@within-7/minto 0.4.0 → 0.4.2

package/dist/services/llm/dispatch.js.map

@@ -1,7 +1,7 @@
 {
   "version": 3,
   "sources": ["../../../src/services/llm/dispatch.ts"],
-  "sourcesContent": ["/**\n * LLM dispatch \u2014 routes queries to the appropriate provider (Anthropic or OpenAI).\n */\nimport { randomUUID } from 'crypto'\n\nimport type { AssistantMessage, UserMessage } from '@query'\nimport { Tool } from '@tool'\nimport type { ToolUseContext } from '@tool'\nimport {\n  getGlobalConfig,\n  ModelProfile,\n} from '@utils/config'\nimport { isGPT5ModelName as isGPT5Model } from '@utils/config'\nimport {\n  debug as debugLogger,\n  markPhase,\n  getCurrentRequest,\n  logErrorWithDiagnosis,\n} from '@utils/debugLogger'\nimport { getModelManager } from '@utils/model'\nimport { withVCR } from '../vcr'\nimport { responseStateManager, getConversationId } from '../responseStateManager'\nimport { generateSystemReminders } from '../systemReminder'\n\nimport { generateMintoContext } from './mintoContext'\nimport { queryAnthropicNative } from './anthropicProvider'\nimport { queryOpenAI } from './openaiProvider'\n\n// \u2500\u2500 queryLLM \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n\nexport async function queryLLM(\n  messages: (UserMessage | AssistantMessage)[],\n  systemPrompt: string[],\n  maxThinkingTokens: number,\n  tools: Tool[],\n  signal: AbortSignal,\n  options: {\n    safeMode: boolean\n    model: string | import('@utils/config').ModelPointerType\n    prependCLISysprompt: boolean\n    toolUseContext?: ToolUseContext\n  },\n): Promise<AssistantMessage> {\n  const modelManager = getModelManager()\n  const modelResolution = modelManager.resolveModelWithInfo(options.model)\n\n  if (!modelResolution.success || !modelResolution.profile) {\n    throw new Error(\n      modelResolution.error || `Failed to resolve model: ${options.model}`,\n    )\n  }\n\n  const modelProfile = modelResolution.profile\n  const resolvedModel = modelProfile.modelName\n\n  // Initialize response state if toolUseContext is provided\n  const toolUseContext = options.toolUseContext\n  if (toolUseContext && !toolUseContext.responseState) {\n    const conversationId = getConversationId(\n      toolUseContext.agentId,\n      toolUseContext.messageId,\n    )\n    const previousResponseId =\n      responseStateManager.getPreviousResponseId(conversationId)\n\n    toolUseContext.responseState = {\n      previousResponseId,\n      conversationId,\n    }\n  }\n\n  debugLogger.api('MODEL_RESOLVED', {\n    inputParam: options.model,\n    resolvedModelName: resolvedModel,\n    provider: modelProfile.provider,\n    isPointer: ['main', 'task', 'reasoning', 'quick'].includes(options.model),\n    hasResponseState: !!toolUseContext?.responseState,\n    conversationId: toolUseContext?.responseState?.conversationId,\n    requestId: getCurrentRequest()?.id,\n  })\n\n  const currentRequest = getCurrentRequest()\n  debugLogger.api('LLM_REQUEST_START', {\n    messageCount: messages.length,\n    systemPromptLength: systemPrompt.join(' ').length,\n    toolCount: tools.length,\n    model: resolvedModel,\n    originalModelParam: options.model,\n    requestId: getCurrentRequest()?.id,\n  })\n\n  markPhase('LLM_CALL')\n\n  try {\n    const result = await withVCR(messages, () =>\n      queryLLMWithPromptCaching(\n        messages,\n        systemPrompt,\n        maxThinkingTokens,\n        tools,\n        signal,\n        { ...options, model: resolvedModel, modelProfile, toolUseContext },\n      ),\n    )\n\n    debugLogger.api('LLM_REQUEST_SUCCESS', {\n      costUSD: result.costUSD,\n      durationMs: result.durationMs,\n      responseLength: result.message.content?.length || 0,\n      requestId: getCurrentRequest()?.id,\n    })\n\n    // Update response state for GPT-5 Responses API continuation\n    if (toolUseContext?.responseState?.conversationId && result.responseId) {\n      responseStateManager.setPreviousResponseId(\n        toolUseContext.responseState.conversationId,\n        result.responseId,\n      )\n\n      debugLogger.api('RESPONSE_STATE_UPDATED', {\n        conversationId: toolUseContext.responseState.conversationId,\n        responseId: result.responseId,\n        requestId: getCurrentRequest()?.id,\n      })\n    }\n\n    return result\n  } catch (error) {\n    logErrorWithDiagnosis(\n      error,\n      {\n        messageCount: messages.length,\n        systemPromptLength: systemPrompt.join(' ').length,\n        model: options.model,\n        toolCount: tools.length,\n        phase: 'LLM_CALL',\n      },\n      currentRequest?.id,\n    )\n\n    throw error\n  }\n}\n\n// \u2500\u2500 formatSystemPromptWithContext \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n\nexport function formatSystemPromptWithContext(\n  systemPrompt: string[],\n  context: { [k: string]: string },\n  agentId?: string,\n): { systemPrompt: string[]; reminders: string } {\n  const enhancedPrompt = [...systemPrompt]\n  let reminders = ''\n\n  // Step 0: Add GPT-5 Agent persistence support for coding tasks\n  const modelManager = getModelManager()\n  const modelProfile = modelManager.getModel('main')\n  if (modelProfile && isGPT5Model(modelProfile.modelName)) {\n    const persistencePrompts = [\n      '\\n# Agent Persistence for Long-Running Coding Tasks',\n      'You are working on a coding project that may involve multiple steps and iterations. Please maintain context and continuity throughout the session:',\n      '- Remember architectural decisions and design patterns established earlier',\n      '- Keep track of file modifications and their relationships',\n      '- Maintain awareness of the overall project structure and goals',\n      '- Reference previous implementations when making related changes',\n      '- Ensure consistency with existing code style and conventions',\n      '- Build incrementally on previous work rather than starting from scratch',\n    ]\n    enhancedPrompt.push(...persistencePrompts)\n  }\n\n  const hasContext = Object.entries(context).length > 0\n\n  if (hasContext) {\n    const mintoContext = generateMintoContext()\n    if (mintoContext) {\n      enhancedPrompt.push('\\n---\\n# \u9879\u76EE\u4E0A\u4E0B\u6587\\n')\n      enhancedPrompt.push(mintoContext)\n      enhancedPrompt.push('\\n---\\n')\n    }\n\n    const reminderMessages = generateSystemReminders(hasContext, agentId)\n    if (reminderMessages.length > 0) {\n      reminders = reminderMessages.map(r => r.content).join('\\n') + '\\n'\n    }\n\n    enhancedPrompt.push(\n      `\\nAs you answer the user's questions, you can use the following context:\\n`,\n    )\n\n    const filteredContext = Object.fromEntries(\n      Object.entries(context).filter(\n        ([key]) => key !== 'projectDocs' && key !== 'userDocs',\n      ),\n    )\n\n    enhancedPrompt.push(\n      ...Object.entries(filteredContext).map(\n        ([key, value]) => `<context name=\"${key}\">${value}</context>`,\n      ),\n    )\n  }\n\n  return { systemPrompt: enhancedPrompt, reminders }\n}\n\n// \u2500\u2500 queryLLMWithPromptCaching (internal dispatcher) \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n\nasync function queryLLMWithPromptCaching(\n  messages: (UserMessage | AssistantMessage)[],\n  systemPrompt: string[],\n  maxThinkingTokens: number,\n  tools: Tool[],\n  signal: AbortSignal,\n  options: {\n    safeMode: boolean\n    model: string\n    prependCLISysprompt: boolean\n    modelProfile?: ModelProfile | null\n    toolUseContext?: ToolUseContext\n  },\n): Promise<AssistantMessage> {\n  const config = getGlobalConfig()\n  const modelManager = getModelManager()\n  const toolUseContext = options.toolUseContext\n\n  const modelProfile = options.modelProfile || modelManager.getModel('main')\n  let provider: string\n\n  if (modelProfile) {\n    provider = modelProfile.provider || config.primaryProvider || 'anthropic'\n  } else {\n    provider = config.primaryProvider || 'anthropic'\n  }\n\n  // Use native Anthropic SDK for Anthropic and some Anthropic-compatible providers\n  if (\n    provider === 'anthropic' ||\n    provider === 'bigdream' ||\n    provider === 'opendev'\n  ) {\n    return queryAnthropicNative(\n      messages,\n      systemPrompt,\n      maxThinkingTokens,\n      tools,\n      signal,\n      { ...options, modelProfile, toolUseContext },\n    )\n  }\n\n  // Use OpenAI-compatible interface for all other providers\n  return queryOpenAI(messages, systemPrompt, maxThinkingTokens, tools, signal, {\n    ...options,\n    modelProfile,\n    toolUseContext,\n  })\n}\n\n// \u2500\u2500 Convenience query functions \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n\nexport async function queryModel(\n  modelPointer: import('@utils/config').ModelPointerType,\n  messages: (UserMessage | AssistantMessage)[],\n  systemPrompt: string[] = [],\n  signal?: AbortSignal,\n): Promise<AssistantMessage> {\n  return queryLLM(\n    messages,\n    systemPrompt,\n    0, // maxThinkingTokens\n    [], // tools\n    signal || new AbortController().signal,\n    {\n      safeMode: false,\n      model: modelPointer,\n      prependCLISysprompt: true,\n    },\n  )\n}\n\nexport async function queryQuick({\n  systemPrompt = [],\n  userPrompt,\n  assistantPrompt,\n  enablePromptCaching = false,\n  signal,\n}: {\n  systemPrompt?: string[]\n  userPrompt: string\n  assistantPrompt?: string\n  enablePromptCaching?: boolean\n  signal?: AbortSignal\n}): Promise<AssistantMessage> {\n  const messages = [\n    {\n      message: { role: 'user', content: userPrompt },\n      type: 'user',\n      uuid: randomUUID(),\n    },\n  ] as (UserMessage | AssistantMessage)[]\n\n  return queryModel('quick', messages, systemPrompt, signal)\n}\n"],
-  "mappings": "AAGA,SAAS,kBAAkB;AAK3B;AAAA,EACE;AAAA,OAEK;AACP,SAAS,mBAAmB,mBAAmB;AAC/C;AAAA,EACE,SAAS;AAAA,EACT;AAAA,EACA;AAAA,EACA;AAAA,OACK;AACP,SAAS,uBAAuB;AAChC,SAAS,eAAe;AACxB,SAAS,sBAAsB,yBAAyB;AACxD,SAAS,+BAA+B;AAExC,SAAS,4BAA4B;AACrC,SAAS,4BAA4B;AACrC,SAAS,mBAAmB;AAI5B,eAAsB,SACpB,UACA,cACA,mBACA,OACA,QACA,SAM2B;AAC3B,QAAM,eAAe,gBAAgB;AACrC,QAAM,kBAAkB,aAAa,qBAAqB,QAAQ,KAAK;AAEvE,MAAI,CAAC,gBAAgB,WAAW,CAAC,gBAAgB,SAAS;AACxD,UAAM,IAAI;AAAA,MACR,gBAAgB,SAAS,4BAA4B,QAAQ,KAAK;AAAA,IACpE;AAAA,EACF;AAEA,QAAM,eAAe,gBAAgB;AACrC,QAAM,gBAAgB,aAAa;AAGnC,QAAM,iBAAiB,QAAQ;AAC/B,MAAI,kBAAkB,CAAC,eAAe,eAAe;AACnD,UAAM,iBAAiB;AAAA,MACrB,eAAe;AAAA,MACf,eAAe;AAAA,IACjB;AACA,UAAM,qBACJ,qBAAqB,sBAAsB,cAAc;AAE3D,mBAAe,gBAAgB;AAAA,MAC7B;AAAA,MACA;AAAA,IACF;AAAA,EACF;AAEA,cAAY,IAAI,kBAAkB;AAAA,IAChC,YAAY,QAAQ;AAAA,IACpB,mBAAmB;AAAA,IACnB,UAAU,aAAa;AAAA,IACvB,WAAW,CAAC,QAAQ,QAAQ,aAAa,OAAO,EAAE,SAAS,QAAQ,KAAK;AAAA,IACxE,kBAAkB,CAAC,CAAC,gBAAgB;AAAA,IACpC,gBAAgB,gBAAgB,eAAe;AAAA,IAC/C,WAAW,kBAAkB,GAAG;AAAA,EAClC,CAAC;AAED,QAAM,iBAAiB,kBAAkB;AACzC,cAAY,IAAI,qBAAqB;AAAA,IACnC,cAAc,SAAS;AAAA,IACvB,oBAAoB,aAAa,KAAK,GAAG,EAAE;AAAA,IAC3C,WAAW,MAAM;AAAA,IACjB,OAAO;AAAA,IACP,oBAAoB,QAAQ;AAAA,IAC5B,WAAW,kBAAkB,GAAG;AAAA,EAClC,CAAC;AAED,YAAU,UAAU;AAEpB,MAAI;AACF,UAAM,SAAS,MAAM;AAAA,MAAQ;AAAA,MAAU,MACrC;AAAA,QACE;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA,EAAE,GAAG,SAAS,OAAO,eAAe,cAAc,eAAe;AAAA,MACnE;AAAA,IACF;AAEA,gBAAY,IAAI,uBAAuB;AAAA,MACrC,SAAS,OAAO;AAAA,MAChB,YAAY,OAAO;AAAA,MACnB,gBAAgB,OAAO,QAAQ,SAAS,UAAU;AAAA,MAClD,WAAW,kBAAkB,GAAG;AAAA,IAClC,CAAC;AAGD,QAAI,gBAAgB,eAAe,kBAAkB,OAAO,YAAY;AACtE,2BAAqB;AAAA,QACnB,eAAe,cAAc;AAAA,QAC7B,OAAO;AAAA,MACT;AAEA,kBAAY,IAAI,0BAA0B;AAAA,QACxC,gBAAgB,eAAe,cAAc;AAAA,QAC7C,YAAY,OAAO;AAAA,QACnB,WAAW,kBAAkB,GAAG;AAAA,MAClC,CAAC;AAAA,IACH;AAEA,WAAO;AAAA,EACT,SAAS,OAAO;AACd;AAAA,MACE;AAAA,MACA;AAAA,QACE,cAAc,SAAS;AAAA,QACvB,oBAAoB,aAAa,KAAK,GAAG,EAAE;AAAA,QAC3C,OAAO,QAAQ;AAAA,QACf,WAAW,MAAM;AAAA,QACjB,OAAO;AAAA,MACT;AAAA,MACA,gBAAgB;AAAA,IAClB;AAEA,UAAM;AAAA,EACR;AACF;AAIO,SAAS,8BACd,cACA,SACA,SAC+C;AAC/C,QAAM,iBAAiB,CAAC,GAAG,YAAY;AACvC,MAAI,YAAY;AAGhB,QAAM,eAAe,gBAAgB;AACrC,QAAM,eAAe,aAAa,SAAS,MAAM;AACjD,MAAI,gBAAgB,YAAY,aAAa,SAAS,GAAG;AACvD,UAAM,qBAAqB;AAAA,MACzB;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF;AACA,mBAAe,KAAK,GAAG,kBAAkB;AAAA,EAC3C;AAEA,QAAM,aAAa,OAAO,QAAQ,OAAO,EAAE,SAAS;AAEpD,MAAI,YAAY;AACd,UAAM,eAAe,qBAAqB;AAC1C,QAAI,cAAc;AAChB,qBAAe,KAAK,2CAAkB;AACtC,qBAAe,KAAK,YAAY;AAChC,qBAAe,KAAK,SAAS;AAAA,IAC/B;AAEA,UAAM,mBAAmB,wBAAwB,YAAY,OAAO;AACpE,QAAI,iBAAiB,SAAS,GAAG;AAC/B,kBAAY,iBAAiB,IAAI,OAAK,EAAE,OAAO,EAAE,KAAK,IAAI,IAAI;AAAA,IAChE;AAEA,mBAAe;AAAA,MACb;AAAA;AAAA;AAAA,IACF;AAEA,UAAM,kBAAkB,OAAO;AAAA,MAC7B,OAAO,QAAQ,OAAO,EAAE;AAAA,QACtB,CAAC,CAAC,GAAG,MAAM,QAAQ,iBAAiB,QAAQ;AAAA,MAC9C;AAAA,IACF;AAEA,mBAAe;AAAA,MACb,GAAG,OAAO,QAAQ,eAAe,EAAE;AAAA,QACjC,CAAC,CAAC,KAAK,KAAK,MAAM,kBAAkB,GAAG,KAAK,KAAK;AAAA,MACnD;AAAA,IACF;AAAA,EACF;AAEA,SAAO,EAAE,cAAc,gBAAgB,UAAU;AACnD;AAIA,eAAe,0BACb,UACA,cACA,mBACA,OACA,QACA,SAO2B;AAC3B,QAAM,SAAS,gBAAgB;AAC/B,QAAM,eAAe,gBAAgB;AACrC,QAAM,iBAAiB,QAAQ;AAE/B,QAAM,eAAe,QAAQ,gBAAgB,aAAa,SAAS,MAAM;AACzE,MAAI;AAEJ,MAAI,cAAc;AAChB,eAAW,aAAa,YAAY,OAAO,mBAAmB;AAAA,EAChE,OAAO;AACL,eAAW,OAAO,mBAAmB;AAAA,EACvC;AAGA,MACE,aAAa,eACb,aAAa,cACb,aAAa,WACb;AACA,WAAO;AAAA,MACL;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA,EAAE,GAAG,SAAS,cAAc,eAAe;AAAA,IAC7C;AAAA,EACF;AAGA,SAAO,YAAY,UAAU,cAAc,mBAAmB,OAAO,QAAQ;AAAA,IAC3E,GAAG;AAAA,IACH;AAAA,IACA;AAAA,EACF,CAAC;AACH;AAIA,eAAsB,WACpB,cACA,UACA,eAAyB,CAAC,GAC1B,QAC2B;AAC3B,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA;AAAA;AAAA,IACA,CAAC;AAAA;AAAA,IACD,UAAU,IAAI,gBAAgB,EAAE;AAAA,IAChC;AAAA,MACE,UAAU;AAAA,MACV,OAAO;AAAA,MACP,qBAAqB;AAAA,IACvB;AAAA,EACF;AACF;AAEA,eAAsB,WAAW;AAAA,EAC/B,eAAe,CAAC;AAAA,EAChB;AAAA,EACA;AAAA,EACA,sBAAsB;AAAA,EACtB;AACF,GAM8B;AAC5B,QAAM,WAAW;AAAA,IACf;AAAA,MACE,SAAS,EAAE,MAAM,QAAQ,SAAS,WAAW;AAAA,MAC7C,MAAM;AAAA,MACN,MAAM,WAAW;AAAA,IACnB;AAAA,EACF;AAEA,SAAO,WAAW,SAAS,UAAU,cAAc,MAAM;AAC3D;",
+  "sourcesContent": ["/**\n * LLM dispatch \u2014 routes queries to the appropriate provider (Anthropic or OpenAI).\n */\nimport { randomUUID } from 'crypto'\n\nimport type { AssistantMessage, UserMessage } from '@query'\nimport { Tool } from '@tool'\nimport type { ToolUseContext } from '@tool'\nimport { getGlobalConfig, ModelProfile } from '@utils/config'\nimport { isGPT5ModelName as isGPT5Model } from '@utils/config'\nimport {\n  debug as debugLogger,\n  markPhase,\n  getCurrentRequest,\n  logErrorWithDiagnosis,\n} from '@utils/debugLogger'\nimport { getModelManager } from '@utils/model'\nimport { withVCR } from '../vcr'\nimport {\n  responseStateManager,\n  getConversationId,\n} from '../responseStateManager'\nimport { generateSystemReminders } from '../systemReminder'\nimport { getSessionMemoryBlock } from '../sessionMemoryInjector'\n\nimport { isAutoMemoryEnabled, readMemoryMd } from '@utils/autoMemoryPaths'\nimport { generateMintoContext } from './mintoContext'\nimport { queryAnthropicNative } from './anthropicProvider'\nimport { queryOpenAI } from './openaiProvider'\n\n// \u2500\u2500 queryLLM \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n\nexport async function queryLLM(\n  messages: (UserMessage | AssistantMessage)[],\n  systemPrompt: string[],\n  maxThinkingTokens: number,\n  tools: Tool[],\n  signal: AbortSignal,\n  options: {\n    safeMode: boolean\n    model: string | import('@utils/config').ModelPointerType\n    prependCLISysprompt: boolean\n    toolUseContext?: ToolUseContext\n  },\n): Promise<AssistantMessage> {\n  const modelManager = getModelManager()\n  const modelResolution = modelManager.resolveModelWithInfo(options.model)\n\n  if (!modelResolution.success || !modelResolution.profile) {\n    throw new Error(\n      modelResolution.error || `Failed to resolve model: ${options.model}`,\n    )\n  }\n\n  const modelProfile = modelResolution.profile\n  const resolvedModel = modelProfile.modelName\n\n  // Initialize response state if toolUseContext is provided\n  const toolUseContext = options.toolUseContext\n  if (toolUseContext && !toolUseContext.responseState) {\n    const conversationId = getConversationId(\n      toolUseContext.agentId,\n      toolUseContext.messageId,\n    )\n    const previousResponseId =\n      responseStateManager.getPreviousResponseId(conversationId)\n\n    toolUseContext.responseState = {\n      previousResponseId,\n      conversationId,\n    }\n  }\n\n  debugLogger.api('MODEL_RESOLVED', {\n    inputParam: options.model,\n    resolvedModelName: resolvedModel,\n    provider: modelProfile.provider,\n    isPointer: ['main', 'task', 'reasoning', 'quick', 'compact'].includes(\n      options.model,\n    ),\n    hasResponseState: !!toolUseContext?.responseState,\n    conversationId: toolUseContext?.responseState?.conversationId,\n    requestId: getCurrentRequest()?.id,\n  })\n\n  const currentRequest = getCurrentRequest()\n  debugLogger.api('LLM_REQUEST_START', {\n    messageCount: messages.length,\n    systemPromptLength: systemPrompt.join(' ').length,\n    toolCount: tools.length,\n    model: resolvedModel,\n    originalModelParam: options.model,\n    requestId: getCurrentRequest()?.id,\n  })\n\n  markPhase('LLM_CALL')\n\n  try {\n    const result = await withVCR(messages, () =>\n      queryLLMWithPromptCaching(\n        messages,\n        systemPrompt,\n        maxThinkingTokens,\n        tools,\n        signal,\n        { ...options, model: resolvedModel, modelProfile, toolUseContext },\n      ),\n    )\n\n    debugLogger.api('LLM_REQUEST_SUCCESS', {\n      costUSD: result.costUSD,\n      durationMs: result.durationMs,\n      responseLength: result.message.content?.length || 0,\n      requestId: getCurrentRequest()?.id,\n    })\n\n    // Update response state for GPT-5 Responses API continuation\n    if (toolUseContext?.responseState?.conversationId && result.responseId) {\n      responseStateManager.setPreviousResponseId(\n        toolUseContext.responseState.conversationId,\n        result.responseId,\n      )\n\n      debugLogger.api('RESPONSE_STATE_UPDATED', {\n        conversationId: toolUseContext.responseState.conversationId,\n        responseId: result.responseId,\n        requestId: getCurrentRequest()?.id,\n      })\n    }\n\n    return result\n  } catch (error) {\n    logErrorWithDiagnosis(\n      error,\n      {\n        messageCount: messages.length,\n        systemPromptLength: systemPrompt.join(' ').length,\n        model: options.model,\n        toolCount: tools.length,\n        phase: 'LLM_CALL',\n      },\n      currentRequest?.id,\n    )\n\n    throw error\n  }\n}\n\n// \u2500\u2500 formatSystemPromptWithContext \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n\nexport function formatSystemPromptWithContext(\n  systemPrompt: string[],\n  context: { [k: string]: string },\n  agentId?: string,\n): { systemPrompt: string[]; reminders: string } {\n  const enhancedPrompt = [...systemPrompt]\n  let reminders = ''\n\n  // Step 0: Add GPT-5 Agent persistence support for coding tasks\n  const modelManager = getModelManager()\n  const modelProfile = modelManager.getModel('main')\n  if (modelProfile && isGPT5Model(modelProfile.modelName)) {\n    const persistencePrompts = [\n      '\\n# Agent Persistence for Long-Running Coding Tasks',\n      'You are working on a coding project that may involve multiple steps and iterations. Please maintain context and continuity throughout the session:',\n      '- Remember architectural decisions and design patterns established earlier',\n      '- Keep track of file modifications and their relationships',\n      '- Maintain awareness of the overall project structure and goals',\n      '- Reference previous implementations when making related changes',\n      '- Ensure consistency with existing code style and conventions',\n      '- Build incrementally on previous work rather than starting from scratch',\n    ]\n    enhancedPrompt.push(...persistencePrompts)\n  }\n\n  const hasContext = Object.entries(context).length > 0\n\n  // Inject session memory from previous sessions (if enabled)\n  const sessionMemoryBlock = getSessionMemoryBlock()\n  if (sessionMemoryBlock) {\n    enhancedPrompt.push(sessionMemoryBlock)\n  }\n\n  // Inject MEMORY.md from per-project auto memory directory (CC-compatible)\n  if (isAutoMemoryEnabled()) {\n    const memoryContent = readMemoryMd()\n    if (memoryContent) {\n      enhancedPrompt.push(`\\n# Auto Memory\\n\\n${memoryContent}\\n`)\n    }\n  }\n\n  if (hasContext) {\n    const mintoContext = generateMintoContext()\n    if (mintoContext) {\n      enhancedPrompt.push('\\n---\\n# \u9879\u76EE\u4E0A\u4E0B\u6587\\n')\n      enhancedPrompt.push(mintoContext)\n      enhancedPrompt.push('\\n---\\n')\n    }\n\n    const reminderMessages = generateSystemReminders(hasContext, agentId)\n    if (reminderMessages.length > 0) {\n      reminders = reminderMessages.map(r => r.content).join('\\n') + '\\n'\n    }\n\n    enhancedPrompt.push(\n      `\\nAs you answer the user's questions, you can use the following context:\\n`,\n    )\n\n    const filteredContext = Object.fromEntries(\n      Object.entries(context).filter(\n        ([key]) => key !== 'projectDocs' && key !== 'userDocs',\n      ),\n    )\n\n    enhancedPrompt.push(\n      ...Object.entries(filteredContext).map(\n        ([key, value]) => `<context name=\"${key}\">${value}</context>`,\n      ),\n    )\n  }\n\n  return { systemPrompt: enhancedPrompt, reminders }\n}\n\n// \u2500\u2500 queryLLMWithPromptCaching (internal dispatcher) \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n\nasync function queryLLMWithPromptCaching(\n  messages: (UserMessage | AssistantMessage)[],\n  systemPrompt: string[],\n  maxThinkingTokens: number,\n  tools: Tool[],\n  signal: AbortSignal,\n  options: {\n    safeMode: boolean\n    model: string\n    prependCLISysprompt: boolean\n    modelProfile?: ModelProfile | null\n    toolUseContext?: ToolUseContext\n  },\n): Promise<AssistantMessage> {\n  const config = getGlobalConfig()\n  const modelManager = getModelManager()\n  const toolUseContext = options.toolUseContext\n\n  const modelProfile = options.modelProfile || modelManager.getModel('main')\n  let provider: string\n\n  if (modelProfile) {\n    provider = modelProfile.provider || config.primaryProvider || 'anthropic'\n  } else {\n    provider = config.primaryProvider || 'anthropic'\n  }\n\n  // Use native Anthropic SDK for Anthropic and some Anthropic-compatible providers\n  if (\n    provider === 'anthropic' ||\n    provider === 'bigdream' ||\n    provider === 'opendev'\n  ) {\n    return queryAnthropicNative(\n      messages,\n      systemPrompt,\n      maxThinkingTokens,\n      tools,\n      signal,\n      { ...options, modelProfile, toolUseContext },\n    )\n  }\n\n  // Use OpenAI-compatible interface for all other providers\n  return queryOpenAI(messages, systemPrompt, maxThinkingTokens, tools, signal, {\n    ...options,\n    modelProfile,\n    toolUseContext,\n  })\n}\n\n// \u2500\u2500 Convenience query functions \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n\nexport async function queryModel(\n  modelPointer: import('@utils/config').ModelPointerType,\n  messages: (UserMessage | AssistantMessage)[],\n  systemPrompt: string[] = [],\n  signal?: AbortSignal,\n): Promise<AssistantMessage> {\n  return queryLLM(\n    messages,\n    systemPrompt,\n    0, // maxThinkingTokens\n    [], // tools\n    signal || new AbortController().signal,\n    {\n      safeMode: false,\n      model: modelPointer,\n      prependCLISysprompt: true,\n    },\n  )\n}\n\nexport async function queryQuick({\n  systemPrompt = [],\n  userPrompt,\n  assistantPrompt,\n  enablePromptCaching = false,\n  signal,\n}: {\n  systemPrompt?: string[]\n  userPrompt: string\n  assistantPrompt?: string\n  enablePromptCaching?: boolean\n  signal?: AbortSignal\n}): Promise<AssistantMessage> {\n  const messages = [\n    {\n      message: { role: 'user', content: userPrompt },\n      type: 'user',\n      uuid: randomUUID(),\n    },\n  ] as (UserMessage | AssistantMessage)[]\n\n  return queryModel('quick', messages, systemPrompt, signal)\n}\n"],
+  "mappings": "AAGA,SAAS,kBAAkB;AAK3B,SAAS,uBAAqC;AAC9C,SAAS,mBAAmB,mBAAmB;AAC/C;AAAA,EACE,SAAS;AAAA,EACT;AAAA,EACA;AAAA,EACA;AAAA,OACK;AACP,SAAS,uBAAuB;AAChC,SAAS,eAAe;AACxB;AAAA,EACE;AAAA,EACA;AAAA,OACK;AACP,SAAS,+BAA+B;AACxC,SAAS,6BAA6B;AAEtC,SAAS,qBAAqB,oBAAoB;AAClD,SAAS,4BAA4B;AACrC,SAAS,4BAA4B;AACrC,SAAS,mBAAmB;AAI5B,eAAsB,SACpB,UACA,cACA,mBACA,OACA,QACA,SAM2B;AAC3B,QAAM,eAAe,gBAAgB;AACrC,QAAM,kBAAkB,aAAa,qBAAqB,QAAQ,KAAK;AAEvE,MAAI,CAAC,gBAAgB,WAAW,CAAC,gBAAgB,SAAS;AACxD,UAAM,IAAI;AAAA,MACR,gBAAgB,SAAS,4BAA4B,QAAQ,KAAK;AAAA,IACpE;AAAA,EACF;AAEA,QAAM,eAAe,gBAAgB;AACrC,QAAM,gBAAgB,aAAa;AAGnC,QAAM,iBAAiB,QAAQ;AAC/B,MAAI,kBAAkB,CAAC,eAAe,eAAe;AACnD,UAAM,iBAAiB;AAAA,MACrB,eAAe;AAAA,MACf,eAAe;AAAA,IACjB;AACA,UAAM,qBACJ,qBAAqB,sBAAsB,cAAc;AAE3D,mBAAe,gBAAgB;AAAA,MAC7B;AAAA,MACA;AAAA,IACF;AAAA,EACF;AAEA,cAAY,IAAI,kBAAkB;AAAA,IAChC,YAAY,QAAQ;AAAA,IACpB,mBAAmB;AAAA,IACnB,UAAU,aAAa;AAAA,IACvB,WAAW,CAAC,QAAQ,QAAQ,aAAa,SAAS,SAAS,EAAE;AAAA,MAC3D,QAAQ;AAAA,IACV;AAAA,IACA,kBAAkB,CAAC,CAAC,gBAAgB;AAAA,IACpC,gBAAgB,gBAAgB,eAAe;AAAA,IAC/C,WAAW,kBAAkB,GAAG;AAAA,EAClC,CAAC;AAED,QAAM,iBAAiB,kBAAkB;AACzC,cAAY,IAAI,qBAAqB;AAAA,IACnC,cAAc,SAAS;AAAA,IACvB,oBAAoB,aAAa,KAAK,GAAG,EAAE;AAAA,IAC3C,WAAW,MAAM;AAAA,IACjB,OAAO;AAAA,IACP,oBAAoB,QAAQ;AAAA,IAC5B,WAAW,kBAAkB,GAAG;AAAA,EAClC,CAAC;AAED,YAAU,UAAU;AAEpB,MAAI;AACF,UAAM,SAAS,MAAM;AAAA,MAAQ;AAAA,MAAU,MACrC;AAAA,QACE;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA,EAAE,GAAG,SAAS,OAAO,eAAe,cAAc,eAAe;AAAA,MACnE;AAAA,IACF;AAEA,gBAAY,IAAI,uBAAuB;AAAA,MACrC,SAAS,OAAO;AAAA,MAChB,YAAY,OAAO;AAAA,MACnB,gBAAgB,OAAO,QAAQ,SAAS,UAAU;AAAA,MAClD,WAAW,kBAAkB,GAAG;AAAA,IAClC,CAAC;AAGD,QAAI,gBAAgB,eAAe,kBAAkB,OAAO,YAAY;AACtE,2BAAqB;AAAA,QACnB,eAAe,cAAc;AAAA,QAC7B,OAAO;AAAA,MACT;AAEA,kBAAY,IAAI,0BAA0B;AAAA,QACxC,gBAAgB,eAAe,cAAc;AAAA,QAC7C,YAAY,OAAO;AAAA,QACnB,WAAW,kBAAkB,GAAG;AAAA,MAClC,CAAC;AAAA,IACH;AAEA,WAAO;AAAA,EACT,SAAS,OAAO;AACd;AAAA,MACE;AAAA,MACA;AAAA,QACE,cAAc,SAAS;AAAA,QACvB,oBAAoB,aAAa,KAAK,GAAG,EAAE;AAAA,QAC3C,OAAO,QAAQ;AAAA,QACf,WAAW,MAAM;AAAA,QACjB,OAAO;AAAA,MACT;AAAA,MACA,gBAAgB;AAAA,IAClB;AAEA,UAAM;AAAA,EACR;AACF;AAIO,SAAS,8BACd,cACA,SACA,SAC+C;AAC/C,QAAM,iBAAiB,CAAC,GAAG,YAAY;AACvC,MAAI,YAAY;AAGhB,QAAM,eAAe,gBAAgB;AACrC,QAAM,eAAe,aAAa,SAAS,MAAM;AACjD,MAAI,gBAAgB,YAAY,aAAa,SAAS,GAAG;AACvD,UAAM,qBAAqB;AAAA,MACzB;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF;AACA,mBAAe,KAAK,GAAG,kBAAkB;AAAA,EAC3C;AAEA,QAAM,aAAa,OAAO,QAAQ,OAAO,EAAE,SAAS;AAGpD,QAAM,qBAAqB,sBAAsB;AACjD,MAAI,oBAAoB;AACtB,mBAAe,KAAK,kBAAkB;AAAA,EACxC;AAGA,MAAI,oBAAoB,GAAG;AACzB,UAAM,gBAAgB,aAAa;AACnC,QAAI,eAAe;AACjB,qBAAe,KAAK;AAAA;AAAA;AAAA,EAAsB,aAAa;AAAA,CAAI;AAAA,IAC7D;AAAA,EACF;AAEA,MAAI,YAAY;AACd,UAAM,eAAe,qBAAqB;AAC1C,QAAI,cAAc;AAChB,qBAAe,KAAK,2CAAkB;AACtC,qBAAe,KAAK,YAAY;AAChC,qBAAe,KAAK,SAAS;AAAA,IAC/B;AAEA,UAAM,mBAAmB,wBAAwB,YAAY,OAAO;AACpE,QAAI,iBAAiB,SAAS,GAAG;AAC/B,kBAAY,iBAAiB,IAAI,OAAK,EAAE,OAAO,EAAE,KAAK,IAAI,IAAI;AAAA,IAChE;AAEA,mBAAe;AAAA,MACb;AAAA;AAAA;AAAA,IACF;AAEA,UAAM,kBAAkB,OAAO;AAAA,MAC7B,OAAO,QAAQ,OAAO,EAAE;AAAA,QACtB,CAAC,CAAC,GAAG,MAAM,QAAQ,iBAAiB,QAAQ;AAAA,MAC9C;AAAA,IACF;AAEA,mBAAe;AAAA,MACb,GAAG,OAAO,QAAQ,eAAe,EAAE;AAAA,QACjC,CAAC,CAAC,KAAK,KAAK,MAAM,kBAAkB,GAAG,KAAK,KAAK;AAAA,MACnD;AAAA,IACF;AAAA,EACF;AAEA,SAAO,EAAE,cAAc,gBAAgB,UAAU;AACnD;AAIA,eAAe,0BACb,UACA,cACA,mBACA,OACA,QACA,SAO2B;AAC3B,QAAM,SAAS,gBAAgB;AAC/B,QAAM,eAAe,gBAAgB;AACrC,QAAM,iBAAiB,QAAQ;AAE/B,QAAM,eAAe,QAAQ,gBAAgB,aAAa,SAAS,MAAM;AACzE,MAAI;AAEJ,MAAI,cAAc;AAChB,eAAW,aAAa,YAAY,OAAO,mBAAmB;AAAA,EAChE,OAAO;AACL,eAAW,OAAO,mBAAmB;AAAA,EACvC;AAGA,MACE,aAAa,eACb,aAAa,cACb,aAAa,WACb;AACA,WAAO;AAAA,MACL;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA,EAAE,GAAG,SAAS,cAAc,eAAe;AAAA,IAC7C;AAAA,EACF;AAGA,SAAO,YAAY,UAAU,cAAc,mBAAmB,OAAO,QAAQ;AAAA,IAC3E,GAAG;AAAA,IACH;AAAA,IACA;AAAA,EACF,CAAC;AACH;AAIA,eAAsB,WACpB,cACA,UACA,eAAyB,CAAC,GAC1B,QAC2B;AAC3B,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA;AAAA;AAAA,IACA,CAAC;AAAA;AAAA,IACD,UAAU,IAAI,gBAAgB,EAAE;AAAA,IAChC;AAAA,MACE,UAAU;AAAA,MACV,OAAO;AAAA,MACP,qBAAqB;AAAA,IACvB;AAAA,EACF;AACF;AAEA,eAAsB,WAAW;AAAA,EAC/B,eAAe,CAAC;AAAA,EAChB;AAAA,EACA;AAAA,EACA,sBAAsB;AAAA,EACtB;AACF,GAM8B;AAC5B,QAAM,WAAW;AAAA,IACf;AAAA,MACE,SAAS,EAAE,MAAM,QAAQ,SAAS,WAAW;AAAA,MAC7C,MAAM;AAAA,MACN,MAAM,WAAW;AAAA,IACnB;AAAA,EACF;AAEA,SAAO,WAAW,SAAS,UAAU,cAAc,MAAM;AAC3D;",
   "names": []
 }

package/dist/services/llm/openaiProvider.js

@@ -1,14 +1,9 @@
 import { randomUUID } from "crypto";
 import { nanoid } from "nanoid";
 import { getToolDescriptionAsync } from "../../Tool.js";
-import {
-  getGlobalConfig
-} from "../../utils/config.js";
+import { getGlobalConfig } from "../../utils/config.js";
 import { isGPT5ModelName as isGPT5Model } from "../../utils/config.js";
-import {
-  createAssistantAPIErrorMessage,
-  normalizeContentFromAPI
-} from "../../utils/messages.js";
+import { normalizeContentFromAPI } from "../../utils/messages.js";
 import {
   debug as debugLogger,
   getCurrentRequest,
@@ -19,7 +14,13 @@ import { getModelManager } from "../../utils/model.js";
 import { zodToJsonSchema } from "zod-to-json-schema";
 import { ModelAdapterFactory } from "../modelAdapterFactory.js";
 import { getCLISyspromptPrefix } from "../../constants/prompts/index.js";
-import { setStreamingState } from "../../utils/streamingState.js";
+import { setStreamingState, getStreamingState } from "../../utils/streamingState.js";
+import {
+  beginTokenTracking,
+  addPreparedChars,
+  addReceivedChars,
+  setFinalTokens
+} from "../../utils/tokenProgress.js";
 import {
   getCompletionWithProfile,
   getGPT5CompletionWithProfile
@@ -29,18 +30,12 @@ import { addRetryEventToTranscript } from "../../utils/agentTranscripts.js";
 import { generateMintoContext } from "./mintoContext.js";
 import { splitSysPromptPrefix } from "./anthropicProvider.js";
 import {
-  API_ERROR_MESSAGE_PREFIX,
-  PROMPT_TOO_LONG_ERROR_MESSAGE,
-  CREDIT_BALANCE_TOO_LOW_ERROR_MESSAGE,
-  INVALID_API_KEY_ERROR_MESSAGE,
   PROMPT_CACHING_ENABLED,
   MAIN_QUERY_TEMPERATURE,
-  SONNET_COST_PER_MILLION_INPUT_TOKENS,
-  SONNET_COST_PER_MILLION_OUTPUT_TOKENS,
   withRetry,
   getMaxTokensFromProfile,
-  getModelInputTokenCostUSD,
-  getModelOutputTokenCostUSD,
+  calculateCostUSD,
+  getAssistantMessageFromError,
   addToTotalCost,
   recordTokenUsage,
   logError
@@ -150,6 +145,7 @@ function messageReducer(previous, item) {
 }
 async function handleMessageStream(stream, signal) {
   const streamStartTime = Date.now();
+  setStreamingState({ receivedChars: 0 });
   let ttftMs;
   let chunkCount = 0;
   let errorCount = 0;
@@ -191,13 +187,20 @@ async function handleMessageStream(stream, signal) {
         }
         if (!usage && chunk.usage) {
           usage = chunk.usage;
-          setStreamingState({
-            inputTokens: chunk.usage.prompt_tokens,
-            outputTokens: chunk.usage.completion_tokens
+          setFinalTokens({
+            input: chunk.usage.prompt_tokens,
+            output: chunk.usage.completion_tokens
           });
         }
         message = messageReducer(message, chunk);
         if (chunk?.choices?.[0]?.delta?.content) {
+          const textChunk = chunk.choices[0].delta.content;
+          addReceivedChars(textChunk.length);
+          const prevText = getStreamingState().streamingText || "";
+          setStreamingState({
+            streamingText: prevText + textChunk,
+            phase: "generating"
+          });
           if (!ttftMs) {
             ttftMs = Date.now() - streamStartTime;
             debugLogger.api("OPENAI_STREAM_FIRST_TOKEN", {
@@ -290,20 +293,6 @@ function convertOpenAIResponseToAnthropic(response, _tools) {
       });
     }
   }
-  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",
@@ -320,33 +309,11 @@ function convertOpenAIResponseToAnthropic(response, _tools) {
   };
   return finalMessage;
 }
-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") {
-      debugLogger.error("OPENAI_API_ERROR", {
-        message: error.message,
-        stack: error.stack
-      });
-    }
-    return createAssistantAPIErrorMessage(
-      `${API_ERROR_MESSAGE_PREFIX}: ${error.message}`
-    );
-  }
-  return createAssistantAPIErrorMessage(API_ERROR_MESSAGE_PREFIX);
-}
 async function queryOpenAI(messages, systemPrompt, maxThinkingTokens, tools, signal, options) {
   const config = getGlobalConfig();
   const modelManager = getModelManager();
   const toolUseContext = options?.toolUseContext;
+  beginTokenTracking();
   const modelProfile = options?.modelProfile || modelManager.getModel("main");
   let model;
   const currentRequest = getCurrentRequest();
@@ -367,7 +334,6 @@ async function queryOpenAI(messages, systemPrompt, maxThinkingTokens, tools, sig
     model = options?.model || modelProfile?.modelName || "";
   }
   if (options?.prependCLISysprompt) {
-    const [firstSyspromptBlock] = splitSysPromptPrefix(systemPrompt);
     systemPrompt = [getCLISyspromptPrefix(), ...systemPrompt];
   }
   const system = splitSysPromptPrefix(systemPrompt).map(
@@ -377,17 +343,20 @@ async function queryOpenAI(messages, systemPrompt, maxThinkingTokens, tools, sig
       type: "text"
     })
   );
+  addPreparedChars(JSON.stringify(system).length);
   const toolSchemas = await Promise.all(
-    tools.map(
-      async (_) => ({
+    tools.map(async (_) => {
+      const schema = {
         type: "function",
         function: {
           name: _.name,
           description: await getToolDescriptionAsync(_),
           parameters: "inputJSONSchema" in _ && _.inputJSONSchema ? _.inputJSONSchema : zodToJsonSchema(_.inputSchema)
         }
-      })
-    )
+      };
+      addPreparedChars(JSON.stringify(schema).length);
+      return schema;
+    })
   );
   const openaiSystem = system.map(
     (s) => ({
@@ -396,6 +365,7 @@ async function queryOpenAI(messages, systemPrompt, maxThinkingTokens, tools, sig
     })
   );
   const openaiMessages = convertAnthropicMessagesToOpenAIMessages(messages);
+  addPreparedChars(JSON.stringify(openaiMessages).length);
   const startIncludingRetries = Date.now();
   logSystemPromptConstruction({
     basePrompt: systemPrompt.join("\n"),
@@ -432,7 +402,7 @@ async function queryOpenAI(messages, systemPrompt, maxThinkingTokens, tools, sig
           opts.tools = toolSchemas;
           opts.tool_choice = "auto";
         }
-        const reasoningEffort = await getReasoningEffort(modelProfile, messages);
+        const reasoningEffort = isReasoningModel ? await getReasoningEffort(modelProfile, messages) : void 0;
         if (reasoningEffort) {
           opts.reasoning_effort = reasoningEffort;
         }
@@ -445,12 +415,11 @@ async function queryOpenAI(messages, systemPrompt, maxThinkingTokens, tools, sig
             apiKeyExists: !!modelProfile.apiKey,
             requestId: getCurrentRequest()?.id
           });
-          const USE_NEW_ADAPTER_SYSTEM = process.env.USE_NEW_ADAPTERS !== "false";
-          if (USE_NEW_ADAPTER_SYSTEM) {
+          if (ModelAdapterFactory.shouldUseResponsesAPI(modelProfile)) {
             const adapter = ModelAdapterFactory.createAdapter(modelProfile);
             const unifiedParams = {
               messages: openaiMessages,
-              systemPrompt: openaiSystem.map((s) => s.content),
+              systemPrompt: openaiSystem.map((s2) => s2.content),
               tools,
               maxTokens: getMaxTokensFromProfile(modelProfile),
               stream: config.stream,
@@ -458,78 +427,84 @@ async function queryOpenAI(messages, systemPrompt, maxThinkingTokens, tools, sig
               temperature: isGPT5Model(model) || model.startsWith("o3") || model.startsWith("o4-mini") ? void 0 : MAIN_QUERY_TEMPERATURE,
               previousResponseId: toolUseContext?.responseState?.previousResponseId,
               verbosity: "high"
-              // High verbosity for coding tasks
             };
             const request = adapter.createRequest(unifiedParams);
-            if (ModelAdapterFactory.shouldUseResponsesAPI(modelProfile)) {
-              const { callGPT5ResponsesAPI } = await import("../openai.js");
-              const response2 = await callGPT5ResponsesAPI(
-                modelProfile,
-                request,
-                signal
-              );
-              const unifiedResponse = adapter.parseResponse(response2);
-              const apiMessage = {
-                role: "assistant",
-                content: unifiedResponse.content,
-                tool_calls: unifiedResponse.toolCalls,
-                usage: {
-                  prompt_tokens: unifiedResponse.usage.promptTokens,
-                  completion_tokens: unifiedResponse.usage.completionTokens
-                }
-              };
-              const assistantMsg = {
-                type: "assistant",
-                message: apiMessage,
-                costUSD: 0,
-                // Will be calculated later
-                durationMs: Date.now() - start,
-                uuid: `${Date.now()}-${Math.random().toString(36).substr(2, 9)}`,
-                responseId: unifiedResponse.responseId
-                // For state management
-              };
-              return assistantMsg;
-            } else {
-              const s = await getCompletionWithProfile(
-                modelProfile,
-                request,
-                0,
-                10,
-                signal
-              );
-              let finalResponse;
-              if (config.stream) {
-                finalResponse = await handleMessageStream(
-                  s,
-                  signal
-                );
-              } else {
-                finalResponse = s;
-              }
-              const r = convertOpenAIResponseToAnthropic(finalResponse, tools);
-              return r;
-            }
-          } else {
-            const completionFunction = isGPT5Model(modelProfile.modelName) ? getGPT5CompletionWithProfile : getCompletionWithProfile;
-            const s = await completionFunction(
+            const { callGPT5ResponsesAPI } = await import("../openai.js");
+            const response2 = await callGPT5ResponsesAPI(
               modelProfile,
-              opts,
-              0,
-              10,
+              request,
               signal
             );
-            let finalResponse;
-            if (opts.stream) {
-              finalResponse = await handleMessageStream(
-                s,
-                signal
-              );
-            } else {
-              finalResponse = s;
-            }
-            const r = convertOpenAIResponseToAnthropic(finalResponse, tools);
-            return r;
+            const unifiedResponse = adapter.parseResponse(response2);
+            const apiMessage = {
+              role: "assistant",
+              content: unifiedResponse.content,
+              tool_calls: unifiedResponse.toolCalls,
+              usage: {
+                prompt_tokens: unifiedResponse.usage.promptTokens,
+                completion_tokens: unifiedResponse.usage.completionTokens
+              }
+            };
+            const respDurationMs = Date.now() - start;
+            const respInputTokens = unifiedResponse.usage.promptTokens;
+            const respOutputTokens = unifiedResponse.usage.completionTokens;
+            const respCostUSD = calculateCostUSD({
+              model,
+              inputTokens: respInputTokens,
+              outputTokens: respOutputTokens,
+              cacheReadTokens: 0,
+              cacheCreationTokens: 0
+            });
+            addToTotalCost(respCostUSD, respDurationMs);
+            recordTokenUsage(
+              {
+                inputTokens: respInputTokens,
+                outputTokens: respOutputTokens,
+                cacheCreationTokens: 0,
+                cacheReadTokens: 0
+              },
+              respCostUSD,
+              model,
+              toolUseContext ? {
+                agentId: toolUseContext.agentId,
+                toolUseId: toolUseContext.toolUseId,
+                model
+              } : void 0
+            );
+            logLLMInteraction({
+              systemPrompt: systemPrompt.join("\n"),
+              messages: [...openaiSystem, ...openaiMessages],
+              response: apiMessage,
+              usage: {
+                inputTokens: respInputTokens,
+                outputTokens: respOutputTokens
+              },
+              timing: { start, end: Date.now() },
+              apiFormat: "openai"
+            });
+            const assistantMsg = {
+              type: "assistant",
+              message: apiMessage,
+              costUSD: respCostUSD,
+              durationMs: respDurationMs,
+              uuid: randomUUID(),
+              responseId: unifiedResponse.responseId
+            };
+            return assistantMsg;
           }
+          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 r = convertOpenAIResponseToAnthropic(finalResponse, tools);
+          return r;
         } else {
           const errorDetails = {
             modelProfileExists: !!modelProfile,
@@ -558,17 +533,21 @@ async function queryOpenAI(messages, systemPrompt, maxThinkingTokens, tools, sig
     );
   } catch (error) {
     logError(error);
-    return getAssistantMessageFromError(error);
+    return getAssistantMessageFromError(error, "OPENAI");
   }
   const durationMs = Date.now() - start;
   const durationMsIncludingRetries = Date.now() - startIncludingRetries;
   const inputTokens = response.usage?.prompt_tokens ?? 0;
   const outputTokens = response.usage?.completion_tokens ?? 0;
   const cacheReadInputTokens = response.usage?.prompt_token_details?.cached_tokens ?? 0;
-  const cacheCreationInputTokens = response.usage?.prompt_token_details?.cached_tokens ?? 0;
-  const inputCostPerMillion = getModelInputTokenCostUSD(model) * 1e6;
-  const outputCostPerMillion = getModelOutputTokenCostUSD(model) * 1e6;
-  const costUSD = inputTokens / 1e6 * (inputCostPerMillion || SONNET_COST_PER_MILLION_INPUT_TOKENS) + outputTokens / 1e6 * (outputCostPerMillion || SONNET_COST_PER_MILLION_OUTPUT_TOKENS) + cacheReadInputTokens / 1e6 * (inputCostPerMillion || SONNET_COST_PER_MILLION_INPUT_TOKENS) * 0.1 + cacheCreationInputTokens / 1e6 * (inputCostPerMillion || SONNET_COST_PER_MILLION_INPUT_TOKENS);
+  const cacheCreationInputTokens = 0;
+  const costUSD = calculateCostUSD({
+    model,
+    inputTokens,
+    outputTokens,
+    cacheReadTokens: cacheReadInputTokens,
+    cacheCreationTokens: cacheCreationInputTokens
+  });
   addToTotalCost(costUSD, durationMsIncludingRetries);
   recordTokenUsage(
     {