && inputJSONBuffers.has(stopIndex)) {
+          const jsonStr = inputJSONBuffers.get(stopIndex) ?? "";
+          if (block.input === void 0) {
+            const trimmed = jsonStr.trim();
+            if (trimmed.length === 0) {
+              block.input = {};
+            } else {
+              block.input = parseToolUsePartialJsonOrThrow(jsonStr) ?? {};
+            }
+          }
+          inputJSONBuffers.delete(stopIndex);
+        }
+        break;
+      case "message_stop":
+        inputJSONBuffers.clear();
+        break;
+    }
+    if (event.type === "message_stop") {
+      break;
+    }
+  }
+  if (!finalResponse || !messageStartEvent) {
+    throw new Error("Stream ended without proper message structure");
+  }
+  finalResponse = {
+    ...messageStartEvent.message,
+    content: contentBlocks.filter(Boolean),
+    stop_reason: stopReason,
+    stop_sequence: stopSequence,
+    usage: {
+      ...messageStartEvent.message.usage,
+      ...usage
+    }
+  };
+  return finalResponse;
+}
+
+// packages/core/src/ai/llm/anthropic/cost.ts
+function getModelInputTokenCostUSD(model) {
+  for (const providerModels of Object.values(models_default)) {
+    const modelInfo = providerModels.find((m) => m.model === model);
+    if (modelInfo) {
+      return modelInfo.input_cost_per_token || 0;
+    }
+  }
+  return 3e-6;
+}
+function getModelOutputTokenCostUSD(model) {
+  for (const providerModels of Object.values(models_default)) {
+    const modelInfo = providerModels.find((m) => m.model === model);
+    if (modelInfo) {
+      return modelInfo.output_cost_per_token || 0;
+    }
+  }
+  return 15e-6;
+}
+
+// packages/core/src/ai/llm/anthropic/native.ts
+async function queryAnthropicNative(messages, systemPrompt, maxThinkingTokens, tools, signal, options) {
+  const config = getGlobalConfig();
+  const modelManager = getModelManager();
+  const toolUseContext = options?.toolUseContext;
+  const modelProfile = options?.modelProfile || modelManager.getModel("main");
+  let anthropic;
+  let model;
+  let provider;
+  debug.api("MODEL_CONFIG_ANTHROPIC", {
+    modelProfileFound: !!modelProfile,
+    modelProfileId: modelProfile?.modelName,
+    modelProfileName: modelProfile?.name,
+    modelProfileModelName: modelProfile?.modelName,
+    modelProfileProvider: modelProfile?.provider,
     modelProfileBaseURL: modelProfile?.baseURL,
     modelProfileApiKeyExists: !!modelProfile?.apiKey,
     optionsModel: options?.model,
@@ -2408,7 +2632,7 @@ async function queryAnthropicNative(messages, systemPrompt, maxThinkingTokens, t
       if (modelProfile.baseURL) {
         clientConfig.baseURL = modelProfile.baseURL;
       }
-      anthropic = new Anthropic(clientConfig);
+      anthropic = new Anthropic2(clientConfig);
     } else {
       anthropic = getAnthropicClient(model);
     }
@@ -2439,7 +2663,7 @@ async function queryAnthropicNative(messages, systemPrompt, maxThinkingTokens, t
       async (tool) => ({
         name: tool.name,
         description: getToolDescription(tool),
-        input_schema: "inputJSONSchema" in tool && tool.inputJSONSchema ? tool.inputJSONSchema : zodToJsonSchema4(tool.inputSchema)
+        input_schema: "inputJSONSchema" in tool && tool.inputJSONSchema ? tool.inputJSONSchema : zodToJsonSchema5(tool.inputSchema)
       })
     )
   );
@@ -2450,6 +2674,7 @@ async function queryAnthropicNative(messages, systemPrompt, maxThinkingTokens, t
     basePrompt: systemPrompt.join("\n"),
     kodeContext: generateKodeContext() || "",
     reminders: [],
+    // 这里可以从 generateSystemReminders 获取
     finalPrompt: systemPrompt.join("\n")
   });
   let start = Date.now();
@@ -2462,7 +2687,7 @@ async function queryAnthropicNative(messages, systemPrompt, maxThinkingTokens, t
         start = Date.now();
         const params = {
           model,
-          max_tokens: options?.maxTokens ?? getMaxTokensFromProfile(modelProfile),
+          max_tokens: options?.maxTokens ?? getMaxTokensFromProfile2(modelProfile),
           messages: processedMessages,
           system: processedSystem,
           tools: toolSchemas.length > 0 ? toolSchemas : void 0,
@@ -2471,148 +2696,37 @@ async function queryAnthropicNative(messages, systemPrompt, maxThinkingTokens, t
           ...options?.stopSequences && options.stopSequences.length > 0 ? { stop_sequences: options.stopSequences } : {}
         };
         if (maxThinkingTokens > 0) {
-          ;
           params.extra_headers = {
             "anthropic-beta": "max-tokens-3-5-sonnet-2024-07-15"
           };
-          params.thinking = { max_tokens: maxThinkingTokens };
+          params.thinking = {
+            type: "enabled",
+            budget_tokens: maxThinkingTokens
+          };
         }
-        debug.api("ANTHROPIC_API_CALL_START_STREAMING", {
-          endpoint: modelProfile?.baseURL || "DEFAULT_ANTHROPIC",
-          model,
-          provider,
-          apiKeyConfigured: !!modelProfile?.apiKey,
-          apiKeyPrefix: modelProfile?.apiKey ? modelProfile.apiKey.substring(0, 8) : null,
-          maxTokens: params.max_tokens,
-          temperature: options?.temperature ?? MAIN_QUERY_TEMPERATURE,
-          params,
-          messageCount: params.messages?.length || 0,
-          streamMode: true,
-          toolsCount: toolSchemas.length,
-          thinkingTokens: maxThinkingTokens,
-          timestamp: (/* @__PURE__ */ new Date()).toISOString(),
-          modelProfileId: modelProfile?.modelName,
-          modelProfileName: modelProfile?.name
-        });
         if (config.stream) {
-          const stream = await anthropic.beta.messages.create(
-            {
-              ...params,
-              stream: true
-            },
-            {
-              signal
-            }
+          debug.api("ANTHROPIC_API_CALL_START_STREAMING", {
+            endpoint: modelProfile?.baseURL || "DEFAULT_ANTHROPIC",
+            model,
+            provider,
+            apiKeyConfigured: !!modelProfile?.apiKey,
+            apiKeyPrefix: modelProfile?.apiKey ? modelProfile.apiKey.substring(0, 8) : null,
+            maxTokens: params.max_tokens,
+            temperature: options?.temperature ?? MAIN_QUERY_TEMPERATURE,
+            params,
+            messageCount: params.messages?.length || 0,
+            streamMode: true,
+            toolsCount: toolSchemas.length,
+            thinkingTokens: maxThinkingTokens,
+            timestamp: (/* @__PURE__ */ new Date()).toISOString(),
+            modelProfileId: modelProfile?.modelName,
+            modelProfileName: modelProfile?.name
+          });
+          return await createAnthropicStreamingMessage(
+            anthropic,
+            params,
+            signal
           );
-          let finalResponse = null;
-          let messageStartEvent = null;
-          const contentBlocks = [];
-          const inputJSONBuffers = /* @__PURE__ */ new Map();
-          let usage = null;
-          let stopReason = null;
-          let stopSequence = null;
-          let hasMarkedStreaming = false;
-          for await (const event of stream) {
-            if (signal.aborted) {
-              debug.flow("STREAM_ABORTED", {
-                eventType: event.type,
-                timestamp: Date.now()
-              });
-              throw new Error("Request was cancelled");
-            }
-            switch (event.type) {
-              case "message_start":
-                messageStartEvent = event;
-                finalResponse = {
-                  ...event.message,
-                  content: []
-                };
-                break;
-              case "content_block_start":
-                contentBlocks[event.index] = { ...event.content_block };
-                const contentBlockType = event.content_block.type;
-                if (contentBlockType === "tool_use" || contentBlockType === "server_tool_use" || contentBlockType === "mcp_tool_use") {
-                  setRequestStatus({
-                    kind: "tool",
-                    detail: event.content_block.name
-                  });
-                  inputJSONBuffers.set(event.index, "");
-                }
-                break;
-              case "content_block_delta":
-                const blockIndex = event.index;
-                if (!contentBlocks[blockIndex]) {
-                  contentBlocks[blockIndex] = {
-                    type: event.delta.type === "text_delta" ? "text" : "tool_use",
-                    text: event.delta.type === "text_delta" ? "" : void 0
-                  };
-                  if (event.delta.type === "input_json_delta") {
-                    inputJSONBuffers.set(blockIndex, "");
-                  }
-                }
-                if (event.delta.type === "text_delta") {
-                  if (!hasMarkedStreaming) {
-                    setRequestStatus({ kind: "streaming" });
-                    hasMarkedStreaming = true;
-                  }
-                  contentBlocks[blockIndex].text += event.delta.text;
-                } else if (event.delta.type === "input_json_delta") {
-                  const currentBuffer = inputJSONBuffers.get(blockIndex) || "";
-                  const nextBuffer = currentBuffer + event.delta.partial_json;
-                  inputJSONBuffers.set(blockIndex, nextBuffer);
-                  const trimmed = nextBuffer.trim();
-                  if (trimmed.length === 0) {
-                    contentBlocks[blockIndex].input = {};
-                    break;
-                  }
-                  contentBlocks[blockIndex].input = parseToolUsePartialJsonOrThrow(nextBuffer) ?? {};
-                }
-                break;
-              case "message_delta":
-                if (event.delta.stop_reason)
-                  stopReason = event.delta.stop_reason;
-                if (event.delta.stop_sequence)
-                  stopSequence = event.delta.stop_sequence;
-                if (event.usage) usage = { ...usage, ...event.usage };
-                break;
-              case "content_block_stop":
-                const stopIndex = event.index;
-                const block = contentBlocks[stopIndex];
-                if ((block?.type === "tool_use" || block?.type === "server_tool_use" || block?.type === "mcp_tool_use") && inputJSONBuffers.has(stopIndex)) {
-                  const jsonStr = inputJSONBuffers.get(stopIndex) ?? "";
-                  if (block.input === void 0) {
-                    const trimmed = jsonStr.trim();
-                    if (trimmed.length === 0) {
-                      block.input = {};
-                    } else {
-                      block.input = parseToolUsePartialJsonOrThrow(jsonStr) ?? {};
-                    }
-                  }
-                  inputJSONBuffers.delete(stopIndex);
-                }
-                break;
-              case "message_stop":
-                inputJSONBuffers.clear();
-                break;
-            }
-            if (event.type === "message_stop") {
-              break;
-            }
-          }
-          if (!finalResponse || !messageStartEvent) {
-            throw new Error("Stream ended without proper message structure");
-          }
-          finalResponse = {
-            ...messageStartEvent.message,
-            content: contentBlocks.filter(Boolean),
-            stop_reason: stopReason,
-            stop_sequence: stopSequence,
-            usage: {
-              ...messageStartEvent.message.usage,
-              ...usage
-            }
-          };
-          return finalResponse;
         } else {
           debug.api("ANTHROPIC_API_CALL_START_NON_STREAMING", {
             endpoint: modelProfile?.baseURL || "DEFAULT_ANTHROPIC",
@@ -2632,6 +2746,7 @@ async function queryAnthropicNative(messages, systemPrompt, maxThinkingTokens, t
           });
           return await anthropic.beta.messages.create(params, {
             signal
+            // ← CRITICAL: Connect the AbortSignal to API call
           });
         }
       },
@@ -2640,7 +2755,7 @@ async function queryAnthropicNative(messages, systemPrompt, maxThinkingTokens, t
     debug.api("ANTHROPIC_API_CALL_SUCCESS", {
       content: response.content
     });
-    const ttftMs = start - Date.now();
+    const ttftMs = Date.now() - start;
     const durationMs = Date.now() - startIncludingRetries;
     const content = response.content.map((block) => {
       if (block.type === "text") {
@@ -2670,9 +2785,10 @@ async function queryAnthropicNative(messages, systemPrompt, maxThinkingTokens, t
         usage: response.usage
       },
       type: "assistant",
-      uuid: nanoid(),
+      uuid: nanoid3(),
       durationMs,
       costUSD: 0
+      // Will be calculated below
     };
     const systemMessages = system.map((block) => ({
       role: "system",
@@ -2704,374 +2820,466 @@ async function queryAnthropicNative(messages, systemPrompt, maxThinkingTokens, t
     return getAssistantMessageFromError(error);
   }
 }
-function getAssistantMessageFromError(error) {
-  if (error instanceof Error && error.message.includes("prompt is too long")) {
-    return createAssistantAPIErrorMessage(PROMPT_TOO_LONG_ERROR_MESSAGE);
-  }
-  if (error instanceof Error && error.message.includes("Your credit balance is too low")) {
-    return createAssistantAPIErrorMessage(CREDIT_BALANCE_TOO_LOW_ERROR_MESSAGE);
+
+// packages/core/src/services/vcr.ts
+import { createHash } from "crypto";
+import { mkdirSync, readFileSync, writeFileSync } from "fs";
+import { dirname } from "path";
+import { existsSync } from "fs";
+import * as path from "path";
+import { mapValues } from "lodash-es";
+async function withVCR(messages, f) {
+  if (process.env.NODE_ENV !== "test") {
+    return await f();
   }
-  if (error instanceof Error && error.message.toLowerCase().includes("x-api-key")) {
-    return createAssistantAPIErrorMessage(INVALID_API_KEY_ERROR_MESSAGE);
+  const dehydratedInput = mapMessages(
+    messages.map((_) => _.message.content),
+    dehydrateValue
+  );
+  const filename = `./fixtures/${dehydratedInput.map((_) => createHash("sha1").update(JSON.stringify(_)).digest("hex").slice(0, 6)).join("-")}.json`;
+  if (existsSync(filename)) {
+    const cached = JSON.parse(readFileSync(filename, "utf-8"));
+    return mapAssistantMessage(cached.output, hydrateValue);
   }
-  if (error instanceof Error) {
-    if (process.env.NODE_ENV === "development") {
-      debug.error("ANTHROPIC_API_ERROR", {
-        message: error.message,
-        stack: error.stack
-      });
-    }
-    return createAssistantAPIErrorMessage(
-      `${API_ERROR_MESSAGE_PREFIX}: ${error.message}`
+  if (env.isCI) {
+    process.stderr.write(
+      `Anthropic API fixture missing. Re-run bun test locally, then commit the result. ${JSON.stringify({ input: dehydratedInput }, null, 2)}
+`
     );
   }
-  return createAssistantAPIErrorMessage(API_ERROR_MESSAGE_PREFIX);
-}
-function addCacheBreakpoints(messages) {
-  return messages.map((msg, index) => {
-    return msg.type === "user" ? userMessageToMessageParam(msg, index > messages.length - 3) : assistantMessageToMessageParam(msg, index > messages.length - 3);
-  });
-}
-async function queryOpenAI(messages, systemPrompt, maxThinkingTokens, tools, signal, options) {
-  const config = getGlobalConfig();
-  const modelManager = getModelManager();
-  const toolUseContext = options?.toolUseContext;
-  const modelProfile = options?.modelProfile || modelManager.getModel("main");
-  let model;
-  const currentRequest = getCurrentRequest();
-  debug.api("MODEL_CONFIG_OPENAI", {
-    modelProfileFound: !!modelProfile,
-    modelProfileId: modelProfile?.modelName,
-    modelProfileName: modelProfile?.name,
-    modelProfileModelName: modelProfile?.modelName,
-    modelProfileProvider: modelProfile?.provider,
-    modelProfileBaseURL: modelProfile?.baseURL,
-    modelProfileApiKeyExists: !!modelProfile?.apiKey,
-    optionsModel: options?.model,
-    requestId: getCurrentRequest()?.id
-  });
-  if (modelProfile) {
-    model = modelProfile.modelName;
-  } else {
-    model = options?.model || modelProfile?.modelName || "";
+  const result = await f();
+  if (env.isCI) {
+    return result;
   }
-  if (options?.prependCLISysprompt) {
-    const [firstSyspromptBlock] = splitSysPromptPrefix(systemPrompt);
-    systemPrompt = [getCLISyspromptPrefix() + systemPrompt];
+  if (!existsSync(dirname(filename))) {
+    mkdirSync(dirname(filename), { recursive: true });
   }
-  const system = splitSysPromptPrefix(systemPrompt).map(
-    (_) => ({
-      ...PROMPT_CACHING_ENABLED ? { cache_control: { type: "ephemeral" } } : {},
-      text: _,
-      type: "text"
-    })
-  );
-  const toolSchemas = await Promise.all(
-    tools.map(
-      async (_) => ({
-        type: "function",
-        function: {
-          name: _.name,
-          description: await _.prompt({
-            safeMode: options?.safeMode
-          }),
-          parameters: "inputJSONSchema" in _ && _.inputJSONSchema ? _.inputJSONSchema : zodToJsonSchema4(_.inputSchema)
-        }
-      })
+  writeFileSync(
+    filename,
+    JSON.stringify(
+      {
+        input: dehydratedInput,
+        output: mapAssistantMessage(result, dehydrateValue)
+      },
+      null,
+      2
     )
   );
-  const openaiSystem = system.map(
-    (s) => ({
-      role: "system",
-      content: s.text
-    })
-  );
-  const openaiMessages = convertAnthropicMessagesToOpenAIMessages2(messages);
-  logSystemPromptConstruction({
-    basePrompt: systemPrompt.join("\n"),
-    kodeContext: generateKodeContext() || "",
-    reminders: [],
-    finalPrompt: systemPrompt.join("\n")
-  });
-  let start = Date.now();
-  let adapterContext = null;
-  if (modelProfile && modelProfile.modelName) {
-    debug.api("CHECKING_ADAPTER_SYSTEM", {
-      modelProfileName: modelProfile.modelName,
-      modelName: modelProfile.modelName,
-      provider: modelProfile.provider,
-      requestId: getCurrentRequest()?.id
-    });
-    const USE_NEW_ADAPTER_SYSTEM = process.env.USE_NEW_ADAPTERS !== "false";
-    if (USE_NEW_ADAPTER_SYSTEM) {
-      const shouldUseResponses = ModelAdapterFactory.shouldUseResponsesAPI(modelProfile);
-      if (shouldUseResponses) {
-        const adapter = ModelAdapterFactory.createAdapter(modelProfile);
-        const reasoningEffort = await getReasoningEffort(modelProfile, messages);
-        let verbosity = "medium";
-        const modelNameLower = modelProfile.modelName.toLowerCase();
-        if (modelNameLower.includes("high")) {
-          verbosity = "high";
-        } else if (modelNameLower.includes("low")) {
-          verbosity = "low";
-        }
-        const unifiedParams = {
-          messages: openaiMessages,
-          systemPrompt: openaiSystem.map((s) => s.content),
-          tools,
-          maxTokens: options?.maxTokens ?? getMaxTokensFromProfile(modelProfile),
-          stream: config.stream,
-          reasoningEffort,
-          temperature: options?.temperature ?? (isGPT5Model(model) ? 1 : MAIN_QUERY_TEMPERATURE),
-          previousResponseId: toolUseContext?.responseState?.previousResponseId,
-          verbosity,
-          ...options?.stopSequences && options.stopSequences.length > 0 ? { stopSequences: options.stopSequences } : {}
-        };
-        adapterContext = {
-          adapter,
-          request: adapter.createRequest(unifiedParams),
-          shouldUseResponses: true
-        };
-      }
+  return result;
+}
+function mapMessages(messages, f) {
+  return messages.map((_) => {
+    if (typeof _ === "string") {
+      return f(_);
     }
-  }
-  let queryResult;
-  let startIncludingRetries = Date.now();
-  try {
-    queryResult = await withRetry(
-      async () => {
-        start = Date.now();
-        if (adapterContext) {
-          if (adapterContext.shouldUseResponses) {
-            const { callGPT5ResponsesAPI } = await import("./openai-RRCWW33N.js");
-            const response = await callGPT5ResponsesAPI(
-              modelProfile,
-              adapterContext.request,
-              signal
-            );
-            const unifiedResponse = await adapterContext.adapter.parseResponse(response);
-            const assistantMessage2 = buildAssistantMessageFromUnifiedResponse(
-              unifiedResponse,
-              start
-            );
-            assistantMessage2.message.usage = normalizeUsage(
-              assistantMessage2.message.usage
-            );
+    return _.map((_2) => {
+      switch (_2.type) {
+        case "tool_result":
+          if (typeof _2.content === "string") {
+            return { ..._2, content: f(_2.content) };
+          }
+          if (Array.isArray(_2.content)) {
             return {
-              assistantMessage: assistantMessage2,
-              rawResponse: unifiedResponse,
-              apiFormat: "openai"
+              ..._2,
+              content: _2.content.map((_3) => {
+                switch (_3.type) {
+                  case "text":
+                    return { ..._3, text: f(_3.text) };
+                  case "image":
+                    return _3;
+                }
+              })
             };
           }
-          const s2 = await getCompletionWithProfile(
-            modelProfile,
-            adapterContext.request,
-            0,
-            10,
-            signal
-          );
-          let finalResponse2;
-          if (config.stream) {
-            finalResponse2 = await handleMessageStream(
-              s2,
-              signal
-            );
-          } else {
-            finalResponse2 = s2;
-          }
-          const message2 = convertOpenAIResponseToAnthropic(finalResponse2, tools);
-          const assistantMsg2 = {
-            type: "assistant",
-            message: message2,
-            costUSD: 0,
-            durationMs: Date.now() - start,
-            uuid: `${Date.now()}-${Math.random().toString(36).substr(2, 9)}`
-          };
+          return _2;
+        case "text":
+          return { ..._2, text: f(_2.text) };
+        case "tool_use":
           return {
-            assistantMessage: assistantMsg2,
-            rawResponse: finalResponse2,
-            apiFormat: "openai"
-          };
-        }
-        const maxTokens = options?.maxTokens ?? getMaxTokensFromProfile(modelProfile);
-        const isGPT5 = isGPT5Model(model);
-        const opts = {
-          model,
-          ...isGPT5 ? { max_completion_tokens: maxTokens } : { max_tokens: maxTokens },
-          messages: [...openaiSystem, ...openaiMessages],
-          temperature: options?.temperature ?? (isGPT5 ? 1 : MAIN_QUERY_TEMPERATURE)
-        };
-        if (options?.stopSequences && options.stopSequences.length > 0) {
-          ;
-          opts.stop = options.stopSequences;
-        }
-        if (config.stream) {
-          ;
-          opts.stream = true;
-          opts.stream_options = {
-            include_usage: true
+            ..._2,
+            input: mapValues(_2.input, f)
           };
+        case "image":
+          return _2;
+      }
+    });
+  });
+}
+function mapAssistantMessage(message, f) {
+  return {
+    durationMs: "DURATION",
+    costUSD: "COST",
+    uuid: "UUID",
+    message: {
+      ...message.message,
+      content: message.message.content.map((_) => {
+        switch (_.type) {
+          case "text":
+            return {
+              ..._,
+              text: f(_.text),
+              citations: _.citations || []
+            };
+          // Ensure citations
+          case "tool_use":
+            return {
+              ..._,
+              input: mapValues(_.input, f)
+            };
+          default:
+            return _;
         }
-        if (toolSchemas.length > 0) {
-          opts.tools = toolSchemas;
-          opts.tool_choice = "auto";
-        }
-        const reasoningEffort = await getReasoningEffort(modelProfile, messages);
-        if (reasoningEffort) {
-          opts.reasoning_effort = reasoningEffort;
-        }
-        const completionFunction = isGPT5Model(modelProfile?.modelName || "") ? getGPT5CompletionWithProfile : getCompletionWithProfile;
-        const s = await completionFunction(modelProfile, opts, 0, 10, signal);
-        let finalResponse;
-        if (opts.stream) {
-          finalResponse = await handleMessageStream(
-            s,
-            signal
-          );
-        } else {
-          finalResponse = s;
-        }
-        const message = convertOpenAIResponseToAnthropic(finalResponse, tools);
-        const assistantMsg = {
-          type: "assistant",
-          message,
-          costUSD: 0,
-          durationMs: Date.now() - start,
-          uuid: `${Date.now()}-${Math.random().toString(36).substr(2, 9)}`
-        };
-        return {
-          assistantMessage: assistantMsg,
-          rawResponse: finalResponse,
-          apiFormat: "openai"
-        };
+      }).filter(Boolean)
+    },
+    type: "assistant"
+  };
+}
+function dehydrateValue(s) {
+  if (typeof s !== "string") {
+    return s;
+  }
+  const s1 = s.replace(/num_files="\d+"/g, 'num_files="[NUM]"').replace(/duration_ms="\d+"/g, 'duration_ms="[DURATION]"').replace(/cost_usd="\d+"/g, 'cost_usd="[COST]"').replace(/\//g, path.sep).replaceAll(getCwd(), "[CWD]");
+  if (s1.includes("Files modified by user:")) {
+    return "Files modified by user: [FILES]";
+  }
+  return s1;
+}
+function hydrateValue(s) {
+  if (typeof s !== "string") {
+    return s;
+  }
+  return s.replaceAll("[NUM]", "1").replaceAll("[DURATION]", "100").replaceAll("[CWD]", getCwd());
+}
+
+// packages/core/src/services/responseStateManager.ts
+var ResponseStateManager = class {
+  conversationStates = /* @__PURE__ */ new Map();
+  // Cache cleanup after 1 hour of inactivity
+  CLEANUP_INTERVAL = 60 * 60 * 1e3;
+  constructor() {
+    setInterval(() => {
+      this.cleanup();
+    }, this.CLEANUP_INTERVAL);
+  }
+  /**
+   * Set the previous response ID for a conversation
+   */
+  setPreviousResponseId(conversationId, responseId) {
+    this.conversationStates.set(conversationId, {
+      previousResponseId: responseId,
+      lastUpdate: Date.now()
+    });
+  }
+  /**
+   * Get the previous response ID for a conversation
+   */
+  getPreviousResponseId(conversationId) {
+    const state = this.conversationStates.get(conversationId);
+    if (state) {
+      state.lastUpdate = Date.now();
+      return state.previousResponseId;
+    }
+    return void 0;
+  }
+  /**
+   * Clear state for a conversation
+   */
+  clearConversation(conversationId) {
+    this.conversationStates.delete(conversationId);
+  }
+  /**
+   * Clear all conversation states
+   */
+  clearAll() {
+    this.conversationStates.clear();
+  }
+  /**
+   * Clean up stale conversations
+   */
+  cleanup() {
+    const now = Date.now();
+    for (const [conversationId, state] of this.conversationStates.entries()) {
+      if (now - state.lastUpdate > this.CLEANUP_INTERVAL) {
+        this.conversationStates.delete(conversationId);
+      }
+    }
+  }
+  /**
+   * Get current state size (for debugging/monitoring)
+   */
+  getStateSize() {
+    return this.conversationStates.size;
+  }
+};
+var responseStateManager = new ResponseStateManager();
+function getConversationId(agentId, messageId) {
+  return agentId || messageId || `conv_${Date.now()}_${Math.random().toString(36).substr(2, 9)}`;
+}
+
+// packages/core/src/ai/llm/apiKey.ts
+import "@anthropic-ai/sdk/shims/node";
+import Anthropic3 from "@anthropic-ai/sdk";
+async function fetchAnthropicModels(baseURL, apiKey) {
+  try {
+    const modelsURL = baseURL ? `${baseURL.replace(/\/+$/, "")}/v1/models` : "https://api.anthropic.com/v1/models";
+    const response = await fetch(modelsURL, {
+      method: "GET",
+      headers: {
+        "x-api-key": apiKey,
+        "anthropic-version": "2023-06-01",
+        "User-Agent": USER_AGENT
+      }
+    });
+    if (!response.ok) {
+      if (response.status === 401) {
+        throw new Error(
+          "Invalid API key. Please check your Anthropic API key and try again."
+        );
+      } else if (response.status === 403) {
+        throw new Error(
+          "API key does not have permission to access models. Please check your API key permissions."
+        );
+      } else if (response.status === 429) {
+        throw new Error(
+          "Too many requests. Please wait a moment and try again."
+        );
+      } else if (response.status >= 500) {
+        throw new Error(
+          "Anthropic service is temporarily unavailable. Please try again later."
+        );
+      } else {
+        throw new Error(
+          `Unable to connect to Anthropic API (${response.status}). Please check your internet connection and API key.`
+        );
+      }
+    }
+    const data = await response.json();
+    return data.data || [];
+  } catch (error) {
+    if (error instanceof Error && error.message.includes("API key") || error instanceof Error && error.message.includes("Anthropic")) {
+      throw error;
+    }
+    logError(error);
+    debug.warn("ANTHROPIC_MODELS_FETCH_FAILED", {
+      error: error instanceof Error ? error.message : String(error)
+    });
+    throw new Error(
+      "Unable to connect to Anthropic API. Please check your internet connection and try again."
+    );
+  }
+}
+async function verifyApiKey(apiKey, baseURL, provider) {
+  if (!apiKey) {
+    return false;
+  }
+  if (provider && provider !== "anthropic") {
+    try {
+      const headers = {
+        Authorization: `Bearer ${apiKey}`,
+        "Content-Type": "application/json"
+      };
+      if (!baseURL) {
+        debug.warn("API_VERIFICATION_MISSING_BASE_URL", { provider });
+        return false;
+      }
+      const modelsURL = `${baseURL.replace(/\/+$/, "")}/models`;
+      const response = await fetch(modelsURL, {
+        method: "GET",
+        headers
+      });
+      return response.ok;
+    } catch (error) {
+      logError(error);
+      debug.warn("API_VERIFICATION_FAILED", {
+        provider,
+        error: error instanceof Error ? error.message : String(error)
+      });
+      return false;
+    }
+  }
+  const clientConfig = {
+    apiKey,
+    dangerouslyAllowBrowser: true,
+    maxRetries: 3,
+    defaultHeaders: {
+      "User-Agent": USER_AGENT
+    }
+  };
+  if (baseURL && (provider === "anthropic" || provider === "minimax-coding")) {
+    clientConfig.baseURL = baseURL;
+  }
+  const anthropic = new Anthropic3(clientConfig);
+  try {
+    await withRetry(
+      async () => {
+        const model = "claude-sonnet-4-20250514";
+        const messages = [{ role: "user", content: "test" }];
+        await anthropic.messages.create({
+          model,
+          max_tokens: 1e3,
+          // Simple test token limit for API verification
+          messages,
+          temperature: 0
+        });
+        return true;
       },
-      { signal }
+      { maxRetries: 2 }
+      // Use fewer retries for API key verification
     );
+    return true;
   } catch (error) {
     logError(error);
-    return getAssistantMessageFromError(error);
+    if (error instanceof Error && error.message.includes(
+      '{"type":"error","error":{"type":"authentication_error","message":"invalid x-api-key"}}'
+    )) {
+      return false;
+    }
+    throw error;
   }
-  const durationMs = Date.now() - start;
-  const durationMsIncludingRetries = Date.now() - startIncludingRetries;
-  const assistantMessage = queryResult.assistantMessage;
-  assistantMessage.message.content = normalizeContentFromAPI(
-    assistantMessage.message.content || []
-  );
-  const normalizedUsage = normalizeUsage(assistantMessage.message.usage);
-  assistantMessage.message.usage = normalizedUsage;
-  const inputTokens = normalizedUsage.input_tokens ?? 0;
-  const outputTokens = normalizedUsage.output_tokens ?? 0;
-  const cacheReadInputTokens = normalizedUsage.cache_read_input_tokens ?? 0;
-  const cacheCreationInputTokens = normalizedUsage.cache_creation_input_tokens ?? 0;
-  const costUSD = inputTokens / 1e6 * SONNET_COST_PER_MILLION_INPUT_TOKENS + outputTokens / 1e6 * SONNET_COST_PER_MILLION_OUTPUT_TOKENS + cacheReadInputTokens / 1e6 * SONNET_COST_PER_MILLION_PROMPT_CACHE_READ_TOKENS + cacheCreationInputTokens / 1e6 * SONNET_COST_PER_MILLION_PROMPT_CACHE_WRITE_TOKENS;
-  addToTotalCost(costUSD, durationMsIncludingRetries);
-  logLLMInteraction({
-    systemPrompt: systemPrompt.join("\n"),
-    messages: [...openaiSystem, ...openaiMessages],
-    response: assistantMessage.message || queryResult.rawResponse,
-    usage: {
-      inputTokens,
-      outputTokens
-    },
-    timing: {
-      start,
-      end: Date.now()
-    },
-    apiFormat: queryResult.apiFormat
-  });
-  assistantMessage.costUSD = costUSD;
-  assistantMessage.durationMs = durationMs;
-  assistantMessage.uuid = assistantMessage.uuid || randomUUID();
-  return assistantMessage;
 }
-function getMaxTokensFromProfile(modelProfile) {
-  return modelProfile?.maxTokens || 8e3;
-}
-function buildAssistantMessageFromUnifiedResponse(unifiedResponse, startTime) {
-  const contentBlocks = [...unifiedResponse.content || []];
-  if (unifiedResponse.toolCalls && unifiedResponse.toolCalls.length > 0) {
-    for (const toolCall of unifiedResponse.toolCalls) {
-      const tool = toolCall.function;
-      const toolName = tool?.name;
-      let toolArgs = {};
-      try {
-        toolArgs = tool?.arguments ? JSON.parse(tool.arguments) : {};
-      } catch (e) {
-      }
-      contentBlocks.push({
-        type: "tool_use",
-        input: toolArgs,
-        name: toolName,
-        id: toolCall.id?.length > 0 ? toolCall.id : nanoid()
-      });
+
+// packages/core/src/ai/llm.ts
+async function queryLLM(messages, systemPrompt, maxThinkingTokens, tools, signal, options) {
+  const modelManager = options.__testModelManager ?? getModelManager();
+  const modelResolution = modelManager.resolveModelWithInfo(options.model);
+  if (!modelResolution.success || !modelResolution.profile) {
+    const fallbackProfile = modelManager.resolveModel(options.model);
+    if (!fallbackProfile) {
+      throw new Error(
+        modelResolution.error || `Failed to resolve model: ${options.model}`
+      );
     }
+    debug.warn("MODEL_RESOLUTION_FALLBACK", {
+      inputParam: options.model,
+      error: modelResolution.error,
+      fallbackModelName: fallbackProfile.modelName,
+      fallbackProvider: fallbackProfile.provider,
+      requestId: getCurrentRequest()?.id
+    });
+    modelResolution.success = true;
+    modelResolution.profile = fallbackProfile;
   }
-  return {
-    type: "assistant",
-    message: {
-      role: "assistant",
-      content: contentBlocks,
-      usage: {
-        input_tokens: unifiedResponse.usage?.promptTokens ?? unifiedResponse.usage?.input_tokens ?? 0,
-        output_tokens: unifiedResponse.usage?.completionTokens ?? unifiedResponse.usage?.output_tokens ?? 0,
-        prompt_tokens: unifiedResponse.usage?.promptTokens ?? unifiedResponse.usage?.input_tokens ?? 0,
-        completion_tokens: unifiedResponse.usage?.completionTokens ?? unifiedResponse.usage?.output_tokens ?? 0,
-        promptTokens: unifiedResponse.usage?.promptTokens ?? unifiedResponse.usage?.input_tokens ?? 0,
-        completionTokens: unifiedResponse.usage?.completionTokens ?? unifiedResponse.usage?.output_tokens ?? 0,
-        totalTokens: unifiedResponse.usage?.totalTokens ?? (unifiedResponse.usage?.promptTokens ?? unifiedResponse.usage?.input_tokens ?? 0) + (unifiedResponse.usage?.completionTokens ?? unifiedResponse.usage?.output_tokens ?? 0)
-      }
-    },
-    costUSD: 0,
-    durationMs: Date.now() - startTime,
-    uuid: `${Date.now()}-${Math.random().toString(36).substr(2, 9)}`,
-    responseId: unifiedResponse.responseId
-  };
-}
-function normalizeUsage(usage) {
-  if (!usage) {
-    return {
-      input_tokens: 0,
-      output_tokens: 0,
-      cache_read_input_tokens: 0,
-      cache_creation_input_tokens: 0
+  const modelProfile = modelResolution.profile;
+  const resolvedModel = modelProfile.modelName;
+  const toolUseContext = options.toolUseContext;
+  if (toolUseContext && !toolUseContext.responseState) {
+    const conversationId = getConversationId(
+      toolUseContext.agentId,
+      toolUseContext.messageId
+    );
+    const previousResponseId = responseStateManager.getPreviousResponseId(conversationId);
+    toolUseContext.responseState = {
+      previousResponseId,
+      conversationId
     };
   }
-  const inputTokens = usage.input_tokens ?? usage.prompt_tokens ?? usage.inputTokens ?? 0;
-  const outputTokens = usage.output_tokens ?? usage.completion_tokens ?? usage.outputTokens ?? 0;
-  const cacheReadInputTokens = usage.cache_read_input_tokens ?? usage.prompt_token_details?.cached_tokens ?? usage.cacheReadInputTokens ?? 0;
-  const cacheCreationInputTokens = usage.cache_creation_input_tokens ?? usage.cacheCreatedInputTokens ?? 0;
-  return {
-    ...usage,
-    input_tokens: inputTokens,
-    output_tokens: outputTokens,
-    cache_read_input_tokens: cacheReadInputTokens,
-    cache_creation_input_tokens: cacheCreationInputTokens
-  };
-}
-function getModelInputTokenCostUSD(model) {
-  for (const providerModels of Object.values(models_default)) {
-    const modelInfo = providerModels.find((m) => m.model === model);
-    if (modelInfo) {
-      return modelInfo.input_cost_per_token || 0;
+  debug.api("MODEL_RESOLVED", {
+    inputParam: options.model,
+    resolvedModelName: resolvedModel,
+    provider: modelProfile.provider,
+    isPointer: ["main", "task", "compact", "quick"].includes(options.model),
+    hasResponseState: !!toolUseContext?.responseState,
+    conversationId: toolUseContext?.responseState?.conversationId,
+    requestId: getCurrentRequest()?.id
+  });
+  const currentRequest = getCurrentRequest();
+  debug.api("LLM_REQUEST_START", {
+    messageCount: messages.length,
+    systemPromptLength: systemPrompt.join(" ").length,
+    toolCount: tools.length,
+    model: resolvedModel,
+    originalModelParam: options.model,
+    requestId: getCurrentRequest()?.id
+  });
+  markPhase("LLM_CALL");
+  try {
+    const queryFn = options.__testQueryLLMWithPromptCaching ?? queryLLMWithPromptCaching;
+    const cleanOptions = { ...options };
+    delete cleanOptions.__testModelManager;
+    delete cleanOptions.__testQueryLLMWithPromptCaching;
+    const runQuery = () => queryFn(
+      messages,
+      systemPrompt,
+      maxThinkingTokens,
+      tools,
+      signal,
+      {
+        ...cleanOptions,
+        model: resolvedModel,
+        modelProfile,
+        toolUseContext
+      }
+      // Pass resolved ModelProfile and toolUseContext
+    );
+    const result = options.__testQueryLLMWithPromptCaching ? await runQuery() : await withVCR(messages, runQuery);
+    debug.api("LLM_REQUEST_SUCCESS", {
+      costUSD: result.costUSD,
+      durationMs: result.durationMs,
+      responseLength: result.message.content?.length || 0,
+      requestId: getCurrentRequest()?.id
+    });
+    if (toolUseContext?.responseState?.conversationId && result.responseId) {
+      responseStateManager.setPreviousResponseId(
+        toolUseContext.responseState.conversationId,
+        result.responseId
+      );
+      debug.api("RESPONSE_STATE_UPDATED", {
+        conversationId: toolUseContext.responseState.conversationId,
+        responseId: result.responseId,
+        requestId: getCurrentRequest()?.id
+      });
     }
+    return result;
+  } catch (error) {
+    logErrorWithDiagnosis(
+      error,
+      {
+        messageCount: messages.length,
+        systemPromptLength: systemPrompt.join(" ").length,
+        model: options.model,
+        toolCount: tools.length,
+        phase: "LLM_CALL"
+      },
+      currentRequest?.id
+    );
+    throw error;
   }
-  return 3e-6;
 }
-function getModelOutputTokenCostUSD(model) {
-  for (const providerModels of Object.values(models_default)) {
-    const modelInfo = providerModels.find((m) => m.model === model);
-    if (modelInfo) {
-      return modelInfo.output_cost_per_token || 0;
-    }
+async function queryLLMWithPromptCaching(messages, systemPrompt, maxThinkingTokens, tools, signal, options) {
+  const config = getGlobalConfig();
+  const modelManager = getModelManager();
+  const toolUseContext = options.toolUseContext;
+  const modelProfile = options.modelProfile || modelManager.getModel("main");
+  let provider;
+  if (modelProfile) {
+    provider = modelProfile.provider || config.primaryProvider || "anthropic";
+  } else {
+    provider = config.primaryProvider || "anthropic";
   }
-  return 15e-6;
+  if (provider === "anthropic" || provider === "bigdream" || provider === "opendev" || provider === "minimax-coding") {
+    return queryAnthropicNative(
+      messages,
+      systemPrompt,
+      maxThinkingTokens,
+      tools,
+      signal,
+      { ...options, modelProfile, toolUseContext }
+    );
+  }
+  return queryOpenAI(messages, systemPrompt, maxThinkingTokens, tools, signal, {
+    ...options,
+    modelProfile,
+    toolUseContext
+  });
 }
 async function queryModel(modelPointer, messages, systemPrompt = [], signal) {
   return queryLLM(
     messages,
     systemPrompt,
     0,
+    // maxThinkingTokens
     [],
+    // tools
     signal || new AbortController().signal,
     {
       safeMode: false,
@@ -3091,7 +3299,7 @@ async function queryQuick({
     {
       message: { role: "user", content: userPrompt },
       type: "user",
-      uuid: randomUUID()
+      uuid: randomUUID4()
     }
   ];
   return queryModel("quick", messages, systemPrompt, signal);