@within-7/minto 0.3.10 → 0.4.0

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

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../../src/services/llm/anthropicProvider.ts"],
+  "sourcesContent": ["/**\n * Anthropic provider \u2014 client management, API verification, cache control,\n * and the native Anthropic streaming/non-streaming query path.\n */\nimport '@anthropic-ai/sdk/shims/node'\nimport Anthropic from '@anthropic-ai/sdk'\nimport { AnthropicBedrock } from '@anthropic-ai/bedrock-sdk'\nimport { AnthropicVertex } from '@anthropic-ai/vertex-sdk'\nimport chalk from 'chalk'\nimport { nanoid } from 'nanoid'\nimport { UUID } from 'crypto'\nimport 'dotenv/config'\n\nimport type { AssistantMessage, UserMessage } from '@query'\nimport { Tool, getToolDescriptionAsync } from '@tool'\nimport type { ToolUseContext } from '@tool'\nimport {\n  getAnthropicApiKey,\n  getGlobalConfig,\n  ModelProfile,\n} from '@utils/config'\nimport { USER_AGENT } from '@utils/http'\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 { USE_BEDROCK, USE_VERTEX, getVertexRegionForModel } from '@utils/model'\nimport { zodToJsonSchema } from 'zod-to-json-schema'\nimport { getCLISyspromptPrefix } from '@constants/prompts'\nimport { ContentBlock } from '@anthropic-ai/sdk/resources/messages/messages'\nimport type {\n  MessageParam,\n  TextBlockParam,\n} from '@anthropic-ai/sdk/resources/index.mjs'\nimport { setStreamingState, getStreamingState } from '@utils/streamingState'\nimport type { ThinkingMetadata } from '@minto-types/thinking'\nimport { addRetryEventToTranscript } from '@utils/agentTranscripts'\nimport type { TokenTrackingContext } from '@core/tokenStats'\n\nimport { generateMintoContext } from './mintoContext'\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  getMetadata,\n  withRetry,\n  getMaxTokensFromProfile,\n  getModelInputTokenCostUSD,\n  getModelOutputTokenCostUSD,\n  addToTotalCost,\n  recordTokenUsage,\n  logError,\n} from './types'\n\n// \u2500\u2500 Anthropic client management \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\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\nlet anthropicClient: Anthropic | AnthropicBedrock | AnthropicVertex | null =\n  null\n\n/**\n * Get the Anthropic client, creating it if it doesn't exist\n */\nexport function getAnthropicClient(\n  model?: string,\n  overrideProfile?: ModelProfile | null,\n): Anthropic | AnthropicBedrock | AnthropicVertex {\n  const config = getGlobalConfig()\n  const provider = config.primaryProvider\n\n  // If an override profile is provided, always create a fresh client\n  // to use the correct API key and baseURL for sub-agents\n  if (overrideProfile) {\n    const overrideHeaders: { [key: string]: string } = {\n      'x-app': 'cli',\n      'User-Agent': USER_AGENT,\n    }\n    return new Anthropic({\n      apiKey: overrideProfile.apiKey || '',\n      dangerouslyAllowBrowser: true,\n      maxRetries: 0,\n      timeout: parseInt(process.env.API_TIMEOUT_MS || String(60 * 1000), 10),\n      defaultHeaders: overrideHeaders,\n      ...(overrideProfile.baseURL && { baseURL: overrideProfile.baseURL }),\n    })\n  }\n\n  // Reset client if provider has changed to ensure correct configuration\n  if (anthropicClient && provider) {\n    // Always recreate client for provider-specific configurations\n    anthropicClient = null\n  }\n\n  if (anthropicClient) {\n    return anthropicClient\n  }\n\n  const region = getVertexRegionForModel(model)\n\n  const defaultHeaders: { [key: string]: string } = {\n    'x-app': 'cli',\n    'User-Agent': USER_AGENT,\n  }\n  if (process.env.ANTHROPIC_AUTH_TOKEN) {\n    defaultHeaders['Authorization'] =\n      `Bearer ${process.env.ANTHROPIC_AUTH_TOKEN}`\n  }\n\n  const ARGS = {\n    defaultHeaders,\n    maxRetries: 0, // Disabled auto-retry in favor of manual implementation\n    timeout: parseInt(process.env.API_TIMEOUT_MS || String(60 * 1000), 10),\n  }\n  if (USE_BEDROCK) {\n    const client = new AnthropicBedrock(ARGS)\n    anthropicClient = client\n    return client\n  }\n  if (USE_VERTEX) {\n    const vertexArgs = {\n      ...ARGS,\n      region: region || process.env.CLOUD_ML_REGION || 'us-east5',\n    }\n    const client = new AnthropicVertex(vertexArgs)\n    anthropicClient = client\n    return client\n  }\n\n  // Get appropriate API key and baseURL from ModelProfile\n  const modelManager = getModelManager()\n  const modelProfile = modelManager.getModel('main')\n\n  let apiKey: string\n  let baseURL: string | undefined\n\n  if (modelProfile) {\n    apiKey = modelProfile.apiKey || ''\n    baseURL = modelProfile.baseURL\n  } else {\n    // Fallback to default anthropic if no ModelProfile\n    apiKey = getAnthropicApiKey()\n    baseURL = undefined\n  }\n\n  if (process.env.USER_TYPE === 'ant' && !apiKey && provider === 'anthropic') {\n    console.error(\n      chalk.red(\n        '[ANT-ONLY] Please set the ANTHROPIC_API_KEY environment variable to use the CLI. To create a new key, go to https://console.anthropic.com/settings/keys.',\n      ),\n    )\n  }\n\n  // Create client with custom baseURL for BigDream/OpenDev\n  // Anthropic SDK will append the appropriate paths (like /v1/messages)\n  const clientConfig = {\n    apiKey,\n    dangerouslyAllowBrowser: true,\n    ...ARGS,\n    ...(baseURL && { baseURL }), // Use baseURL directly, SDK will handle API versioning\n  }\n\n  anthropicClient = new Anthropic(clientConfig)\n  return anthropicClient\n}\n\n/**\n * Reset the Anthropic client to null, forcing a new client to be created on next use\n */\nexport function resetAnthropicClient(): void {\n  anthropicClient = null\n}\n\n// \u2500\u2500 API verification \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\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 * Fetch available models from Anthropic API\n */\nexport async function fetchAnthropicModels(\n  baseURL: string,\n  apiKey: string,\n): Promise<any[]> {\n  try {\n    // Use provided baseURL or default to official Anthropic API\n    const modelsURL = baseURL\n      ? `${baseURL.replace(/\\/+$/, '')}/v1/models`\n      : 'https://api.anthropic.com/v1/models'\n\n    const response = await fetch(modelsURL, {\n      method: 'GET',\n      headers: {\n        'x-api-key': apiKey,\n        'anthropic-version': '2023-06-01',\n        'User-Agent': USER_AGENT,\n      },\n      signal: AbortSignal.timeout(30_000),\n    })\n\n    if (!response.ok) {\n      // Provide user-friendly error messages based on status code\n      if (response.status === 401) {\n        throw new Error(\n          'Invalid API key. Please check your Anthropic API key and try again.',\n        )\n      } else if (response.status === 403) {\n        throw new Error(\n          'API key does not have permission to access models. Please check your API key permissions.',\n        )\n      } else if (response.status === 429) {\n        throw new Error(\n          'Too many requests. Please wait a moment and try again.',\n        )\n      } else if (response.status >= 500) {\n        throw new Error(\n          'Anthropic service is temporarily unavailable. Please try again later.',\n        )\n      } else {\n        throw new Error(\n          `Unable to connect to Anthropic API (${response.status}). Please check your internet connection and API key.`,\n        )\n      }\n    }\n\n    const data = await response.json()\n    return data.data || []\n  } catch (error) {\n    // If it's already our custom error, pass it through\n    if (\n      (error instanceof Error && error.message.includes('API key')) ||\n      (error instanceof Error && error.message.includes('Anthropic'))\n    ) {\n      throw error\n    }\n\n    // For network errors or other issues\n    logError(error)\n    throw new Error(\n      'Unable to connect to Anthropic API. Please check your internet connection and try again.',\n    )\n  }\n}\n\nexport async function verifyApiKey(\n  apiKey: string,\n  baseURL?: string,\n  provider?: string,\n): Promise<boolean> {\n  if (!apiKey) {\n    return false\n  }\n\n  // For non-Anthropic providers, use OpenAI-compatible verification\n  if (provider && provider !== 'anthropic') {\n    try {\n      const headers: Record<string, string> = {\n        Authorization: `Bearer ${apiKey}`,\n        'Content-Type': 'application/json',\n      }\n\n      if (!baseURL) {\n        debugLogger.warn('API_VERIFY', {\n          error: 'No baseURL provided for non-Anthropic provider',\n        })\n        return false\n      }\n\n      const modelsURL = `${baseURL.replace(/\\/+$/, '')}/models`\n\n      const response = await fetch(modelsURL, {\n        method: 'GET',\n        headers,\n        signal: AbortSignal.timeout(30_000),\n      })\n\n      return response.ok\n    } catch (error) {\n      debugLogger.warn('API_VERIFY', {\n        error: 'non-Anthropic provider verification failed',\n        details: String(error),\n      })\n      return false\n    }\n  }\n\n  // For Anthropic and Anthropic-compatible APIs\n  const clientConfig: any = {\n    apiKey,\n    dangerouslyAllowBrowser: true,\n    maxRetries: 3,\n    defaultHeaders: {\n      'User-Agent': USER_AGENT,\n    },\n  }\n\n  // Only add baseURL for true Anthropic-compatible APIs\n  if (\n    baseURL &&\n    (provider === 'anthropic' ||\n      provider === 'bigdream' ||\n      provider === 'opendev')\n  ) {\n    clientConfig.baseURL = baseURL\n  }\n\n  const anthropic = new Anthropic(clientConfig)\n\n  try {\n    await withRetry(\n      async () => {\n        const model = 'claude-sonnet-4-20250514'\n        const messages: MessageParam[] = [{ role: 'user', content: 'test' }]\n        await anthropic.messages.create({\n          model,\n          max_tokens: 1000, // Simple test token limit for API verification\n          messages,\n          temperature: 0,\n          metadata: getMetadata(),\n        })\n        return true\n      },\n      { maxRetries: 2 }, // Use fewer retries for API key verification\n    )\n    return true\n  } catch (error) {\n    logError(error)\n    // Check for authentication error\n    if (\n      error instanceof Error &&\n      error.message.includes(\n        '{\"type\":\"error\",\"error\":{\"type\":\"authentication_error\",\"message\":\"invalid x-api-key\"}}',\n      )\n    ) {\n      return false\n    }\n    throw error\n  }\n}\n\n// \u2500\u2500 Cache management \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\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 * Manage cache control to ensure it doesn't exceed Claude's 4 cache block limit\n */\nfunction applyCacheControlWithLimits(\n  systemBlocks: TextBlockParam[],\n  messageParams: MessageParam[],\n): { systemBlocks: TextBlockParam[]; messageParams: MessageParam[] } {\n  if (!PROMPT_CACHING_ENABLED) {\n    return { systemBlocks, messageParams }\n  }\n\n  const maxCacheBlocks = 4\n  let usedCacheBlocks = 0\n\n  // 1. Prioritize adding cache to system prompts (highest priority)\n  const processedSystemBlocks = systemBlocks.map((block, _index) => {\n    if (usedCacheBlocks < maxCacheBlocks && block.text.length > 1000) {\n      usedCacheBlocks++\n      return {\n        ...block,\n        cache_control: { type: 'ephemeral' as const },\n      }\n    }\n    const { cache_control, ...blockWithoutCache } = block\n    return blockWithoutCache\n  })\n\n  // 2. Add cache to message content based on priority\n  const processedMessageParams = messageParams.map((message, messageIndex) => {\n    if (Array.isArray(message.content)) {\n      const processedContent = message.content.map(\n        (contentBlock, blockIndex) => {\n          // Determine whether this content block should be cached\n          const shouldCache =\n            usedCacheBlocks < maxCacheBlocks &&\n            contentBlock.type === 'text' &&\n            typeof contentBlock.text === 'string' &&\n            // Long documents (over 2000 characters)\n            (contentBlock.text.length > 2000 ||\n              // Last content block of the last message (may be important context)\n              (messageIndex === messageParams.length - 1 &&\n                blockIndex === message.content.length - 1 &&\n                contentBlock.text.length > 500))\n\n          if (shouldCache) {\n            usedCacheBlocks++\n            return {\n              ...contentBlock,\n              cache_control: { type: 'ephemeral' as const },\n            }\n          }\n\n          // Remove existing cache_control\n          const { cache_control, ...blockWithoutCache } = contentBlock as any\n          return blockWithoutCache\n        },\n      )\n\n      return {\n        ...message,\n        content: processedContent,\n      }\n    }\n\n    return message\n  })\n\n  return {\n    systemBlocks: processedSystemBlocks,\n    messageParams: processedMessageParams,\n  }\n}\n\nexport function userMessageToMessageParam(\n  message: UserMessage,\n  addCache = false,\n): MessageParam {\n  if (addCache) {\n    if (typeof message.message.content === 'string') {\n      return {\n        role: 'user',\n        content: [\n          {\n            type: 'text',\n            text: message.message.content,\n          },\n        ],\n      }\n    } else {\n      return {\n        role: 'user',\n        content: message.message.content.map(_ => ({ ..._ })),\n      }\n    }\n  }\n  return {\n    role: 'user',\n    content: message.message.content,\n  }\n}\n\nexport function assistantMessageToMessageParam(\n  message: AssistantMessage,\n  addCache = false,\n): MessageParam {\n  if (addCache) {\n    if (typeof message.message.content === 'string') {\n      return {\n        role: 'assistant',\n        content: [\n          {\n            type: 'text',\n            text: message.message.content,\n          },\n        ],\n      }\n    } else {\n      return {\n        role: 'assistant',\n        content: message.message.content.map(_ => ({ ..._ })),\n      }\n    }\n  }\n  return {\n    role: 'assistant',\n    content: message.message.content,\n  }\n}\n\nexport function splitSysPromptPrefix(systemPrompt: string[]): string[] {\n  // split out the first block of the system prompt as the \"prefix\" for API\n  const systemPromptFirstBlock = systemPrompt[0] || ''\n  const systemPromptRest = systemPrompt.slice(1)\n  return [systemPromptFirstBlock, systemPromptRest.join('\\n')].filter(Boolean)\n}\n\nfunction addCacheBreakpoints(\n  messages: (UserMessage | AssistantMessage)[],\n): MessageParam[] {\n  return messages.map((msg, index) => {\n    return msg.type === 'user'\n      ? userMessageToMessageParam(msg, index > messages.length - 3)\n      : assistantMessageToMessageParam(msg, index > messages.length - 3)\n  })\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('ANTHROPIC_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 queryAnthropicNative \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\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 queryAnthropicNative(\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 anthropic: Anthropic | AnthropicBedrock | AnthropicVertex\n  let model: string\n  let provider: string\n\n  debugLogger.api('MODEL_CONFIG_ANTHROPIC', {\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    provider = modelProfile.provider || config.primaryProvider || 'anthropic'\n\n    if (\n      modelProfile.provider === 'anthropic' ||\n      modelProfile.provider === 'bigdream' ||\n      modelProfile.provider === 'opendev'\n    ) {\n      const clientConfig: any = {\n        apiKey: modelProfile.apiKey,\n        dangerouslyAllowBrowser: true,\n        maxRetries: 0,\n        timeout: parseInt(process.env.API_TIMEOUT_MS || String(60 * 1000), 10),\n        defaultHeaders: {\n          'x-app': 'cli',\n          'User-Agent': USER_AGENT,\n        },\n      }\n\n      if (modelProfile.baseURL) {\n        clientConfig.baseURL = modelProfile.baseURL\n      }\n\n      anthropic = new Anthropic(clientConfig)\n    } else {\n      anthropic = getAnthropicClient(model)\n    }\n  } else {\n    const errorDetails = {\n      modelProfileExists: !!modelProfile,\n      modelProfileModelName: modelProfile?.modelName,\n      requestedModel: options?.model,\n      requestId: getCurrentRequest()?.id,\n    }\n    debugLogger.error('ANTHROPIC_FALLBACK_ERROR', errorDetails)\n    throw new Error(\n      `No valid ModelProfile available for Anthropic provider. Please configure model through /model command. Debug: ${JSON.stringify(errorDetails)}`,\n    )\n  }\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      text: _,\n      type: 'text',\n    }),\n  )\n\n  const toolSchemas = await Promise.all(\n    tools.map(\n      async tool =>\n        ({\n          name: tool.name,\n          description: await getToolDescriptionAsync(tool),\n          input_schema:\n            'inputJSONSchema' in tool && tool.inputJSONSchema\n              ? tool.inputJSONSchema\n              : zodToJsonSchema(tool.inputSchema),\n        }) as unknown as Anthropic.Beta.Messages.BetaTool,\n    ),\n  )\n\n  const anthropicMessages = addCacheBreakpoints(messages)\n\n  const { systemBlocks: processedSystem, messageParams: processedMessages } =\n    applyCacheControlWithLimits(system, anthropicMessages)\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\n        const params: Anthropic.Beta.Messages.MessageCreateParams = {\n          model,\n          max_tokens: getMaxTokensFromProfile(modelProfile),\n          messages: processedMessages,\n          system: processedSystem,\n          tools: toolSchemas.length > 0 ? toolSchemas : undefined,\n          tool_choice: toolSchemas.length > 0 ? { type: 'auto' } : undefined,\n        }\n\n        if (maxThinkingTokens > 0) {\n          ;(params as any).extra_headers = {\n            'anthropic-beta': 'max-tokens-3-5-sonnet-2024-07-15',\n          }\n          ;(params as any).thinking = { max_tokens: maxThinkingTokens }\n        }\n\n        debugLogger.api('ANTHROPIC_API_CALL_START_STREAMING', {\n          endpoint: modelProfile?.baseURL || 'DEFAULT_ANTHROPIC',\n          model,\n          provider,\n          apiKeyConfigured: !!modelProfile?.apiKey,\n          apiKeyPrefix: modelProfile?.apiKey\n            ? modelProfile.apiKey.substring(0, 8)\n            : null,\n          maxTokens: params.max_tokens,\n          temperature: MAIN_QUERY_TEMPERATURE,\n          params: params,\n          messageCount: params.messages?.length || 0,\n          streamMode: true,\n          toolsCount: toolSchemas.length,\n          thinkingTokens: maxThinkingTokens,\n          timestamp: new Date().toISOString(),\n          modelProfileId: modelProfile?.modelName,\n          modelProfileName: modelProfile?.name,\n        })\n\n        if (config.stream) {\n          const stream = await anthropic.beta.messages.create(\n            {\n              ...params,\n              stream: true,\n            },\n            {\n              signal: signal,\n            },\n          )\n\n          let finalResponse: any | null = null\n          let messageStartEvent: any = null\n          const contentBlocks: any[] = []\n          const inputJSONBuffers = new Map<number, string>()\n          let usage: any = null\n          let stopReason: string | null = null\n          let stopSequence: string | null = null\n\n          for await (const event of stream) {\n            if (signal.aborted) {\n              debugLogger.flow('STREAM_ABORTED', {\n                eventType: event.type,\n                timestamp: Date.now(),\n              })\n              throw new Error('Request was cancelled')\n            }\n\n            switch (event.type) {\n              case 'message_start':\n                messageStartEvent = event\n                finalResponse = {\n                  ...event.message,\n                  content: [],\n                }\n                setStreamingState({\n                  inputTokens: event.message.usage?.input_tokens,\n                  receivedChars: 0,\n                })\n                break\n\n              case 'content_block_start':\n                contentBlocks[event.index] = { ...event.content_block }\n                if (event.content_block.type === 'tool_use') {\n                  inputJSONBuffers.set(event.index, '')\n                }\n                if (event.content_block.type === 'thinking') {\n                  const thinkingStartTime = Date.now()\n                  setStreamingState({\n                    phase: 'deep_thinking',\n                    tokenCount: 0,\n                    thinkingMaxTokens: maxThinkingTokens,\n                    thinkingStartTime,\n                  })\n                  debugLogger.flow('THINKING_BLOCK_START', {\n                    index: event.index,\n                    maxTokens: maxThinkingTokens,\n                    startTime: thinkingStartTime,\n                  })\n                }\n                break\n\n              case 'content_block_delta':\n                const blockIndex = event.index\n\n                if (!contentBlocks[blockIndex]) {\n                  contentBlocks[blockIndex] = {\n                    type:\n                      event.delta.type === 'text_delta' ? 'text' : 'tool_use',\n                    text: event.delta.type === 'text_delta' ? '' : undefined,\n                  }\n                  if (event.delta.type === 'input_json_delta') {\n                    inputJSONBuffers.set(blockIndex, '')\n                  }\n                }\n\n                if (event.delta.type === 'text_delta') {\n                  contentBlocks[blockIndex].text += event.delta.text\n                  const currentState = getStreamingState()\n                  setStreamingState({\n                    receivedChars:\n                      (currentState.receivedChars || 0) +\n                      event.delta.text.length,\n                  })\n                } else if (event.delta.type === 'input_json_delta') {\n                  const currentBuffer = inputJSONBuffers.get(blockIndex) || ''\n                  inputJSONBuffers.set(\n                    blockIndex,\n                    currentBuffer + event.delta.partial_json,\n                  )\n                }\n                // Handle thinking_delta events to update token count\n                else if (event.delta.type === 'thinking_delta') {\n                  const currentThinking =\n                    contentBlocks[blockIndex].thinking || ''\n                  const updatedThinking =\n                    currentThinking + (event.delta.thinking || '')\n                  contentBlocks[blockIndex] = {\n                    ...contentBlocks[blockIndex],\n                    thinking: updatedThinking,\n                  }\n\n                  setStreamingState({\n                    phase: 'deep_thinking',\n                    tokenCount: updatedThinking.length,\n                    thinkingMaxTokens: maxThinkingTokens,\n                  })\n\n                  debugLogger.trace('THINKING_DELTA', {\n                    index: blockIndex,\n                    tokenCount: updatedThinking.length,\n                  })\n                }\n                break\n\n              case 'message_delta':\n                if (event.delta.stop_reason)\n                  stopReason = event.delta.stop_reason\n                if (event.delta.stop_sequence)\n                  stopSequence = event.delta.stop_sequence\n                if (event.usage) {\n                  usage = { ...usage, ...event.usage }\n                  setStreamingState({\n                    outputTokens: event.usage.output_tokens,\n                  })\n                }\n                break\n\n              case 'content_block_stop':\n                const stopIndex = event.index\n                const block = contentBlocks[stopIndex]\n\n                if (block?.type === 'thinking') {\n                  const streamState = getStreamingState()\n                  const thinkingDurationMs = streamState.thinkingStartTime\n                    ? Date.now() - streamState.thinkingStartTime\n                    : 0\n                  setStreamingState({\n                    phase: 'generating',\n                    thinkingDurationMs,\n                  })\n                  debugLogger.flow('THINKING_BLOCK_COMPLETE', {\n                    index: stopIndex,\n                    finalTokenCount: block.thinking?.length || 0,\n                    durationMs: thinkingDurationMs,\n                  })\n                }\n\n                if (\n                  block?.type === 'tool_use' &&\n                  inputJSONBuffers.has(stopIndex)\n                ) {\n                  const jsonStr = inputJSONBuffers.get(stopIndex)\n                  if (jsonStr) {\n                    try {\n                      block.input = JSON.parse(jsonStr)\n                    } catch (error) {\n                      const errorMsg =\n                        error instanceof Error ? error.message : String(error)\n                      debugLogger.error('JSON_PARSE_ERROR', {\n                        blockIndex: stopIndex,\n                        jsonStr:\n                          jsonStr.length > 500\n                            ? jsonStr.slice(0, 500) + '...'\n                            : jsonStr,\n                        error: errorMsg,\n                      })\n                      block.input = {\n                        __parse_error__: true,\n                        __error_message__: `JSON parse failed: ${errorMsg}`,\n                        __raw_json_preview__:\n                          jsonStr.length > 200\n                            ? jsonStr.slice(0, 200) + '...'\n                            : jsonStr,\n                      }\n                    }\n                    inputJSONBuffers.delete(stopIndex)\n                  }\n                }\n                break\n\n              case 'message_stop':\n                inputJSONBuffers.clear()\n                break\n            }\n\n            if (event.type === 'message_stop') {\n              break\n            }\n          }\n\n          if (!finalResponse || !messageStartEvent) {\n            throw new Error('Stream ended without proper message structure')\n          }\n\n          // Construct the final response\n          finalResponse = {\n            ...messageStartEvent.message,\n            content: contentBlocks.filter(Boolean),\n            stop_reason: stopReason,\n            stop_sequence: stopSequence,\n            usage: {\n              ...messageStartEvent.message.usage,\n              ...usage,\n            },\n          }\n\n          return finalResponse\n        } else {\n          debugLogger.api('ANTHROPIC_API_CALL_START_NON_STREAMING', {\n            endpoint: modelProfile?.baseURL || 'DEFAULT_ANTHROPIC',\n            model,\n            provider,\n            apiKeyConfigured: !!modelProfile?.apiKey,\n            apiKeyPrefix: modelProfile?.apiKey\n              ? modelProfile.apiKey.substring(0, 8)\n              : null,\n            maxTokens: params.max_tokens,\n            temperature: MAIN_QUERY_TEMPERATURE,\n            messageCount: params.messages?.length || 0,\n            streamMode: false,\n            toolsCount: toolSchemas.length,\n            thinkingTokens: maxThinkingTokens,\n            timestamp: new Date().toISOString(),\n            modelProfileId: modelProfile?.modelName,\n            modelProfileName: modelProfile?.name,\n          })\n\n          return await anthropic.beta.messages.create(params, {\n            signal: signal,\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\n    debugLogger.api('ANTHROPIC_API_CALL_SUCCESS', {\n      content: response.content,\n    })\n\n    const ttftMs = start - Date.now()\n    const durationMs = Date.now() - startIncludingRetries\n\n    const content = response.content.map((block: ContentBlock) => {\n      if (block.type === 'text') {\n        return {\n          type: 'text' as const,\n          text: block.text,\n        }\n      } else if (block.type === 'tool_use') {\n        return {\n          type: 'tool_use' as const,\n          id: block.id,\n          name: block.name,\n          input: block.input,\n        }\n      }\n      return block\n    })\n\n    // Build thinking metadata if thinking blocks are present\n    const thinkingBlocks = content.filter(\n      (block): block is ContentBlock & { type: 'thinking'; thinking: string } =>\n        block.type === 'thinking',\n    )\n    const streamState = getStreamingState()\n    const thinkingMetadata: ThinkingMetadata | undefined =\n      thinkingBlocks.length > 0\n        ? {\n            charCount: thinkingBlocks.reduce(\n              (sum, b) => sum + (b.thinking?.length || 0),\n              0,\n            ),\n            durationMs: streamState.thinkingDurationMs || 0,\n            startTime: streamState.thinkingStartTime || Date.now(),\n            isComplete: true,\n          }\n        : undefined\n\n    const assistantMessage: AssistantMessage = {\n      message: {\n        id: response.id,\n        content,\n        model: response.model,\n        role: 'assistant',\n        stop_reason: response.stop_reason,\n        stop_sequence: response.stop_sequence,\n        type: 'message',\n        usage: response.usage,\n      },\n      type: 'assistant',\n      uuid: nanoid() as UUID,\n      durationMs,\n      costUSD: 0, // Will be calculated below\n      thinkingMetadata,\n    }\n\n    const systemMessages = system.map(block => ({\n      role: 'system',\n      content: block.text,\n    }))\n\n    logLLMInteraction({\n      systemPrompt: systemPrompt.join('\\n'),\n      messages: [...systemMessages, ...anthropicMessages],\n      response: response,\n      usage: response.usage\n        ? {\n            inputTokens: response.usage.input_tokens,\n            outputTokens: response.usage.output_tokens,\n          }\n        : undefined,\n      timing: {\n        start: start,\n        end: Date.now(),\n      },\n      apiFormat: 'anthropic',\n    })\n\n    // Calculate cost using native Anthropic usage data\n    const inputTokens = response.usage.input_tokens\n    const outputTokens = response.usage.output_tokens\n    const cacheCreationInputTokens =\n      response.usage.cache_creation_input_tokens ?? 0\n    const cacheReadInputTokens = response.usage.cache_read_input_tokens ?? 0\n\n    const costUSD =\n      (inputTokens / 1_000_000) * getModelInputTokenCostUSD(model) +\n      (outputTokens / 1_000_000) * getModelOutputTokenCostUSD(model) +\n      (cacheCreationInputTokens / 1_000_000) *\n        getModelInputTokenCostUSD(model) +\n      (cacheReadInputTokens / 1_000_000) *\n        (getModelInputTokenCostUSD(model) * 0.1) // Cache reads are 10% of input cost\n\n    assistantMessage.costUSD = costUSD\n    addToTotalCost(costUSD, durationMs)\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    return assistantMessage\n  } catch (error) {\n    return getAssistantMessageFromError(error)\n  }\n}\n"],
+  "mappings": "AAIA,OAAO;AACP,OAAO,eAAe;AACtB,SAAS,wBAAwB;AACjC,SAAS,uBAAuB;AAChC,OAAO,WAAW;AAClB,SAAS,cAAc;AAEvB,OAAO;AAGP,SAAe,+BAA+B;AAE9C;AAAA,EACE;AAAA,EACA;AAAA,OAEK;AACP,SAAS,kBAAkB;AAC3B;AAAA,EACE;AAAA,OAEK;AACP;AAAA,EACE,SAAS;AAAA,EACT;AAAA,EACA;AAAA,EACA;AAAA,OACK;AACP,SAAS,uBAAuB;AAChC,SAAS,aAAa,YAAY,+BAA+B;AACjE,SAAS,uBAAuB;AAChC,SAAS,6BAA6B;AAMtC,SAAS,mBAAmB,yBAAyB;AAErD,SAAS,iCAAiC;AAG1C,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,OACK;AAIP,IAAI,kBACF;AAKK,SAAS,mBACd,OACA,iBACgD;AAChD,QAAM,SAAS,gBAAgB;AAC/B,QAAM,WAAW,OAAO;AAIxB,MAAI,iBAAiB;AACnB,UAAM,kBAA6C;AAAA,MACjD,SAAS;AAAA,MACT,cAAc;AAAA,IAChB;AACA,WAAO,IAAI,UAAU;AAAA,MACnB,QAAQ,gBAAgB,UAAU;AAAA,MAClC,yBAAyB;AAAA,MACzB,YAAY;AAAA,MACZ,SAAS,SAAS,QAAQ,IAAI,kBAAkB,OAAO,KAAK,GAAI,GAAG,EAAE;AAAA,MACrE,gBAAgB;AAAA,MAChB,GAAI,gBAAgB,WAAW,EAAE,SAAS,gBAAgB,QAAQ;AAAA,IACpE,CAAC;AAAA,EACH;AAGA,MAAI,mBAAmB,UAAU;AAE/B,sBAAkB;AAAA,EACpB;AAEA,MAAI,iBAAiB;AACnB,WAAO;AAAA,EACT;AAEA,QAAM,SAAS,wBAAwB,KAAK;AAE5C,QAAM,iBAA4C;AAAA,IAChD,SAAS;AAAA,IACT,cAAc;AAAA,EAChB;AACA,MAAI,QAAQ,IAAI,sBAAsB;AACpC,mBAAe,eAAe,IAC5B,UAAU,QAAQ,IAAI,oBAAoB;AAAA,EAC9C;AAEA,QAAM,OAAO;AAAA,IACX;AAAA,IACA,YAAY;AAAA;AAAA,IACZ,SAAS,SAAS,QAAQ,IAAI,kBAAkB,OAAO,KAAK,GAAI,GAAG,EAAE;AAAA,EACvE;AACA,MAAI,aAAa;AACf,UAAM,SAAS,IAAI,iBAAiB,IAAI;AACxC,sBAAkB;AAClB,WAAO;AAAA,EACT;AACA,MAAI,YAAY;AACd,UAAM,aAAa;AAAA,MACjB,GAAG;AAAA,MACH,QAAQ,UAAU,QAAQ,IAAI,mBAAmB;AAAA,IACnD;AACA,UAAM,SAAS,IAAI,gBAAgB,UAAU;AAC7C,sBAAkB;AAClB,WAAO;AAAA,EACT;AAGA,QAAM,eAAe,gBAAgB;AACrC,QAAM,eAAe,aAAa,SAAS,MAAM;AAEjD,MAAI;AACJ,MAAI;AAEJ,MAAI,cAAc;AAChB,aAAS,aAAa,UAAU;AAChC,cAAU,aAAa;AAAA,EACzB,OAAO;AAEL,aAAS,mBAAmB;AAC5B,cAAU;AAAA,EACZ;AAEA,MAAI,QAAQ,IAAI,cAAc,SAAS,CAAC,UAAU,aAAa,aAAa;AAC1E,YAAQ;AAAA,MACN,MAAM;AAAA,QACJ;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAIA,QAAM,eAAe;AAAA,IACnB;AAAA,IACA,yBAAyB;AAAA,IACzB,GAAG;AAAA,IACH,GAAI,WAAW,EAAE,QAAQ;AAAA;AAAA,EAC3B;AAEA,oBAAkB,IAAI,UAAU,YAAY;AAC5C,SAAO;AACT;AAKO,SAAS,uBAA6B;AAC3C,oBAAkB;AACpB;AAOA,eAAsB,qBACpB,SACA,QACgB;AAChB,MAAI;AAEF,UAAM,YAAY,UACd,GAAG,QAAQ,QAAQ,QAAQ,EAAE,CAAC,eAC9B;AAEJ,UAAM,WAAW,MAAM,MAAM,WAAW;AAAA,MACtC,QAAQ;AAAA,MACR,SAAS;AAAA,QACP,aAAa;AAAA,QACb,qBAAqB;AAAA,QACrB,cAAc;AAAA,MAChB;AAAA,MACA,QAAQ,YAAY,QAAQ,GAAM;AAAA,IACpC,CAAC;AAED,QAAI,CAAC,SAAS,IAAI;AAEhB,UAAI,SAAS,WAAW,KAAK;AAC3B,cAAM,IAAI;AAAA,UACR;AAAA,QACF;AAAA,MACF,WAAW,SAAS,WAAW,KAAK;AAClC,cAAM,IAAI;AAAA,UACR;AAAA,QACF;AAAA,MACF,WAAW,SAAS,WAAW,KAAK;AAClC,cAAM,IAAI;AAAA,UACR;AAAA,QACF;AAAA,MACF,WAAW,SAAS,UAAU,KAAK;AACjC,cAAM,IAAI;AAAA,UACR;AAAA,QACF;AAAA,MACF,OAAO;AACL,cAAM,IAAI;AAAA,UACR,uCAAuC,SAAS,MAAM;AAAA,QACxD;AAAA,MACF;AAAA,IACF;AAEA,UAAM,OAAO,MAAM,SAAS,KAAK;AACjC,WAAO,KAAK,QAAQ,CAAC;AAAA,EACvB,SAAS,OAAO;AAEd,QACG,iBAAiB,SAAS,MAAM,QAAQ,SAAS,SAAS,KAC1D,iBAAiB,SAAS,MAAM,QAAQ,SAAS,WAAW,GAC7D;AACA,YAAM;AAAA,IACR;AAGA,aAAS,KAAK;AACd,UAAM,IAAI;AAAA,MACR;AAAA,IACF;AAAA,EACF;AACF;AAEA,eAAsB,aACpB,QACA,SACA,UACkB;AAClB,MAAI,CAAC,QAAQ;AACX,WAAO;AAAA,EACT;AAGA,MAAI,YAAY,aAAa,aAAa;AACxC,QAAI;AACF,YAAM,UAAkC;AAAA,QACtC,eAAe,UAAU,MAAM;AAAA,QAC/B,gBAAgB;AAAA,MAClB;AAEA,UAAI,CAAC,SAAS;AACZ,oBAAY,KAAK,cAAc;AAAA,UAC7B,OAAO;AAAA,QACT,CAAC;AACD,eAAO;AAAA,MACT;AAEA,YAAM,YAAY,GAAG,QAAQ,QAAQ,QAAQ,EAAE,CAAC;AAEhD,YAAM,WAAW,MAAM,MAAM,WAAW;AAAA,QACtC,QAAQ;AAAA,QACR;AAAA,QACA,QAAQ,YAAY,QAAQ,GAAM;AAAA,MACpC,CAAC;AAED,aAAO,SAAS;AAAA,IAClB,SAAS,OAAO;AACd,kBAAY,KAAK,cAAc;AAAA,QAC7B,OAAO;AAAA,QACP,SAAS,OAAO,KAAK;AAAA,MACvB,CAAC;AACD,aAAO;AAAA,IACT;AAAA,EACF;AAGA,QAAM,eAAoB;AAAA,IACxB;AAAA,IACA,yBAAyB;AAAA,IACzB,YAAY;AAAA,IACZ,gBAAgB;AAAA,MACd,cAAc;AAAA,IAChB;AAAA,EACF;AAGA,MACE,YACC,aAAa,eACZ,aAAa,cACb,aAAa,YACf;AACA,iBAAa,UAAU;AAAA,EACzB;AAEA,QAAM,YAAY,IAAI,UAAU,YAAY;AAE5C,MAAI;AACF,UAAM;AAAA,MACJ,YAAY;AACV,cAAM,QAAQ;AACd,cAAM,WAA2B,CAAC,EAAE,MAAM,QAAQ,SAAS,OAAO,CAAC;AACnE,cAAM,UAAU,SAAS,OAAO;AAAA,UAC9B;AAAA,UACA,YAAY;AAAA;AAAA,UACZ;AAAA,UACA,aAAa;AAAA,UACb,UAAU,YAAY;AAAA,QACxB,CAAC;AACD,eAAO;AAAA,MACT;AAAA,MACA,EAAE,YAAY,EAAE;AAAA;AAAA,IAClB;AACA,WAAO;AAAA,EACT,SAAS,OAAO;AACd,aAAS,KAAK;AAEd,QACE,iBAAiB,SACjB,MAAM,QAAQ;AAAA,MACZ;AAAA,IACF,GACA;AACA,aAAO;AAAA,IACT;AACA,UAAM;AAAA,EACR;AACF;AAOA,SAAS,4BACP,cACA,eACmE;AACnE,MAAI,CAAC,wBAAwB;AAC3B,WAAO,EAAE,cAAc,cAAc;AAAA,EACvC;AAEA,QAAM,iBAAiB;AACvB,MAAI,kBAAkB;AAGtB,QAAM,wBAAwB,aAAa,IAAI,CAAC,OAAO,WAAW;AAChE,QAAI,kBAAkB,kBAAkB,MAAM,KAAK,SAAS,KAAM;AAChE;AACA,aAAO;AAAA,QACL,GAAG;AAAA,QACH,eAAe,EAAE,MAAM,YAAqB;AAAA,MAC9C;AAAA,IACF;AACA,UAAM,EAAE,eAAe,GAAG,kBAAkB,IAAI;AAChD,WAAO;AAAA,EACT,CAAC;AAGD,QAAM,yBAAyB,cAAc,IAAI,CAAC,SAAS,iBAAiB;AAC1E,QAAI,MAAM,QAAQ,QAAQ,OAAO,GAAG;AAClC,YAAM,mBAAmB,QAAQ,QAAQ;AAAA,QACvC,CAAC,cAAc,eAAe;AAE5B,gBAAM,cACJ,kBAAkB,kBAClB,aAAa,SAAS,UACtB,OAAO,aAAa,SAAS;AAAA,WAE5B,aAAa,KAAK,SAAS;AAAA,UAEzB,iBAAiB,cAAc,SAAS,KACvC,eAAe,QAAQ,QAAQ,SAAS,KACxC,aAAa,KAAK,SAAS;AAEjC,cAAI,aAAa;AACf;AACA,mBAAO;AAAA,cACL,GAAG;AAAA,cACH,eAAe,EAAE,MAAM,YAAqB;AAAA,YAC9C;AAAA,UACF;AAGA,gBAAM,EAAE,eAAe,GAAG,kBAAkB,IAAI;AAChD,iBAAO;AAAA,QACT;AAAA,MACF;AAEA,aAAO;AAAA,QACL,GAAG;AAAA,QACH,SAAS;AAAA,MACX;AAAA,IACF;AAEA,WAAO;AAAA,EACT,CAAC;AAED,SAAO;AAAA,IACL,cAAc;AAAA,IACd,eAAe;AAAA,EACjB;AACF;AAEO,SAAS,0BACd,SACA,WAAW,OACG;AACd,MAAI,UAAU;AACZ,QAAI,OAAO,QAAQ,QAAQ,YAAY,UAAU;AAC/C,aAAO;AAAA,QACL,MAAM;AAAA,QACN,SAAS;AAAA,UACP;AAAA,YACE,MAAM;AAAA,YACN,MAAM,QAAQ,QAAQ;AAAA,UACxB;AAAA,QACF;AAAA,MACF;AAAA,IACF,OAAO;AACL,aAAO;AAAA,QACL,MAAM;AAAA,QACN,SAAS,QAAQ,QAAQ,QAAQ,IAAI,QAAM,EAAE,GAAG,EAAE,EAAE;AAAA,MACtD;AAAA,IACF;AAAA,EACF;AACA,SAAO;AAAA,IACL,MAAM;AAAA,IACN,SAAS,QAAQ,QAAQ;AAAA,EAC3B;AACF;AAEO,SAAS,+BACd,SACA,WAAW,OACG;AACd,MAAI,UAAU;AACZ,QAAI,OAAO,QAAQ,QAAQ,YAAY,UAAU;AAC/C,aAAO;AAAA,QACL,MAAM;AAAA,QACN,SAAS;AAAA,UACP;AAAA,YACE,MAAM;AAAA,YACN,MAAM,QAAQ,QAAQ;AAAA,UACxB;AAAA,QACF;AAAA,MACF;AAAA,IACF,OAAO;AACL,aAAO;AAAA,QACL,MAAM;AAAA,QACN,SAAS,QAAQ,QAAQ,QAAQ,IAAI,QAAM,EAAE,GAAG,EAAE,EAAE;AAAA,MACtD;AAAA,IACF;AAAA,EACF;AACA,SAAO;AAAA,IACL,MAAM;AAAA,IACN,SAAS,QAAQ,QAAQ;AAAA,EAC3B;AACF;AAEO,SAAS,qBAAqB,cAAkC;AAErE,QAAM,yBAAyB,aAAa,CAAC,KAAK;AAClD,QAAM,mBAAmB,aAAa,MAAM,CAAC;AAC7C,SAAO,CAAC,wBAAwB,iBAAiB,KAAK,IAAI,CAAC,EAAE,OAAO,OAAO;AAC7E;AAEA,SAAS,oBACP,UACgB;AAChB,SAAO,SAAS,IAAI,CAAC,KAAK,UAAU;AAClC,WAAO,IAAI,SAAS,SAChB,0BAA0B,KAAK,QAAQ,SAAS,SAAS,CAAC,IAC1D,+BAA+B,KAAK,QAAQ,SAAS,SAAS,CAAC;AAAA,EACrE,CAAC;AACH;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,uBAAuB;AAAA,QACvC,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,qBACpB,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;AACJ,MAAI;AACJ,MAAI;AAEJ,cAAY,IAAI,0BAA0B;AAAA,IACxC,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;AACrB,eAAW,aAAa,YAAY,OAAO,mBAAmB;AAE9D,QACE,aAAa,aAAa,eAC1B,aAAa,aAAa,cAC1B,aAAa,aAAa,WAC1B;AACA,YAAM,eAAoB;AAAA,QACxB,QAAQ,aAAa;AAAA,QACrB,yBAAyB;AAAA,QACzB,YAAY;AAAA,QACZ,SAAS,SAAS,QAAQ,IAAI,kBAAkB,OAAO,KAAK,GAAI,GAAG,EAAE;AAAA,QACrE,gBAAgB;AAAA,UACd,SAAS;AAAA,UACT,cAAc;AAAA,QAChB;AAAA,MACF;AAEA,UAAI,aAAa,SAAS;AACxB,qBAAa,UAAU,aAAa;AAAA,MACtC;AAEA,kBAAY,IAAI,UAAU,YAAY;AAAA,IACxC,OAAO;AACL,kBAAY,mBAAmB,KAAK;AAAA,IACtC;AAAA,EACF,OAAO;AACL,UAAM,eAAe;AAAA,MACnB,oBAAoB,CAAC,CAAC;AAAA,MACtB,uBAAuB,cAAc;AAAA,MACrC,gBAAgB,SAAS;AAAA,MACzB,WAAW,kBAAkB,GAAG;AAAA,IAClC;AACA,gBAAY,MAAM,4BAA4B,YAAY;AAC1D,UAAM,IAAI;AAAA,MACR,iHAAiH,KAAK,UAAU,YAAY,CAAC;AAAA,IAC/I;AAAA,EACF;AAGA,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,MAAM;AAAA,MACN,MAAM;AAAA,IACR;AAAA,EACF;AAEA,QAAM,cAAc,MAAM,QAAQ;AAAA,IAChC,MAAM;AAAA,MACJ,OAAM,UACH;AAAA,QACC,MAAM,KAAK;AAAA,QACX,aAAa,MAAM,wBAAwB,IAAI;AAAA,QAC/C,cACE,qBAAqB,QAAQ,KAAK,kBAC9B,KAAK,kBACL,gBAAgB,KAAK,WAAW;AAAA,MACxC;AAAA,IACJ;AAAA,EACF;AAEA,QAAM,oBAAoB,oBAAoB,QAAQ;AAEtD,QAAM,EAAE,cAAc,iBAAiB,eAAe,kBAAkB,IACtE,4BAA4B,QAAQ,iBAAiB;AACvD,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;AAEjB,cAAM,SAAsD;AAAA,UAC1D;AAAA,UACA,YAAY,wBAAwB,YAAY;AAAA,UAChD,UAAU;AAAA,UACV,QAAQ;AAAA,UACR,OAAO,YAAY,SAAS,IAAI,cAAc;AAAA,UAC9C,aAAa,YAAY,SAAS,IAAI,EAAE,MAAM,OAAO,IAAI;AAAA,QAC3D;AAEA,YAAI,oBAAoB,GAAG;AACzB;AAAC,UAAC,OAAe,gBAAgB;AAAA,YAC/B,kBAAkB;AAAA,UACpB;AACC,UAAC,OAAe,WAAW,EAAE,YAAY,kBAAkB;AAAA,QAC9D;AAEA,oBAAY,IAAI,sCAAsC;AAAA,UACpD,UAAU,cAAc,WAAW;AAAA,UACnC;AAAA,UACA;AAAA,UACA,kBAAkB,CAAC,CAAC,cAAc;AAAA,UAClC,cAAc,cAAc,SACxB,aAAa,OAAO,UAAU,GAAG,CAAC,IAClC;AAAA,UACJ,WAAW,OAAO;AAAA,UAClB,aAAa;AAAA,UACb;AAAA,UACA,cAAc,OAAO,UAAU,UAAU;AAAA,UACzC,YAAY;AAAA,UACZ,YAAY,YAAY;AAAA,UACxB,gBAAgB;AAAA,UAChB,YAAW,oBAAI,KAAK,GAAE,YAAY;AAAA,UAClC,gBAAgB,cAAc;AAAA,UAC9B,kBAAkB,cAAc;AAAA,QAClC,CAAC;AAED,YAAI,OAAO,QAAQ;AACjB,gBAAM,SAAS,MAAM,UAAU,KAAK,SAAS;AAAA,YAC3C;AAAA,cACE,GAAG;AAAA,cACH,QAAQ;AAAA,YACV;AAAA,YACA;AAAA,cACE;AAAA,YACF;AAAA,UACF;AAEA,cAAI,gBAA4B;AAChC,cAAI,oBAAyB;AAC7B,gBAAM,gBAAuB,CAAC;AAC9B,gBAAM,mBAAmB,oBAAI,IAAoB;AACjD,cAAI,QAAa;AACjB,cAAI,aAA4B;AAChC,cAAI,eAA8B;AAElC,2BAAiB,SAAS,QAAQ;AAChC,gBAAI,OAAO,SAAS;AAClB,0BAAY,KAAK,kBAAkB;AAAA,gBACjC,WAAW,MAAM;AAAA,gBACjB,WAAW,KAAK,IAAI;AAAA,cACtB,CAAC;AACD,oBAAM,IAAI,MAAM,uBAAuB;AAAA,YACzC;AAEA,oBAAQ,MAAM,MAAM;AAAA,cAClB,KAAK;AACH,oCAAoB;AACpB,gCAAgB;AAAA,kBACd,GAAG,MAAM;AAAA,kBACT,SAAS,CAAC;AAAA,gBACZ;AACA,kCAAkB;AAAA,kBAChB,aAAa,MAAM,QAAQ,OAAO;AAAA,kBAClC,eAAe;AAAA,gBACjB,CAAC;AACD;AAAA,cAEF,KAAK;AACH,8BAAc,MAAM,KAAK,IAAI,EAAE,GAAG,MAAM,cAAc;AACtD,oBAAI,MAAM,cAAc,SAAS,YAAY;AAC3C,mCAAiB,IAAI,MAAM,OAAO,EAAE;AAAA,gBACtC;AACA,oBAAI,MAAM,cAAc,SAAS,YAAY;AAC3C,wBAAM,oBAAoB,KAAK,IAAI;AACnC,oCAAkB;AAAA,oBAChB,OAAO;AAAA,oBACP,YAAY;AAAA,oBACZ,mBAAmB;AAAA,oBACnB;AAAA,kBACF,CAAC;AACD,8BAAY,KAAK,wBAAwB;AAAA,oBACvC,OAAO,MAAM;AAAA,oBACb,WAAW;AAAA,oBACX,WAAW;AAAA,kBACb,CAAC;AAAA,gBACH;AACA;AAAA,cAEF,KAAK;AACH,sBAAM,aAAa,MAAM;AAEzB,oBAAI,CAAC,cAAc,UAAU,GAAG;AAC9B,gCAAc,UAAU,IAAI;AAAA,oBAC1B,MACE,MAAM,MAAM,SAAS,eAAe,SAAS;AAAA,oBAC/C,MAAM,MAAM,MAAM,SAAS,eAAe,KAAK;AAAA,kBACjD;AACA,sBAAI,MAAM,MAAM,SAAS,oBAAoB;AAC3C,qCAAiB,IAAI,YAAY,EAAE;AAAA,kBACrC;AAAA,gBACF;AAEA,oBAAI,MAAM,MAAM,SAAS,cAAc;AACrC,gCAAc,UAAU,EAAE,QAAQ,MAAM,MAAM;AAC9C,wBAAM,eAAe,kBAAkB;AACvC,oCAAkB;AAAA,oBAChB,gBACG,aAAa,iBAAiB,KAC/B,MAAM,MAAM,KAAK;AAAA,kBACrB,CAAC;AAAA,gBACH,WAAW,MAAM,MAAM,SAAS,oBAAoB;AAClD,wBAAM,gBAAgB,iBAAiB,IAAI,UAAU,KAAK;AAC1D,mCAAiB;AAAA,oBACf;AAAA,oBACA,gBAAgB,MAAM,MAAM;AAAA,kBAC9B;AAAA,gBACF,WAES,MAAM,MAAM,SAAS,kBAAkB;AAC9C,wBAAM,kBACJ,cAAc,UAAU,EAAE,YAAY;AACxC,wBAAM,kBACJ,mBAAmB,MAAM,MAAM,YAAY;AAC7C,gCAAc,UAAU,IAAI;AAAA,oBAC1B,GAAG,cAAc,UAAU;AAAA,oBAC3B,UAAU;AAAA,kBACZ;AAEA,oCAAkB;AAAA,oBAChB,OAAO;AAAA,oBACP,YAAY,gBAAgB;AAAA,oBAC5B,mBAAmB;AAAA,kBACrB,CAAC;AAED,8BAAY,MAAM,kBAAkB;AAAA,oBAClC,OAAO;AAAA,oBACP,YAAY,gBAAgB;AAAA,kBAC9B,CAAC;AAAA,gBACH;AACA;AAAA,cAEF,KAAK;AACH,oBAAI,MAAM,MAAM;AACd,+BAAa,MAAM,MAAM;AAC3B,oBAAI,MAAM,MAAM;AACd,iCAAe,MAAM,MAAM;AAC7B,oBAAI,MAAM,OAAO;AACf,0BAAQ,EAAE,GAAG,OAAO,GAAG,MAAM,MAAM;AACnC,oCAAkB;AAAA,oBAChB,cAAc,MAAM,MAAM;AAAA,kBAC5B,CAAC;AAAA,gBACH;AACA;AAAA,cAEF,KAAK;AACH,sBAAM,YAAY,MAAM;AACxB,sBAAM,QAAQ,cAAc,SAAS;AAErC,oBAAI,OAAO,SAAS,YAAY;AAC9B,wBAAMA,eAAc,kBAAkB;AACtC,wBAAM,qBAAqBA,aAAY,oBACnC,KAAK,IAAI,IAAIA,aAAY,oBACzB;AACJ,oCAAkB;AAAA,oBAChB,OAAO;AAAA,oBACP;AAAA,kBACF,CAAC;AACD,8BAAY,KAAK,2BAA2B;AAAA,oBAC1C,OAAO;AAAA,oBACP,iBAAiB,MAAM,UAAU,UAAU;AAAA,oBAC3C,YAAY;AAAA,kBACd,CAAC;AAAA,gBACH;AAEA,oBACE,OAAO,SAAS,cAChB,iBAAiB,IAAI,SAAS,GAC9B;AACA,wBAAM,UAAU,iBAAiB,IAAI,SAAS;AAC9C,sBAAI,SAAS;AACX,wBAAI;AACF,4BAAM,QAAQ,KAAK,MAAM,OAAO;AAAA,oBAClC,SAAS,OAAO;AACd,4BAAM,WACJ,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK;AACvD,kCAAY,MAAM,oBAAoB;AAAA,wBACpC,YAAY;AAAA,wBACZ,SACE,QAAQ,SAAS,MACb,QAAQ,MAAM,GAAG,GAAG,IAAI,QACxB;AAAA,wBACN,OAAO;AAAA,sBACT,CAAC;AACD,4BAAM,QAAQ;AAAA,wBACZ,iBAAiB;AAAA,wBACjB,mBAAmB,sBAAsB,QAAQ;AAAA,wBACjD,sBACE,QAAQ,SAAS,MACb,QAAQ,MAAM,GAAG,GAAG,IAAI,QACxB;AAAA,sBACR;AAAA,oBACF;AACA,qCAAiB,OAAO,SAAS;AAAA,kBACnC;AAAA,gBACF;AACA;AAAA,cAEF,KAAK;AACH,iCAAiB,MAAM;AACvB;AAAA,YACJ;AAEA,gBAAI,MAAM,SAAS,gBAAgB;AACjC;AAAA,YACF;AAAA,UACF;AAEA,cAAI,CAAC,iBAAiB,CAAC,mBAAmB;AACxC,kBAAM,IAAI,MAAM,+CAA+C;AAAA,UACjE;AAGA,0BAAgB;AAAA,YACd,GAAG,kBAAkB;AAAA,YACrB,SAAS,cAAc,OAAO,OAAO;AAAA,YACrC,aAAa;AAAA,YACb,eAAe;AAAA,YACf,OAAO;AAAA,cACL,GAAG,kBAAkB,QAAQ;AAAA,cAC7B,GAAG;AAAA,YACL;AAAA,UACF;AAEA,iBAAO;AAAA,QACT,OAAO;AACL,sBAAY,IAAI,0CAA0C;AAAA,YACxD,UAAU,cAAc,WAAW;AAAA,YACnC;AAAA,YACA;AAAA,YACA,kBAAkB,CAAC,CAAC,cAAc;AAAA,YAClC,cAAc,cAAc,SACxB,aAAa,OAAO,UAAU,GAAG,CAAC,IAClC;AAAA,YACJ,WAAW,OAAO;AAAA,YAClB,aAAa;AAAA,YACb,cAAc,OAAO,UAAU,UAAU;AAAA,YACzC,YAAY;AAAA,YACZ,YAAY,YAAY;AAAA,YACxB,gBAAgB;AAAA,YAChB,YAAW,oBAAI,KAAK,GAAE,YAAY;AAAA,YAClC,gBAAgB,cAAc;AAAA,YAC9B,kBAAkB,cAAc;AAAA,UAClC,CAAC;AAED,iBAAO,MAAM,UAAU,KAAK,SAAS,OAAO,QAAQ;AAAA,YAClD;AAAA,UACF,CAAC;AAAA,QACH;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;AAEA,gBAAY,IAAI,8BAA8B;AAAA,MAC5C,SAAS,SAAS;AAAA,IACpB,CAAC;AAED,UAAM,SAAS,QAAQ,KAAK,IAAI;AAChC,UAAM,aAAa,KAAK,IAAI,IAAI;AAEhC,UAAM,UAAU,SAAS,QAAQ,IAAI,CAAC,UAAwB;AAC5D,UAAI,MAAM,SAAS,QAAQ;AACzB,eAAO;AAAA,UACL,MAAM;AAAA,UACN,MAAM,MAAM;AAAA,QACd;AAAA,MACF,WAAW,MAAM,SAAS,YAAY;AACpC,eAAO;AAAA,UACL,MAAM;AAAA,UACN,IAAI,MAAM;AAAA,UACV,MAAM,MAAM;AAAA,UACZ,OAAO,MAAM;AAAA,QACf;AAAA,MACF;AACA,aAAO;AAAA,IACT,CAAC;AAGD,UAAM,iBAAiB,QAAQ;AAAA,MAC7B,CAAC,UACC,MAAM,SAAS;AAAA,IACnB;AACA,UAAM,cAAc,kBAAkB;AACtC,UAAM,mBACJ,eAAe,SAAS,IACpB;AAAA,MACE,WAAW,eAAe;AAAA,QACxB,CAAC,KAAK,MAAM,OAAO,EAAE,UAAU,UAAU;AAAA,QACzC;AAAA,MACF;AAAA,MACA,YAAY,YAAY,sBAAsB;AAAA,MAC9C,WAAW,YAAY,qBAAqB,KAAK,IAAI;AAAA,MACrD,YAAY;AAAA,IACd,IACA;AAEN,UAAM,mBAAqC;AAAA,MACzC,SAAS;AAAA,QACP,IAAI,SAAS;AAAA,QACb;AAAA,QACA,OAAO,SAAS;AAAA,QAChB,MAAM;AAAA,QACN,aAAa,SAAS;AAAA,QACtB,eAAe,SAAS;AAAA,QACxB,MAAM;AAAA,QACN,OAAO,SAAS;AAAA,MAClB;AAAA,MACA,MAAM;AAAA,MACN,MAAM,OAAO;AAAA,MACb;AAAA,MACA,SAAS;AAAA;AAAA,MACT;AAAA,IACF;AAEA,UAAM,iBAAiB,OAAO,IAAI,YAAU;AAAA,MAC1C,MAAM;AAAA,MACN,SAAS,MAAM;AAAA,IACjB,EAAE;AAEF,sBAAkB;AAAA,MAChB,cAAc,aAAa,KAAK,IAAI;AAAA,MACpC,UAAU,CAAC,GAAG,gBAAgB,GAAG,iBAAiB;AAAA,MAClD;AAAA,MACA,OAAO,SAAS,QACZ;AAAA,QACE,aAAa,SAAS,MAAM;AAAA,QAC5B,cAAc,SAAS,MAAM;AAAA,MAC/B,IACA;AAAA,MACJ,QAAQ;AAAA,QACN;AAAA,QACA,KAAK,KAAK,IAAI;AAAA,MAChB;AAAA,MACA,WAAW;AAAA,IACb,CAAC;AAGD,UAAM,cAAc,SAAS,MAAM;AACnC,UAAM,eAAe,SAAS,MAAM;AACpC,UAAM,2BACJ,SAAS,MAAM,+BAA+B;AAChD,UAAM,uBAAuB,SAAS,MAAM,2BAA2B;AAEvE,UAAM,UACH,cAAc,MAAa,0BAA0B,KAAK,IAC1D,eAAe,MAAa,2BAA2B,KAAK,IAC5D,2BAA2B,MAC1B,0BAA0B,KAAK,IAChC,uBAAuB,OACrB,0BAA0B,KAAK,IAAI;AAExC,qBAAiB,UAAU;AAC3B,mBAAe,SAAS,UAAU;AAGlC;AAAA,MACE;AAAA,QACE;AAAA,QACA;AAAA,QACA,qBAAqB;AAAA,QACrB,iBAAiB;AAAA,MACnB;AAAA,MACA;AAAA,MACA;AAAA,MACA,iBACK;AAAA,QACC,SAAS,eAAe;AAAA,QACxB,WAAW,eAAe;AAAA,QAC1B;AAAA,MACF,IACA;AAAA,IACN;AAEA,WAAO;AAAA,EACT,SAAS,OAAO;AACd,WAAO,6BAA6B,KAAK;AAAA,EAC3C;AACF;",
+  "names": ["streamState"]
+}

package/dist/services/llm/dispatch.js

@@ -0,0 +1,218 @@
+import { randomUUID } from "crypto";
+import {
+  getGlobalConfig
+} from "../../utils/config.js";
+import { isGPT5ModelName as isGPT5Model } from "../../utils/config.js";
+import {
+  debug as debugLogger,
+  markPhase,
+  getCurrentRequest,
+  logErrorWithDiagnosis
+} from "../../utils/debugLogger.js";
+import { getModelManager } from "../../utils/model.js";
+import { withVCR } from "../vcr.js";
+import { responseStateManager, getConversationId } from "../responseStateManager.js";
+import { generateSystemReminders } from "../systemReminder.js";
+import { generateMintoContext } from "./mintoContext.js";
+import { queryAnthropicNative } from "./anthropicProvider.js";
+import { queryOpenAI } from "./openaiProvider.js";
+async function queryLLM(messages, systemPrompt, maxThinkingTokens, tools, signal, options) {
+  const modelManager = getModelManager();
+  const modelResolution = modelManager.resolveModelWithInfo(options.model);
+  if (!modelResolution.success || !modelResolution.profile) {
+    throw new Error(
+      modelResolution.error || `Failed to resolve model: ${options.model}`
+    );
+  }
+  const modelProfile = modelResolution.profile;
+  const resolvedModel = modelProfile.modelName;
+  const toolUseContext = options.toolUseContext;
+  if (toolUseContext && !toolUseContext.responseState) {
+    const conversationId = getConversationId(
+      toolUseContext.agentId,
+      toolUseContext.messageId
+    );
+    const previousResponseId = responseStateManager.getPreviousResponseId(conversationId);
+    toolUseContext.responseState = {
+      previousResponseId,
+      conversationId
+    };
+  }
+  debugLogger.api("MODEL_RESOLVED", {
+    inputParam: options.model,
+    resolvedModelName: resolvedModel,
+    provider: modelProfile.provider,
+    isPointer: ["main", "task", "reasoning", "quick"].includes(options.model),
+    hasResponseState: !!toolUseContext?.responseState,
+    conversationId: toolUseContext?.responseState?.conversationId,
+    requestId: getCurrentRequest()?.id
+  });
+  const currentRequest = getCurrentRequest();
+  debugLogger.api("LLM_REQUEST_START", {
+    messageCount: messages.length,
+    systemPromptLength: systemPrompt.join(" ").length,
+    toolCount: tools.length,
+    model: resolvedModel,
+    originalModelParam: options.model,
+    requestId: getCurrentRequest()?.id
+  });
+  markPhase("LLM_CALL");
+  try {
+    const result = await withVCR(
+      messages,
+      () => queryLLMWithPromptCaching(
+        messages,
+        systemPrompt,
+        maxThinkingTokens,
+        tools,
+        signal,
+        { ...options, model: resolvedModel, modelProfile, toolUseContext }
+      )
+    );
+    debugLogger.api("LLM_REQUEST_SUCCESS", {
+      costUSD: result.costUSD,
+      durationMs: result.durationMs,
+      responseLength: result.message.content?.length || 0,
+      requestId: getCurrentRequest()?.id
+    });
+    if (toolUseContext?.responseState?.conversationId && result.responseId) {
+      responseStateManager.setPreviousResponseId(
+        toolUseContext.responseState.conversationId,
+        result.responseId
+      );
+      debugLogger.api("RESPONSE_STATE_UPDATED", {
+        conversationId: toolUseContext.responseState.conversationId,
+        responseId: result.responseId,
+        requestId: getCurrentRequest()?.id
+      });
+    }
+    return result;
+  } catch (error) {
+    logErrorWithDiagnosis(
+      error,
+      {
+        messageCount: messages.length,
+        systemPromptLength: systemPrompt.join(" ").length,
+        model: options.model,
+        toolCount: tools.length,
+        phase: "LLM_CALL"
+      },
+      currentRequest?.id
+    );
+    throw error;
+  }
+}
+function formatSystemPromptWithContext(systemPrompt, context, agentId) {
+  const enhancedPrompt = [...systemPrompt];
+  let reminders = "";
+  const modelManager = getModelManager();
+  const modelProfile = modelManager.getModel("main");
+  if (modelProfile && isGPT5Model(modelProfile.modelName)) {
+    const persistencePrompts = [
+      "\n# Agent Persistence for Long-Running Coding Tasks",
+      "You are working on a coding project that may involve multiple steps and iterations. Please maintain context and continuity throughout the session:",
+      "- Remember architectural decisions and design patterns established earlier",
+      "- Keep track of file modifications and their relationships",
+      "- Maintain awareness of the overall project structure and goals",
+      "- Reference previous implementations when making related changes",
+      "- Ensure consistency with existing code style and conventions",
+      "- Build incrementally on previous work rather than starting from scratch"
+    ];
+    enhancedPrompt.push(...persistencePrompts);
+  }
+  const hasContext = Object.entries(context).length > 0;
+  if (hasContext) {
+    const mintoContext = generateMintoContext();
+    if (mintoContext) {
+      enhancedPrompt.push("\n---\n# \u9879\u76EE\u4E0A\u4E0B\u6587\n");
+      enhancedPrompt.push(mintoContext);
+      enhancedPrompt.push("\n---\n");
+    }
+    const reminderMessages = generateSystemReminders(hasContext, agentId);
+    if (reminderMessages.length > 0) {
+      reminders = reminderMessages.map((r) => r.content).join("\n") + "\n";
+    }
+    enhancedPrompt.push(
+      `
+As you answer the user's questions, you can use the following context:
+`
+    );
+    const filteredContext = Object.fromEntries(
+      Object.entries(context).filter(
+        ([key]) => key !== "projectDocs" && key !== "userDocs"
+      )
+    );
+    enhancedPrompt.push(
+      ...Object.entries(filteredContext).map(
+        ([key, value]) => `<context name="${key}">${value}</context>`
+      )
+    );
+  }
+  return { systemPrompt: enhancedPrompt, reminders };
+}
+async function queryLLMWithPromptCaching(messages, systemPrompt, maxThinkingTokens, tools, signal, options) {
+  const config = getGlobalConfig();
+  const modelManager = getModelManager();
+  const toolUseContext = options.toolUseContext;
+  const modelProfile = options.modelProfile || modelManager.getModel("main");
+  let provider;
+  if (modelProfile) {
+    provider = modelProfile.provider || config.primaryProvider || "anthropic";
+  } else {
+    provider = config.primaryProvider || "anthropic";
+  }
+  if (provider === "anthropic" || provider === "bigdream" || provider === "opendev") {
+    return queryAnthropicNative(
+      messages,
+      systemPrompt,
+      maxThinkingTokens,
+      tools,
+      signal,
+      { ...options, modelProfile, toolUseContext }
+    );
+  }
+  return queryOpenAI(messages, systemPrompt, maxThinkingTokens, tools, signal, {
+    ...options,
+    modelProfile,
+    toolUseContext
+  });
+}
+async function queryModel(modelPointer, messages, systemPrompt = [], signal) {
+  return queryLLM(
+    messages,
+    systemPrompt,
+    0,
+    // maxThinkingTokens
+    [],
+    // tools
+    signal || new AbortController().signal,
+    {
+      safeMode: false,
+      model: modelPointer,
+      prependCLISysprompt: true
+    }
+  );
+}
+async function queryQuick({
+  systemPrompt = [],
+  userPrompt,
+  assistantPrompt,
+  enablePromptCaching = false,
+  signal
+}) {
+  const messages = [
+    {
+      message: { role: "user", content: userPrompt },
+      type: "user",
+      uuid: randomUUID()
+    }
+  ];
+  return queryModel("quick", messages, systemPrompt, signal);
+}
+export {
+  formatSystemPromptWithContext,
+  queryLLM,
+  queryModel,
+  queryQuick
+};
+//# sourceMappingURL=dispatch.js.map

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

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

package/dist/services/llm/index.js

@@ -0,0 +1,44 @@
+import {
+  API_ERROR_MESSAGE_PREFIX,
+  PROMPT_TOO_LONG_ERROR_MESSAGE,
+  CREDIT_BALANCE_TOO_LOW_ERROR_MESSAGE,
+  INVALID_API_KEY_ERROR_MESSAGE,
+  NO_CONTENT_MESSAGE,
+  MAIN_QUERY_TEMPERATURE
+} from "./types.js";
+import { generateMintoContext, refreshMintoContext } from "./mintoContext.js";
+import {
+  fetchAnthropicModels,
+  verifyApiKey,
+  getAnthropicClient,
+  resetAnthropicClient,
+  userMessageToMessageParam,
+  assistantMessageToMessageParam
+} from "./anthropicProvider.js";
+import {
+  queryLLM,
+  formatSystemPromptWithContext,
+  queryModel,
+  queryQuick
+} from "./dispatch.js";
+export {
+  API_ERROR_MESSAGE_PREFIX,
+  CREDIT_BALANCE_TOO_LOW_ERROR_MESSAGE,
+  INVALID_API_KEY_ERROR_MESSAGE,
+  MAIN_QUERY_TEMPERATURE,
+  NO_CONTENT_MESSAGE,
+  PROMPT_TOO_LONG_ERROR_MESSAGE,
+  assistantMessageToMessageParam,
+  fetchAnthropicModels,
+  formatSystemPromptWithContext,
+  generateMintoContext,
+  getAnthropicClient,
+  queryLLM,
+  queryModel,
+  queryQuick,
+  refreshMintoContext,
+  resetAnthropicClient,
+  userMessageToMessageParam,
+  verifyApiKey
+};
+//# sourceMappingURL=index.js.map

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

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../../src/services/llm/index.ts"],
+  "sourcesContent": ["/**\n * LLM service layer \u2014 barrel re-exports.\n *\n * All public symbols that were previously exported from `@services/claude`\n * are re-exported here so existing importers continue to work.\n */\n\n// types & constants\nexport {\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  NO_CONTENT_MESSAGE,\n  MAIN_QUERY_TEMPERATURE,\n} from './types'\n\n// MintoContext\nexport { generateMintoContext, refreshMintoContext } from './mintoContext'\n\n// Anthropic provider\nexport {\n  fetchAnthropicModels,\n  verifyApiKey,\n  getAnthropicClient,\n  resetAnthropicClient,\n  userMessageToMessageParam,\n  assistantMessageToMessageParam,\n} from './anthropicProvider'\n\n// Dispatch (query entry points)\nexport {\n  queryLLM,\n  formatSystemPromptWithContext,\n  queryModel,\n  queryQuick,\n} from './dispatch'\n"],
+  "mappings": "AAQA;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OACK;AAGP,SAAS,sBAAsB,2BAA2B;AAG1D;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OACK;AAGP;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OACK;",
+  "names": []
+}

package/dist/services/llm/mintoContext.js

@@ -0,0 +1,69 @@
+import { getProjectDocs } from "../../context.js";
+import { debug as debugLogger } from "../../utils/debugLogger.js";
+class MintoContextManager {
+  static instance;
+  projectDocsCache = "";
+  cacheInitialized = false;
+  initPromise = null;
+  constructor() {
+  }
+  static getInstance() {
+    if (!MintoContextManager.instance) {
+      MintoContextManager.instance = new MintoContextManager();
+    }
+    return MintoContextManager.instance;
+  }
+  async initialize() {
+    if (this.cacheInitialized) return;
+    if (this.initPromise) {
+      return this.initPromise;
+    }
+    this.initPromise = this.loadProjectDocs();
+    await this.initPromise;
+  }
+  async loadProjectDocs() {
+    try {
+      const projectDocs = await getProjectDocs();
+      this.projectDocsCache = projectDocs || "";
+      this.cacheInitialized = true;
+      if (process.env.NODE_ENV === "development") {
+        debugLogger.info("MINTO_CONTEXT_LOADED", {
+          characters: this.projectDocsCache.length
+        });
+      }
+    } catch (error) {
+      debugLogger.warn("MINTO_CONTEXT_LOAD_FAILED", { error: String(error) });
+      this.projectDocsCache = "";
+      this.cacheInitialized = true;
+    }
+  }
+  getMintoContext() {
+    if (!this.cacheInitialized) {
+      this.initialize().catch((error) => {
+        debugLogger.warn("MINTO_CONTEXT_INIT_FAILED", { error: String(error) });
+      });
+      return "";
+    }
+    return this.projectDocsCache;
+  }
+  async refreshCache() {
+    this.cacheInitialized = false;
+    this.initPromise = null;
+    await this.initialize();
+  }
+}
+const mintoContextManager = MintoContextManager.getInstance();
+mintoContextManager.initialize().catch((error) => {
+  debugLogger.warn("MINTO_CONTEXT_MODULE_INIT_FAILED", { error: String(error) });
+});
+const generateMintoContext = () => {
+  return mintoContextManager.getMintoContext();
+};
+const refreshMintoContext = async () => {
+  await mintoContextManager.refreshCache();
+};
+export {
+  generateMintoContext,
+  refreshMintoContext
+};
+//# sourceMappingURL=mintoContext.js.map

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

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../../src/services/llm/mintoContext.ts"],
+  "sourcesContent": ["/**\n * MintoContext manager \u2014 synchronous cache for project documentation.\n */\nimport { getProjectDocs } from '@context'\nimport { debug as debugLogger } from '@utils/debugLogger'\n\nclass MintoContextManager {\n  private static instance: MintoContextManager\n  private projectDocsCache: string = ''\n  private cacheInitialized: boolean = false\n  private initPromise: Promise<void> | null = null\n\n  private constructor() {}\n\n  public static getInstance(): MintoContextManager {\n    if (!MintoContextManager.instance) {\n      MintoContextManager.instance = new MintoContextManager()\n    }\n    return MintoContextManager.instance\n  }\n\n  public async initialize(): Promise<void> {\n    if (this.cacheInitialized) return\n\n    if (this.initPromise) {\n      return this.initPromise\n    }\n\n    this.initPromise = this.loadProjectDocs()\n    await this.initPromise\n  }\n\n  private async loadProjectDocs(): Promise<void> {\n    try {\n      const projectDocs = await getProjectDocs()\n      this.projectDocsCache = projectDocs || ''\n      this.cacheInitialized = true\n\n      if (process.env.NODE_ENV === 'development') {\n        debugLogger.info('MINTO_CONTEXT_LOADED', {\n          characters: this.projectDocsCache.length,\n        })\n      }\n    } catch (error) {\n      debugLogger.warn('MINTO_CONTEXT_LOAD_FAILED', { error: String(error) })\n      this.projectDocsCache = ''\n      this.cacheInitialized = true\n    }\n  }\n\n  public getMintoContext(): string {\n    if (!this.cacheInitialized) {\n      this.initialize().catch(error => {\n        debugLogger.warn('MINTO_CONTEXT_INIT_FAILED', { error: String(error) })\n      })\n      return ''\n    }\n    return this.projectDocsCache\n  }\n\n  public async refreshCache(): Promise<void> {\n    this.cacheInitialized = false\n    this.initPromise = null\n    await this.initialize()\n  }\n}\n\nconst mintoContextManager = MintoContextManager.getInstance()\n\n// Initialize asynchronously on module load\nmintoContextManager.initialize().catch(error => {\n  debugLogger.warn('MINTO_CONTEXT_MODULE_INIT_FAILED', { error: String(error) })\n})\n\nexport const generateMintoContext = (): string => {\n  return mintoContextManager.getMintoContext()\n}\n\nexport const refreshMintoContext = async (): Promise<void> => {\n  await mintoContextManager.refreshCache()\n}\n"],
+  "mappings": "AAGA,SAAS,sBAAsB;AAC/B,SAAS,SAAS,mBAAmB;AAErC,MAAM,oBAAoB;AAAA,EACxB,OAAe;AAAA,EACP,mBAA2B;AAAA,EAC3B,mBAA4B;AAAA,EAC5B,cAAoC;AAAA,EAEpC,cAAc;AAAA,EAAC;AAAA,EAEvB,OAAc,cAAmC;AAC/C,QAAI,CAAC,oBAAoB,UAAU;AACjC,0BAAoB,WAAW,IAAI,oBAAoB;AAAA,IACzD;AACA,WAAO,oBAAoB;AAAA,EAC7B;AAAA,EAEA,MAAa,aAA4B;AACvC,QAAI,KAAK,iBAAkB;AAE3B,QAAI,KAAK,aAAa;AACpB,aAAO,KAAK;AAAA,IACd;AAEA,SAAK,cAAc,KAAK,gBAAgB;AACxC,UAAM,KAAK;AAAA,EACb;AAAA,EAEA,MAAc,kBAAiC;AAC7C,QAAI;AACF,YAAM,cAAc,MAAM,eAAe;AACzC,WAAK,mBAAmB,eAAe;AACvC,WAAK,mBAAmB;AAExB,UAAI,QAAQ,IAAI,aAAa,eAAe;AAC1C,oBAAY,KAAK,wBAAwB;AAAA,UACvC,YAAY,KAAK,iBAAiB;AAAA,QACpC,CAAC;AAAA,MACH;AAAA,IACF,SAAS,OAAO;AACd,kBAAY,KAAK,6BAA6B,EAAE,OAAO,OAAO,KAAK,EAAE,CAAC;AACtE,WAAK,mBAAmB;AACxB,WAAK,mBAAmB;AAAA,IAC1B;AAAA,EACF;AAAA,EAEO,kBAA0B;AAC/B,QAAI,CAAC,KAAK,kBAAkB;AAC1B,WAAK,WAAW,EAAE,MAAM,WAAS;AAC/B,oBAAY,KAAK,6BAA6B,EAAE,OAAO,OAAO,KAAK,EAAE,CAAC;AAAA,MACxE,CAAC;AACD,aAAO;AAAA,IACT;AACA,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,MAAa,eAA8B;AACzC,SAAK,mBAAmB;AACxB,SAAK,cAAc;AACnB,UAAM,KAAK,WAAW;AAAA,EACxB;AACF;AAEA,MAAM,sBAAsB,oBAAoB,YAAY;AAG5D,oBAAoB,WAAW,EAAE,MAAM,WAAS;AAC9C,cAAY,KAAK,oCAAoC,EAAE,OAAO,OAAO,KAAK,EAAE,CAAC;AAC/E,CAAC;AAEM,MAAM,uBAAuB,MAAc;AAChD,SAAO,oBAAoB,gBAAgB;AAC7C;AAEO,MAAM,sBAAsB,YAA2B;AAC5D,QAAM,oBAAoB,aAAa;AACzC;",
+  "names": []
+}