@within-7/minto 0.4.1 → 0.4.2

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

@@ -1,7 +1,7 @@
 {
   "version": 3,
   "sources": ["../../../src/services/llm/openaiProvider.ts"],
-  "sourcesContent": ["/**\n * OpenAI-compatible provider \u2014 message conversion, streaming, and query path.\n */\nimport { randomUUID } from 'crypto'\nimport { nanoid } from 'nanoid'\n\nimport type { AssistantMessage, UserMessage } from '@query'\nimport { Tool, getToolDescriptionAsync } 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  createAssistantAPIErrorMessage,\n  normalizeContentFromAPI,\n} from '@utils/messages'\nimport {\n  debug as debugLogger,\n  getCurrentRequest,\n  logLLMInteraction,\n  logSystemPromptConstruction,\n} from '@utils/debugLogger'\nimport { getModelManager } from '@utils/model'\nimport { zodToJsonSchema } from 'zod-to-json-schema'\nimport { ModelAdapterFactory } from '../modelAdapterFactory'\nimport { UnifiedRequestParams } from '@minto-types/modelCapabilities'\nimport { getCLISyspromptPrefix } from '@constants/prompts'\nimport OpenAI from 'openai'\nimport type { ChatCompletionStream } from 'openai/lib/ChatCompletionStream'\nimport { ContentBlock } from '@anthropic-ai/sdk/resources/messages/messages'\nimport type { TextBlockParam } from '@anthropic-ai/sdk/resources/index.mjs'\nimport { setStreamingState } from '@utils/streamingState'\nimport {\n  getCompletionWithProfile,\n  getGPT5CompletionWithProfile,\n} from '../openai'\nimport { getReasoningEffort } from '@utils/thinking'\nimport { addRetryEventToTranscript } from '@utils/agentTranscripts'\nimport type { TokenTrackingContext } from '@core/tokenStats'\n\nimport { generateMintoContext } from './mintoContext'\nimport { splitSysPromptPrefix } from './anthropicProvider'\nimport {\n  API_ERROR_MESSAGE_PREFIX,\n  PROMPT_TOO_LONG_ERROR_MESSAGE,\n  CREDIT_BALANCE_TOO_LOW_ERROR_MESSAGE,\n  INVALID_API_KEY_ERROR_MESSAGE,\n  PROMPT_CACHING_ENABLED,\n  MAIN_QUERY_TEMPERATURE,\n  SONNET_COST_PER_MILLION_INPUT_TOKENS,\n  SONNET_COST_PER_MILLION_OUTPUT_TOKENS,\n  withRetry,\n  getMaxTokensFromProfile,\n  getModelInputTokenCostUSD,\n  getModelOutputTokenCostUSD,\n  addToTotalCost,\n  recordTokenUsage,\n  logError,\n} from './types'\n\n// \u2500\u2500 OpenAI message conversion \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\nfunction convertAnthropicMessagesToOpenAIMessages(\n  messages: (UserMessage | AssistantMessage)[],\n): (\n  | OpenAI.ChatCompletionMessageParam\n  | OpenAI.ChatCompletionToolMessageParam\n)[] {\n  const openaiMessages: (\n    | OpenAI.ChatCompletionMessageParam\n    | OpenAI.ChatCompletionToolMessageParam\n  )[] = []\n\n  const toolResults: Record<string, OpenAI.ChatCompletionToolMessageParam> = {}\n\n  for (const message of messages) {\n    let contentBlocks = []\n    if (typeof message.message.content === 'string') {\n      contentBlocks = [\n        {\n          type: 'text',\n          text: message.message.content,\n        },\n      ]\n    } else if (!Array.isArray(message.message.content)) {\n      contentBlocks = [message.message.content]\n    } else {\n      contentBlocks = message.message.content\n    }\n\n    for (const block of contentBlocks) {\n      if (block.type === 'text') {\n        openaiMessages.push({\n          role: message.message.role,\n          content: block.text,\n        })\n      } else if (block.type === 'tool_use') {\n        openaiMessages.push({\n          role: 'assistant',\n          content: undefined,\n          tool_calls: [\n            {\n              type: 'function',\n              function: {\n                name: block.name,\n                arguments: JSON.stringify(block.input),\n              },\n              id: block.id,\n            },\n          ],\n        })\n      } else if (block.type === 'tool_result') {\n        // Ensure content is always a string for role:tool messages\n        let toolContent = block.content\n        if (typeof toolContent !== 'string') {\n          toolContent = JSON.stringify(toolContent)\n        }\n\n        toolResults[block.tool_use_id] = {\n          role: 'tool',\n          content: toolContent,\n          tool_call_id: block.tool_use_id,\n        }\n      }\n    }\n  }\n\n  const finalMessages: (\n    | OpenAI.ChatCompletionMessageParam\n    | OpenAI.ChatCompletionToolMessageParam\n  )[] = []\n\n  for (const message of openaiMessages) {\n    finalMessages.push(message)\n\n    if ('tool_calls' in message && message.tool_calls) {\n      for (const toolCall of message.tool_calls) {\n        if (toolResults[toolCall.id]) {\n          finalMessages.push(toolResults[toolCall.id])\n        }\n      }\n    }\n  }\n\n  return finalMessages\n}\n\nfunction messageReducer(\n  previous: OpenAI.ChatCompletionMessage,\n  item: OpenAI.ChatCompletionChunk,\n): OpenAI.ChatCompletionMessage {\n  const reduce = (acc: any, delta: OpenAI.ChatCompletionChunk.Choice.Delta) => {\n    acc = { ...acc }\n    for (const [key, value] of Object.entries(delta)) {\n      if (acc[key] === undefined || acc[key] === null) {\n        acc[key] = value\n        if (Array.isArray(acc[key])) {\n          for (const arr of acc[key]) {\n            delete arr.index\n          }\n        }\n      } else if (typeof acc[key] === 'string' && typeof value === 'string') {\n        acc[key] += value\n      } else if (typeof acc[key] === 'number' && typeof value === 'number') {\n        acc[key] = value\n      } else if (Array.isArray(acc[key]) && Array.isArray(value)) {\n        const accArray = acc[key]\n        for (let i = 0; i < value.length; i++) {\n          const { index: rawIndex, ...chunkTool } = value[i]\n          // Default index to 0 when missing (Gemini doesn't include index in tool call deltas)\n          const index = rawIndex ?? 0\n          if (index - accArray.length > 1) {\n            throw new Error(\n              `Error: An array has an empty value when tool_calls are constructed. tool_calls: ${accArray}; tool: ${value}`,\n            )\n          }\n          accArray[index] = reduce(accArray[index], chunkTool)\n        }\n      } else if (typeof acc[key] === 'object' && typeof value === 'object') {\n        acc[key] = reduce(acc[key], value)\n      }\n    }\n    return acc\n  }\n\n  const choice = item.choices?.[0]\n  if (!choice) {\n    return previous\n  }\n  return reduce(previous, choice.delta) as OpenAI.ChatCompletionMessage\n}\n\nasync function handleMessageStream(\n  stream: ChatCompletionStream,\n  signal?: AbortSignal,\n): Promise<OpenAI.ChatCompletion> {\n  const streamStartTime = Date.now()\n  let ttftMs: number | undefined\n  let chunkCount = 0\n  let errorCount = 0\n\n  debugLogger.api('OPENAI_STREAM_START', {\n    streamStartTime: String(streamStartTime),\n  })\n\n  let message = {} as OpenAI.ChatCompletionMessage\n\n  let id, model, created, object, usage\n  try {\n    for await (const chunk of stream) {\n      if (signal?.aborted) {\n        debugLogger.flow('OPENAI_STREAM_ABORTED', {\n          chunkCount,\n          timestamp: Date.now(),\n        })\n        throw new Error('Request was cancelled')\n      }\n\n      chunkCount++\n\n      try {\n        if (!id) {\n          id = chunk.id\n          debugLogger.api('OPENAI_STREAM_ID_RECEIVED', {\n            id,\n            chunkNumber: String(chunkCount),\n          })\n        }\n        if (!model) {\n          model = chunk.model\n          debugLogger.api('OPENAI_STREAM_MODEL_RECEIVED', {\n            model,\n            chunkNumber: String(chunkCount),\n          })\n        }\n        if (!created) {\n          created = chunk.created\n        }\n        if (!object) {\n          object = chunk.object\n        }\n        if (!usage && chunk.usage) {\n          usage = chunk.usage\n          setStreamingState({\n            inputTokens: chunk.usage.prompt_tokens,\n            outputTokens: chunk.usage.completion_tokens,\n          })\n        }\n\n        message = messageReducer(message, chunk)\n\n        if (chunk?.choices?.[0]?.delta?.content) {\n          if (!ttftMs) {\n            ttftMs = Date.now() - streamStartTime\n            debugLogger.api('OPENAI_STREAM_FIRST_TOKEN', {\n              ttftMs: String(ttftMs),\n              chunkNumber: String(chunkCount),\n            })\n          }\n        }\n      } catch (chunkError) {\n        errorCount++\n        debugLogger.error('OPENAI_STREAM_CHUNK_ERROR', {\n          chunkNumber: String(chunkCount),\n          errorMessage:\n            chunkError instanceof Error\n              ? chunkError.message\n              : String(chunkError),\n          errorType:\n            chunkError instanceof Error\n              ? chunkError.constructor.name\n              : typeof chunkError,\n        })\n      }\n    }\n\n    debugLogger.api('OPENAI_STREAM_COMPLETE', {\n      totalChunks: String(chunkCount),\n      errorCount: String(errorCount),\n      totalDuration: String(Date.now() - streamStartTime),\n      ttftMs: String(ttftMs || 0),\n      finalMessageId: id || 'undefined',\n    })\n  } catch (streamError) {\n    debugLogger.error('OPENAI_STREAM_FATAL_ERROR', {\n      totalChunks: String(chunkCount),\n      errorCount: String(errorCount),\n      errorMessage:\n        streamError instanceof Error\n          ? streamError.message\n          : String(streamError),\n      errorType:\n        streamError instanceof Error\n          ? streamError.constructor.name\n          : typeof streamError,\n    })\n    throw streamError\n  }\n  return {\n    id,\n    created,\n    model,\n    object,\n    choices: [\n      {\n        index: 0,\n        message,\n        finish_reason: 'stop',\n        logprobs: undefined,\n      },\n    ],\n    usage,\n  }\n}\n\nfunction convertOpenAIResponseToAnthropic(\n  response: OpenAI.ChatCompletion,\n  _tools?: Tool[],\n) {\n  let contentBlocks: ContentBlock[] = []\n  const message = response.choices?.[0]?.message\n  if (!message) {\n    return {\n      role: 'assistant',\n      content: [],\n      stop_reason: response.choices?.[0]?.finish_reason,\n      type: 'message',\n      usage: response.usage,\n    }\n  }\n\n  if (message?.tool_calls) {\n    for (const toolCall of message.tool_calls) {\n      const tool = toolCall.function\n      const toolName = tool?.name\n      let toolArgs = {}\n      let parseError: string | null = null\n      try {\n        toolArgs = tool?.arguments ? JSON.parse(tool.arguments) : {}\n      } catch (e) {\n        parseError = e instanceof Error ? e.message : String(e)\n        debugLogger.warn('TOOL_ARGUMENTS_PARSE_ERROR', {\n          toolName,\n          rawArguments: tool?.arguments?.substring(0, 200),\n          errorMessage: parseError,\n        })\n      }\n\n      if (parseError && Object.keys(toolArgs).length === 0) {\n        debugLogger.error('TOOL_CALL_EMPTY_ARGUMENTS', {\n          toolName,\n          parseError,\n          suggestion:\n            'Model may have sent malformed JSON. The tool call will likely fail validation.',\n        })\n      }\n\n      contentBlocks.push({\n        type: 'tool_use',\n        input: toolArgs,\n        name: toolName,\n        id: toolCall.id?.length > 0 ? toolCall.id : nanoid(),\n      })\n    }\n  }\n\n  if ((message as any).reasoning) {\n    contentBlocks.push({\n      type: 'thinking',\n      thinking: (message as any).reasoning,\n      signature: '',\n    })\n  }\n\n  if ((message as any).reasoning_content) {\n    contentBlocks.push({\n      type: 'thinking',\n      thinking: (message as any).reasoning_content,\n      signature: '',\n    })\n  }\n\n  if (message.content) {\n    contentBlocks.push({\n      type: 'text',\n      text: message?.content,\n      citations: [],\n    })\n  }\n\n  const finalMessage = {\n    role: 'assistant',\n    content: contentBlocks,\n    stop_reason: response.choices?.[0]?.finish_reason,\n    type: 'message',\n    usage: response.usage,\n  }\n\n  return finalMessage\n}\n\n// \u2500\u2500 Error helpers \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\nfunction getAssistantMessageFromError(error: unknown): AssistantMessage {\n  if (error instanceof Error && error.message.includes('prompt is too long')) {\n    return createAssistantAPIErrorMessage(PROMPT_TOO_LONG_ERROR_MESSAGE)\n  }\n  if (\n    error instanceof Error &&\n    error.message.includes('Your credit balance is too low')\n  ) {\n    return createAssistantAPIErrorMessage(CREDIT_BALANCE_TOO_LOW_ERROR_MESSAGE)\n  }\n  if (\n    error instanceof Error &&\n    error.message.toLowerCase().includes('x-api-key')\n  ) {\n    return createAssistantAPIErrorMessage(INVALID_API_KEY_ERROR_MESSAGE)\n  }\n  if (error instanceof Error) {\n    if (process.env.NODE_ENV === 'development') {\n      debugLogger.error('OPENAI_API_ERROR', {\n        message: error.message,\n        stack: error.stack,\n      })\n    }\n    return createAssistantAPIErrorMessage(\n      `${API_ERROR_MESSAGE_PREFIX}: ${error.message}`,\n    )\n  }\n  return createAssistantAPIErrorMessage(API_ERROR_MESSAGE_PREFIX)\n}\n\n// \u2500\u2500 queryOpenAI \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 queryOpenAI(\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 model: string\n\n  const currentRequest = getCurrentRequest()\n  debugLogger.api('MODEL_CONFIG_OPENAI', {\n    modelProfileFound: !!modelProfile,\n    modelProfileId: modelProfile?.modelName,\n    modelProfileName: modelProfile?.name,\n    modelProfileModelName: modelProfile?.modelName,\n    modelProfileProvider: modelProfile?.provider,\n    modelProfileBaseURL: modelProfile?.baseURL,\n    modelProfileApiKeyExists: !!modelProfile?.apiKey,\n    optionsModel: options?.model,\n    requestId: getCurrentRequest()?.id,\n  })\n\n  if (modelProfile) {\n    model = modelProfile.modelName\n  } else {\n    model = options?.model || modelProfile?.modelName || ''\n  }\n  // Prepend system prompt block for easy API identification\n  if (options?.prependCLISysprompt) {\n    const [firstSyspromptBlock] = splitSysPromptPrefix(systemPrompt)\n    systemPrompt = [getCLISyspromptPrefix(), ...systemPrompt]\n  }\n\n  const system: TextBlockParam[] = splitSysPromptPrefix(systemPrompt).map(\n    _ => ({\n      ...(PROMPT_CACHING_ENABLED\n        ? { cache_control: { type: 'ephemeral' } }\n        : {}),\n      text: _,\n      type: 'text',\n    }),\n  )\n\n  const toolSchemas = await Promise.all(\n    tools.map(\n      async _ =>\n        ({\n          type: 'function',\n          function: {\n            name: _.name,\n            description: await getToolDescriptionAsync(_),\n            parameters:\n              'inputJSONSchema' in _ && _.inputJSONSchema\n                ? _.inputJSONSchema\n                : zodToJsonSchema(_.inputSchema),\n          },\n        }) as OpenAI.ChatCompletionTool,\n    ),\n  )\n\n  const openaiSystem = system.map(\n    s =>\n      ({\n        role: 'system',\n        content: s.text,\n      }) as OpenAI.ChatCompletionMessageParam,\n  )\n\n  const openaiMessages = convertAnthropicMessagesToOpenAIMessages(messages)\n  const startIncludingRetries = Date.now()\n\n  logSystemPromptConstruction({\n    basePrompt: systemPrompt.join('\\n'),\n    mintoContext: generateMintoContext() || '',\n    reminders: [],\n    finalPrompt: systemPrompt.join('\\n'),\n  })\n\n  let start = Date.now()\n  let attemptNumber = 0\n  let response\n\n  try {\n    response = await withRetry(\n      async attempt => {\n        attemptNumber = attempt\n        start = Date.now()\n        const maxTokens = getMaxTokensFromProfile(modelProfile)\n        const isGPT5 = isGPT5Model(model)\n\n        // Determine if this model requires temperature to be omitted entirely\n        const isReasoningModel =\n          isGPT5 ||\n          model.startsWith('o3') ||\n          model.startsWith('o4-mini') ||\n          model === 'o1' ||\n          model === 'o1-mini' ||\n          model === 'o1-preview'\n\n        const opts: OpenAI.ChatCompletionCreateParams = {\n          model,\n\n          ...(isGPT5\n            ? { max_completion_tokens: maxTokens }\n            : { max_tokens: maxTokens }),\n          messages: [...openaiSystem, ...openaiMessages],\n\n          // Reasoning models (GPT-5, o3, o4-mini) require temperature to be omitted\n          ...(!isReasoningModel && { temperature: MAIN_QUERY_TEMPERATURE }),\n        }\n        if (config.stream) {\n          ;(opts as OpenAI.ChatCompletionCreateParams).stream = true\n          opts.stream_options = {\n            include_usage: true,\n          }\n        }\n\n        if (toolSchemas.length > 0) {\n          opts.tools = toolSchemas\n          opts.tool_choice = 'auto'\n        }\n        const reasoningEffort = await getReasoningEffort(modelProfile, messages)\n        if (reasoningEffort) {\n          opts.reasoning_effort = reasoningEffort\n        }\n\n        if (modelProfile && modelProfile.modelName) {\n          debugLogger.api('USING_MODEL_PROFILE_PATH', {\n            modelProfileName: modelProfile.modelName,\n            modelName: modelProfile.modelName,\n            provider: modelProfile.provider,\n            baseURL: modelProfile.baseURL,\n            apiKeyExists: !!modelProfile.apiKey,\n            requestId: getCurrentRequest()?.id,\n          })\n\n          // Enable new adapter system with environment variable\n          const USE_NEW_ADAPTER_SYSTEM =\n            process.env.USE_NEW_ADAPTERS !== 'false'\n\n          if (USE_NEW_ADAPTER_SYSTEM) {\n            // New adapter system\n            const adapter = ModelAdapterFactory.createAdapter(modelProfile)\n\n            // Build unified request parameters\n            const unifiedParams: UnifiedRequestParams = {\n              messages: openaiMessages,\n              systemPrompt: openaiSystem.map(s => s.content as string),\n              tools: tools,\n              maxTokens: getMaxTokensFromProfile(modelProfile),\n              stream: config.stream,\n              reasoningEffort: reasoningEffort as any,\n              temperature:\n                isGPT5Model(model) ||\n                model.startsWith('o3') ||\n                model.startsWith('o4-mini')\n                  ? undefined\n                  : MAIN_QUERY_TEMPERATURE,\n              previousResponseId:\n                toolUseContext?.responseState?.previousResponseId,\n              verbosity: 'high', // High verbosity for coding tasks\n            }\n\n            // Create request using adapter\n            const request = adapter.createRequest(unifiedParams)\n\n            // Determine which API to use\n            if (ModelAdapterFactory.shouldUseResponsesAPI(modelProfile)) {\n              // Use Responses API for GPT-5 and similar models\n              const { callGPT5ResponsesAPI } = await import('../openai')\n              const response = await callGPT5ResponsesAPI(\n                modelProfile,\n                request,\n                signal,\n              )\n              const unifiedResponse = adapter.parseResponse(response)\n\n              // Convert unified response back to Anthropic format\n              const apiMessage = {\n                role: 'assistant' as const,\n                content: unifiedResponse.content,\n                tool_calls: unifiedResponse.toolCalls,\n                usage: {\n                  prompt_tokens: unifiedResponse.usage.promptTokens,\n                  completion_tokens: unifiedResponse.usage.completionTokens,\n                },\n              }\n              const assistantMsg: AssistantMessage = {\n                type: 'assistant',\n                message: apiMessage as any,\n                costUSD: 0, // Will be calculated later\n                durationMs: Date.now() - start,\n                uuid: `${Date.now()}-${Math.random().toString(36).substr(2, 9)}` as any,\n                responseId: unifiedResponse.responseId, // For state management\n              }\n              return assistantMsg\n            } else {\n              // Use existing Chat Completions flow\n              const s = await getCompletionWithProfile(\n                modelProfile,\n                request,\n                0,\n                10,\n                signal,\n              )\n              let finalResponse\n              if (config.stream) {\n                finalResponse = await handleMessageStream(\n                  s as ChatCompletionStream,\n                  signal,\n                )\n              } else {\n                finalResponse = s\n              }\n              const r = convertOpenAIResponseToAnthropic(finalResponse, tools)\n              return r\n            }\n          } else {\n            // Legacy system (preserved for fallback)\n            const completionFunction = isGPT5Model(modelProfile.modelName)\n              ? getGPT5CompletionWithProfile\n              : getCompletionWithProfile\n            const s = await completionFunction(\n              modelProfile,\n              opts,\n              0,\n              10,\n              signal,\n            )\n            let finalResponse\n            if (opts.stream) {\n              finalResponse = await handleMessageStream(\n                s as ChatCompletionStream,\n                signal,\n              )\n            } else {\n              finalResponse = s\n            }\n            const r = convertOpenAIResponseToAnthropic(finalResponse, tools)\n            return r\n          }\n        } else {\n          const errorDetails = {\n            modelProfileExists: !!modelProfile,\n            modelProfileId: modelProfile?.modelName,\n            modelNameExists: !!modelProfile?.modelName,\n            requestedModel: model,\n            requestId: getCurrentRequest()?.id,\n          }\n          debugLogger.error('NO_VALID_MODEL_PROFILE', errorDetails)\n          throw new Error(\n            `No valid ModelProfile available for model: ${model}. Please configure model through /model command. Debug: ${JSON.stringify(errorDetails)}`,\n          )\n        }\n      },\n      {\n        signal,\n        onRetry: toolUseContext?.agentId\n          ? ({ attempt, maxRetries, error, delayMs }) => {\n              addRetryEventToTranscript(toolUseContext.agentId!, {\n                attempt,\n                maxRetries,\n                errorMessage: error.message,\n                delayMs,\n              })\n            }\n          : undefined,\n      },\n    )\n  } catch (error) {\n    logError(error)\n    return getAssistantMessageFromError(error)\n  }\n  const durationMs = Date.now() - start\n  const durationMsIncludingRetries = Date.now() - startIncludingRetries\n\n  const inputTokens = response.usage?.prompt_tokens ?? 0\n  const outputTokens = response.usage?.completion_tokens ?? 0\n  const cacheReadInputTokens =\n    response.usage?.prompt_token_details?.cached_tokens ?? 0\n  const cacheCreationInputTokens =\n    response.usage?.prompt_token_details?.cached_tokens ?? 0\n\n  // Use per-model pricing from models registry, falling back to defaults for unknown models\n  const inputCostPerMillion = getModelInputTokenCostUSD(model) * 1_000_000\n  const outputCostPerMillion = getModelOutputTokenCostUSD(model) * 1_000_000\n  const costUSD =\n    (inputTokens / 1_000_000) *\n      (inputCostPerMillion || SONNET_COST_PER_MILLION_INPUT_TOKENS) +\n    (outputTokens / 1_000_000) *\n      (outputCostPerMillion || SONNET_COST_PER_MILLION_OUTPUT_TOKENS) +\n    (cacheReadInputTokens / 1_000_000) *\n      (inputCostPerMillion || SONNET_COST_PER_MILLION_INPUT_TOKENS) *\n      0.1 +\n    (cacheCreationInputTokens / 1_000_000) *\n      (inputCostPerMillion || SONNET_COST_PER_MILLION_INPUT_TOKENS)\n\n  addToTotalCost(costUSD, durationMsIncludingRetries)\n\n  // Record token usage to unified stats manager\n  recordTokenUsage(\n    {\n      inputTokens,\n      outputTokens,\n      cacheCreationTokens: cacheCreationInputTokens,\n      cacheReadTokens: cacheReadInputTokens,\n    },\n    costUSD,\n    model,\n    toolUseContext\n      ? ({\n          agentId: toolUseContext.agentId,\n          toolUseId: toolUseContext.toolUseId,\n          model,\n        } as TokenTrackingContext)\n      : undefined,\n  )\n\n  // Log LLM interaction (OpenAI path)\n  logLLMInteraction({\n    systemPrompt: systemPrompt.join('\\n'),\n    messages: [...openaiSystem, ...openaiMessages],\n    response: response,\n    usage: {\n      inputTokens: inputTokens,\n      outputTokens: outputTokens,\n    },\n    timing: {\n      start: start,\n      end: Date.now(),\n    },\n    apiFormat: 'openai',\n  })\n\n  return {\n    message: {\n      ...response,\n      content: normalizeContentFromAPI(response.content),\n      usage: {\n        input_tokens: inputTokens,\n        output_tokens: outputTokens,\n        cache_read_input_tokens: cacheReadInputTokens,\n        cache_creation_input_tokens: 0,\n      },\n    },\n    costUSD,\n    durationMs,\n    type: 'assistant',\n    uuid: randomUUID(),\n  }\n}\n"],
-  "mappings": "AAGA,SAAS,kBAAkB;AAC3B,SAAS,cAAc;AAGvB,SAAe,+BAA+B;AAE9C;AAAA,EACE;AAAA,OAEK;AACP,SAAS,mBAAmB,mBAAmB;AAC/C;AAAA,EACE;AAAA,EACA;AAAA,OACK;AACP;AAAA,EACE,SAAS;AAAA,EACT;AAAA,EACA;AAAA,EACA;AAAA,OACK;AACP,SAAS,uBAAuB;AAChC,SAAS,uBAAuB;AAChC,SAAS,2BAA2B;AAEpC,SAAS,6BAA6B;AAKtC,SAAS,yBAAyB;AAClC;AAAA,EACE;AAAA,EACA;AAAA,OACK;AACP,SAAS,0BAA0B;AACnC,SAAS,iCAAiC;AAG1C,SAAS,4BAA4B;AACrC,SAAS,4BAA4B;AACrC;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OACK;AAIP,SAAS,yCACP,UAIE;AACF,QAAM,iBAGA,CAAC;AAEP,QAAM,cAAqE,CAAC;AAE5E,aAAW,WAAW,UAAU;AAC9B,QAAI,gBAAgB,CAAC;AACrB,QAAI,OAAO,QAAQ,QAAQ,YAAY,UAAU;AAC/C,sBAAgB;AAAA,QACd;AAAA,UACE,MAAM;AAAA,UACN,MAAM,QAAQ,QAAQ;AAAA,QACxB;AAAA,MACF;AAAA,IACF,WAAW,CAAC,MAAM,QAAQ,QAAQ,QAAQ,OAAO,GAAG;AAClD,sBAAgB,CAAC,QAAQ,QAAQ,OAAO;AAAA,IAC1C,OAAO;AACL,sBAAgB,QAAQ,QAAQ;AAAA,IAClC;AAEA,eAAW,SAAS,eAAe;AACjC,UAAI,MAAM,SAAS,QAAQ;AACzB,uBAAe,KAAK;AAAA,UAClB,MAAM,QAAQ,QAAQ;AAAA,UACtB,SAAS,MAAM;AAAA,QACjB,CAAC;AAAA,MACH,WAAW,MAAM,SAAS,YAAY;AACpC,uBAAe,KAAK;AAAA,UAClB,MAAM;AAAA,UACN,SAAS;AAAA,UACT,YAAY;AAAA,YACV;AAAA,cACE,MAAM;AAAA,cACN,UAAU;AAAA,gBACR,MAAM,MAAM;AAAA,gBACZ,WAAW,KAAK,UAAU,MAAM,KAAK;AAAA,cACvC;AAAA,cACA,IAAI,MAAM;AAAA,YACZ;AAAA,UACF;AAAA,QACF,CAAC;AAAA,MACH,WAAW,MAAM,SAAS,eAAe;AAEvC,YAAI,cAAc,MAAM;AACxB,YAAI,OAAO,gBAAgB,UAAU;AACnC,wBAAc,KAAK,UAAU,WAAW;AAAA,QAC1C;AAEA,oBAAY,MAAM,WAAW,IAAI;AAAA,UAC/B,MAAM;AAAA,UACN,SAAS;AAAA,UACT,cAAc,MAAM;AAAA,QACtB;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAEA,QAAM,gBAGA,CAAC;AAEP,aAAW,WAAW,gBAAgB;AACpC,kBAAc,KAAK,OAAO;AAE1B,QAAI,gBAAgB,WAAW,QAAQ,YAAY;AACjD,iBAAW,YAAY,QAAQ,YAAY;AACzC,YAAI,YAAY,SAAS,EAAE,GAAG;AAC5B,wBAAc,KAAK,YAAY,SAAS,EAAE,CAAC;AAAA,QAC7C;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAEA,SAAO;AACT;AAEA,SAAS,eACP,UACA,MAC8B;AAC9B,QAAM,SAAS,CAAC,KAAU,UAAmD;AAC3E,UAAM,EAAE,GAAG,IAAI;AACf,eAAW,CAAC,KAAK,KAAK,KAAK,OAAO,QAAQ,KAAK,GAAG;AAChD,UAAI,IAAI,GAAG,MAAM,UAAa,IAAI,GAAG,MAAM,MAAM;AAC/C,YAAI,GAAG,IAAI;AACX,YAAI,MAAM,QAAQ,IAAI,GAAG,CAAC,GAAG;AAC3B,qBAAW,OAAO,IAAI,GAAG,GAAG;AAC1B,mBAAO,IAAI;AAAA,UACb;AAAA,QACF;AAAA,MACF,WAAW,OAAO,IAAI,GAAG,MAAM,YAAY,OAAO,UAAU,UAAU;AACpE,YAAI,GAAG,KAAK;AAAA,MACd,WAAW,OAAO,IAAI,GAAG,MAAM,YAAY,OAAO,UAAU,UAAU;AACpE,YAAI,GAAG,IAAI;AAAA,MACb,WAAW,MAAM,QAAQ,IAAI,GAAG,CAAC,KAAK,MAAM,QAAQ,KAAK,GAAG;AAC1D,cAAM,WAAW,IAAI,GAAG;AACxB,iBAAS,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;AACrC,gBAAM,EAAE,OAAO,UAAU,GAAG,UAAU,IAAI,MAAM,CAAC;AAEjD,gBAAM,QAAQ,YAAY;AAC1B,cAAI,QAAQ,SAAS,SAAS,GAAG;AAC/B,kBAAM,IAAI;AAAA,cACR,mFAAmF,QAAQ,WAAW,KAAK;AAAA,YAC7G;AAAA,UACF;AACA,mBAAS,KAAK,IAAI,OAAO,SAAS,KAAK,GAAG,SAAS;AAAA,QACrD;AAAA,MACF,WAAW,OAAO,IAAI,GAAG,MAAM,YAAY,OAAO,UAAU,UAAU;AACpE,YAAI,GAAG,IAAI,OAAO,IAAI,GAAG,GAAG,KAAK;AAAA,MACnC;AAAA,IACF;AACA,WAAO;AAAA,EACT;AAEA,QAAM,SAAS,KAAK,UAAU,CAAC;AAC/B,MAAI,CAAC,QAAQ;AACX,WAAO;AAAA,EACT;AACA,SAAO,OAAO,UAAU,OAAO,KAAK;AACtC;AAEA,eAAe,oBACb,QACA,QACgC;AAChC,QAAM,kBAAkB,KAAK,IAAI;AACjC,MAAI;AACJ,MAAI,aAAa;AACjB,MAAI,aAAa;AAEjB,cAAY,IAAI,uBAAuB;AAAA,IACrC,iBAAiB,OAAO,eAAe;AAAA,EACzC,CAAC;AAED,MAAI,UAAU,CAAC;AAEf,MAAI,IAAI,OAAO,SAAS,QAAQ;AAChC,MAAI;AACF,qBAAiB,SAAS,QAAQ;AAChC,UAAI,QAAQ,SAAS;AACnB,oBAAY,KAAK,yBAAyB;AAAA,UACxC;AAAA,UACA,WAAW,KAAK,IAAI;AAAA,QACtB,CAAC;AACD,cAAM,IAAI,MAAM,uBAAuB;AAAA,MACzC;AAEA;AAEA,UAAI;AACF,YAAI,CAAC,IAAI;AACP,eAAK,MAAM;AACX,sBAAY,IAAI,6BAA6B;AAAA,YAC3C;AAAA,YACA,aAAa,OAAO,UAAU;AAAA,UAChC,CAAC;AAAA,QACH;AACA,YAAI,CAAC,OAAO;AACV,kBAAQ,MAAM;AACd,sBAAY,IAAI,gCAAgC;AAAA,YAC9C;AAAA,YACA,aAAa,OAAO,UAAU;AAAA,UAChC,CAAC;AAAA,QACH;AACA,YAAI,CAAC,SAAS;AACZ,oBAAU,MAAM;AAAA,QAClB;AACA,YAAI,CAAC,QAAQ;AACX,mBAAS,MAAM;AAAA,QACjB;AACA,YAAI,CAAC,SAAS,MAAM,OAAO;AACzB,kBAAQ,MAAM;AACd,4BAAkB;AAAA,YAChB,aAAa,MAAM,MAAM;AAAA,YACzB,cAAc,MAAM,MAAM;AAAA,UAC5B,CAAC;AAAA,QACH;AAEA,kBAAU,eAAe,SAAS,KAAK;AAEvC,YAAI,OAAO,UAAU,CAAC,GAAG,OAAO,SAAS;AACvC,cAAI,CAAC,QAAQ;AACX,qBAAS,KAAK,IAAI,IAAI;AACtB,wBAAY,IAAI,6BAA6B;AAAA,cAC3C,QAAQ,OAAO,MAAM;AAAA,cACrB,aAAa,OAAO,UAAU;AAAA,YAChC,CAAC;AAAA,UACH;AAAA,QACF;AAAA,MACF,SAAS,YAAY;AACnB;AACA,oBAAY,MAAM,6BAA6B;AAAA,UAC7C,aAAa,OAAO,UAAU;AAAA,UAC9B,cACE,sBAAsB,QAClB,WAAW,UACX,OAAO,UAAU;AAAA,UACvB,WACE,sBAAsB,QAClB,WAAW,YAAY,OACvB,OAAO;AAAA,QACf,CAAC;AAAA,MACH;AAAA,IACF;AAEA,gBAAY,IAAI,0BAA0B;AAAA,MACxC,aAAa,OAAO,UAAU;AAAA,MAC9B,YAAY,OAAO,UAAU;AAAA,MAC7B,eAAe,OAAO,KAAK,IAAI,IAAI,eAAe;AAAA,MAClD,QAAQ,OAAO,UAAU,CAAC;AAAA,MAC1B,gBAAgB,MAAM;AAAA,IACxB,CAAC;AAAA,EACH,SAAS,aAAa;AACpB,gBAAY,MAAM,6BAA6B;AAAA,MAC7C,aAAa,OAAO,UAAU;AAAA,MAC9B,YAAY,OAAO,UAAU;AAAA,MAC7B,cACE,uBAAuB,QACnB,YAAY,UACZ,OAAO,WAAW;AAAA,MACxB,WACE,uBAAuB,QACnB,YAAY,YAAY,OACxB,OAAO;AAAA,IACf,CAAC;AACD,UAAM;AAAA,EACR;AACA,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA,SAAS;AAAA,MACP;AAAA,QACE,OAAO;AAAA,QACP;AAAA,QACA,eAAe;AAAA,QACf,UAAU;AAAA,MACZ;AAAA,IACF;AAAA,IACA;AAAA,EACF;AACF;AAEA,SAAS,iCACP,UACA,QACA;AACA,MAAI,gBAAgC,CAAC;AACrC,QAAM,UAAU,SAAS,UAAU,CAAC,GAAG;AACvC,MAAI,CAAC,SAAS;AACZ,WAAO;AAAA,MACL,MAAM;AAAA,MACN,SAAS,CAAC;AAAA,MACV,aAAa,SAAS,UAAU,CAAC,GAAG;AAAA,MACpC,MAAM;AAAA,MACN,OAAO,SAAS;AAAA,IAClB;AAAA,EACF;AAEA,MAAI,SAAS,YAAY;AACvB,eAAW,YAAY,QAAQ,YAAY;AACzC,YAAM,OAAO,SAAS;AACtB,YAAM,WAAW,MAAM;AACvB,UAAI,WAAW,CAAC;AAChB,UAAI,aAA4B;AAChC,UAAI;AACF,mBAAW,MAAM,YAAY,KAAK,MAAM,KAAK,SAAS,IAAI,CAAC;AAAA,MAC7D,SAAS,GAAG;AACV,qBAAa,aAAa,QAAQ,EAAE,UAAU,OAAO,CAAC;AACtD,oBAAY,KAAK,8BAA8B;AAAA,UAC7C;AAAA,UACA,cAAc,MAAM,WAAW,UAAU,GAAG,GAAG;AAAA,UAC/C,cAAc;AAAA,QAChB,CAAC;AAAA,MACH;AAEA,UAAI,cAAc,OAAO,KAAK,QAAQ,EAAE,WAAW,GAAG;AACpD,oBAAY,MAAM,6BAA6B;AAAA,UAC7C;AAAA,UACA;AAAA,UACA,YACE;AAAA,QACJ,CAAC;AAAA,MACH;AAEA,oBAAc,KAAK;AAAA,QACjB,MAAM;AAAA,QACN,OAAO;AAAA,QACP,MAAM;AAAA,QACN,IAAI,SAAS,IAAI,SAAS,IAAI,SAAS,KAAK,OAAO;AAAA,MACrD,CAAC;AAAA,IACH;AAAA,EACF;AAEA,MAAK,QAAgB,WAAW;AAC9B,kBAAc,KAAK;AAAA,MACjB,MAAM;AAAA,MACN,UAAW,QAAgB;AAAA,MAC3B,WAAW;AAAA,IACb,CAAC;AAAA,EACH;AAEA,MAAK,QAAgB,mBAAmB;AACtC,kBAAc,KAAK;AAAA,MACjB,MAAM;AAAA,MACN,UAAW,QAAgB;AAAA,MAC3B,WAAW;AAAA,IACb,CAAC;AAAA,EACH;AAEA,MAAI,QAAQ,SAAS;AACnB,kBAAc,KAAK;AAAA,MACjB,MAAM;AAAA,MACN,MAAM,SAAS;AAAA,MACf,WAAW,CAAC;AAAA,IACd,CAAC;AAAA,EACH;AAEA,QAAM,eAAe;AAAA,IACnB,MAAM;AAAA,IACN,SAAS;AAAA,IACT,aAAa,SAAS,UAAU,CAAC,GAAG;AAAA,IACpC,MAAM;AAAA,IACN,OAAO,SAAS;AAAA,EAClB;AAEA,SAAO;AACT;AAIA,SAAS,6BAA6B,OAAkC;AACtE,MAAI,iBAAiB,SAAS,MAAM,QAAQ,SAAS,oBAAoB,GAAG;AAC1E,WAAO,+BAA+B,6BAA6B;AAAA,EACrE;AACA,MACE,iBAAiB,SACjB,MAAM,QAAQ,SAAS,gCAAgC,GACvD;AACA,WAAO,+BAA+B,oCAAoC;AAAA,EAC5E;AACA,MACE,iBAAiB,SACjB,MAAM,QAAQ,YAAY,EAAE,SAAS,WAAW,GAChD;AACA,WAAO,+BAA+B,6BAA6B;AAAA,EACrE;AACA,MAAI,iBAAiB,OAAO;AAC1B,QAAI,QAAQ,IAAI,aAAa,eAAe;AAC1C,kBAAY,MAAM,oBAAoB;AAAA,QACpC,SAAS,MAAM;AAAA,QACf,OAAO,MAAM;AAAA,MACf,CAAC;AAAA,IACH;AACA,WAAO;AAAA,MACL,GAAG,wBAAwB,KAAK,MAAM,OAAO;AAAA,IAC/C;AAAA,EACF;AACA,SAAO,+BAA+B,wBAAwB;AAChE;AAIA,eAAsB,YACpB,UACA,cACA,mBACA,OACA,QACA,SAO2B;AAC3B,QAAM,SAAS,gBAAgB;AAC/B,QAAM,eAAe,gBAAgB;AACrC,QAAM,iBAAiB,SAAS;AAEhC,QAAM,eAAe,SAAS,gBAAgB,aAAa,SAAS,MAAM;AAC1E,MAAI;AAEJ,QAAM,iBAAiB,kBAAkB;AACzC,cAAY,IAAI,uBAAuB;AAAA,IACrC,mBAAmB,CAAC,CAAC;AAAA,IACrB,gBAAgB,cAAc;AAAA,IAC9B,kBAAkB,cAAc;AAAA,IAChC,uBAAuB,cAAc;AAAA,IACrC,sBAAsB,cAAc;AAAA,IACpC,qBAAqB,cAAc;AAAA,IACnC,0BAA0B,CAAC,CAAC,cAAc;AAAA,IAC1C,cAAc,SAAS;AAAA,IACvB,WAAW,kBAAkB,GAAG;AAAA,EAClC,CAAC;AAED,MAAI,cAAc;AAChB,YAAQ,aAAa;AAAA,EACvB,OAAO;AACL,YAAQ,SAAS,SAAS,cAAc,aAAa;AAAA,EACvD;AAEA,MAAI,SAAS,qBAAqB;AAChC,UAAM,CAAC,mBAAmB,IAAI,qBAAqB,YAAY;AAC/D,mBAAe,CAAC,sBAAsB,GAAG,GAAG,YAAY;AAAA,EAC1D;AAEA,QAAM,SAA2B,qBAAqB,YAAY,EAAE;AAAA,IAClE,QAAM;AAAA,MACJ,GAAI,yBACA,EAAE,eAAe,EAAE,MAAM,YAAY,EAAE,IACvC,CAAC;AAAA,MACL,MAAM;AAAA,MACN,MAAM;AAAA,IACR;AAAA,EACF;AAEA,QAAM,cAAc,MAAM,QAAQ;AAAA,IAChC,MAAM;AAAA,MACJ,OAAM,OACH;AAAA,QACC,MAAM;AAAA,QACN,UAAU;AAAA,UACR,MAAM,EAAE;AAAA,UACR,aAAa,MAAM,wBAAwB,CAAC;AAAA,UAC5C,YACE,qBAAqB,KAAK,EAAE,kBACxB,EAAE,kBACF,gBAAgB,EAAE,WAAW;AAAA,QACrC;AAAA,MACF;AAAA,IACJ;AAAA,EACF;AAEA,QAAM,eAAe,OAAO;AAAA,IAC1B,QACG;AAAA,MACC,MAAM;AAAA,MACN,SAAS,EAAE;AAAA,IACb;AAAA,EACJ;AAEA,QAAM,iBAAiB,yCAAyC,QAAQ;AACxE,QAAM,wBAAwB,KAAK,IAAI;AAEvC,8BAA4B;AAAA,IAC1B,YAAY,aAAa,KAAK,IAAI;AAAA,IAClC,cAAc,qBAAqB,KAAK;AAAA,IACxC,WAAW,CAAC;AAAA,IACZ,aAAa,aAAa,KAAK,IAAI;AAAA,EACrC,CAAC;AAED,MAAI,QAAQ,KAAK,IAAI;AACrB,MAAI,gBAAgB;AACpB,MAAI;AAEJ,MAAI;AACF,eAAW,MAAM;AAAA,MACf,OAAM,YAAW;AACf,wBAAgB;AAChB,gBAAQ,KAAK,IAAI;AACjB,cAAM,YAAY,wBAAwB,YAAY;AACtD,cAAM,SAAS,YAAY,KAAK;AAGhC,cAAM,mBACJ,UACA,MAAM,WAAW,IAAI,KACrB,MAAM,WAAW,SAAS,KAC1B,UAAU,QACV,UAAU,aACV,UAAU;AAEZ,cAAM,OAA0C;AAAA,UAC9C;AAAA,UAEA,GAAI,SACA,EAAE,uBAAuB,UAAU,IACnC,EAAE,YAAY,UAAU;AAAA,UAC5B,UAAU,CAAC,GAAG,cAAc,GAAG,cAAc;AAAA;AAAA,UAG7C,GAAI,CAAC,oBAAoB,EAAE,aAAa,uBAAuB;AAAA,QACjE;AACA,YAAI,OAAO,QAAQ;AACjB;AAAC,UAAC,KAA2C,SAAS;AACtD,eAAK,iBAAiB;AAAA,YACpB,eAAe;AAAA,UACjB;AAAA,QACF;AAEA,YAAI,YAAY,SAAS,GAAG;AAC1B,eAAK,QAAQ;AACb,eAAK,cAAc;AAAA,QACrB;AACA,cAAM,kBAAkB,MAAM,mBAAmB,cAAc,QAAQ;AACvE,YAAI,iBAAiB;AACnB,eAAK,mBAAmB;AAAA,QAC1B;AAEA,YAAI,gBAAgB,aAAa,WAAW;AAC1C,sBAAY,IAAI,4BAA4B;AAAA,YAC1C,kBAAkB,aAAa;AAAA,YAC/B,WAAW,aAAa;AAAA,YACxB,UAAU,aAAa;AAAA,YACvB,SAAS,aAAa;AAAA,YACtB,cAAc,CAAC,CAAC,aAAa;AAAA,YAC7B,WAAW,kBAAkB,GAAG;AAAA,UAClC,CAAC;AAGD,gBAAM,yBACJ,QAAQ,IAAI,qBAAqB;AAEnC,cAAI,wBAAwB;AAE1B,kBAAM,UAAU,oBAAoB,cAAc,YAAY;AAG9D,kBAAM,gBAAsC;AAAA,cAC1C,UAAU;AAAA,cACV,cAAc,aAAa,IAAI,OAAK,EAAE,OAAiB;AAAA,cACvD;AAAA,cACA,WAAW,wBAAwB,YAAY;AAAA,cAC/C,QAAQ,OAAO;AAAA,cACf;AAAA,cACA,aACE,YAAY,KAAK,KACjB,MAAM,WAAW,IAAI,KACrB,MAAM,WAAW,SAAS,IACtB,SACA;AAAA,cACN,oBACE,gBAAgB,eAAe;AAAA,cACjC,WAAW;AAAA;AAAA,YACb;AAGA,kBAAM,UAAU,QAAQ,cAAc,aAAa;AAGnD,gBAAI,oBAAoB,sBAAsB,YAAY,GAAG;AAE3D,oBAAM,EAAE,qBAAqB,IAAI,MAAM,OAAO,WAAW;AACzD,oBAAMA,YAAW,MAAM;AAAA,gBACrB;AAAA,gBACA;AAAA,gBACA;AAAA,cACF;AACA,oBAAM,kBAAkB,QAAQ,cAAcA,SAAQ;AAGtD,oBAAM,aAAa;AAAA,gBACjB,MAAM;AAAA,gBACN,SAAS,gBAAgB;AAAA,gBACzB,YAAY,gBAAgB;AAAA,gBAC5B,OAAO;AAAA,kBACL,eAAe,gBAAgB,MAAM;AAAA,kBACrC,mBAAmB,gBAAgB,MAAM;AAAA,gBAC3C;AAAA,cACF;AACA,oBAAM,eAAiC;AAAA,gBACrC,MAAM;AAAA,gBACN,SAAS;AAAA,gBACT,SAAS;AAAA;AAAA,gBACT,YAAY,KAAK,IAAI,IAAI;AAAA,gBACzB,MAAM,GAAG,KAAK,IAAI,CAAC,IAAI,KAAK,OAAO,EAAE,SAAS,EAAE,EAAE,OAAO,GAAG,CAAC,CAAC;AAAA,gBAC9D,YAAY,gBAAgB;AAAA;AAAA,cAC9B;AACA,qBAAO;AAAA,YACT,OAAO;AAEL,oBAAM,IAAI,MAAM;AAAA,gBACd;AAAA,gBACA;AAAA,gBACA;AAAA,gBACA;AAAA,gBACA;AAAA,cACF;AACA,kBAAI;AACJ,kBAAI,OAAO,QAAQ;AACjB,gCAAgB,MAAM;AAAA,kBACpB;AAAA,kBACA;AAAA,gBACF;AAAA,cACF,OAAO;AACL,gCAAgB;AAAA,cAClB;AACA,oBAAM,IAAI,iCAAiC,eAAe,KAAK;AAC/D,qBAAO;AAAA,YACT;AAAA,UACF,OAAO;AAEL,kBAAM,qBAAqB,YAAY,aAAa,SAAS,IACzD,+BACA;AACJ,kBAAM,IAAI,MAAM;AAAA,cACd;AAAA,cACA;AAAA,cACA;AAAA,cACA;AAAA,cACA;AAAA,YACF;AACA,gBAAI;AACJ,gBAAI,KAAK,QAAQ;AACf,8BAAgB,MAAM;AAAA,gBACpB;AAAA,gBACA;AAAA,cACF;AAAA,YACF,OAAO;AACL,8BAAgB;AAAA,YAClB;AACA,kBAAM,IAAI,iCAAiC,eAAe,KAAK;AAC/D,mBAAO;AAAA,UACT;AAAA,QACF,OAAO;AACL,gBAAM,eAAe;AAAA,YACnB,oBAAoB,CAAC,CAAC;AAAA,YACtB,gBAAgB,cAAc;AAAA,YAC9B,iBAAiB,CAAC,CAAC,cAAc;AAAA,YACjC,gBAAgB;AAAA,YAChB,WAAW,kBAAkB,GAAG;AAAA,UAClC;AACA,sBAAY,MAAM,0BAA0B,YAAY;AACxD,gBAAM,IAAI;AAAA,YACR,8CAA8C,KAAK,2DAA2D,KAAK,UAAU,YAAY,CAAC;AAAA,UAC5I;AAAA,QACF;AAAA,MACF;AAAA,MACA;AAAA,QACE;AAAA,QACA,SAAS,gBAAgB,UACrB,CAAC,EAAE,SAAS,YAAY,OAAO,QAAQ,MAAM;AAC3C,oCAA0B,eAAe,SAAU;AAAA,YACjD;AAAA,YACA;AAAA,YACA,cAAc,MAAM;AAAA,YACpB;AAAA,UACF,CAAC;AAAA,QACH,IACA;AAAA,MACN;AAAA,IACF;AAAA,EACF,SAAS,OAAO;AACd,aAAS,KAAK;AACd,WAAO,6BAA6B,KAAK;AAAA,EAC3C;AACA,QAAM,aAAa,KAAK,IAAI,IAAI;AAChC,QAAM,6BAA6B,KAAK,IAAI,IAAI;AAEhD,QAAM,cAAc,SAAS,OAAO,iBAAiB;AACrD,QAAM,eAAe,SAAS,OAAO,qBAAqB;AAC1D,QAAM,uBACJ,SAAS,OAAO,sBAAsB,iBAAiB;AACzD,QAAM,2BACJ,SAAS,OAAO,sBAAsB,iBAAiB;AAGzD,QAAM,sBAAsB,0BAA0B,KAAK,IAAI;AAC/D,QAAM,uBAAuB,2BAA2B,KAAK,IAAI;AACjE,QAAM,UACH,cAAc,OACZ,uBAAuB,wCACzB,eAAe,OACb,wBAAwB,yCAC1B,uBAAuB,OACrB,uBAAuB,wCACxB,MACD,2BAA2B,OACzB,uBAAuB;AAE5B,iBAAe,SAAS,0BAA0B;AAGlD;AAAA,IACE;AAAA,MACE;AAAA,MACA;AAAA,MACA,qBAAqB;AAAA,MACrB,iBAAiB;AAAA,IACnB;AAAA,IACA;AAAA,IACA;AAAA,IACA,iBACK;AAAA,MACC,SAAS,eAAe;AAAA,MACxB,WAAW,eAAe;AAAA,MAC1B;AAAA,IACF,IACA;AAAA,EACN;AAGA,oBAAkB;AAAA,IAChB,cAAc,aAAa,KAAK,IAAI;AAAA,IACpC,UAAU,CAAC,GAAG,cAAc,GAAG,cAAc;AAAA,IAC7C;AAAA,IACA,OAAO;AAAA,MACL;AAAA,MACA;AAAA,IACF;AAAA,IACA,QAAQ;AAAA,MACN;AAAA,MACA,KAAK,KAAK,IAAI;AAAA,IAChB;AAAA,IACA,WAAW;AAAA,EACb,CAAC;AAED,SAAO;AAAA,IACL,SAAS;AAAA,MACP,GAAG;AAAA,MACH,SAAS,wBAAwB,SAAS,OAAO;AAAA,MACjD,OAAO;AAAA,QACL,cAAc;AAAA,QACd,eAAe;AAAA,QACf,yBAAyB;AAAA,QACzB,6BAA6B;AAAA,MAC/B;AAAA,IACF;AAAA,IACA;AAAA,IACA;AAAA,IACA,MAAM;AAAA,IACN,MAAM,WAAW;AAAA,EACnB;AACF;",
-  "names": ["response"]
+  "sourcesContent": ["/**\n * OpenAI-compatible provider \u2014 message conversion, streaming, and query path.\n */\nimport { randomUUID } from 'crypto'\nimport { nanoid } from 'nanoid'\n\nimport type { AssistantMessage, UserMessage } from '@query'\nimport { Tool, getToolDescriptionAsync } from '@tool'\nimport type { ToolUseContext } from '@tool'\nimport { getGlobalConfig, ModelProfile } from '@utils/config'\nimport { isGPT5ModelName as isGPT5Model } from '@utils/config'\nimport { normalizeContentFromAPI } from '@utils/messages'\nimport {\n  debug as debugLogger,\n  getCurrentRequest,\n  logLLMInteraction,\n  logSystemPromptConstruction,\n} from '@utils/debugLogger'\nimport { getModelManager } from '@utils/model'\nimport { zodToJsonSchema } from 'zod-to-json-schema'\nimport { ModelAdapterFactory } from '../modelAdapterFactory'\nimport { UnifiedRequestParams } from '@minto-types/modelCapabilities'\nimport { getCLISyspromptPrefix } from '@constants/prompts'\nimport OpenAI from 'openai'\nimport type { ChatCompletionStream } from 'openai/lib/ChatCompletionStream'\nimport { ContentBlock } from '@anthropic-ai/sdk/resources/messages/messages'\nimport type { TextBlockParam } from '@anthropic-ai/sdk/resources/index.mjs'\nimport { setStreamingState, getStreamingState } from '@utils/streamingState'\nimport {\n  beginTokenTracking,\n  addPreparedChars,\n  addReceivedChars,\n  setFinalTokens,\n} from '@utils/tokenProgress'\nimport {\n  getCompletionWithProfile,\n  getGPT5CompletionWithProfile,\n} from '../openai'\nimport { getReasoningEffort } from '@utils/thinking'\nimport { addRetryEventToTranscript } from '@utils/agentTranscripts'\nimport type { TokenTrackingContext } from '@core/tokenStats'\n\nimport { generateMintoContext } from './mintoContext'\nimport { splitSysPromptPrefix } from './anthropicProvider'\nimport {\n  PROMPT_CACHING_ENABLED,\n  MAIN_QUERY_TEMPERATURE,\n  withRetry,\n  getMaxTokensFromProfile,\n  calculateCostUSD,\n  getAssistantMessageFromError,\n  addToTotalCost,\n  recordTokenUsage,\n  logError,\n} from './types'\n\n// \u2500\u2500 OpenAI message conversion \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\nfunction convertAnthropicMessagesToOpenAIMessages(\n  messages: (UserMessage | AssistantMessage)[],\n): (\n  | OpenAI.ChatCompletionMessageParam\n  | OpenAI.ChatCompletionToolMessageParam\n)[] {\n  const openaiMessages: (\n    | OpenAI.ChatCompletionMessageParam\n    | OpenAI.ChatCompletionToolMessageParam\n  )[] = []\n\n  const toolResults: Record<string, OpenAI.ChatCompletionToolMessageParam> = {}\n\n  for (const message of messages) {\n    let contentBlocks = []\n    if (typeof message.message.content === 'string') {\n      contentBlocks = [\n        {\n          type: 'text',\n          text: message.message.content,\n        },\n      ]\n    } else if (!Array.isArray(message.message.content)) {\n      contentBlocks = [message.message.content]\n    } else {\n      contentBlocks = message.message.content\n    }\n\n    for (const block of contentBlocks) {\n      if (block.type === 'text') {\n        openaiMessages.push({\n          role: message.message.role,\n          content: block.text,\n        })\n      } else if (block.type === 'tool_use') {\n        openaiMessages.push({\n          role: 'assistant',\n          content: undefined,\n          tool_calls: [\n            {\n              type: 'function',\n              function: {\n                name: block.name,\n                arguments: JSON.stringify(block.input),\n              },\n              id: block.id,\n            },\n          ],\n        })\n      } else if (block.type === 'tool_result') {\n        // Ensure content is always a string for role:tool messages\n        let toolContent = block.content\n        if (typeof toolContent !== 'string') {\n          toolContent = JSON.stringify(toolContent)\n        }\n\n        toolResults[block.tool_use_id] = {\n          role: 'tool',\n          content: toolContent,\n          tool_call_id: block.tool_use_id,\n        }\n      }\n    }\n  }\n\n  const finalMessages: (\n    | OpenAI.ChatCompletionMessageParam\n    | OpenAI.ChatCompletionToolMessageParam\n  )[] = []\n\n  for (const message of openaiMessages) {\n    finalMessages.push(message)\n\n    if ('tool_calls' in message && message.tool_calls) {\n      for (const toolCall of message.tool_calls) {\n        if (toolResults[toolCall.id]) {\n          finalMessages.push(toolResults[toolCall.id])\n        }\n      }\n    }\n  }\n\n  return finalMessages\n}\n\nfunction messageReducer(\n  previous: OpenAI.ChatCompletionMessage,\n  item: OpenAI.ChatCompletionChunk,\n): OpenAI.ChatCompletionMessage {\n  const reduce = (acc: any, delta: OpenAI.ChatCompletionChunk.Choice.Delta) => {\n    acc = { ...acc }\n    for (const [key, value] of Object.entries(delta)) {\n      if (acc[key] === undefined || acc[key] === null) {\n        acc[key] = value\n        if (Array.isArray(acc[key])) {\n          for (const arr of acc[key]) {\n            delete arr.index\n          }\n        }\n      } else if (typeof acc[key] === 'string' && typeof value === 'string') {\n        acc[key] += value\n      } else if (typeof acc[key] === 'number' && typeof value === 'number') {\n        acc[key] = value\n      } else if (Array.isArray(acc[key]) && Array.isArray(value)) {\n        const accArray = acc[key]\n        for (let i = 0; i < value.length; i++) {\n          const { index: rawIndex, ...chunkTool } = value[i]\n          // Default index to 0 when missing (Gemini doesn't include index in tool call deltas)\n          const index = rawIndex ?? 0\n          if (index - accArray.length > 1) {\n            throw new Error(\n              `Error: An array has an empty value when tool_calls are constructed. tool_calls: ${accArray}; tool: ${value}`,\n            )\n          }\n          accArray[index] = reduce(accArray[index], chunkTool)\n        }\n      } else if (typeof acc[key] === 'object' && typeof value === 'object') {\n        acc[key] = reduce(acc[key], value)\n      }\n    }\n    return acc\n  }\n\n  const choice = item.choices?.[0]\n  if (!choice) {\n    return previous\n  }\n  return reduce(previous, choice.delta) as OpenAI.ChatCompletionMessage\n}\n\nasync function handleMessageStream(\n  stream: ChatCompletionStream,\n  signal?: AbortSignal,\n): Promise<OpenAI.ChatCompletion> {\n  const streamStartTime = Date.now()\n  setStreamingState({ receivedChars: 0 })\n  let ttftMs: number | undefined\n  let chunkCount = 0\n  let errorCount = 0\n\n  debugLogger.api('OPENAI_STREAM_START', {\n    streamStartTime: String(streamStartTime),\n  })\n\n  let message = {} as OpenAI.ChatCompletionMessage\n\n  let id, model, created, object, usage\n  try {\n    for await (const chunk of stream) {\n      if (signal?.aborted) {\n        debugLogger.flow('OPENAI_STREAM_ABORTED', {\n          chunkCount,\n          timestamp: Date.now(),\n        })\n        throw new Error('Request was cancelled')\n      }\n\n      chunkCount++\n\n      try {\n        if (!id) {\n          id = chunk.id\n          debugLogger.api('OPENAI_STREAM_ID_RECEIVED', {\n            id,\n            chunkNumber: String(chunkCount),\n          })\n        }\n        if (!model) {\n          model = chunk.model\n          debugLogger.api('OPENAI_STREAM_MODEL_RECEIVED', {\n            model,\n            chunkNumber: String(chunkCount),\n          })\n        }\n        if (!created) {\n          created = chunk.created\n        }\n        if (!object) {\n          object = chunk.object\n        }\n        if (!usage && chunk.usage) {\n          usage = chunk.usage\n          setFinalTokens({\n            input: chunk.usage.prompt_tokens,\n            output: chunk.usage.completion_tokens,\n          })\n        }\n\n        message = messageReducer(message, chunk)\n\n        if (chunk?.choices?.[0]?.delta?.content) {\n          const textChunk = chunk.choices[0].delta.content!\n          addReceivedChars(textChunk.length)\n          // Append to streaming text buffer for typewriter preview\n          const prevText = getStreamingState().streamingText || ''\n          setStreamingState({\n            streamingText: prevText + textChunk,\n            phase: 'generating',\n          })\n          if (!ttftMs) {\n            ttftMs = Date.now() - streamStartTime\n            debugLogger.api('OPENAI_STREAM_FIRST_TOKEN', {\n              ttftMs: String(ttftMs),\n              chunkNumber: String(chunkCount),\n            })\n          }\n        }\n      } catch (chunkError) {\n        errorCount++\n        debugLogger.error('OPENAI_STREAM_CHUNK_ERROR', {\n          chunkNumber: String(chunkCount),\n          errorMessage:\n            chunkError instanceof Error\n              ? chunkError.message\n              : String(chunkError),\n          errorType:\n            chunkError instanceof Error\n              ? chunkError.constructor.name\n              : typeof chunkError,\n        })\n      }\n    }\n\n    debugLogger.api('OPENAI_STREAM_COMPLETE', {\n      totalChunks: String(chunkCount),\n      errorCount: String(errorCount),\n      totalDuration: String(Date.now() - streamStartTime),\n      ttftMs: String(ttftMs || 0),\n      finalMessageId: id || 'undefined',\n    })\n  } catch (streamError) {\n    debugLogger.error('OPENAI_STREAM_FATAL_ERROR', {\n      totalChunks: String(chunkCount),\n      errorCount: String(errorCount),\n      errorMessage:\n        streamError instanceof Error\n          ? streamError.message\n          : String(streamError),\n      errorType:\n        streamError instanceof Error\n          ? streamError.constructor.name\n          : typeof streamError,\n    })\n    throw streamError\n  }\n  return {\n    id,\n    created,\n    model,\n    object,\n    choices: [\n      {\n        index: 0,\n        message,\n        finish_reason: 'stop',\n        logprobs: undefined,\n      },\n    ],\n    usage,\n  }\n}\n\nfunction convertOpenAIResponseToAnthropic(\n  response: OpenAI.ChatCompletion,\n  _tools?: Tool[],\n) {\n  let contentBlocks: ContentBlock[] = []\n  const message = response.choices?.[0]?.message\n  if (!message) {\n    return {\n      role: 'assistant',\n      content: [],\n      stop_reason: response.choices?.[0]?.finish_reason,\n      type: 'message',\n      usage: response.usage,\n    }\n  }\n\n  if (message?.tool_calls) {\n    for (const toolCall of message.tool_calls) {\n      const tool = toolCall.function\n      const toolName = tool?.name\n      let toolArgs = {}\n      let parseError: string | null = null\n      try {\n        toolArgs = tool?.arguments ? JSON.parse(tool.arguments) : {}\n      } catch (e) {\n        parseError = e instanceof Error ? e.message : String(e)\n        debugLogger.warn('TOOL_ARGUMENTS_PARSE_ERROR', {\n          toolName,\n          rawArguments: tool?.arguments?.substring(0, 200),\n          errorMessage: parseError,\n        })\n      }\n\n      if (parseError && Object.keys(toolArgs).length === 0) {\n        debugLogger.error('TOOL_CALL_EMPTY_ARGUMENTS', {\n          toolName,\n          parseError,\n          suggestion:\n            'Model may have sent malformed JSON. The tool call will likely fail validation.',\n        })\n      }\n\n      contentBlocks.push({\n        type: 'tool_use',\n        input: toolArgs,\n        name: toolName,\n        id: toolCall.id?.length > 0 ? toolCall.id : nanoid(),\n      })\n    }\n  }\n\n  // Non-standard reasoning/reasoning_content fields from some providers\n  // (e.g., DeepSeek, QwQ, GLM) are intentionally ignored \u2014 they cause\n  // repetitive thinking display across tool-use turns. The model's actual\n  // response comes through standard content/tool_calls fields.\n\n  if (message.content) {\n    contentBlocks.push({\n      type: 'text',\n      text: message?.content,\n      citations: [],\n    })\n  }\n\n  const finalMessage = {\n    role: 'assistant',\n    content: contentBlocks,\n    stop_reason: response.choices?.[0]?.finish_reason,\n    type: 'message',\n    usage: response.usage,\n  }\n\n  return finalMessage\n}\n\n// Error helpers: uses shared getAssistantMessageFromError from types.ts\n\n// \u2500\u2500 queryOpenAI \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 queryOpenAI(\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  beginTokenTracking()\n\n  const modelProfile = options?.modelProfile || modelManager.getModel('main')\n  let model: string\n\n  const currentRequest = getCurrentRequest()\n  debugLogger.api('MODEL_CONFIG_OPENAI', {\n    modelProfileFound: !!modelProfile,\n    modelProfileId: modelProfile?.modelName,\n    modelProfileName: modelProfile?.name,\n    modelProfileModelName: modelProfile?.modelName,\n    modelProfileProvider: modelProfile?.provider,\n    modelProfileBaseURL: modelProfile?.baseURL,\n    modelProfileApiKeyExists: !!modelProfile?.apiKey,\n    optionsModel: options?.model,\n    requestId: getCurrentRequest()?.id,\n  })\n\n  if (modelProfile) {\n    model = modelProfile.modelName\n  } else {\n    model = options?.model || modelProfile?.modelName || ''\n  }\n  // Prepend system prompt block for easy API identification\n  if (options?.prependCLISysprompt) {\n    systemPrompt = [getCLISyspromptPrefix(), ...systemPrompt]\n  }\n\n  const system: TextBlockParam[] = splitSysPromptPrefix(systemPrompt).map(\n    _ => ({\n      ...(PROMPT_CACHING_ENABLED\n        ? { cache_control: { type: 'ephemeral' } }\n        : {}),\n      text: _,\n      type: 'text',\n    }),\n  )\n\n  // Track system prompt preparation progress\n  addPreparedChars(JSON.stringify(system).length)\n\n  const toolSchemas = await Promise.all(\n    tools.map(async _ => {\n      const schema = {\n        type: 'function',\n        function: {\n          name: _.name,\n          description: await getToolDescriptionAsync(_),\n          parameters:\n            'inputJSONSchema' in _ && _.inputJSONSchema\n              ? _.inputJSONSchema\n              : zodToJsonSchema(_.inputSchema),\n        },\n      } as OpenAI.ChatCompletionTool\n      addPreparedChars(JSON.stringify(schema).length)\n      return schema\n    }),\n  )\n\n  const openaiSystem = system.map(\n    s =>\n      ({\n        role: 'system',\n        content: s.text,\n      }) as OpenAI.ChatCompletionMessageParam,\n  )\n\n  const openaiMessages = convertAnthropicMessagesToOpenAIMessages(messages)\n  addPreparedChars(JSON.stringify(openaiMessages).length)\n\n  const startIncludingRetries = Date.now()\n\n  logSystemPromptConstruction({\n    basePrompt: systemPrompt.join('\\n'),\n    mintoContext: generateMintoContext() || '',\n    reminders: [],\n    finalPrompt: systemPrompt.join('\\n'),\n  })\n\n  let start = Date.now()\n  let attemptNumber = 0\n  let response\n\n  try {\n    response = await withRetry(\n      async attempt => {\n        attemptNumber = attempt\n        start = Date.now()\n        const maxTokens = getMaxTokensFromProfile(modelProfile)\n        const isGPT5 = isGPT5Model(model)\n\n        // Determine if this model requires temperature to be omitted entirely\n        const isReasoningModel =\n          isGPT5 ||\n          model.startsWith('o3') ||\n          model.startsWith('o4-mini') ||\n          model === 'o1' ||\n          model === 'o1-mini' ||\n          model === 'o1-preview'\n\n        const opts: OpenAI.ChatCompletionCreateParams = {\n          model,\n\n          ...(isGPT5\n            ? { max_completion_tokens: maxTokens }\n            : { max_tokens: maxTokens }),\n          messages: [...openaiSystem, ...openaiMessages],\n\n          // Reasoning models (GPT-5, o3, o4-mini) require temperature to be omitted\n          ...(!isReasoningModel && { temperature: MAIN_QUERY_TEMPERATURE }),\n        }\n        if (config.stream) {\n          ;(opts as OpenAI.ChatCompletionCreateParams).stream = true\n          opts.stream_options = {\n            include_usage: true,\n          }\n        }\n\n        if (toolSchemas.length > 0) {\n          opts.tools = toolSchemas\n          opts.tool_choice = 'auto'\n        }\n        // Only send reasoning_effort for models that explicitly support it\n        // (e.g., o3, o4-mini, GPT-5). Sending it to models like glm-5 or\n        // deepseek causes a 400 \"UnsupportedParamsError\" from proxies.\n        const reasoningEffort = isReasoningModel\n          ? await getReasoningEffort(modelProfile, messages)\n          : undefined\n        if (reasoningEffort) {\n          opts.reasoning_effort = reasoningEffort\n        }\n\n        if (modelProfile && modelProfile.modelName) {\n          debugLogger.api('USING_MODEL_PROFILE_PATH', {\n            modelProfileName: modelProfile.modelName,\n            modelName: modelProfile.modelName,\n            provider: modelProfile.provider,\n            baseURL: modelProfile.baseURL,\n            apiKeyExists: !!modelProfile.apiKey,\n            requestId: getCurrentRequest()?.id,\n          })\n\n          // Only use adapter system for models that need the Responses API\n          // (e.g., GPT-5). For standard Chat Completions models, use opts\n          // directly \u2014 opts already has full async tool descriptions and\n          // properly formatted messages. The adapter would rebuild these\n          // from raw Tool[] objects using sync getToolDescription(), which\n          // returns degraded fallback descriptions for async tools.\n          if (ModelAdapterFactory.shouldUseResponsesAPI(modelProfile)) {\n            const adapter = ModelAdapterFactory.createAdapter(modelProfile)\n\n            const unifiedParams: UnifiedRequestParams = {\n              messages: openaiMessages,\n              systemPrompt: openaiSystem.map(s => s.content as string),\n              tools: tools,\n              maxTokens: getMaxTokensFromProfile(modelProfile),\n              stream: config.stream,\n              reasoningEffort: reasoningEffort as any,\n              temperature:\n                isGPT5Model(model) ||\n                model.startsWith('o3') ||\n                model.startsWith('o4-mini')\n                  ? undefined\n                  : MAIN_QUERY_TEMPERATURE,\n              previousResponseId:\n                toolUseContext?.responseState?.previousResponseId,\n              verbosity: 'high',\n            }\n\n            const request = adapter.createRequest(unifiedParams)\n\n            const { callGPT5ResponsesAPI } = await import('../openai')\n            const response = await callGPT5ResponsesAPI(\n              modelProfile,\n              request,\n              signal,\n            )\n            const unifiedResponse = adapter.parseResponse(response)\n\n            const apiMessage = {\n              role: 'assistant' as const,\n              content: unifiedResponse.content,\n              tool_calls: unifiedResponse.toolCalls,\n              usage: {\n                prompt_tokens: unifiedResponse.usage.promptTokens,\n                completion_tokens: unifiedResponse.usage.completionTokens,\n              },\n            }\n            const respDurationMs = Date.now() - start\n            const respInputTokens = unifiedResponse.usage.promptTokens\n            const respOutputTokens = unifiedResponse.usage.completionTokens\n            const respCostUSD = calculateCostUSD({\n              model,\n              inputTokens: respInputTokens,\n              outputTokens: respOutputTokens,\n              cacheReadTokens: 0,\n              cacheCreationTokens: 0,\n            })\n            addToTotalCost(respCostUSD, respDurationMs)\n            recordTokenUsage(\n              {\n                inputTokens: respInputTokens,\n                outputTokens: respOutputTokens,\n                cacheCreationTokens: 0,\n                cacheReadTokens: 0,\n              },\n              respCostUSD,\n              model,\n              toolUseContext\n                ? ({\n                    agentId: toolUseContext.agentId,\n                    toolUseId: toolUseContext.toolUseId,\n                    model,\n                  } as TokenTrackingContext)\n                : undefined,\n            )\n            logLLMInteraction({\n              systemPrompt: systemPrompt.join('\\n'),\n              messages: [...openaiSystem, ...openaiMessages],\n              response: apiMessage,\n              usage: {\n                inputTokens: respInputTokens,\n                outputTokens: respOutputTokens,\n              },\n              timing: { start, end: Date.now() },\n              apiFormat: 'openai',\n            })\n            const assistantMsg: AssistantMessage = {\n              type: 'assistant',\n              message: apiMessage as any,\n              costUSD: respCostUSD,\n              durationMs: respDurationMs,\n              uuid: randomUUID(),\n              responseId: unifiedResponse.responseId,\n            }\n            return assistantMsg\n          }\n\n          // Chat Completions path \u2014 use pre-built opts with full async\n          // tool descriptions, correct messages, and proper streaming config.\n          const completionFunction = isGPT5Model(modelProfile.modelName)\n            ? getGPT5CompletionWithProfile\n            : getCompletionWithProfile\n          const s = await completionFunction(modelProfile, opts, 0, 10, signal)\n          let finalResponse\n          if (opts.stream) {\n            finalResponse = await handleMessageStream(\n              s as ChatCompletionStream,\n              signal,\n            )\n          } else {\n            finalResponse = s\n          }\n          const r = convertOpenAIResponseToAnthropic(finalResponse, tools)\n          return r\n        } else {\n          const errorDetails = {\n            modelProfileExists: !!modelProfile,\n            modelProfileId: modelProfile?.modelName,\n            modelNameExists: !!modelProfile?.modelName,\n            requestedModel: model,\n            requestId: getCurrentRequest()?.id,\n          }\n          debugLogger.error('NO_VALID_MODEL_PROFILE', errorDetails)\n          throw new Error(\n            `No valid ModelProfile available for model: ${model}. Please configure model through /model command. Debug: ${JSON.stringify(errorDetails)}`,\n          )\n        }\n      },\n      {\n        signal,\n        onRetry: toolUseContext?.agentId\n          ? ({ attempt, maxRetries, error, delayMs }) => {\n              addRetryEventToTranscript(toolUseContext.agentId!, {\n                attempt,\n                maxRetries,\n                errorMessage: error.message,\n                delayMs,\n              })\n            }\n          : undefined,\n      },\n    )\n  } catch (error) {\n    logError(error)\n    return getAssistantMessageFromError(error, 'OPENAI')\n  }\n  const durationMs = Date.now() - start\n  const durationMsIncludingRetries = Date.now() - startIncludingRetries\n\n  const inputTokens = response.usage?.prompt_tokens ?? 0\n  const outputTokens = response.usage?.completion_tokens ?? 0\n  const cacheReadInputTokens =\n    response.usage?.prompt_token_details?.cached_tokens ?? 0\n  // OpenAI does not report cache creation tokens separately\n  const cacheCreationInputTokens = 0\n\n  const costUSD = calculateCostUSD({\n    model,\n    inputTokens,\n    outputTokens,\n    cacheReadTokens: cacheReadInputTokens,\n    cacheCreationTokens: cacheCreationInputTokens,\n  })\n\n  addToTotalCost(costUSD, durationMsIncludingRetries)\n\n  // Record token usage to unified stats manager\n  recordTokenUsage(\n    {\n      inputTokens,\n      outputTokens,\n      cacheCreationTokens: cacheCreationInputTokens,\n      cacheReadTokens: cacheReadInputTokens,\n    },\n    costUSD,\n    model,\n    toolUseContext\n      ? ({\n          agentId: toolUseContext.agentId,\n          toolUseId: toolUseContext.toolUseId,\n          model,\n        } as TokenTrackingContext)\n      : undefined,\n  )\n\n  // Log LLM interaction (OpenAI path)\n  logLLMInteraction({\n    systemPrompt: systemPrompt.join('\\n'),\n    messages: [...openaiSystem, ...openaiMessages],\n    response: response,\n    usage: {\n      inputTokens: inputTokens,\n      outputTokens: outputTokens,\n    },\n    timing: {\n      start: start,\n      end: Date.now(),\n    },\n    apiFormat: 'openai',\n  })\n\n  return {\n    message: {\n      ...response,\n      content: normalizeContentFromAPI(response.content),\n      usage: {\n        input_tokens: inputTokens,\n        output_tokens: outputTokens,\n        cache_read_input_tokens: cacheReadInputTokens,\n        cache_creation_input_tokens: 0,\n      },\n    },\n    costUSD,\n    durationMs,\n    type: 'assistant',\n    uuid: randomUUID(),\n  }\n}\n"],
+  "mappings": "AAGA,SAAS,kBAAkB;AAC3B,SAAS,cAAc;AAGvB,SAAe,+BAA+B;AAE9C,SAAS,uBAAqC;AAC9C,SAAS,mBAAmB,mBAAmB;AAC/C,SAAS,+BAA+B;AACxC;AAAA,EACE,SAAS;AAAA,EACT;AAAA,EACA;AAAA,EACA;AAAA,OACK;AACP,SAAS,uBAAuB;AAChC,SAAS,uBAAuB;AAChC,SAAS,2BAA2B;AAEpC,SAAS,6BAA6B;AAKtC,SAAS,mBAAmB,yBAAyB;AACrD;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OACK;AACP;AAAA,EACE;AAAA,EACA;AAAA,OACK;AACP,SAAS,0BAA0B;AACnC,SAAS,iCAAiC;AAG1C,SAAS,4BAA4B;AACrC,SAAS,4BAA4B;AACrC;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OACK;AAIP,SAAS,yCACP,UAIE;AACF,QAAM,iBAGA,CAAC;AAEP,QAAM,cAAqE,CAAC;AAE5E,aAAW,WAAW,UAAU;AAC9B,QAAI,gBAAgB,CAAC;AACrB,QAAI,OAAO,QAAQ,QAAQ,YAAY,UAAU;AAC/C,sBAAgB;AAAA,QACd;AAAA,UACE,MAAM;AAAA,UACN,MAAM,QAAQ,QAAQ;AAAA,QACxB;AAAA,MACF;AAAA,IACF,WAAW,CAAC,MAAM,QAAQ,QAAQ,QAAQ,OAAO,GAAG;AAClD,sBAAgB,CAAC,QAAQ,QAAQ,OAAO;AAAA,IAC1C,OAAO;AACL,sBAAgB,QAAQ,QAAQ;AAAA,IAClC;AAEA,eAAW,SAAS,eAAe;AACjC,UAAI,MAAM,SAAS,QAAQ;AACzB,uBAAe,KAAK;AAAA,UAClB,MAAM,QAAQ,QAAQ;AAAA,UACtB,SAAS,MAAM;AAAA,QACjB,CAAC;AAAA,MACH,WAAW,MAAM,SAAS,YAAY;AACpC,uBAAe,KAAK;AAAA,UAClB,MAAM;AAAA,UACN,SAAS;AAAA,UACT,YAAY;AAAA,YACV;AAAA,cACE,MAAM;AAAA,cACN,UAAU;AAAA,gBACR,MAAM,MAAM;AAAA,gBACZ,WAAW,KAAK,UAAU,MAAM,KAAK;AAAA,cACvC;AAAA,cACA,IAAI,MAAM;AAAA,YACZ;AAAA,UACF;AAAA,QACF,CAAC;AAAA,MACH,WAAW,MAAM,SAAS,eAAe;AAEvC,YAAI,cAAc,MAAM;AACxB,YAAI,OAAO,gBAAgB,UAAU;AACnC,wBAAc,KAAK,UAAU,WAAW;AAAA,QAC1C;AAEA,oBAAY,MAAM,WAAW,IAAI;AAAA,UAC/B,MAAM;AAAA,UACN,SAAS;AAAA,UACT,cAAc,MAAM;AAAA,QACtB;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAEA,QAAM,gBAGA,CAAC;AAEP,aAAW,WAAW,gBAAgB;AACpC,kBAAc,KAAK,OAAO;AAE1B,QAAI,gBAAgB,WAAW,QAAQ,YAAY;AACjD,iBAAW,YAAY,QAAQ,YAAY;AACzC,YAAI,YAAY,SAAS,EAAE,GAAG;AAC5B,wBAAc,KAAK,YAAY,SAAS,EAAE,CAAC;AAAA,QAC7C;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAEA,SAAO;AACT;AAEA,SAAS,eACP,UACA,MAC8B;AAC9B,QAAM,SAAS,CAAC,KAAU,UAAmD;AAC3E,UAAM,EAAE,GAAG,IAAI;AACf,eAAW,CAAC,KAAK,KAAK,KAAK,OAAO,QAAQ,KAAK,GAAG;AAChD,UAAI,IAAI,GAAG,MAAM,UAAa,IAAI,GAAG,MAAM,MAAM;AAC/C,YAAI,GAAG,IAAI;AACX,YAAI,MAAM,QAAQ,IAAI,GAAG,CAAC,GAAG;AAC3B,qBAAW,OAAO,IAAI,GAAG,GAAG;AAC1B,mBAAO,IAAI;AAAA,UACb;AAAA,QACF;AAAA,MACF,WAAW,OAAO,IAAI,GAAG,MAAM,YAAY,OAAO,UAAU,UAAU;AACpE,YAAI,GAAG,KAAK;AAAA,MACd,WAAW,OAAO,IAAI,GAAG,MAAM,YAAY,OAAO,UAAU,UAAU;AACpE,YAAI,GAAG,IAAI;AAAA,MACb,WAAW,MAAM,QAAQ,IAAI,GAAG,CAAC,KAAK,MAAM,QAAQ,KAAK,GAAG;AAC1D,cAAM,WAAW,IAAI,GAAG;AACxB,iBAAS,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;AACrC,gBAAM,EAAE,OAAO,UAAU,GAAG,UAAU,IAAI,MAAM,CAAC;AAEjD,gBAAM,QAAQ,YAAY;AAC1B,cAAI,QAAQ,SAAS,SAAS,GAAG;AAC/B,kBAAM,IAAI;AAAA,cACR,mFAAmF,QAAQ,WAAW,KAAK;AAAA,YAC7G;AAAA,UACF;AACA,mBAAS,KAAK,IAAI,OAAO,SAAS,KAAK,GAAG,SAAS;AAAA,QACrD;AAAA,MACF,WAAW,OAAO,IAAI,GAAG,MAAM,YAAY,OAAO,UAAU,UAAU;AACpE,YAAI,GAAG,IAAI,OAAO,IAAI,GAAG,GAAG,KAAK;AAAA,MACnC;AAAA,IACF;AACA,WAAO;AAAA,EACT;AAEA,QAAM,SAAS,KAAK,UAAU,CAAC;AAC/B,MAAI,CAAC,QAAQ;AACX,WAAO;AAAA,EACT;AACA,SAAO,OAAO,UAAU,OAAO,KAAK;AACtC;AAEA,eAAe,oBACb,QACA,QACgC;AAChC,QAAM,kBAAkB,KAAK,IAAI;AACjC,oBAAkB,EAAE,eAAe,EAAE,CAAC;AACtC,MAAI;AACJ,MAAI,aAAa;AACjB,MAAI,aAAa;AAEjB,cAAY,IAAI,uBAAuB;AAAA,IACrC,iBAAiB,OAAO,eAAe;AAAA,EACzC,CAAC;AAED,MAAI,UAAU,CAAC;AAEf,MAAI,IAAI,OAAO,SAAS,QAAQ;AAChC,MAAI;AACF,qBAAiB,SAAS,QAAQ;AAChC,UAAI,QAAQ,SAAS;AACnB,oBAAY,KAAK,yBAAyB;AAAA,UACxC;AAAA,UACA,WAAW,KAAK,IAAI;AAAA,QACtB,CAAC;AACD,cAAM,IAAI,MAAM,uBAAuB;AAAA,MACzC;AAEA;AAEA,UAAI;AACF,YAAI,CAAC,IAAI;AACP,eAAK,MAAM;AACX,sBAAY,IAAI,6BAA6B;AAAA,YAC3C;AAAA,YACA,aAAa,OAAO,UAAU;AAAA,UAChC,CAAC;AAAA,QACH;AACA,YAAI,CAAC,OAAO;AACV,kBAAQ,MAAM;AACd,sBAAY,IAAI,gCAAgC;AAAA,YAC9C;AAAA,YACA,aAAa,OAAO,UAAU;AAAA,UAChC,CAAC;AAAA,QACH;AACA,YAAI,CAAC,SAAS;AACZ,oBAAU,MAAM;AAAA,QAClB;AACA,YAAI,CAAC,QAAQ;AACX,mBAAS,MAAM;AAAA,QACjB;AACA,YAAI,CAAC,SAAS,MAAM,OAAO;AACzB,kBAAQ,MAAM;AACd,yBAAe;AAAA,YACb,OAAO,MAAM,MAAM;AAAA,YACnB,QAAQ,MAAM,MAAM;AAAA,UACtB,CAAC;AAAA,QACH;AAEA,kBAAU,eAAe,SAAS,KAAK;AAEvC,YAAI,OAAO,UAAU,CAAC,GAAG,OAAO,SAAS;AACvC,gBAAM,YAAY,MAAM,QAAQ,CAAC,EAAE,MAAM;AACzC,2BAAiB,UAAU,MAAM;AAEjC,gBAAM,WAAW,kBAAkB,EAAE,iBAAiB;AACtD,4BAAkB;AAAA,YAChB,eAAe,WAAW;AAAA,YAC1B,OAAO;AAAA,UACT,CAAC;AACD,cAAI,CAAC,QAAQ;AACX,qBAAS,KAAK,IAAI,IAAI;AACtB,wBAAY,IAAI,6BAA6B;AAAA,cAC3C,QAAQ,OAAO,MAAM;AAAA,cACrB,aAAa,OAAO,UAAU;AAAA,YAChC,CAAC;AAAA,UACH;AAAA,QACF;AAAA,MACF,SAAS,YAAY;AACnB;AACA,oBAAY,MAAM,6BAA6B;AAAA,UAC7C,aAAa,OAAO,UAAU;AAAA,UAC9B,cACE,sBAAsB,QAClB,WAAW,UACX,OAAO,UAAU;AAAA,UACvB,WACE,sBAAsB,QAClB,WAAW,YAAY,OACvB,OAAO;AAAA,QACf,CAAC;AAAA,MACH;AAAA,IACF;AAEA,gBAAY,IAAI,0BAA0B;AAAA,MACxC,aAAa,OAAO,UAAU;AAAA,MAC9B,YAAY,OAAO,UAAU;AAAA,MAC7B,eAAe,OAAO,KAAK,IAAI,IAAI,eAAe;AAAA,MAClD,QAAQ,OAAO,UAAU,CAAC;AAAA,MAC1B,gBAAgB,MAAM;AAAA,IACxB,CAAC;AAAA,EACH,SAAS,aAAa;AACpB,gBAAY,MAAM,6BAA6B;AAAA,MAC7C,aAAa,OAAO,UAAU;AAAA,MAC9B,YAAY,OAAO,UAAU;AAAA,MAC7B,cACE,uBAAuB,QACnB,YAAY,UACZ,OAAO,WAAW;AAAA,MACxB,WACE,uBAAuB,QACnB,YAAY,YAAY,OACxB,OAAO;AAAA,IACf,CAAC;AACD,UAAM;AAAA,EACR;AACA,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA,SAAS;AAAA,MACP;AAAA,QACE,OAAO;AAAA,QACP;AAAA,QACA,eAAe;AAAA,QACf,UAAU;AAAA,MACZ;AAAA,IACF;AAAA,IACA;AAAA,EACF;AACF;AAEA,SAAS,iCACP,UACA,QACA;AACA,MAAI,gBAAgC,CAAC;AACrC,QAAM,UAAU,SAAS,UAAU,CAAC,GAAG;AACvC,MAAI,CAAC,SAAS;AACZ,WAAO;AAAA,MACL,MAAM;AAAA,MACN,SAAS,CAAC;AAAA,MACV,aAAa,SAAS,UAAU,CAAC,GAAG;AAAA,MACpC,MAAM;AAAA,MACN,OAAO,SAAS;AAAA,IAClB;AAAA,EACF;AAEA,MAAI,SAAS,YAAY;AACvB,eAAW,YAAY,QAAQ,YAAY;AACzC,YAAM,OAAO,SAAS;AACtB,YAAM,WAAW,MAAM;AACvB,UAAI,WAAW,CAAC;AAChB,UAAI,aAA4B;AAChC,UAAI;AACF,mBAAW,MAAM,YAAY,KAAK,MAAM,KAAK,SAAS,IAAI,CAAC;AAAA,MAC7D,SAAS,GAAG;AACV,qBAAa,aAAa,QAAQ,EAAE,UAAU,OAAO,CAAC;AACtD,oBAAY,KAAK,8BAA8B;AAAA,UAC7C;AAAA,UACA,cAAc,MAAM,WAAW,UAAU,GAAG,GAAG;AAAA,UAC/C,cAAc;AAAA,QAChB,CAAC;AAAA,MACH;AAEA,UAAI,cAAc,OAAO,KAAK,QAAQ,EAAE,WAAW,GAAG;AACpD,oBAAY,MAAM,6BAA6B;AAAA,UAC7C;AAAA,UACA;AAAA,UACA,YACE;AAAA,QACJ,CAAC;AAAA,MACH;AAEA,oBAAc,KAAK;AAAA,QACjB,MAAM;AAAA,QACN,OAAO;AAAA,QACP,MAAM;AAAA,QACN,IAAI,SAAS,IAAI,SAAS,IAAI,SAAS,KAAK,OAAO;AAAA,MACrD,CAAC;AAAA,IACH;AAAA,EACF;AAOA,MAAI,QAAQ,SAAS;AACnB,kBAAc,KAAK;AAAA,MACjB,MAAM;AAAA,MACN,MAAM,SAAS;AAAA,MACf,WAAW,CAAC;AAAA,IACd,CAAC;AAAA,EACH;AAEA,QAAM,eAAe;AAAA,IACnB,MAAM;AAAA,IACN,SAAS;AAAA,IACT,aAAa,SAAS,UAAU,CAAC,GAAG;AAAA,IACpC,MAAM;AAAA,IACN,OAAO,SAAS;AAAA,EAClB;AAEA,SAAO;AACT;AAMA,eAAsB,YACpB,UACA,cACA,mBACA,OACA,QACA,SAO2B;AAC3B,QAAM,SAAS,gBAAgB;AAC/B,QAAM,eAAe,gBAAgB;AACrC,QAAM,iBAAiB,SAAS;AAEhC,qBAAmB;AAEnB,QAAM,eAAe,SAAS,gBAAgB,aAAa,SAAS,MAAM;AAC1E,MAAI;AAEJ,QAAM,iBAAiB,kBAAkB;AACzC,cAAY,IAAI,uBAAuB;AAAA,IACrC,mBAAmB,CAAC,CAAC;AAAA,IACrB,gBAAgB,cAAc;AAAA,IAC9B,kBAAkB,cAAc;AAAA,IAChC,uBAAuB,cAAc;AAAA,IACrC,sBAAsB,cAAc;AAAA,IACpC,qBAAqB,cAAc;AAAA,IACnC,0BAA0B,CAAC,CAAC,cAAc;AAAA,IAC1C,cAAc,SAAS;AAAA,IACvB,WAAW,kBAAkB,GAAG;AAAA,EAClC,CAAC;AAED,MAAI,cAAc;AAChB,YAAQ,aAAa;AAAA,EACvB,OAAO;AACL,YAAQ,SAAS,SAAS,cAAc,aAAa;AAAA,EACvD;AAEA,MAAI,SAAS,qBAAqB;AAChC,mBAAe,CAAC,sBAAsB,GAAG,GAAG,YAAY;AAAA,EAC1D;AAEA,QAAM,SAA2B,qBAAqB,YAAY,EAAE;AAAA,IAClE,QAAM;AAAA,MACJ,GAAI,yBACA,EAAE,eAAe,EAAE,MAAM,YAAY,EAAE,IACvC,CAAC;AAAA,MACL,MAAM;AAAA,MACN,MAAM;AAAA,IACR;AAAA,EACF;AAGA,mBAAiB,KAAK,UAAU,MAAM,EAAE,MAAM;AAE9C,QAAM,cAAc,MAAM,QAAQ;AAAA,IAChC,MAAM,IAAI,OAAM,MAAK;AACnB,YAAM,SAAS;AAAA,QACb,MAAM;AAAA,QACN,UAAU;AAAA,UACR,MAAM,EAAE;AAAA,UACR,aAAa,MAAM,wBAAwB,CAAC;AAAA,UAC5C,YACE,qBAAqB,KAAK,EAAE,kBACxB,EAAE,kBACF,gBAAgB,EAAE,WAAW;AAAA,QACrC;AAAA,MACF;AACA,uBAAiB,KAAK,UAAU,MAAM,EAAE,MAAM;AAC9C,aAAO;AAAA,IACT,CAAC;AAAA,EACH;AAEA,QAAM,eAAe,OAAO;AAAA,IAC1B,QACG;AAAA,MACC,MAAM;AAAA,MACN,SAAS,EAAE;AAAA,IACb;AAAA,EACJ;AAEA,QAAM,iBAAiB,yCAAyC,QAAQ;AACxE,mBAAiB,KAAK,UAAU,cAAc,EAAE,MAAM;AAEtD,QAAM,wBAAwB,KAAK,IAAI;AAEvC,8BAA4B;AAAA,IAC1B,YAAY,aAAa,KAAK,IAAI;AAAA,IAClC,cAAc,qBAAqB,KAAK;AAAA,IACxC,WAAW,CAAC;AAAA,IACZ,aAAa,aAAa,KAAK,IAAI;AAAA,EACrC,CAAC;AAED,MAAI,QAAQ,KAAK,IAAI;AACrB,MAAI,gBAAgB;AACpB,MAAI;AAEJ,MAAI;AACF,eAAW,MAAM;AAAA,MACf,OAAM,YAAW;AACf,wBAAgB;AAChB,gBAAQ,KAAK,IAAI;AACjB,cAAM,YAAY,wBAAwB,YAAY;AACtD,cAAM,SAAS,YAAY,KAAK;AAGhC,cAAM,mBACJ,UACA,MAAM,WAAW,IAAI,KACrB,MAAM,WAAW,SAAS,KAC1B,UAAU,QACV,UAAU,aACV,UAAU;AAEZ,cAAM,OAA0C;AAAA,UAC9C;AAAA,UAEA,GAAI,SACA,EAAE,uBAAuB,UAAU,IACnC,EAAE,YAAY,UAAU;AAAA,UAC5B,UAAU,CAAC,GAAG,cAAc,GAAG,cAAc;AAAA;AAAA,UAG7C,GAAI,CAAC,oBAAoB,EAAE,aAAa,uBAAuB;AAAA,QACjE;AACA,YAAI,OAAO,QAAQ;AACjB;AAAC,UAAC,KAA2C,SAAS;AACtD,eAAK,iBAAiB;AAAA,YACpB,eAAe;AAAA,UACjB;AAAA,QACF;AAEA,YAAI,YAAY,SAAS,GAAG;AAC1B,eAAK,QAAQ;AACb,eAAK,cAAc;AAAA,QACrB;AAIA,cAAM,kBAAkB,mBACpB,MAAM,mBAAmB,cAAc,QAAQ,IAC/C;AACJ,YAAI,iBAAiB;AACnB,eAAK,mBAAmB;AAAA,QAC1B;AAEA,YAAI,gBAAgB,aAAa,WAAW;AAC1C,sBAAY,IAAI,4BAA4B;AAAA,YAC1C,kBAAkB,aAAa;AAAA,YAC/B,WAAW,aAAa;AAAA,YACxB,UAAU,aAAa;AAAA,YACvB,SAAS,aAAa;AAAA,YACtB,cAAc,CAAC,CAAC,aAAa;AAAA,YAC7B,WAAW,kBAAkB,GAAG;AAAA,UAClC,CAAC;AAQD,cAAI,oBAAoB,sBAAsB,YAAY,GAAG;AAC3D,kBAAM,UAAU,oBAAoB,cAAc,YAAY;AAE9D,kBAAM,gBAAsC;AAAA,cAC1C,UAAU;AAAA,cACV,cAAc,aAAa,IAAI,CAAAA,OAAKA,GAAE,OAAiB;AAAA,cACvD;AAAA,cACA,WAAW,wBAAwB,YAAY;AAAA,cAC/C,QAAQ,OAAO;AAAA,cACf;AAAA,cACA,aACE,YAAY,KAAK,KACjB,MAAM,WAAW,IAAI,KACrB,MAAM,WAAW,SAAS,IACtB,SACA;AAAA,cACN,oBACE,gBAAgB,eAAe;AAAA,cACjC,WAAW;AAAA,YACb;AAEA,kBAAM,UAAU,QAAQ,cAAc,aAAa;AAEnD,kBAAM,EAAE,qBAAqB,IAAI,MAAM,OAAO,WAAW;AACzD,kBAAMC,YAAW,MAAM;AAAA,cACrB;AAAA,cACA;AAAA,cACA;AAAA,YACF;AACA,kBAAM,kBAAkB,QAAQ,cAAcA,SAAQ;AAEtD,kBAAM,aAAa;AAAA,cACjB,MAAM;AAAA,cACN,SAAS,gBAAgB;AAAA,cACzB,YAAY,gBAAgB;AAAA,cAC5B,OAAO;AAAA,gBACL,eAAe,gBAAgB,MAAM;AAAA,gBACrC,mBAAmB,gBAAgB,MAAM;AAAA,cAC3C;AAAA,YACF;AACA,kBAAM,iBAAiB,KAAK,IAAI,IAAI;AACpC,kBAAM,kBAAkB,gBAAgB,MAAM;AAC9C,kBAAM,mBAAmB,gBAAgB,MAAM;AAC/C,kBAAM,cAAc,iBAAiB;AAAA,cACnC;AAAA,cACA,aAAa;AAAA,cACb,cAAc;AAAA,cACd,iBAAiB;AAAA,cACjB,qBAAqB;AAAA,YACvB,CAAC;AACD,2BAAe,aAAa,cAAc;AAC1C;AAAA,cACE;AAAA,gBACE,aAAa;AAAA,gBACb,cAAc;AAAA,gBACd,qBAAqB;AAAA,gBACrB,iBAAiB;AAAA,cACnB;AAAA,cACA;AAAA,cACA;AAAA,cACA,iBACK;AAAA,gBACC,SAAS,eAAe;AAAA,gBACxB,WAAW,eAAe;AAAA,gBAC1B;AAAA,cACF,IACA;AAAA,YACN;AACA,8BAAkB;AAAA,cAChB,cAAc,aAAa,KAAK,IAAI;AAAA,cACpC,UAAU,CAAC,GAAG,cAAc,GAAG,cAAc;AAAA,cAC7C,UAAU;AAAA,cACV,OAAO;AAAA,gBACL,aAAa;AAAA,gBACb,cAAc;AAAA,cAChB;AAAA,cACA,QAAQ,EAAE,OAAO,KAAK,KAAK,IAAI,EAAE;AAAA,cACjC,WAAW;AAAA,YACb,CAAC;AACD,kBAAM,eAAiC;AAAA,cACrC,MAAM;AAAA,cACN,SAAS;AAAA,cACT,SAAS;AAAA,cACT,YAAY;AAAA,cACZ,MAAM,WAAW;AAAA,cACjB,YAAY,gBAAgB;AAAA,YAC9B;AACA,mBAAO;AAAA,UACT;AAIA,gBAAM,qBAAqB,YAAY,aAAa,SAAS,IACzD,+BACA;AACJ,gBAAM,IAAI,MAAM,mBAAmB,cAAc,MAAM,GAAG,IAAI,MAAM;AACpE,cAAI;AACJ,cAAI,KAAK,QAAQ;AACf,4BAAgB,MAAM;AAAA,cACpB;AAAA,cACA;AAAA,YACF;AAAA,UACF,OAAO;AACL,4BAAgB;AAAA,UAClB;AACA,gBAAM,IAAI,iCAAiC,eAAe,KAAK;AAC/D,iBAAO;AAAA,QACT,OAAO;AACL,gBAAM,eAAe;AAAA,YACnB,oBAAoB,CAAC,CAAC;AAAA,YACtB,gBAAgB,cAAc;AAAA,YAC9B,iBAAiB,CAAC,CAAC,cAAc;AAAA,YACjC,gBAAgB;AAAA,YAChB,WAAW,kBAAkB,GAAG;AAAA,UAClC;AACA,sBAAY,MAAM,0BAA0B,YAAY;AACxD,gBAAM,IAAI;AAAA,YACR,8CAA8C,KAAK,2DAA2D,KAAK,UAAU,YAAY,CAAC;AAAA,UAC5I;AAAA,QACF;AAAA,MACF;AAAA,MACA;AAAA,QACE;AAAA,QACA,SAAS,gBAAgB,UACrB,CAAC,EAAE,SAAS,YAAY,OAAO,QAAQ,MAAM;AAC3C,oCAA0B,eAAe,SAAU;AAAA,YACjD;AAAA,YACA;AAAA,YACA,cAAc,MAAM;AAAA,YACpB;AAAA,UACF,CAAC;AAAA,QACH,IACA;AAAA,MACN;AAAA,IACF;AAAA,EACF,SAAS,OAAO;AACd,aAAS,KAAK;AACd,WAAO,6BAA6B,OAAO,QAAQ;AAAA,EACrD;AACA,QAAM,aAAa,KAAK,IAAI,IAAI;AAChC,QAAM,6BAA6B,KAAK,IAAI,IAAI;AAEhD,QAAM,cAAc,SAAS,OAAO,iBAAiB;AACrD,QAAM,eAAe,SAAS,OAAO,qBAAqB;AAC1D,QAAM,uBACJ,SAAS,OAAO,sBAAsB,iBAAiB;AAEzD,QAAM,2BAA2B;AAEjC,QAAM,UAAU,iBAAiB;AAAA,IAC/B;AAAA,IACA;AAAA,IACA;AAAA,IACA,iBAAiB;AAAA,IACjB,qBAAqB;AAAA,EACvB,CAAC;AAED,iBAAe,SAAS,0BAA0B;AAGlD;AAAA,IACE;AAAA,MACE;AAAA,MACA;AAAA,MACA,qBAAqB;AAAA,MACrB,iBAAiB;AAAA,IACnB;AAAA,IACA;AAAA,IACA;AAAA,IACA,iBACK;AAAA,MACC,SAAS,eAAe;AAAA,MACxB,WAAW,eAAe;AAAA,MAC1B;AAAA,IACF,IACA;AAAA,EACN;AAGA,oBAAkB;AAAA,IAChB,cAAc,aAAa,KAAK,IAAI;AAAA,IACpC,UAAU,CAAC,GAAG,cAAc,GAAG,cAAc;AAAA,IAC7C;AAAA,IACA,OAAO;AAAA,MACL;AAAA,MACA;AAAA,IACF;AAAA,IACA,QAAQ;AAAA,MACN;AAAA,MACA,KAAK,KAAK,IAAI;AAAA,IAChB;AAAA,IACA,WAAW;AAAA,EACb,CAAC;AAED,SAAO;AAAA,IACL,SAAS;AAAA,MACP,GAAG;AAAA,MACH,SAAS,wBAAwB,SAAS,OAAO;AAAA,MACjD,OAAO;AAAA,QACL,cAAc;AAAA,QACd,eAAe;AAAA,QACf,yBAAyB;AAAA,QACzB,6BAA6B;AAAA,MAC/B;AAAA,IACF;AAAA,IACA;AAAA,IACA;AAAA,IACA,MAAM;AAAA,IACN,MAAM,WAAW;AAAA,EACnB;AACF;",
+  "names": ["s", "response"]
 }

package/dist/services/llm/types.js

@@ -1,3 +1,4 @@
+import { randomUUID } from "crypto";
 import { APIConnectionError, APIError } from "@anthropic-ai/sdk";
 import { addToTotalCost } from "../../cost-tracker.js";
 import { recordTokenUsage } from "../../core/tokenStatsManager.js";
@@ -8,9 +9,7 @@ import {
   abortableDelay,
   getRetryDelay as sharedGetRetryDelay
 } from "../../utils/async.js";
-import {
-  debug as debugLogger
-} from "../../utils/debugLogger.js";
+import { debug as debugLogger } from "../../utils/debugLogger.js";
 import { setStreamingState } from "../../utils/streamingState.js";
 import { addRetryEventToTranscript } from "../../utils/agentTranscripts.js";
 const API_ERROR_MESSAGE_PREFIX = "API Error";
@@ -24,7 +23,7 @@ const SONNET_COST_PER_MILLION_OUTPUT_TOKENS = 15;
 const SONNET_COST_PER_MILLION_PROMPT_CACHE_WRITE_TOKENS = 3.75;
 const SONNET_COST_PER_MILLION_PROMPT_CACHE_READ_TOKENS = 0.3;
 const MAIN_QUERY_TEMPERATURE = 1;
-const MAX_RETRIES = process.env.USER_TYPE === "SWE_BENCH" ? 100 : 10;
+const MAX_RETRIES = process.env.USER_TYPE === "SWE_BENCH" ? 100 : 5;
 const BASE_DELAY_MS = 500;
 function getMetadata() {
   return {
@@ -71,7 +70,8 @@ async function withRetry(operation, options = {}) {
         phase: "retrying",
         retryCount: attempt,
         maxRetries,
-        errorName: error.name
+        errorName: error.name,
+        retryDelayMs: delayMs
       });
       debugLogger.flow("API_RETRY_ATTEMPT", {
         attempt,
@@ -111,23 +111,93 @@ async function withRetry(operation, options = {}) {
 function getMaxTokensFromProfile(modelProfile) {
   return modelProfile?.maxTokens || 8e3;
 }
-function getModelInputTokenCostUSD(model) {
+function findModelCostInfo(model) {
   for (const providerModels of Object.values(models)) {
-    const modelInfo = providerModels.find((m) => m.model === model);
-    if (modelInfo) {
-      return modelInfo.input_cost_per_token || 0;
-    }
+    const modelInfo = providerModels.find(
+      (m) => m.model === model
+    );
+    if (modelInfo) return modelInfo;
   }
+  return void 0;
+}
+function getModelInputTokenCostUSD(model) {
+  const info = findModelCostInfo(model);
+  if (info?.input_cost_per_token) return info.input_cost_per_token;
+  debugLogger.warn("UNKNOWN_MODEL_COST", { model, direction: "input" });
   return 3e-6;
 }
 function getModelOutputTokenCostUSD(model) {
-  for (const providerModels of Object.values(models)) {
-    const modelInfo = providerModels.find((m) => m.model === model);
-    if (modelInfo) {
-      return modelInfo.output_cost_per_token || 0;
+  const info = findModelCostInfo(model);
+  if (info?.output_cost_per_token) return info.output_cost_per_token;
+  debugLogger.warn("UNKNOWN_MODEL_COST", { model, direction: "output" });
+  return 15e-6;
+}
+function getModelCacheReadCostUSD(model) {
+  const info = findModelCostInfo(model);
+  if (info?.cache_read_input_token_cost) return info.cache_read_input_token_cost;
+  return getModelInputTokenCostUSD(model) * 0.1;
+}
+function getModelCacheCreationCostUSD(model) {
+  const info = findModelCostInfo(model);
+  if (info?.cache_creation_input_token_cost !== void 0)
+    return info.cache_creation_input_token_cost;
+  return getModelInputTokenCostUSD(model) * 1.25;
+}
+function calculateCostUSD(params) {
+  return params.inputTokens * getModelInputTokenCostUSD(params.model) + params.outputTokens * getModelOutputTokenCostUSD(params.model) + params.cacheReadTokens * getModelCacheReadCostUSD(params.model) + params.cacheCreationTokens * getModelCacheCreationCostUSD(params.model);
+}
+function createAssistantAPIErrorMessage(content) {
+  return {
+    type: "assistant",
+    costUSD: 0,
+    durationMs: 0,
+    uuid: randomUUID(),
+    isApiErrorMessage: true,
+    message: {
+      id: randomUUID(),
+      model: "<synthetic>",
+      role: "assistant",
+      stop_reason: "stop_sequence",
+      stop_sequence: "",
+      type: "message",
+      usage: {
+        input_tokens: 0,
+        output_tokens: 0,
+        cache_creation_input_tokens: 0,
+        cache_read_input_tokens: 0
+      },
+      content: [
+        {
+          type: "text",
+          text: content === "" ? NO_CONTENT_MESSAGE : content,
+          citations: []
+        }
+      ]
     }
+  };
+}
+function getAssistantMessageFromError(error, providerTag = "LLM") {
+  if (error instanceof Error && error.message.includes("prompt is too long")) {
+    return createAssistantAPIErrorMessage(PROMPT_TOO_LONG_ERROR_MESSAGE);
   }
-  return 15e-6;
+  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(`${providerTag}_API_ERROR`, {
+        message: error.message,
+        stack: error.stack
+      });
+    }
+    return createAssistantAPIErrorMessage(
+      `${API_ERROR_MESSAGE_PREFIX}: ${error.message}`
+    );
+  }
+  return createAssistantAPIErrorMessage(API_ERROR_MESSAGE_PREFIX);
 }
 export {
   API_ERROR_MESSAGE_PREFIX,
@@ -144,9 +214,13 @@ export {
   SONNET_COST_PER_MILLION_PROMPT_CACHE_WRITE_TOKENS,
   addRetryEventToTranscript,
   addToTotalCost,
+  calculateCostUSD,
   debugLogger,
+  getAssistantMessageFromError,
   getMaxTokensFromProfile,
   getMetadata,
+  getModelCacheCreationCostUSD,
+  getModelCacheReadCostUSD,
   getModelInputTokenCostUSD,
   getModelOutputTokenCostUSD,
   logError,

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

@@ -1,7 +1,7 @@
 {
   "version": 3,
   "sources": ["../../../src/services/llm/types.ts"],
-  "sourcesContent": ["/**\n * Shared types, constants, and utility functions for LLM providers.\n */\nimport { APIConnectionError, APIError } from '@anthropic-ai/sdk'\n\nimport { addToTotalCost } from '@costTracker'\nimport { recordTokenUsage } from '@core/tokenStatsManager'\nimport type { TokenTrackingContext } from '@core/tokenStats'\nimport models from '@constants/models'\nimport { getOrCreateUserID } from '@utils/config'\nimport { logError, SESSION_ID } from '@utils/log'\nimport {\n  abortableDelay,\n  getRetryDelay as sharedGetRetryDelay,\n} from '@utils/async'\nimport {\n  debug as debugLogger,\n} from '@utils/debugLogger'\nimport { setStreamingState } from '@utils/streamingState'\nimport { addRetryEventToTranscript } from '@utils/agentTranscripts'\n\n// \u2500\u2500 Error message constants \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 const API_ERROR_MESSAGE_PREFIX = 'API Error'\nexport const PROMPT_TOO_LONG_ERROR_MESSAGE = 'Prompt is too long'\nexport const CREDIT_BALANCE_TOO_LOW_ERROR_MESSAGE = 'Credit balance is too low'\nexport const INVALID_API_KEY_ERROR_MESSAGE =\n  'Invalid API key \u00B7 Please run /login'\nexport const NO_CONTENT_MESSAGE = '(no content)'\n\n// \u2500\u2500 Cost constants \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\nconst PROMPT_CACHING_ENABLED_VALUE = !process.env.DISABLE_PROMPT_CACHING\nexport { PROMPT_CACHING_ENABLED_VALUE as PROMPT_CACHING_ENABLED }\n\n// @see https://docs.anthropic.com/en/docs/about-claude/models#model-comparison-table\nexport const SONNET_COST_PER_MILLION_INPUT_TOKENS = 3\nexport const SONNET_COST_PER_MILLION_OUTPUT_TOKENS = 15\nexport const SONNET_COST_PER_MILLION_PROMPT_CACHE_WRITE_TOKENS = 3.75\nexport const SONNET_COST_PER_MILLION_PROMPT_CACHE_READ_TOKENS = 0.3\n\nexport const MAIN_QUERY_TEMPERATURE = 1 // to get more variation for binary feedback\n\n// \u2500\u2500 Retry logic \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 const MAX_RETRIES = process.env.USER_TYPE === 'SWE_BENCH' ? 100 : 10\nconst BASE_DELAY_MS = 500\n\nexport interface RetryOptions {\n  maxRetries?: number\n  signal?: AbortSignal\n  /** Callback for retry events. If provided, suppresses default debugLogger output. */\n  onRetry?: (info: {\n    attempt: number\n    maxRetries: number\n    error: Error\n    delayMs: number\n  }) => void\n}\n\nexport function getMetadata() {\n  return {\n    user_id: `${getOrCreateUserID()}_${SESSION_ID}`,\n  }\n}\n\n/**\n * Calculate retry delay \u2014 delegates to shared utility with Retry-After header support\n */\nfunction getRetryDelay(\n  attempt: number,\n  retryAfterHeader?: string | null,\n): number {\n  return sharedGetRetryDelay(attempt, retryAfterHeader)\n}\n\nexport function shouldRetry(error: APIError): boolean {\n  // Check for overloaded errors first and only retry for SWE_BENCH\n  if (error.message?.includes('\"type\":\"overloaded_error\"')) {\n    return process.env.USER_TYPE === 'SWE_BENCH'\n  }\n\n  // Note this is not a standard header.\n  const shouldRetryHeader = error.headers?.['x-should-retry']\n\n  // If the server explicitly says whether or not to retry, obey.\n  if (shouldRetryHeader === 'true') return true\n  if (shouldRetryHeader === 'false') return false\n\n  if (error instanceof APIConnectionError) {\n    return true\n  }\n\n  if (!error.status) return false\n\n  // Retry on request timeouts.\n  if (error.status === 408) return true\n\n  // Retry on lock timeouts.\n  if (error.status === 409) return true\n\n  // Retry on rate limits.\n  if (error.status === 429) return true\n\n  // Retry internal errors.\n  if (error.status && error.status >= 500) return true\n\n  return false\n}\n\nexport async function withRetry<T>(\n  operation: (attempt: number) => Promise<T>,\n  options: RetryOptions = {},\n): Promise<T> {\n  const maxRetries = options.maxRetries ?? MAX_RETRIES\n  let lastError: unknown\n\n  for (let attempt = 1; attempt <= maxRetries + 1; attempt++) {\n    try {\n      return await operation(attempt)\n    } catch (error) {\n      lastError = error\n      // Only retry if the error indicates we should\n      if (\n        attempt > maxRetries ||\n        !(error instanceof APIError) ||\n        !shouldRetry(error)\n      ) {\n        throw error\n      }\n\n      if (options.signal?.aborted) {\n        throw new Error('Request cancelled by user')\n      }\n\n      // Get retry-after header if available\n      const retryAfter = error.headers?.['retry-after'] ?? null\n      const delayMs = getRetryDelay(attempt, retryAfter)\n\n      // Update streaming state to show retry in UI\n      setStreamingState({\n        phase: 'retrying',\n        retryCount: attempt,\n        maxRetries: maxRetries,\n        errorName: error.name,\n      })\n\n      debugLogger.flow('API_RETRY_ATTEMPT', {\n        attempt,\n        maxRetries,\n        errorName: error.name,\n        errorMessage: error.message,\n        delayMs,\n      })\n\n      // Use callback if provided, otherwise fallback to debugLogger\n      if (options.onRetry) {\n        options.onRetry({\n          attempt,\n          maxRetries,\n          error,\n          delayMs,\n        })\n      } else {\n        debugLogger.warn('API_RETRY', {\n          errorName: error.name,\n          errorMessage: error.message,\n          delayMs,\n          attempt,\n          maxRetries,\n        })\n      }\n\n      try {\n        await abortableDelay(delayMs, options.signal)\n      } catch (delayError) {\n        // If aborted during delay, throw the error to stop retrying\n        if (delayError.message === 'Request was aborted') {\n          throw new Error('Request cancelled by user')\n        }\n        throw delayError\n      }\n    }\n  }\n\n  throw lastError\n}\n\n// \u2500\u2500 Cost utility 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\u2500\u2500\u2500\u2500\u2500\n\nexport function getMaxTokensFromProfile(modelProfile: any): number {\n  // Use ModelProfile maxTokens or reasonable default\n  return modelProfile?.maxTokens || 8000\n}\n\nexport function getModelInputTokenCostUSD(model: string): number {\n  // Find the model in the models object\n  for (const providerModels of Object.values(models)) {\n    const modelInfo = providerModels.find((m: any) => m.model === model)\n    if (modelInfo) {\n      return modelInfo.input_cost_per_token || 0\n    }\n  }\n  // Default fallback cost for unknown models\n  return 0.000003 // Default to Claude 3 Haiku cost\n}\n\nexport function getModelOutputTokenCostUSD(model: string): number {\n  // Find the model in the models object\n  for (const providerModels of Object.values(models)) {\n    const modelInfo = providerModels.find((m: any) => m.model === model)\n    if (modelInfo) {\n      return modelInfo.output_cost_per_token || 0\n    }\n  }\n  // Default fallback cost for unknown models\n  return 0.000015 // Default to Claude 3 Haiku cost\n}\n\n// Re-export commonly used dependencies for other llm modules\nexport {\n  addToTotalCost,\n  recordTokenUsage,\n  type TokenTrackingContext,\n  logError,\n  debugLogger,\n  addRetryEventToTranscript,\n}\n"],
-  "mappings": "AAGA,SAAS,oBAAoB,gBAAgB;AAE7C,SAAS,sBAAsB;AAC/B,SAAS,wBAAwB;AAEjC,OAAO,YAAY;AACnB,SAAS,yBAAyB;AAClC,SAAS,UAAU,kBAAkB;AACrC;AAAA,EACE;AAAA,EACA,iBAAiB;AAAA,OACZ;AACP;AAAA,EACE,SAAS;AAAA,OACJ;AACP,SAAS,yBAAyB;AAClC,SAAS,iCAAiC;AAInC,MAAM,2BAA2B;AACjC,MAAM,gCAAgC;AACtC,MAAM,uCAAuC;AAC7C,MAAM,gCACX;AACK,MAAM,qBAAqB;AAIlC,MAAM,+BAA+B,CAAC,QAAQ,IAAI;AAI3C,MAAM,uCAAuC;AAC7C,MAAM,wCAAwC;AAC9C,MAAM,oDAAoD;AAC1D,MAAM,mDAAmD;AAEzD,MAAM,yBAAyB;AAI/B,MAAM,cAAc,QAAQ,IAAI,cAAc,cAAc,MAAM;AACzE,MAAM,gBAAgB;AAcf,SAAS,cAAc;AAC5B,SAAO;AAAA,IACL,SAAS,GAAG,kBAAkB,CAAC,IAAI,UAAU;AAAA,EAC/C;AACF;AAKA,SAAS,cACP,SACA,kBACQ;AACR,SAAO,oBAAoB,SAAS,gBAAgB;AACtD;AAEO,SAAS,YAAY,OAA0B;AAEpD,MAAI,MAAM,SAAS,SAAS,2BAA2B,GAAG;AACxD,WAAO,QAAQ,IAAI,cAAc;AAAA,EACnC;AAGA,QAAM,oBAAoB,MAAM,UAAU,gBAAgB;AAG1D,MAAI,sBAAsB,OAAQ,QAAO;AACzC,MAAI,sBAAsB,QAAS,QAAO;AAE1C,MAAI,iBAAiB,oBAAoB;AACvC,WAAO;AAAA,EACT;AAEA,MAAI,CAAC,MAAM,OAAQ,QAAO;AAG1B,MAAI,MAAM,WAAW,IAAK,QAAO;AAGjC,MAAI,MAAM,WAAW,IAAK,QAAO;AAGjC,MAAI,MAAM,WAAW,IAAK,QAAO;AAGjC,MAAI,MAAM,UAAU,MAAM,UAAU,IAAK,QAAO;AAEhD,SAAO;AACT;AAEA,eAAsB,UACpB,WACA,UAAwB,CAAC,GACb;AACZ,QAAM,aAAa,QAAQ,cAAc;AACzC,MAAI;AAEJ,WAAS,UAAU,GAAG,WAAW,aAAa,GAAG,WAAW;AAC1D,QAAI;AACF,aAAO,MAAM,UAAU,OAAO;AAAA,IAChC,SAAS,OAAO;AACd,kBAAY;AAEZ,UACE,UAAU,cACV,EAAE,iBAAiB,aACnB,CAAC,YAAY,KAAK,GAClB;AACA,cAAM;AAAA,MACR;AAEA,UAAI,QAAQ,QAAQ,SAAS;AAC3B,cAAM,IAAI,MAAM,2BAA2B;AAAA,MAC7C;AAGA,YAAM,aAAa,MAAM,UAAU,aAAa,KAAK;AACrD,YAAM,UAAU,cAAc,SAAS,UAAU;AAGjD,wBAAkB;AAAA,QAChB,OAAO;AAAA,QACP,YAAY;AAAA,QACZ;AAAA,QACA,WAAW,MAAM;AAAA,MACnB,CAAC;AAED,kBAAY,KAAK,qBAAqB;AAAA,QACpC;AAAA,QACA;AAAA,QACA,WAAW,MAAM;AAAA,QACjB,cAAc,MAAM;AAAA,QACpB;AAAA,MACF,CAAC;AAGD,UAAI,QAAQ,SAAS;AACnB,gBAAQ,QAAQ;AAAA,UACd;AAAA,UACA;AAAA,UACA;AAAA,UACA;AAAA,QACF,CAAC;AAAA,MACH,OAAO;AACL,oBAAY,KAAK,aAAa;AAAA,UAC5B,WAAW,MAAM;AAAA,UACjB,cAAc,MAAM;AAAA,UACpB;AAAA,UACA;AAAA,UACA;AAAA,QACF,CAAC;AAAA,MACH;AAEA,UAAI;AACF,cAAM,eAAe,SAAS,QAAQ,MAAM;AAAA,MAC9C,SAAS,YAAY;AAEnB,YAAI,WAAW,YAAY,uBAAuB;AAChD,gBAAM,IAAI,MAAM,2BAA2B;AAAA,QAC7C;AACA,cAAM;AAAA,MACR;AAAA,IACF;AAAA,EACF;AAEA,QAAM;AACR;AAIO,SAAS,wBAAwB,cAA2B;AAEjE,SAAO,cAAc,aAAa;AACpC;AAEO,SAAS,0BAA0B,OAAuB;AAE/D,aAAW,kBAAkB,OAAO,OAAO,MAAM,GAAG;AAClD,UAAM,YAAY,eAAe,KAAK,CAAC,MAAW,EAAE,UAAU,KAAK;AACnE,QAAI,WAAW;AACb,aAAO,UAAU,wBAAwB;AAAA,IAC3C;AAAA,EACF;AAEA,SAAO;AACT;AAEO,SAAS,2BAA2B,OAAuB;AAEhE,aAAW,kBAAkB,OAAO,OAAO,MAAM,GAAG;AAClD,UAAM,YAAY,eAAe,KAAK,CAAC,MAAW,EAAE,UAAU,KAAK;AACnE,QAAI,WAAW;AACb,aAAO,UAAU,yBAAyB;AAAA,IAC5C;AAAA,EACF;AAEA,SAAO;AACT;",
+  "sourcesContent": ["/**\n * Shared types, constants, and utility functions for LLM providers.\n */\nimport { randomUUID } from 'crypto'\nimport { APIConnectionError, APIError } from '@anthropic-ai/sdk'\n\nimport { addToTotalCost } from '@costTracker'\nimport { recordTokenUsage } from '@core/tokenStatsManager'\nimport type { TokenTrackingContext } from '@core/tokenStats'\nimport models from '@constants/models'\nimport { getOrCreateUserID } from '@utils/config'\nimport { logError, SESSION_ID } from '@utils/log'\nimport {\n  abortableDelay,\n  getRetryDelay as sharedGetRetryDelay,\n} from '@utils/async'\nimport { debug as debugLogger } from '@utils/debugLogger'\nimport { setStreamingState } from '@utils/streamingState'\nimport { addRetryEventToTranscript } from '@utils/agentTranscripts'\n\n// \u2500\u2500 Error message constants \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 const API_ERROR_MESSAGE_PREFIX = 'API Error'\nexport const PROMPT_TOO_LONG_ERROR_MESSAGE = 'Prompt is too long'\nexport const CREDIT_BALANCE_TOO_LOW_ERROR_MESSAGE = 'Credit balance is too low'\nexport const INVALID_API_KEY_ERROR_MESSAGE =\n  'Invalid API key \u00B7 Please run /login'\nexport const NO_CONTENT_MESSAGE = '(no content)'\n\n// \u2500\u2500 Cost constants \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\nconst PROMPT_CACHING_ENABLED_VALUE = !process.env.DISABLE_PROMPT_CACHING\nexport { PROMPT_CACHING_ENABLED_VALUE as PROMPT_CACHING_ENABLED }\n\n// @see https://docs.anthropic.com/en/docs/about-claude/models#model-comparison-table\nexport const SONNET_COST_PER_MILLION_INPUT_TOKENS = 3\nexport const SONNET_COST_PER_MILLION_OUTPUT_TOKENS = 15\nexport const SONNET_COST_PER_MILLION_PROMPT_CACHE_WRITE_TOKENS = 3.75\nexport const SONNET_COST_PER_MILLION_PROMPT_CACHE_READ_TOKENS = 0.3\n\nexport const MAIN_QUERY_TEMPERATURE = 1 // to get more variation for binary feedback\n\n// \u2500\u2500 Retry logic \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 const MAX_RETRIES = process.env.USER_TYPE === 'SWE_BENCH' ? 100 : 5\nconst BASE_DELAY_MS = 500\n\nexport interface RetryOptions {\n  maxRetries?: number\n  signal?: AbortSignal\n  /** Callback for retry events. If provided, suppresses default debugLogger output. */\n  onRetry?: (info: {\n    attempt: number\n    maxRetries: number\n    error: Error\n    delayMs: number\n  }) => void\n}\n\nexport function getMetadata() {\n  return {\n    user_id: `${getOrCreateUserID()}_${SESSION_ID}`,\n  }\n}\n\n/**\n * Calculate retry delay \u2014 delegates to shared utility with Retry-After header support\n */\nfunction getRetryDelay(\n  attempt: number,\n  retryAfterHeader?: string | null,\n): number {\n  return sharedGetRetryDelay(attempt, retryAfterHeader)\n}\n\nexport function shouldRetry(error: APIError): boolean {\n  // Check for overloaded errors first and only retry for SWE_BENCH\n  if (error.message?.includes('\"type\":\"overloaded_error\"')) {\n    return process.env.USER_TYPE === 'SWE_BENCH'\n  }\n\n  // Note this is not a standard header.\n  const shouldRetryHeader = error.headers?.['x-should-retry']\n\n  // If the server explicitly says whether or not to retry, obey.\n  if (shouldRetryHeader === 'true') return true\n  if (shouldRetryHeader === 'false') return false\n\n  if (error instanceof APIConnectionError) {\n    return true\n  }\n\n  if (!error.status) return false\n\n  // Retry on request timeouts.\n  if (error.status === 408) return true\n\n  // Retry on lock timeouts.\n  if (error.status === 409) return true\n\n  // Retry on rate limits.\n  if (error.status === 429) return true\n\n  // Retry internal errors.\n  if (error.status && error.status >= 500) return true\n\n  return false\n}\n\nexport async function withRetry<T>(\n  operation: (attempt: number) => Promise<T>,\n  options: RetryOptions = {},\n): Promise<T> {\n  const maxRetries = options.maxRetries ?? MAX_RETRIES\n  let lastError: unknown\n\n  for (let attempt = 1; attempt <= maxRetries + 1; attempt++) {\n    try {\n      return await operation(attempt)\n    } catch (error) {\n      lastError = error\n      // Only retry if the error indicates we should\n      if (\n        attempt > maxRetries ||\n        !(error instanceof APIError) ||\n        !shouldRetry(error)\n      ) {\n        throw error\n      }\n\n      if (options.signal?.aborted) {\n        throw new Error('Request cancelled by user')\n      }\n\n      // Get retry-after header if available\n      const retryAfter = error.headers?.['retry-after'] ?? null\n      const delayMs = getRetryDelay(attempt, retryAfter)\n\n      // Update streaming state to show retry in UI\n      setStreamingState({\n        phase: 'retrying',\n        retryCount: attempt,\n        maxRetries: maxRetries,\n        errorName: error.name,\n        retryDelayMs: delayMs,\n      })\n\n      debugLogger.flow('API_RETRY_ATTEMPT', {\n        attempt,\n        maxRetries,\n        errorName: error.name,\n        errorMessage: error.message,\n        delayMs,\n      })\n\n      // Use callback if provided, otherwise fallback to debugLogger\n      if (options.onRetry) {\n        options.onRetry({\n          attempt,\n          maxRetries,\n          error,\n          delayMs,\n        })\n      } else {\n        debugLogger.warn('API_RETRY', {\n          errorName: error.name,\n          errorMessage: error.message,\n          delayMs,\n          attempt,\n          maxRetries,\n        })\n      }\n\n      try {\n        await abortableDelay(delayMs, options.signal)\n      } catch (delayError) {\n        // If aborted during delay, throw the error to stop retrying\n        if (delayError.message === 'Request was aborted') {\n          throw new Error('Request cancelled by user')\n        }\n        throw delayError\n      }\n    }\n  }\n\n  throw lastError\n}\n\n// \u2500\u2500 Cost utility 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\u2500\u2500\u2500\u2500\u2500\n\nexport function getMaxTokensFromProfile(modelProfile: {\n  maxTokens?: number\n}): number {\n  // Use ModelProfile maxTokens or reasonable default\n  return modelProfile?.maxTokens || 8000\n}\n\n/** Model cost field interface for type-safe lookups */\ninterface ModelCostInfo {\n  model?: string\n  input_cost_per_token?: number\n  output_cost_per_token?: number\n  cache_read_input_token_cost?: number\n  cache_creation_input_token_cost?: number\n}\n\n/**\n * Look up a model's cost info from the models registry.\n * Returns the first match across all providers, or undefined if not found.\n */\nfunction findModelCostInfo(model: string): ModelCostInfo | undefined {\n  for (const providerModels of Object.values(models)) {\n    const modelInfo = providerModels.find(\n      (m: { model?: string }) => m.model === model,\n    )\n    if (modelInfo) return modelInfo as ModelCostInfo\n  }\n  return undefined\n}\n\nexport function getModelInputTokenCostUSD(model: string): number {\n  const info = findModelCostInfo(model)\n  if (info?.input_cost_per_token) return info.input_cost_per_token\n  debugLogger.warn('UNKNOWN_MODEL_COST', { model, direction: 'input' })\n  return 0.000003 // Default to Claude Sonnet cost\n}\n\nexport function getModelOutputTokenCostUSD(model: string): number {\n  const info = findModelCostInfo(model)\n  if (info?.output_cost_per_token) return info.output_cost_per_token\n  debugLogger.warn('UNKNOWN_MODEL_COST', { model, direction: 'output' })\n  return 0.000015 // Default to Claude Sonnet cost\n}\n\n/**\n * Get per-token cache read cost. Falls back to 10% of input cost if not\n * explicitly defined in the models registry.\n */\nexport function getModelCacheReadCostUSD(model: string): number {\n  const info = findModelCostInfo(model)\n  if (info?.cache_read_input_token_cost) return info.cache_read_input_token_cost\n  // Default: 10% of input cost (Anthropic standard cache pricing)\n  return getModelInputTokenCostUSD(model) * 0.1\n}\n\n/**\n * Get per-token cache creation cost. Falls back to 125% of input cost if not\n * explicitly defined in the models registry.\n */\nexport function getModelCacheCreationCostUSD(model: string): number {\n  const info = findModelCostInfo(model)\n  if (info?.cache_creation_input_token_cost !== undefined)\n    return info.cache_creation_input_token_cost\n  // Default: 125% of input cost (Anthropic standard cache pricing)\n  return getModelInputTokenCostUSD(model) * 1.25\n}\n\n/**\n * Unified cost calculation for all providers.\n * Uses per-token costs from the models registry.\n * Formula: tokens * cost_per_token (all costs are per-token in USD).\n */\nexport function calculateCostUSD(params: {\n  model: string\n  inputTokens: number\n  outputTokens: number\n  cacheReadTokens: number\n  cacheCreationTokens: number\n}): number {\n  return (\n    params.inputTokens * getModelInputTokenCostUSD(params.model) +\n    params.outputTokens * getModelOutputTokenCostUSD(params.model) +\n    params.cacheReadTokens * getModelCacheReadCostUSD(params.model) +\n    params.cacheCreationTokens * getModelCacheCreationCostUSD(params.model)\n  )\n}\n\n// \u2500\u2500 Shared error handling \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\n/**\n * Convert an API error into an AssistantMessage for UI display.\n * Shared between Anthropic and OpenAI providers to avoid duplication.\n */\n// Inline helper to avoid circular dependency with @utils/messages\nfunction createAssistantAPIErrorMessage(\n  content: string,\n): import('@query').AssistantMessage {\n  return {\n    type: 'assistant',\n    costUSD: 0,\n    durationMs: 0,\n    uuid: randomUUID(),\n    isApiErrorMessage: true,\n    message: {\n      id: randomUUID(),\n      model: '<synthetic>',\n      role: 'assistant' as const,\n      stop_reason: 'stop_sequence',\n      stop_sequence: '',\n      type: 'message',\n      usage: {\n        input_tokens: 0,\n        output_tokens: 0,\n        cache_creation_input_tokens: 0,\n        cache_read_input_tokens: 0,\n      },\n      content: [\n        {\n          type: 'text' as const,\n          text: content === '' ? NO_CONTENT_MESSAGE : content,\n          citations: [],\n        },\n      ],\n    },\n  }\n}\n\nexport function getAssistantMessageFromError(\n  error: unknown,\n  providerTag: string = 'LLM',\n): import('@query').AssistantMessage {\n  if (error instanceof Error && error.message.includes('prompt is too long')) {\n    return createAssistantAPIErrorMessage(PROMPT_TOO_LONG_ERROR_MESSAGE)\n  }\n  if (\n    error instanceof Error &&\n    error.message.includes('Your credit balance is too low')\n  ) {\n    return createAssistantAPIErrorMessage(CREDIT_BALANCE_TOO_LOW_ERROR_MESSAGE)\n  }\n  if (\n    error instanceof Error &&\n    error.message.toLowerCase().includes('x-api-key')\n  ) {\n    return createAssistantAPIErrorMessage(INVALID_API_KEY_ERROR_MESSAGE)\n  }\n  if (error instanceof Error) {\n    if (process.env.NODE_ENV === 'development') {\n      debugLogger.error(`${providerTag}_API_ERROR`, {\n        message: error.message,\n        stack: error.stack,\n      })\n    }\n    return createAssistantAPIErrorMessage(\n      `${API_ERROR_MESSAGE_PREFIX}: ${error.message}`,\n    )\n  }\n  return createAssistantAPIErrorMessage(API_ERROR_MESSAGE_PREFIX)\n}\n\n// Re-export commonly used dependencies for other llm modules\nexport {\n  addToTotalCost,\n  recordTokenUsage,\n  type TokenTrackingContext,\n  logError,\n  debugLogger,\n  addRetryEventToTranscript,\n}\n"],
+  "mappings": "AAGA,SAAS,kBAAkB;AAC3B,SAAS,oBAAoB,gBAAgB;AAE7C,SAAS,sBAAsB;AAC/B,SAAS,wBAAwB;AAEjC,OAAO,YAAY;AACnB,SAAS,yBAAyB;AAClC,SAAS,UAAU,kBAAkB;AACrC;AAAA,EACE;AAAA,EACA,iBAAiB;AAAA,OACZ;AACP,SAAS,SAAS,mBAAmB;AACrC,SAAS,yBAAyB;AAClC,SAAS,iCAAiC;AAInC,MAAM,2BAA2B;AACjC,MAAM,gCAAgC;AACtC,MAAM,uCAAuC;AAC7C,MAAM,gCACX;AACK,MAAM,qBAAqB;AAIlC,MAAM,+BAA+B,CAAC,QAAQ,IAAI;AAI3C,MAAM,uCAAuC;AAC7C,MAAM,wCAAwC;AAC9C,MAAM,oDAAoD;AAC1D,MAAM,mDAAmD;AAEzD,MAAM,yBAAyB;AAI/B,MAAM,cAAc,QAAQ,IAAI,cAAc,cAAc,MAAM;AACzE,MAAM,gBAAgB;AAcf,SAAS,cAAc;AAC5B,SAAO;AAAA,IACL,SAAS,GAAG,kBAAkB,CAAC,IAAI,UAAU;AAAA,EAC/C;AACF;AAKA,SAAS,cACP,SACA,kBACQ;AACR,SAAO,oBAAoB,SAAS,gBAAgB;AACtD;AAEO,SAAS,YAAY,OAA0B;AAEpD,MAAI,MAAM,SAAS,SAAS,2BAA2B,GAAG;AACxD,WAAO,QAAQ,IAAI,cAAc;AAAA,EACnC;AAGA,QAAM,oBAAoB,MAAM,UAAU,gBAAgB;AAG1D,MAAI,sBAAsB,OAAQ,QAAO;AACzC,MAAI,sBAAsB,QAAS,QAAO;AAE1C,MAAI,iBAAiB,oBAAoB;AACvC,WAAO;AAAA,EACT;AAEA,MAAI,CAAC,MAAM,OAAQ,QAAO;AAG1B,MAAI,MAAM,WAAW,IAAK,QAAO;AAGjC,MAAI,MAAM,WAAW,IAAK,QAAO;AAGjC,MAAI,MAAM,WAAW,IAAK,QAAO;AAGjC,MAAI,MAAM,UAAU,MAAM,UAAU,IAAK,QAAO;AAEhD,SAAO;AACT;AAEA,eAAsB,UACpB,WACA,UAAwB,CAAC,GACb;AACZ,QAAM,aAAa,QAAQ,cAAc;AACzC,MAAI;AAEJ,WAAS,UAAU,GAAG,WAAW,aAAa,GAAG,WAAW;AAC1D,QAAI;AACF,aAAO,MAAM,UAAU,OAAO;AAAA,IAChC,SAAS,OAAO;AACd,kBAAY;AAEZ,UACE,UAAU,cACV,EAAE,iBAAiB,aACnB,CAAC,YAAY,KAAK,GAClB;AACA,cAAM;AAAA,MACR;AAEA,UAAI,QAAQ,QAAQ,SAAS;AAC3B,cAAM,IAAI,MAAM,2BAA2B;AAAA,MAC7C;AAGA,YAAM,aAAa,MAAM,UAAU,aAAa,KAAK;AACrD,YAAM,UAAU,cAAc,SAAS,UAAU;AAGjD,wBAAkB;AAAA,QAChB,OAAO;AAAA,QACP,YAAY;AAAA,QACZ;AAAA,QACA,WAAW,MAAM;AAAA,QACjB,cAAc;AAAA,MAChB,CAAC;AAED,kBAAY,KAAK,qBAAqB;AAAA,QACpC;AAAA,QACA;AAAA,QACA,WAAW,MAAM;AAAA,QACjB,cAAc,MAAM;AAAA,QACpB;AAAA,MACF,CAAC;AAGD,UAAI,QAAQ,SAAS;AACnB,gBAAQ,QAAQ;AAAA,UACd;AAAA,UACA;AAAA,UACA;AAAA,UACA;AAAA,QACF,CAAC;AAAA,MACH,OAAO;AACL,oBAAY,KAAK,aAAa;AAAA,UAC5B,WAAW,MAAM;AAAA,UACjB,cAAc,MAAM;AAAA,UACpB;AAAA,UACA;AAAA,UACA;AAAA,QACF,CAAC;AAAA,MACH;AAEA,UAAI;AACF,cAAM,eAAe,SAAS,QAAQ,MAAM;AAAA,MAC9C,SAAS,YAAY;AAEnB,YAAI,WAAW,YAAY,uBAAuB;AAChD,gBAAM,IAAI,MAAM,2BAA2B;AAAA,QAC7C;AACA,cAAM;AAAA,MACR;AAAA,IACF;AAAA,EACF;AAEA,QAAM;AACR;AAIO,SAAS,wBAAwB,cAE7B;AAET,SAAO,cAAc,aAAa;AACpC;AAeA,SAAS,kBAAkB,OAA0C;AACnE,aAAW,kBAAkB,OAAO,OAAO,MAAM,GAAG;AAClD,UAAM,YAAY,eAAe;AAAA,MAC/B,CAAC,MAA0B,EAAE,UAAU;AAAA,IACzC;AACA,QAAI,UAAW,QAAO;AAAA,EACxB;AACA,SAAO;AACT;AAEO,SAAS,0BAA0B,OAAuB;AAC/D,QAAM,OAAO,kBAAkB,KAAK;AACpC,MAAI,MAAM,qBAAsB,QAAO,KAAK;AAC5C,cAAY,KAAK,sBAAsB,EAAE,OAAO,WAAW,QAAQ,CAAC;AACpE,SAAO;AACT;AAEO,SAAS,2BAA2B,OAAuB;AAChE,QAAM,OAAO,kBAAkB,KAAK;AACpC,MAAI,MAAM,sBAAuB,QAAO,KAAK;AAC7C,cAAY,KAAK,sBAAsB,EAAE,OAAO,WAAW,SAAS,CAAC;AACrE,SAAO;AACT;AAMO,SAAS,yBAAyB,OAAuB;AAC9D,QAAM,OAAO,kBAAkB,KAAK;AACpC,MAAI,MAAM,4BAA6B,QAAO,KAAK;AAEnD,SAAO,0BAA0B,KAAK,IAAI;AAC5C;AAMO,SAAS,6BAA6B,OAAuB;AAClE,QAAM,OAAO,kBAAkB,KAAK;AACpC,MAAI,MAAM,oCAAoC;AAC5C,WAAO,KAAK;AAEd,SAAO,0BAA0B,KAAK,IAAI;AAC5C;AAOO,SAAS,iBAAiB,QAMtB;AACT,SACE,OAAO,cAAc,0BAA0B,OAAO,KAAK,IAC3D,OAAO,eAAe,2BAA2B,OAAO,KAAK,IAC7D,OAAO,kBAAkB,yBAAyB,OAAO,KAAK,IAC9D,OAAO,sBAAsB,6BAA6B,OAAO,KAAK;AAE1E;AASA,SAAS,+BACP,SACmC;AACnC,SAAO;AAAA,IACL,MAAM;AAAA,IACN,SAAS;AAAA,IACT,YAAY;AAAA,IACZ,MAAM,WAAW;AAAA,IACjB,mBAAmB;AAAA,IACnB,SAAS;AAAA,MACP,IAAI,WAAW;AAAA,MACf,OAAO;AAAA,MACP,MAAM;AAAA,MACN,aAAa;AAAA,MACb,eAAe;AAAA,MACf,MAAM;AAAA,MACN,OAAO;AAAA,QACL,cAAc;AAAA,QACd,eAAe;AAAA,QACf,6BAA6B;AAAA,QAC7B,yBAAyB;AAAA,MAC3B;AAAA,MACA,SAAS;AAAA,QACP;AAAA,UACE,MAAM;AAAA,UACN,MAAM,YAAY,KAAK,qBAAqB;AAAA,UAC5C,WAAW,CAAC;AAAA,QACd;AAAA,MACF;AAAA,IACF;AAAA,EACF;AACF;AAEO,SAAS,6BACd,OACA,cAAsB,OACa;AACnC,MAAI,iBAAiB,SAAS,MAAM,QAAQ,SAAS,oBAAoB,GAAG;AAC1E,WAAO,+BAA+B,6BAA6B;AAAA,EACrE;AACA,MACE,iBAAiB,SACjB,MAAM,QAAQ,SAAS,gCAAgC,GACvD;AACA,WAAO,+BAA+B,oCAAoC;AAAA,EAC5E;AACA,MACE,iBAAiB,SACjB,MAAM,QAAQ,YAAY,EAAE,SAAS,WAAW,GAChD;AACA,WAAO,+BAA+B,6BAA6B;AAAA,EACrE;AACA,MAAI,iBAAiB,OAAO;AAC1B,QAAI,QAAQ,IAAI,aAAa,eAAe;AAC1C,kBAAY,MAAM,GAAG,WAAW,cAAc;AAAA,QAC5C,SAAS,MAAM;AAAA,QACf,OAAO,MAAM;AAAA,MACf,CAAC;AAAA,IACH;AACA,WAAO;AAAA,MACL,GAAG,wBAAwB,KAAK,MAAM,OAAO;AAAA,IAC/C;AAAA,EACF;AACA,SAAO,+BAA+B,wBAAwB;AAChE;",
   "names": []
 }

package/dist/services/mcpClient.js

@@ -4,6 +4,7 @@ import {
   getCurrentProjectConfig,
   saveCurrentProjectConfig,
   getGlobalConfig,
+  getMutableGlobalConfig,
   saveGlobalConfig,
   getMcprcConfig,
   addMcprcServerForTesting,
@@ -35,15 +36,50 @@ const SENSITIVE_ENV_PREFIXES = [
   "ANTHROPIC_",
   "OPENAI_",
   "AWS_SECRET",
+  "AWS_ACCESS_KEY",
+  "AWS_SESSION_TOKEN",
+  "AZURE_",
+  "GOOGLE_API_KEY",
+  "GOOGLE_APPLICATION_CREDENTIALS",
+  "HF_TOKEN",
+  "REPLICATE_API_TOKEN",
+  "HUGGINGFACE_",
+  "COHERE_",
+  "MISTRAL_",
+  "GROQ_",
+  "TOGETHER_",
+  "FIREWORKS_",
+  "PERPLEXITY_",
+  "DEEPSEEK_",
   "GITHUB_TOKEN",
   "GH_TOKEN",
   "MINTO_API"
 ];
+const SENSITIVE_ENV_SUFFIXES = ["_SECRET", "_PASSWORD", "_TOKEN", "_KEY"];
+const SAFE_SYSTEM_ENV_VARS = /* @__PURE__ */ new Set([
+  "HOME",
+  "PATH",
+  "TERM",
+  "SHELL",
+  "USER",
+  "LANG",
+  "EDITOR",
+  "DISPLAY",
+  "SSH_AUTH_SOCK",
+  "GPG_TTY"
+]);
+function isSafeSystemVar(key) {
+  return SAFE_SYSTEM_ENV_VARS.has(key) || key.startsWith("XDG_");
+}
 function filterSensitiveEnv(env) {
   return Object.fromEntries(
-    Object.entries(env).filter(
-      ([k]) => !SENSITIVE_ENV_PREFIXES.some((p) => k.startsWith(p) || k === p)
-    ).filter((e) => e[1] !== void 0)
+    Object.entries(env).filter(([k]) => {
+      if (isSafeSystemVar(k)) return true;
+      if (SENSITIVE_ENV_PREFIXES.some((p) => k.startsWith(p) || k === p))
+        return false;
+      if (SENSITIVE_ENV_SUFFIXES.some((s) => k.endsWith(s))) return false;
+      return true;
+    }).filter((e) => e[1] !== void 0)
   );
 }
 const mcpCircuitBreakers = new CircuitBreakerRegistry({
@@ -108,7 +144,7 @@ function addMcpServer(name, server, scope = "project") {
       }
     }
   } else if (scope === "global") {
-    const config = getGlobalConfig();
+    const config = getMutableGlobalConfig();
     if (!config.mcpServers) {
       config.mcpServers = {};
     }
@@ -241,6 +277,12 @@ async function connectToServer(name, serverRef) {
   });
   await Promise.race([connectPromise, timeoutPromise]);
   if (serverRef.type === "stdio") {
+    const childProc = transport._process;
+    if (childProc?.pid) {
+      const { registerChildPid, unregisterChildPid } = await import("../utils/exit.js");
+      registerChildPid(childProc.pid);
+      childProc.on?.("exit", () => unregisterChildPid(childProc.pid));
+    }
     ;
     transport.stderr?.on("data", (data) => {
       const errorText = data.toString().trim();
@@ -491,6 +533,33 @@ function isLazyMCPEnabled() {
   return !!(process.env.MINTO_ENABLE_TOOL_SEARCH || process.env.ENABLE_TOOL_SEARCH);
 }
 const lazySchemaCache = /* @__PURE__ */ new Map();
+const deferredTools = /* @__PURE__ */ new Map();
+const activatedToolNames = /* @__PURE__ */ new Set();
+function getDeferredMCPTools() {
+  return deferredTools;
+}
+function activateMCPTool(toolFullName) {
+  if (activatedToolNames.has(toolFullName)) {
+    return false;
+  }
+  activatedToolNames.add(toolFullName);
+  deferredTools.delete(toolFullName);
+  try {
+    import("../tools.js").then(({ invalidateToolsCache }) => {
+      invalidateToolsCache();
+    }).catch(() => {
+    });
+  } catch {
+  }
+  return true;
+}
+function isMCPToolActivated(toolFullName) {
+  return activatedToolNames.has(toolFullName);
+}
+function clearDeferredState() {
+  deferredTools.clear();
+  activatedToolNames.clear();
+}
 async function getLazyInputSchema(client, toolName, fullName) {
   if (lazySchemaCache.has(fullName)) {
     return lazySchemaCache.get(fullName);
@@ -792,6 +861,7 @@ async function refreshMCPConnections() {
       getMCPCommands.cache.clear();
     }
     mcpCircuitBreakers.resetAll();
+    clearDeferredState();
     try {
       const { invalidateToolsCache } = await import("../tools.js");
       invalidateToolsCache();
@@ -806,6 +876,7 @@ async function refreshMCPConnections() {
   }
 }
 async function shutdownMCPClients() {
+  stopHealthCheck();
   if (connectedClients.length === 0) return;
   const oldClients = connectedClients;
   connectedClients = [];
@@ -892,12 +963,16 @@ async function readMCPResource(uri, serverName) {
   return null;
 }
 export {
+  activateMCPTool,
   addMcpServer,
+  clearDeferredState,
   configureMcpPool,
   ensureConfigScope,
+  filterSensitiveEnv,
   getAllCircuitBreakerStats,
   getClients,
   getConnectionHealth,
+  getDeferredMCPTools,
   getMCPCommands,
   getMCPTools,
   getMcpPoolConfig,
@@ -906,6 +981,7 @@ export {
   getPoolStats,
   getServerCircuitBreaker,
   getServerHealth,
+  isMCPToolActivated,
   listMCPResources,
   listMCPServers,
   parseEnvVars,