@midscene/core 1.2.2-beta-20260116064919.0 → 1.2.2-beta-20260116083700.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/es/agent/agent.mjs +7 -4
- package/dist/es/agent/agent.mjs.map +1 -1
- package/dist/es/agent/tasks.mjs +4 -2
- package/dist/es/agent/tasks.mjs.map +1 -1
- package/dist/es/agent/utils.mjs +1 -1
- package/dist/es/ai-model/auto-glm/actions.mjs +224 -0
- package/dist/es/ai-model/auto-glm/actions.mjs.map +1 -0
- package/dist/es/ai-model/auto-glm/index.mjs +6 -0
- package/dist/es/ai-model/auto-glm/parser.mjs +239 -0
- package/dist/es/ai-model/auto-glm/parser.mjs.map +1 -0
- package/dist/es/ai-model/auto-glm/planning.mjs +63 -0
- package/dist/es/ai-model/auto-glm/planning.mjs.map +1 -0
- package/dist/es/ai-model/auto-glm/prompt.mjs +222 -0
- package/dist/es/ai-model/auto-glm/prompt.mjs.map +1 -0
- package/dist/es/ai-model/auto-glm/util.mjs +6 -0
- package/dist/es/ai-model/auto-glm/util.mjs.map +1 -0
- package/dist/es/ai-model/index.mjs +2 -1
- package/dist/es/ai-model/inspect.mjs +71 -6
- package/dist/es/ai-model/inspect.mjs.map +1 -1
- package/dist/es/ai-model/llm-planning.mjs +1 -1
- package/dist/es/ai-model/llm-planning.mjs.map +1 -1
- package/dist/es/ai-model/service-caller/index.mjs +5 -0
- package/dist/es/ai-model/service-caller/index.mjs.map +1 -1
- package/dist/es/ai-model/ui-tars-planning.mjs +25 -22
- package/dist/es/ai-model/ui-tars-planning.mjs.map +1 -1
- package/dist/es/common.mjs +15 -2
- package/dist/es/common.mjs.map +1 -1
- package/dist/es/index.mjs.map +1 -1
- package/dist/es/screenshot-item.mjs +9 -4
- package/dist/es/screenshot-item.mjs.map +1 -1
- package/dist/es/service/index.mjs +6 -1
- package/dist/es/service/index.mjs.map +1 -1
- package/dist/es/utils.mjs +2 -2
- package/dist/es/yaml/player.mjs +4 -3
- package/dist/es/yaml/player.mjs.map +1 -1
- package/dist/lib/agent/agent.js +7 -4
- package/dist/lib/agent/agent.js.map +1 -1
- package/dist/lib/agent/tasks.js +3 -1
- package/dist/lib/agent/tasks.js.map +1 -1
- package/dist/lib/agent/utils.js +1 -1
- package/dist/lib/ai-model/auto-glm/actions.js +258 -0
- package/dist/lib/ai-model/auto-glm/actions.js.map +1 -0
- package/dist/lib/ai-model/auto-glm/index.js +63 -0
- package/dist/lib/ai-model/auto-glm/index.js.map +1 -0
- package/dist/lib/ai-model/auto-glm/parser.js +282 -0
- package/dist/lib/ai-model/auto-glm/parser.js.map +1 -0
- package/dist/lib/ai-model/auto-glm/planning.js +97 -0
- package/dist/lib/ai-model/auto-glm/planning.js.map +1 -0
- package/dist/lib/ai-model/auto-glm/prompt.js +259 -0
- package/dist/lib/ai-model/auto-glm/prompt.js.map +1 -0
- package/dist/lib/ai-model/auto-glm/util.js +40 -0
- package/dist/lib/ai-model/auto-glm/util.js.map +1 -0
- package/dist/lib/ai-model/index.js +15 -11
- package/dist/lib/ai-model/inspect.js +70 -5
- package/dist/lib/ai-model/inspect.js.map +1 -1
- package/dist/lib/ai-model/llm-planning.js +1 -1
- package/dist/lib/ai-model/llm-planning.js.map +1 -1
- package/dist/lib/ai-model/service-caller/index.js +5 -0
- package/dist/lib/ai-model/service-caller/index.js.map +1 -1
- package/dist/lib/ai-model/ui-tars-planning.js +25 -22
- package/dist/lib/ai-model/ui-tars-planning.js.map +1 -1
- package/dist/lib/common.js +18 -2
- package/dist/lib/common.js.map +1 -1
- package/dist/lib/index.js.map +1 -1
- package/dist/lib/screenshot-item.js +9 -4
- package/dist/lib/screenshot-item.js.map +1 -1
- package/dist/lib/service/index.js +6 -1
- package/dist/lib/service/index.js.map +1 -1
- package/dist/lib/utils.js +2 -2
- package/dist/lib/yaml/player.js +4 -3
- package/dist/lib/yaml/player.js.map +1 -1
- package/dist/types/ai-model/auto-glm/actions.d.ts +77 -0
- package/dist/types/ai-model/auto-glm/index.d.ts +6 -0
- package/dist/types/ai-model/auto-glm/parser.d.ts +18 -0
- package/dist/types/ai-model/auto-glm/planning.d.ts +9 -0
- package/dist/types/ai-model/auto-glm/prompt.d.ts +27 -0
- package/dist/types/ai-model/auto-glm/util.d.ts +7 -0
- package/dist/types/ai-model/index.d.ts +1 -0
- package/dist/types/common.d.ts +10 -0
- package/dist/types/index.d.ts +1 -0
- package/dist/types/screenshot-item.d.ts +18 -8
- package/dist/types/yaml.d.ts +5 -1
- package/package.json +2 -2
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"ai-model/service-caller/index.mjs","sources":["../../../../src/ai-model/service-caller/index.ts"],"sourcesContent":["import {\n AIResponseFormat,\n type AIUsageInfo,\n type DeepThinkOption,\n} from '@/types';\nimport type { CodeGenerationChunk, StreamingCallback } from '@/types';\nimport {\n type IModelConfig,\n MIDSCENE_LANGFUSE_DEBUG,\n MIDSCENE_LANGSMITH_DEBUG,\n MIDSCENE_MODEL_MAX_TOKENS,\n OPENAI_MAX_TOKENS,\n type TModelFamily,\n type TVlModeTypes,\n type UITarsModelVersion,\n globalConfigManager,\n} from '@midscene/shared/env';\n\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert, ifInBrowser } from '@midscene/shared/utils';\nimport { jsonrepair } from 'jsonrepair';\nimport OpenAI from 'openai';\nimport type { ChatCompletionMessageParam } from 'openai/resources/index';\nimport type { Stream } from 'openai/streaming';\nimport type { AIArgs } from '../../common';\n\nasync function createChatClient({\n modelConfig,\n}: {\n modelConfig: IModelConfig;\n}): Promise<{\n completion: OpenAI.Chat.Completions;\n modelName: string;\n modelDescription: string;\n uiTarsVersion?: UITarsModelVersion;\n vlMode: TVlModeTypes | undefined;\n modelFamily: TModelFamily | undefined;\n}> {\n const {\n socksProxy,\n httpProxy,\n modelName,\n openaiBaseURL,\n openaiApiKey,\n openaiExtraConfig,\n modelDescription,\n uiTarsModelVersion: uiTarsVersion,\n vlMode,\n modelFamily,\n createOpenAIClient,\n timeout,\n } = modelConfig;\n\n let proxyAgent: any = undefined;\n const debugProxy = getDebug('ai:call:proxy');\n\n // Helper function to sanitize proxy URL for logging (remove credentials)\n // Uses URL API instead of regex to avoid ReDoS vulnerabilities\n const sanitizeProxyUrl = (url: string): string => {\n try {\n const parsed = new URL(url);\n if (parsed.username) {\n // Keep username for debugging, hide password for security\n parsed.password = '****';\n return parsed.href;\n }\n return url;\n } catch {\n // If URL parsing fails, return original URL (will be caught later)\n return url;\n }\n };\n\n if (httpProxy) {\n debugProxy('using http proxy', sanitizeProxyUrl(httpProxy));\n if (ifInBrowser) {\n console.warn(\n 'HTTP proxy is configured but not supported in browser environment',\n );\n } else {\n // Dynamic import with variable to avoid bundler static analysis\n const moduleName = 'undici';\n const { ProxyAgent } = await import(moduleName);\n proxyAgent = new ProxyAgent({\n uri: httpProxy,\n // Note: authentication is handled via the URI (e.g., http://user:pass@proxy.com:8080)\n });\n }\n } else if (socksProxy) {\n debugProxy('using socks proxy', sanitizeProxyUrl(socksProxy));\n if (ifInBrowser) {\n console.warn(\n 'SOCKS proxy is configured but not supported in browser environment',\n );\n } else {\n try {\n // Dynamic import with variable to avoid bundler static analysis\n const moduleName = 'fetch-socks';\n const { socksDispatcher } = await import(moduleName);\n // Parse SOCKS proxy URL (e.g., socks5://127.0.0.1:1080)\n const proxyUrl = new URL(socksProxy);\n\n // Validate hostname\n if (!proxyUrl.hostname) {\n throw new Error('SOCKS proxy URL must include a valid hostname');\n }\n\n // Validate and parse port\n const port = Number.parseInt(proxyUrl.port, 10);\n if (!proxyUrl.port || Number.isNaN(port)) {\n throw new Error('SOCKS proxy URL must include a valid port');\n }\n\n // Parse SOCKS version from protocol\n const protocol = proxyUrl.protocol.replace(':', '');\n const socksType =\n protocol === 'socks4' ? 4 : protocol === 'socks5' ? 5 : 5;\n\n proxyAgent = socksDispatcher({\n type: socksType,\n host: proxyUrl.hostname,\n port,\n ...(proxyUrl.username\n ? {\n userId: decodeURIComponent(proxyUrl.username),\n password: decodeURIComponent(proxyUrl.password || ''),\n }\n : {}),\n });\n debugProxy('socks proxy configured successfully', {\n type: socksType,\n host: proxyUrl.hostname,\n port: port,\n });\n } catch (error) {\n console.error('Failed to configure SOCKS proxy:', error);\n throw new Error(\n `Invalid SOCKS proxy URL: ${socksProxy}. Expected format: socks4://host:port, socks5://host:port, or with authentication: socks5://user:pass@host:port`,\n );\n }\n }\n }\n\n const openAIOptions = {\n baseURL: openaiBaseURL,\n apiKey: openaiApiKey,\n // Use fetchOptions.dispatcher for fetch-based SDK instead of httpAgent\n // Note: Type assertion needed due to undici version mismatch between dependencies\n ...(proxyAgent ? { fetchOptions: { dispatcher: proxyAgent as any } } : {}),\n ...openaiExtraConfig,\n ...(typeof timeout === 'number' ? { timeout } : {}),\n dangerouslyAllowBrowser: true,\n };\n\n const baseOpenAI = new OpenAI(openAIOptions);\n\n let openai: OpenAI = baseOpenAI;\n\n // LangSmith wrapper\n if (\n openai &&\n globalConfigManager.getEnvConfigInBoolean(MIDSCENE_LANGSMITH_DEBUG)\n ) {\n if (ifInBrowser) {\n throw new Error('langsmith is not supported in browser');\n }\n console.log('DEBUGGING MODE: langsmith wrapper enabled');\n // Use variable to prevent static analysis by bundlers\n const langsmithModule = 'langsmith/wrappers';\n const { wrapOpenAI } = await import(langsmithModule);\n openai = wrapOpenAI(openai);\n }\n\n // Langfuse wrapper\n if (\n openai &&\n globalConfigManager.getEnvConfigInBoolean(MIDSCENE_LANGFUSE_DEBUG)\n ) {\n if (ifInBrowser) {\n throw new Error('langfuse is not supported in browser');\n }\n console.log('DEBUGGING MODE: langfuse wrapper enabled');\n // Use variable to prevent static analysis by bundlers\n const langfuseModule = 'langfuse';\n const { observeOpenAI } = await import(langfuseModule);\n openai = observeOpenAI(openai);\n }\n\n if (createOpenAIClient) {\n const wrappedClient = await createOpenAIClient(baseOpenAI, openAIOptions);\n\n if (wrappedClient) {\n openai = wrappedClient as OpenAI;\n }\n }\n\n return {\n completion: openai.chat.completions,\n modelName,\n modelDescription,\n uiTarsVersion,\n vlMode,\n modelFamily,\n };\n}\n\nexport async function callAI(\n messages: ChatCompletionMessageParam[],\n modelConfig: IModelConfig,\n options?: {\n stream?: boolean;\n onChunk?: StreamingCallback;\n deepThink?: DeepThinkOption;\n },\n): Promise<{\n content: string;\n reasoning_content?: string;\n usage?: AIUsageInfo;\n isStreamed: boolean;\n}> {\n const {\n completion,\n modelName,\n modelDescription,\n uiTarsVersion,\n vlMode,\n modelFamily,\n } = await createChatClient({\n modelConfig,\n });\n\n const maxTokens =\n globalConfigManager.getEnvConfigValueAsNumber(MIDSCENE_MODEL_MAX_TOKENS) ??\n globalConfigManager.getEnvConfigValueAsNumber(OPENAI_MAX_TOKENS);\n const debugCall = getDebug('ai:call');\n const debugProfileStats = getDebug('ai:profile:stats');\n const debugProfileDetail = getDebug('ai:profile:detail');\n\n const startTime = Date.now();\n const temperature = modelConfig.temperature ?? 0;\n\n const isStreaming = options?.stream && options?.onChunk;\n let content: string | undefined;\n let accumulated = '';\n let accumulatedReasoning = '';\n let usage: OpenAI.CompletionUsage | undefined;\n let timeCost: number | undefined;\n\n const buildUsageInfo = (usageData?: OpenAI.CompletionUsage) => {\n if (!usageData) return undefined;\n\n const cachedInputTokens = (\n usageData as { prompt_tokens_details?: { cached_tokens?: number } }\n )?.prompt_tokens_details?.cached_tokens;\n\n return {\n prompt_tokens: usageData.prompt_tokens ?? 0,\n completion_tokens: usageData.completion_tokens ?? 0,\n total_tokens: usageData.total_tokens ?? 0,\n cached_input: cachedInputTokens ?? 0,\n time_cost: timeCost ?? 0,\n model_name: modelName,\n model_description: modelDescription,\n intent: modelConfig.intent,\n } satisfies AIUsageInfo;\n };\n\n const commonConfig = {\n temperature,\n stream: !!isStreaming,\n max_tokens: maxTokens,\n ...(vlMode === 'qwen2.5-vl' // qwen vl v2 specific config\n ? {\n vl_high_resolution_images: true,\n }\n : {}),\n };\n const {\n config: deepThinkConfig,\n debugMessage,\n warningMessage,\n } = resolveDeepThinkConfig({\n deepThink: options?.deepThink,\n modelFamily,\n });\n if (debugMessage) {\n debugCall(debugMessage);\n }\n if (warningMessage) {\n debugCall(warningMessage);\n console.warn(warningMessage);\n }\n\n try {\n debugCall(\n `sending ${isStreaming ? 'streaming ' : ''}request to ${modelName}`,\n );\n\n if (isStreaming) {\n const stream = (await completion.create(\n {\n model: modelName,\n messages,\n ...commonConfig,\n ...deepThinkConfig,\n },\n {\n stream: true,\n },\n )) as Stream<OpenAI.Chat.Completions.ChatCompletionChunk> & {\n _request_id?: string | null;\n };\n\n for await (const chunk of stream) {\n const content = chunk.choices?.[0]?.delta?.content || '';\n const reasoning_content =\n (chunk.choices?.[0]?.delta as any)?.reasoning_content || '';\n\n // Check for usage info in any chunk (OpenAI provides usage in separate chunks)\n if (chunk.usage) {\n usage = chunk.usage;\n }\n\n if (content || reasoning_content) {\n accumulated += content;\n accumulatedReasoning += reasoning_content;\n const chunkData: CodeGenerationChunk = {\n content,\n reasoning_content,\n accumulated,\n isComplete: false,\n usage: undefined,\n };\n options.onChunk!(chunkData);\n }\n\n // Check if stream is complete\n if (chunk.choices?.[0]?.finish_reason) {\n timeCost = Date.now() - startTime;\n\n // If usage is not available from the stream, provide a basic usage info\n if (!usage) {\n // Estimate token counts based on content length (rough approximation)\n const estimatedTokens = Math.max(\n 1,\n Math.floor(accumulated.length / 4),\n );\n usage = {\n prompt_tokens: estimatedTokens,\n completion_tokens: estimatedTokens,\n total_tokens: estimatedTokens * 2,\n };\n }\n\n // Send final chunk\n const finalChunk: CodeGenerationChunk = {\n content: '',\n accumulated,\n reasoning_content: '',\n isComplete: true,\n usage: buildUsageInfo(usage),\n };\n options.onChunk!(finalChunk);\n break;\n }\n }\n content = accumulated;\n debugProfileStats(\n `streaming model, ${modelName}, mode, ${vlMode || 'default'}, cost-ms, ${timeCost}, temperature, ${temperature ?? ''}`,\n );\n } else {\n // Non-streaming with retry logic\n const retryCount = modelConfig.retryCount ?? 1;\n const retryInterval = modelConfig.retryInterval ?? 2000;\n const maxAttempts = retryCount + 1; // retryCount=1 means 2 total attempts (1 initial + 1 retry)\n\n let lastError: Error | undefined;\n\n for (let attempt = 1; attempt <= maxAttempts; attempt++) {\n try {\n const result = await completion.create({\n model: modelName,\n messages,\n ...commonConfig,\n ...deepThinkConfig,\n } as any);\n\n timeCost = Date.now() - startTime;\n\n debugProfileStats(\n `model, ${modelName}, mode, ${vlMode || 'default'}, ui-tars-version, ${uiTarsVersion}, prompt-tokens, ${result.usage?.prompt_tokens || ''}, completion-tokens, ${result.usage?.completion_tokens || ''}, total-tokens, ${result.usage?.total_tokens || ''}, cost-ms, ${timeCost}, requestId, ${result._request_id || ''}, temperature, ${temperature ?? ''}`,\n );\n\n debugProfileDetail(\n `model usage detail: ${JSON.stringify(result.usage)}`,\n );\n\n if (!result.choices) {\n throw new Error(\n `invalid response from LLM service: ${JSON.stringify(result)}`,\n );\n }\n\n content = result.choices[0].message.content!;\n if (!content) {\n throw new Error('empty content from AI model');\n }\n\n accumulatedReasoning =\n (result.choices[0].message as any)?.reasoning_content || '';\n usage = result.usage;\n break; // Success, exit retry loop\n } catch (error) {\n lastError = error as Error;\n if (attempt < maxAttempts) {\n console.warn(\n `[Midscene] AI call failed (attempt ${attempt}/${maxAttempts}), retrying in ${retryInterval}ms... Error: ${lastError.message}`,\n );\n await new Promise((resolve) => setTimeout(resolve, retryInterval));\n }\n }\n }\n\n if (!content) {\n throw lastError;\n }\n }\n\n debugCall(`response reasoning content: ${accumulatedReasoning}`);\n debugCall(`response content: ${content}`);\n\n // Ensure we always have usage info for streaming responses\n if (isStreaming && !usage) {\n // Estimate token counts based on content length (rough approximation)\n const estimatedTokens = Math.max(\n 1,\n Math.floor((content || '').length / 4),\n );\n usage = {\n prompt_tokens: estimatedTokens,\n completion_tokens: estimatedTokens,\n total_tokens: estimatedTokens * 2,\n } as OpenAI.CompletionUsage;\n }\n\n return {\n content: content || '',\n reasoning_content: accumulatedReasoning || undefined,\n usage: buildUsageInfo(usage),\n isStreamed: !!isStreaming,\n };\n } catch (e: any) {\n console.error(' call AI error', e);\n const newError = new Error(\n `failed to call ${isStreaming ? 'streaming ' : ''}AI model service (${modelName}): ${e.message}\\nTrouble shooting: https://midscenejs.com/model-provider.html`,\n {\n cause: e,\n },\n );\n throw newError;\n }\n}\n\nexport async function callAIWithObjectResponse<T>(\n messages: ChatCompletionMessageParam[],\n modelConfig: IModelConfig,\n options?: {\n deepThink?: DeepThinkOption;\n },\n): Promise<{\n content: T;\n contentString: string;\n usage?: AIUsageInfo;\n reasoning_content?: string;\n}> {\n const response = await callAI(messages, modelConfig, {\n deepThink: options?.deepThink,\n });\n assert(response, 'empty response');\n const vlMode = modelConfig.vlMode;\n const jsonContent = safeParseJson(response.content, vlMode);\n assert(\n typeof jsonContent === 'object',\n `failed to parse json response from model (${modelConfig.modelName}): ${response.content}`,\n );\n return {\n content: jsonContent,\n contentString: response.content,\n usage: response.usage,\n reasoning_content: response.reasoning_content,\n };\n}\n\nexport async function callAIWithStringResponse(\n msgs: AIArgs,\n modelConfig: IModelConfig,\n): Promise<{ content: string; usage?: AIUsageInfo }> {\n const { content, usage } = await callAI(msgs, modelConfig);\n return { content, usage };\n}\n\nexport function extractJSONFromCodeBlock(response: string) {\n try {\n // First, try to match a JSON object directly in the response\n const jsonMatch = response.match(/^\\s*(\\{[\\s\\S]*\\})\\s*$/);\n if (jsonMatch) {\n return jsonMatch[1];\n }\n\n // If no direct JSON object is found, try to extract JSON from a code block\n const codeBlockMatch = response.match(\n /```(?:json)?\\s*(\\{[\\s\\S]*?\\})\\s*```/,\n );\n if (codeBlockMatch) {\n return codeBlockMatch[1];\n }\n\n // If no code block is found, try to find a JSON-like structure in the text\n const jsonLikeMatch = response.match(/\\{[\\s\\S]*\\}/);\n if (jsonLikeMatch) {\n return jsonLikeMatch[0];\n }\n } catch {}\n // If no JSON-like structure is found, return the original response\n return response;\n}\n\nexport function preprocessDoubaoBboxJson(input: string) {\n if (input.includes('bbox')) {\n // when its values like 940 445 969 490, replace all /\\d+\\s+\\d+/g with /$1,$2/g\n while (/\\d+\\s+\\d+/.test(input)) {\n input = input.replace(/(\\d+)\\s+(\\d+)/g, '$1,$2');\n }\n }\n return input;\n}\n\nexport function resolveDeepThinkConfig({\n deepThink,\n modelFamily,\n}: {\n deepThink?: DeepThinkOption;\n modelFamily?: TModelFamily;\n}): {\n config: Record<string, unknown>;\n debugMessage?: string;\n warningMessage?: string;\n} {\n const normalizedDeepThink = deepThink === 'unset' ? undefined : deepThink;\n\n if (normalizedDeepThink === undefined) {\n return { config: {}, debugMessage: undefined };\n }\n\n if (modelFamily === 'qwen3-vl') {\n return {\n config: { enable_thinking: normalizedDeepThink },\n debugMessage: `deepThink mapped to enable_thinking=${normalizedDeepThink} for qwen3-vl`,\n };\n }\n\n if (modelFamily === 'doubao-vision') {\n return {\n config: {\n thinking: { type: normalizedDeepThink ? 'enabled' : 'disabled' },\n },\n debugMessage: `deepThink mapped to thinking.type=${normalizedDeepThink ? 'enabled' : 'disabled'} for doubao-vision`,\n };\n }\n\n if (modelFamily === 'glm-v') {\n return {\n config: {\n thinking: { type: normalizedDeepThink ? 'enabled' : 'disabled' },\n },\n debugMessage: `deepThink mapped to thinking.type=${normalizedDeepThink ? 'enabled' : 'disabled'} for glm-v`,\n };\n }\n\n return {\n config: {},\n debugMessage: `deepThink ignored: unsupported model_family \"${modelFamily ?? 'default'}\"`,\n warningMessage: `The \"deepThink\" option is not supported for model_family \"${modelFamily ?? 'default'}\".`,\n };\n}\n\n/**\n * Normalize a parsed JSON object by trimming whitespace from:\n * 1. All object keys (e.g., \" prompt \" -> \"prompt\")\n * 2. All string values (e.g., \" Tap \" -> \"Tap\")\n * This handles LLM output that may include leading/trailing spaces.\n */\nfunction normalizeJsonObject(obj: any): any {\n // Handle null and undefined\n if (obj === null || obj === undefined) {\n return obj;\n }\n\n // Handle arrays - recursively normalize each element\n if (Array.isArray(obj)) {\n return obj.map((item) => normalizeJsonObject(item));\n }\n\n // Handle objects\n if (typeof obj === 'object') {\n const normalized: any = {};\n\n for (const [key, value] of Object.entries(obj)) {\n // Trim the key to remove leading/trailing spaces\n const trimmedKey = key.trim();\n\n // Recursively normalize the value\n let normalizedValue = normalizeJsonObject(value);\n\n // Trim all string values\n if (typeof normalizedValue === 'string') {\n normalizedValue = normalizedValue.trim();\n }\n\n normalized[trimmedKey] = normalizedValue;\n }\n\n return normalized;\n }\n\n // Handle primitive strings\n if (typeof obj === 'string') {\n return obj.trim();\n }\n\n // Return other primitives as-is\n return obj;\n}\n\nexport function safeParseJson(input: string, vlMode: TVlModeTypes | undefined) {\n const cleanJsonString = extractJSONFromCodeBlock(input);\n // match the point\n if (cleanJsonString?.match(/\\((\\d+),(\\d+)\\)/)) {\n return cleanJsonString\n .match(/\\((\\d+),(\\d+)\\)/)\n ?.slice(1)\n .map(Number);\n }\n\n let parsed: any;\n let lastError: unknown;\n try {\n parsed = JSON.parse(cleanJsonString);\n return normalizeJsonObject(parsed);\n } catch (error) {\n lastError = error;\n }\n try {\n parsed = JSON.parse(jsonrepair(cleanJsonString));\n return normalizeJsonObject(parsed);\n } catch (error) {\n lastError = error;\n }\n\n if (vlMode === 'doubao-vision' || vlMode === 'vlm-ui-tars') {\n const jsonString = preprocessDoubaoBboxJson(cleanJsonString);\n try {\n parsed = JSON.parse(jsonrepair(jsonString));\n return normalizeJsonObject(parsed);\n } catch (error) {\n lastError = error;\n }\n }\n throw Error(\n `failed to parse LLM response into JSON. Error - ${String(\n lastError ?? 'unknown error',\n )}. Response - \\n ${input}`,\n );\n}\n"],"names":["createChatClient","modelConfig","socksProxy","httpProxy","modelName","openaiBaseURL","openaiApiKey","openaiExtraConfig","modelDescription","uiTarsVersion","vlMode","modelFamily","createOpenAIClient","timeout","proxyAgent","debugProxy","getDebug","sanitizeProxyUrl","url","parsed","URL","ifInBrowser","console","moduleName","ProxyAgent","socksDispatcher","proxyUrl","Error","port","Number","protocol","socksType","decodeURIComponent","error","openAIOptions","baseOpenAI","OpenAI","openai","globalConfigManager","MIDSCENE_LANGSMITH_DEBUG","langsmithModule","wrapOpenAI","MIDSCENE_LANGFUSE_DEBUG","langfuseModule","observeOpenAI","wrappedClient","callAI","messages","options","completion","maxTokens","MIDSCENE_MODEL_MAX_TOKENS","OPENAI_MAX_TOKENS","debugCall","debugProfileStats","debugProfileDetail","startTime","Date","temperature","isStreaming","content","accumulated","accumulatedReasoning","usage","timeCost","buildUsageInfo","usageData","cachedInputTokens","commonConfig","deepThinkConfig","debugMessage","warningMessage","resolveDeepThinkConfig","stream","chunk","reasoning_content","chunkData","undefined","estimatedTokens","Math","finalChunk","retryCount","retryInterval","maxAttempts","lastError","attempt","result","JSON","Promise","resolve","setTimeout","e","newError","callAIWithObjectResponse","response","assert","jsonContent","safeParseJson","callAIWithStringResponse","msgs","extractJSONFromCodeBlock","jsonMatch","codeBlockMatch","jsonLikeMatch","preprocessDoubaoBboxJson","input","deepThink","normalizedDeepThink","normalizeJsonObject","obj","Array","item","normalized","key","value","Object","trimmedKey","normalizedValue","cleanJsonString","jsonrepair","jsonString","String"],"mappings":";;;;;AA0BA,eAAeA,iBAAiB,EAC9BC,WAAW,EAGZ;IAQC,MAAM,EACJC,UAAU,EACVC,SAAS,EACTC,SAAS,EACTC,aAAa,EACbC,YAAY,EACZC,iBAAiB,EACjBC,gBAAgB,EAChB,oBAAoBC,aAAa,EACjCC,MAAM,EACNC,WAAW,EACXC,kBAAkB,EAClBC,OAAO,EACR,GAAGZ;IAEJ,IAAIa;IACJ,MAAMC,aAAaC,SAAS;IAI5B,MAAMC,mBAAmB,CAACC;QACxB,IAAI;YACF,MAAMC,SAAS,IAAIC,IAAIF;YACvB,IAAIC,OAAO,QAAQ,EAAE;gBAEnBA,OAAO,QAAQ,GAAG;gBAClB,OAAOA,OAAO,IAAI;YACpB;YACA,OAAOD;QACT,EAAE,OAAM;YAEN,OAAOA;QACT;IACF;IAEA,IAAIf,WAAW;QACbY,WAAW,oBAAoBE,iBAAiBd;QAChD,IAAIkB,aACFC,QAAQ,IAAI,CACV;aAEG;YAEL,MAAMC,aAAa;YACnB,MAAM,EAAEC,UAAU,EAAE,GAAG,MAAM,MAAM,CAACD;YACpCT,aAAa,IAAIU,WAAW;gBAC1B,KAAKrB;YAEP;QACF;IACF,OAAO,IAAID,YAAY;QACrBa,WAAW,qBAAqBE,iBAAiBf;QACjD,IAAImB,aACFC,QAAQ,IAAI,CACV;aAGF,IAAI;YAEF,MAAMC,aAAa;YACnB,MAAM,EAAEE,eAAe,EAAE,GAAG,MAAM,MAAM,CAACF;YAEzC,MAAMG,WAAW,IAAIN,IAAIlB;YAGzB,IAAI,CAACwB,SAAS,QAAQ,EACpB,MAAM,IAAIC,MAAM;YAIlB,MAAMC,OAAOC,OAAO,QAAQ,CAACH,SAAS,IAAI,EAAE;YAC5C,IAAI,CAACA,SAAS,IAAI,IAAIG,OAAO,KAAK,CAACD,OACjC,MAAM,IAAID,MAAM;YAIlB,MAAMG,WAAWJ,SAAS,QAAQ,CAAC,OAAO,CAAC,KAAK;YAChD,MAAMK,YACJD,AAAa,aAAbA,WAAwB,IAAIA,AAAa,aAAbA,WAAwB,IAAI;YAE1DhB,aAAaW,gBAAgB;gBAC3B,MAAMM;gBACN,MAAML,SAAS,QAAQ;gBACvBE;gBACA,GAAIF,SAAS,QAAQ,GACjB;oBACE,QAAQM,mBAAmBN,SAAS,QAAQ;oBAC5C,UAAUM,mBAAmBN,SAAS,QAAQ,IAAI;gBACpD,IACA,CAAC,CAAC;YACR;YACAX,WAAW,uCAAuC;gBAChD,MAAMgB;gBACN,MAAML,SAAS,QAAQ;gBACvB,MAAME;YACR;QACF,EAAE,OAAOK,OAAO;YACdX,QAAQ,KAAK,CAAC,oCAAoCW;YAClD,MAAM,IAAIN,MACR,CAAC,yBAAyB,EAAEzB,WAAW,+GAA+G,CAAC;QAE3J;IAEJ;IAEA,MAAMgC,gBAAgB;QACpB,SAAS7B;QACT,QAAQC;QAGR,GAAIQ,aAAa;YAAE,cAAc;gBAAE,YAAYA;YAAkB;QAAE,IAAI,CAAC,CAAC;QACzE,GAAGP,iBAAiB;QACpB,GAAI,AAAmB,YAAnB,OAAOM,UAAuB;YAAEA;QAAQ,IAAI,CAAC,CAAC;QAClD,yBAAyB;IAC3B;IAEA,MAAMsB,aAAa,IAAIC,SAAOF;IAE9B,IAAIG,SAAiBF;IAGrB,IACEE,UACAC,oBAAoB,qBAAqB,CAACC,2BAC1C;QACA,IAAIlB,aACF,MAAM,IAAIM,MAAM;QAElBL,QAAQ,GAAG,CAAC;QAEZ,MAAMkB,kBAAkB;QACxB,MAAM,EAAEC,UAAU,EAAE,GAAG,MAAM,MAAM,CAACD;QACpCH,SAASI,WAAWJ;IACtB;IAGA,IACEA,UACAC,oBAAoB,qBAAqB,CAACI,0BAC1C;QACA,IAAIrB,aACF,MAAM,IAAIM,MAAM;QAElBL,QAAQ,GAAG,CAAC;QAEZ,MAAMqB,iBAAiB;QACvB,MAAM,EAAEC,aAAa,EAAE,GAAG,MAAM,MAAM,CAACD;QACvCN,SAASO,cAAcP;IACzB;IAEA,IAAIzB,oBAAoB;QACtB,MAAMiC,gBAAgB,MAAMjC,mBAAmBuB,YAAYD;QAE3D,IAAIW,eACFR,SAASQ;IAEb;IAEA,OAAO;QACL,YAAYR,OAAO,IAAI,CAAC,WAAW;QACnCjC;QACAI;QACAC;QACAC;QACAC;IACF;AACF;AAEO,eAAemC,OACpBC,QAAsC,EACtC9C,WAAyB,EACzB+C,OAIC;IAOD,MAAM,EACJC,UAAU,EACV7C,SAAS,EACTI,gBAAgB,EAChBC,aAAa,EACbC,MAAM,EACNC,WAAW,EACZ,GAAG,MAAMX,iBAAiB;QACzBC;IACF;IAEA,MAAMiD,YACJZ,oBAAoB,yBAAyB,CAACa,8BAC9Cb,oBAAoB,yBAAyB,CAACc;IAChD,MAAMC,YAAYrC,SAAS;IAC3B,MAAMsC,oBAAoBtC,SAAS;IACnC,MAAMuC,qBAAqBvC,SAAS;IAEpC,MAAMwC,YAAYC,KAAK,GAAG;IAC1B,MAAMC,cAAczD,YAAY,WAAW,IAAI;IAE/C,MAAM0D,cAAcX,SAAS,UAAUA,SAAS;IAChD,IAAIY;IACJ,IAAIC,cAAc;IAClB,IAAIC,uBAAuB;IAC3B,IAAIC;IACJ,IAAIC;IAEJ,MAAMC,iBAAiB,CAACC;QACtB,IAAI,CAACA,WAAW;QAEhB,MAAMC,oBACJD,WACC,uBAAuB;QAE1B,OAAO;YACL,eAAeA,UAAU,aAAa,IAAI;YAC1C,mBAAmBA,UAAU,iBAAiB,IAAI;YAClD,cAAcA,UAAU,YAAY,IAAI;YACxC,cAAcC,qBAAqB;YACnC,WAAWH,YAAY;YACvB,YAAY5D;YACZ,mBAAmBI;YACnB,QAAQP,YAAY,MAAM;QAC5B;IACF;IAEA,MAAMmE,eAAe;QACnBV;QACA,QAAQ,CAAC,CAACC;QACV,YAAYT;QACZ,GAAIxC,AAAW,iBAAXA,SACA;YACE,2BAA2B;QAC7B,IACA,CAAC,CAAC;IACR;IACA,MAAM,EACJ,QAAQ2D,eAAe,EACvBC,YAAY,EACZC,cAAc,EACf,GAAGC,uBAAuB;QACzB,WAAWxB,SAAS;QACpBrC;IACF;IACA,IAAI2D,cACFjB,UAAUiB;IAEZ,IAAIC,gBAAgB;QAClBlB,UAAUkB;QACVjD,QAAQ,IAAI,CAACiD;IACf;IAEA,IAAI;QACFlB,UACE,CAAC,QAAQ,EAAEM,cAAc,eAAe,GAAG,WAAW,EAAEvD,WAAW;QAGrE,IAAIuD,aAAa;YACf,MAAMc,SAAU,MAAMxB,WAAW,MAAM,CACrC;gBACE,OAAO7C;gBACP2C;gBACA,GAAGqB,YAAY;gBACf,GAAGC,eAAe;YACpB,GACA;gBACE,QAAQ;YACV;YAKF,WAAW,MAAMK,SAASD,OAAQ;gBAChC,MAAMb,UAAUc,MAAM,OAAO,EAAE,CAAC,EAAE,EAAE,OAAO,WAAW;gBACtD,MAAMC,oBACHD,MAAM,OAAO,EAAE,CAAC,EAAE,EAAE,OAAe,qBAAqB;gBAG3D,IAAIA,MAAM,KAAK,EACbX,QAAQW,MAAM,KAAK;gBAGrB,IAAId,WAAWe,mBAAmB;oBAChCd,eAAeD;oBACfE,wBAAwBa;oBACxB,MAAMC,YAAiC;wBACrChB;wBACAe;wBACAd;wBACA,YAAY;wBACZ,OAAOgB;oBACT;oBACA7B,QAAQ,OAAO,CAAE4B;gBACnB;gBAGA,IAAIF,MAAM,OAAO,EAAE,CAAC,EAAE,EAAE,eAAe;oBACrCV,WAAWP,KAAK,GAAG,KAAKD;oBAGxB,IAAI,CAACO,OAAO;wBAEV,MAAMe,kBAAkBC,KAAK,GAAG,CAC9B,GACAA,KAAK,KAAK,CAAClB,YAAY,MAAM,GAAG;wBAElCE,QAAQ;4BACN,eAAee;4BACf,mBAAmBA;4BACnB,cAAcA,AAAkB,IAAlBA;wBAChB;oBACF;oBAGA,MAAME,aAAkC;wBACtC,SAAS;wBACTnB;wBACA,mBAAmB;wBACnB,YAAY;wBACZ,OAAOI,eAAeF;oBACxB;oBACAf,QAAQ,OAAO,CAAEgC;oBACjB;gBACF;YACF;YACApB,UAAUC;YACVP,kBACE,CAAC,iBAAiB,EAAElD,UAAU,QAAQ,EAAEM,UAAU,UAAU,WAAW,EAAEsD,SAAS,eAAe,EAAEN,eAAe,IAAI;QAE1H,OAAO;YAEL,MAAMuB,aAAahF,YAAY,UAAU,IAAI;YAC7C,MAAMiF,gBAAgBjF,YAAY,aAAa,IAAI;YACnD,MAAMkF,cAAcF,aAAa;YAEjC,IAAIG;YAEJ,IAAK,IAAIC,UAAU,GAAGA,WAAWF,aAAaE,UAC5C,IAAI;gBACF,MAAMC,SAAS,MAAMrC,WAAW,MAAM,CAAC;oBACrC,OAAO7C;oBACP2C;oBACA,GAAGqB,YAAY;oBACf,GAAGC,eAAe;gBACpB;gBAEAL,WAAWP,KAAK,GAAG,KAAKD;gBAExBF,kBACE,CAAC,OAAO,EAAElD,UAAU,QAAQ,EAAEM,UAAU,UAAU,mBAAmB,EAAED,cAAc,iBAAiB,EAAE6E,OAAO,KAAK,EAAE,iBAAiB,GAAG,qBAAqB,EAAEA,OAAO,KAAK,EAAE,qBAAqB,GAAG,gBAAgB,EAAEA,OAAO,KAAK,EAAE,gBAAgB,GAAG,WAAW,EAAEtB,SAAS,aAAa,EAAEsB,OAAO,WAAW,IAAI,GAAG,eAAe,EAAE5B,eAAe,IAAI;gBAG9VH,mBACE,CAAC,oBAAoB,EAAEgC,KAAK,SAAS,CAACD,OAAO,KAAK,GAAG;gBAGvD,IAAI,CAACA,OAAO,OAAO,EACjB,MAAM,IAAI3D,MACR,CAAC,mCAAmC,EAAE4D,KAAK,SAAS,CAACD,SAAS;gBAIlE1B,UAAU0B,OAAO,OAAO,CAAC,EAAE,CAAC,OAAO,CAAC,OAAO;gBAC3C,IAAI,CAAC1B,SACH,MAAM,IAAIjC,MAAM;gBAGlBmC,uBACGwB,OAAO,OAAO,CAAC,EAAE,CAAC,OAAO,EAAU,qBAAqB;gBAC3DvB,QAAQuB,OAAO,KAAK;gBACpB;YACF,EAAE,OAAOrD,OAAO;gBACdmD,YAAYnD;gBACZ,IAAIoD,UAAUF,aAAa;oBACzB7D,QAAQ,IAAI,CACV,CAAC,mCAAmC,EAAE+D,QAAQ,CAAC,EAAEF,YAAY,eAAe,EAAED,cAAc,aAAa,EAAEE,UAAU,OAAO,EAAE;oBAEhI,MAAM,IAAII,QAAQ,CAACC,UAAYC,WAAWD,SAASP;gBACrD;YACF;YAGF,IAAI,CAACtB,SACH,MAAMwB;QAEV;QAEA/B,UAAU,CAAC,4BAA4B,EAAES,sBAAsB;QAC/DT,UAAU,CAAC,kBAAkB,EAAEO,SAAS;QAGxC,IAAID,eAAe,CAACI,OAAO;YAEzB,MAAMe,kBAAkBC,KAAK,GAAG,CAC9B,GACAA,KAAK,KAAK,CAAEnB,AAAAA,CAAAA,WAAW,EAAC,EAAG,MAAM,GAAG;YAEtCG,QAAQ;gBACN,eAAee;gBACf,mBAAmBA;gBACnB,cAAcA,AAAkB,IAAlBA;YAChB;QACF;QAEA,OAAO;YACL,SAASlB,WAAW;YACpB,mBAAmBE,wBAAwBe;YAC3C,OAAOZ,eAAeF;YACtB,YAAY,CAAC,CAACJ;QAChB;IACF,EAAE,OAAOgC,GAAQ;QACfrE,QAAQ,KAAK,CAAC,kBAAkBqE;QAChC,MAAMC,WAAW,IAAIjE,MACnB,CAAC,eAAe,EAAEgC,cAAc,eAAe,GAAG,kBAAkB,EAAEvD,UAAU,GAAG,EAAEuF,EAAE,OAAO,CAAC,8DAA8D,CAAC,EAC9J;YACE,OAAOA;QACT;QAEF,MAAMC;IACR;AACF;AAEO,eAAeC,yBACpB9C,QAAsC,EACtC9C,WAAyB,EACzB+C,OAEC;IAOD,MAAM8C,WAAW,MAAMhD,OAAOC,UAAU9C,aAAa;QACnD,WAAW+C,SAAS;IACtB;IACA+C,OAAOD,UAAU;IACjB,MAAMpF,SAAST,YAAY,MAAM;IACjC,MAAM+F,cAAcC,cAAcH,SAAS,OAAO,EAAEpF;IACpDqF,OACE,AAAuB,YAAvB,OAAOC,aACP,CAAC,0CAA0C,EAAE/F,YAAY,SAAS,CAAC,GAAG,EAAE6F,SAAS,OAAO,EAAE;IAE5F,OAAO;QACL,SAASE;QACT,eAAeF,SAAS,OAAO;QAC/B,OAAOA,SAAS,KAAK;QACrB,mBAAmBA,SAAS,iBAAiB;IAC/C;AACF;AAEO,eAAeI,yBACpBC,IAAY,EACZlG,WAAyB;IAEzB,MAAM,EAAE2D,OAAO,EAAEG,KAAK,EAAE,GAAG,MAAMjB,OAAOqD,MAAMlG;IAC9C,OAAO;QAAE2D;QAASG;IAAM;AAC1B;AAEO,SAASqC,yBAAyBN,QAAgB;IACvD,IAAI;QAEF,MAAMO,YAAYP,SAAS,KAAK,CAAC;QACjC,IAAIO,WACF,OAAOA,SAAS,CAAC,EAAE;QAIrB,MAAMC,iBAAiBR,SAAS,KAAK,CACnC;QAEF,IAAIQ,gBACF,OAAOA,cAAc,CAAC,EAAE;QAI1B,MAAMC,gBAAgBT,SAAS,KAAK,CAAC;QACrC,IAAIS,eACF,OAAOA,aAAa,CAAC,EAAE;IAE3B,EAAE,OAAM,CAAC;IAET,OAAOT;AACT;AAEO,SAASU,yBAAyBC,KAAa;IACpD,IAAIA,MAAM,QAAQ,CAAC,SAEjB,MAAO,YAAY,IAAI,CAACA,OACtBA,QAAQA,MAAM,OAAO,CAAC,kBAAkB;IAG5C,OAAOA;AACT;AAEO,SAASjC,uBAAuB,EACrCkC,SAAS,EACT/F,WAAW,EAIZ;IAKC,MAAMgG,sBAAsBD,AAAc,YAAdA,YAAwB7B,SAAY6B;IAEhE,IAAIC,AAAwB9B,WAAxB8B,qBACF,OAAO;QAAE,QAAQ,CAAC;QAAG,cAAc9B;IAAU;IAG/C,IAAIlE,AAAgB,eAAhBA,aACF,OAAO;QACL,QAAQ;YAAE,iBAAiBgG;QAAoB;QAC/C,cAAc,CAAC,oCAAoC,EAAEA,oBAAoB,aAAa,CAAC;IACzF;IAGF,IAAIhG,AAAgB,oBAAhBA,aACF,OAAO;QACL,QAAQ;YACN,UAAU;gBAAE,MAAMgG,sBAAsB,YAAY;YAAW;QACjE;QACA,cAAc,CAAC,kCAAkC,EAAEA,sBAAsB,YAAY,WAAW,kBAAkB,CAAC;IACrH;IAGF,IAAIhG,AAAgB,YAAhBA,aACF,OAAO;QACL,QAAQ;YACN,UAAU;gBAAE,MAAMgG,sBAAsB,YAAY;YAAW;QACjE;QACA,cAAc,CAAC,kCAAkC,EAAEA,sBAAsB,YAAY,WAAW,UAAU,CAAC;IAC7G;IAGF,OAAO;QACL,QAAQ,CAAC;QACT,cAAc,CAAC,6CAA6C,EAAEhG,eAAe,UAAU,CAAC,CAAC;QACzF,gBAAgB,CAAC,0DAA0D,EAAEA,eAAe,UAAU,EAAE,CAAC;IAC3G;AACF;AAQA,SAASiG,oBAAoBC,GAAQ;IAEnC,IAAIA,QAAAA,KACF,OAAOA;IAIT,IAAIC,MAAM,OAAO,CAACD,MAChB,OAAOA,IAAI,GAAG,CAAC,CAACE,OAASH,oBAAoBG;IAI/C,IAAI,AAAe,YAAf,OAAOF,KAAkB;QAC3B,MAAMG,aAAkB,CAAC;QAEzB,KAAK,MAAM,CAACC,KAAKC,MAAM,IAAIC,OAAO,OAAO,CAACN,KAAM;YAE9C,MAAMO,aAAaH,IAAI,IAAI;YAG3B,IAAII,kBAAkBT,oBAAoBM;YAG1C,IAAI,AAA2B,YAA3B,OAAOG,iBACTA,kBAAkBA,gBAAgB,IAAI;YAGxCL,UAAU,CAACI,WAAW,GAAGC;QAC3B;QAEA,OAAOL;IACT;IAGA,IAAI,AAAe,YAAf,OAAOH,KACT,OAAOA,IAAI,IAAI;IAIjB,OAAOA;AACT;AAEO,SAASZ,cAAcQ,KAAa,EAAE/F,MAAgC;IAC3E,MAAM4G,kBAAkBlB,yBAAyBK;IAEjD,IAAIa,iBAAiB,MAAM,oBACzB,OAAOA,gBACJ,KAAK,CAAC,oBACL,MAAM,GACP,IAAIzF;IAGT,IAAIV;IACJ,IAAIiE;IACJ,IAAI;QACFjE,SAASoE,KAAK,KAAK,CAAC+B;QACpB,OAAOV,oBAAoBzF;IAC7B,EAAE,OAAOc,OAAO;QACdmD,YAAYnD;IACd;IACA,IAAI;QACFd,SAASoE,KAAK,KAAK,CAACgC,WAAWD;QAC/B,OAAOV,oBAAoBzF;IAC7B,EAAE,OAAOc,OAAO;QACdmD,YAAYnD;IACd;IAEA,IAAIvB,AAAW,oBAAXA,UAA8BA,AAAW,kBAAXA,QAA0B;QAC1D,MAAM8G,aAAahB,yBAAyBc;QAC5C,IAAI;YACFnG,SAASoE,KAAK,KAAK,CAACgC,WAAWC;YAC/B,OAAOZ,oBAAoBzF;QAC7B,EAAE,OAAOc,OAAO;YACdmD,YAAYnD;QACd;IACF;IACA,MAAMN,MACJ,CAAC,gDAAgD,EAAE8F,OACjDrC,aAAa,iBACb,gBAAgB,EAAEqB,OAAO;AAE/B"}
|
|
1
|
+
{"version":3,"file":"ai-model/service-caller/index.mjs","sources":["../../../../src/ai-model/service-caller/index.ts"],"sourcesContent":["import {\n AIResponseFormat,\n type AIUsageInfo,\n type DeepThinkOption,\n} from '@/types';\nimport type { CodeGenerationChunk, StreamingCallback } from '@/types';\nimport {\n type IModelConfig,\n MIDSCENE_LANGFUSE_DEBUG,\n MIDSCENE_LANGSMITH_DEBUG,\n MIDSCENE_MODEL_MAX_TOKENS,\n OPENAI_MAX_TOKENS,\n type TModelFamily,\n type TVlModeTypes,\n type UITarsModelVersion,\n globalConfigManager,\n} from '@midscene/shared/env';\n\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert, ifInBrowser } from '@midscene/shared/utils';\nimport { jsonrepair } from 'jsonrepair';\nimport OpenAI from 'openai';\nimport type { ChatCompletionMessageParam } from 'openai/resources/index';\nimport type { Stream } from 'openai/streaming';\nimport type { AIArgs } from '../../common';\nimport { isAutoGLM } from '../auto-glm/util';\n\nasync function createChatClient({\n modelConfig,\n}: {\n modelConfig: IModelConfig;\n}): Promise<{\n completion: OpenAI.Chat.Completions;\n modelName: string;\n modelDescription: string;\n uiTarsVersion?: UITarsModelVersion;\n vlMode: TVlModeTypes | undefined;\n modelFamily: TModelFamily | undefined;\n}> {\n const {\n socksProxy,\n httpProxy,\n modelName,\n openaiBaseURL,\n openaiApiKey,\n openaiExtraConfig,\n modelDescription,\n uiTarsModelVersion: uiTarsVersion,\n vlMode,\n modelFamily,\n createOpenAIClient,\n timeout,\n } = modelConfig;\n\n let proxyAgent: any = undefined;\n const debugProxy = getDebug('ai:call:proxy');\n\n // Helper function to sanitize proxy URL for logging (remove credentials)\n // Uses URL API instead of regex to avoid ReDoS vulnerabilities\n const sanitizeProxyUrl = (url: string): string => {\n try {\n const parsed = new URL(url);\n if (parsed.username) {\n // Keep username for debugging, hide password for security\n parsed.password = '****';\n return parsed.href;\n }\n return url;\n } catch {\n // If URL parsing fails, return original URL (will be caught later)\n return url;\n }\n };\n\n if (httpProxy) {\n debugProxy('using http proxy', sanitizeProxyUrl(httpProxy));\n if (ifInBrowser) {\n console.warn(\n 'HTTP proxy is configured but not supported in browser environment',\n );\n } else {\n // Dynamic import with variable to avoid bundler static analysis\n const moduleName = 'undici';\n const { ProxyAgent } = await import(moduleName);\n proxyAgent = new ProxyAgent({\n uri: httpProxy,\n // Note: authentication is handled via the URI (e.g., http://user:pass@proxy.com:8080)\n });\n }\n } else if (socksProxy) {\n debugProxy('using socks proxy', sanitizeProxyUrl(socksProxy));\n if (ifInBrowser) {\n console.warn(\n 'SOCKS proxy is configured but not supported in browser environment',\n );\n } else {\n try {\n // Dynamic import with variable to avoid bundler static analysis\n const moduleName = 'fetch-socks';\n const { socksDispatcher } = await import(moduleName);\n // Parse SOCKS proxy URL (e.g., socks5://127.0.0.1:1080)\n const proxyUrl = new URL(socksProxy);\n\n // Validate hostname\n if (!proxyUrl.hostname) {\n throw new Error('SOCKS proxy URL must include a valid hostname');\n }\n\n // Validate and parse port\n const port = Number.parseInt(proxyUrl.port, 10);\n if (!proxyUrl.port || Number.isNaN(port)) {\n throw new Error('SOCKS proxy URL must include a valid port');\n }\n\n // Parse SOCKS version from protocol\n const protocol = proxyUrl.protocol.replace(':', '');\n const socksType =\n protocol === 'socks4' ? 4 : protocol === 'socks5' ? 5 : 5;\n\n proxyAgent = socksDispatcher({\n type: socksType,\n host: proxyUrl.hostname,\n port,\n ...(proxyUrl.username\n ? {\n userId: decodeURIComponent(proxyUrl.username),\n password: decodeURIComponent(proxyUrl.password || ''),\n }\n : {}),\n });\n debugProxy('socks proxy configured successfully', {\n type: socksType,\n host: proxyUrl.hostname,\n port: port,\n });\n } catch (error) {\n console.error('Failed to configure SOCKS proxy:', error);\n throw new Error(\n `Invalid SOCKS proxy URL: ${socksProxy}. Expected format: socks4://host:port, socks5://host:port, or with authentication: socks5://user:pass@host:port`,\n );\n }\n }\n }\n\n const openAIOptions = {\n baseURL: openaiBaseURL,\n apiKey: openaiApiKey,\n // Use fetchOptions.dispatcher for fetch-based SDK instead of httpAgent\n // Note: Type assertion needed due to undici version mismatch between dependencies\n ...(proxyAgent ? { fetchOptions: { dispatcher: proxyAgent as any } } : {}),\n ...openaiExtraConfig,\n ...(typeof timeout === 'number' ? { timeout } : {}),\n dangerouslyAllowBrowser: true,\n };\n\n const baseOpenAI = new OpenAI(openAIOptions);\n\n let openai: OpenAI = baseOpenAI;\n\n // LangSmith wrapper\n if (\n openai &&\n globalConfigManager.getEnvConfigInBoolean(MIDSCENE_LANGSMITH_DEBUG)\n ) {\n if (ifInBrowser) {\n throw new Error('langsmith is not supported in browser');\n }\n console.log('DEBUGGING MODE: langsmith wrapper enabled');\n // Use variable to prevent static analysis by bundlers\n const langsmithModule = 'langsmith/wrappers';\n const { wrapOpenAI } = await import(langsmithModule);\n openai = wrapOpenAI(openai);\n }\n\n // Langfuse wrapper\n if (\n openai &&\n globalConfigManager.getEnvConfigInBoolean(MIDSCENE_LANGFUSE_DEBUG)\n ) {\n if (ifInBrowser) {\n throw new Error('langfuse is not supported in browser');\n }\n console.log('DEBUGGING MODE: langfuse wrapper enabled');\n // Use variable to prevent static analysis by bundlers\n const langfuseModule = 'langfuse';\n const { observeOpenAI } = await import(langfuseModule);\n openai = observeOpenAI(openai);\n }\n\n if (createOpenAIClient) {\n const wrappedClient = await createOpenAIClient(baseOpenAI, openAIOptions);\n\n if (wrappedClient) {\n openai = wrappedClient as OpenAI;\n }\n }\n\n return {\n completion: openai.chat.completions,\n modelName,\n modelDescription,\n uiTarsVersion,\n vlMode,\n modelFamily,\n };\n}\n\nexport async function callAI(\n messages: ChatCompletionMessageParam[],\n modelConfig: IModelConfig,\n options?: {\n stream?: boolean;\n onChunk?: StreamingCallback;\n deepThink?: DeepThinkOption;\n },\n): Promise<{\n content: string;\n reasoning_content?: string;\n usage?: AIUsageInfo;\n isStreamed: boolean;\n}> {\n const {\n completion,\n modelName,\n modelDescription,\n uiTarsVersion,\n vlMode,\n modelFamily,\n } = await createChatClient({\n modelConfig,\n });\n\n const maxTokens =\n globalConfigManager.getEnvConfigValueAsNumber(MIDSCENE_MODEL_MAX_TOKENS) ??\n globalConfigManager.getEnvConfigValueAsNumber(OPENAI_MAX_TOKENS);\n const debugCall = getDebug('ai:call');\n const debugProfileStats = getDebug('ai:profile:stats');\n const debugProfileDetail = getDebug('ai:profile:detail');\n\n const startTime = Date.now();\n const temperature = modelConfig.temperature ?? 0;\n\n const isStreaming = options?.stream && options?.onChunk;\n let content: string | undefined;\n let accumulated = '';\n let accumulatedReasoning = '';\n let usage: OpenAI.CompletionUsage | undefined;\n let timeCost: number | undefined;\n\n const buildUsageInfo = (usageData?: OpenAI.CompletionUsage) => {\n if (!usageData) return undefined;\n\n const cachedInputTokens = (\n usageData as { prompt_tokens_details?: { cached_tokens?: number } }\n )?.prompt_tokens_details?.cached_tokens;\n\n return {\n prompt_tokens: usageData.prompt_tokens ?? 0,\n completion_tokens: usageData.completion_tokens ?? 0,\n total_tokens: usageData.total_tokens ?? 0,\n cached_input: cachedInputTokens ?? 0,\n time_cost: timeCost ?? 0,\n model_name: modelName,\n model_description: modelDescription,\n intent: modelConfig.intent,\n } satisfies AIUsageInfo;\n };\n\n const commonConfig = {\n temperature,\n stream: !!isStreaming,\n max_tokens: maxTokens,\n ...(vlMode === 'qwen2.5-vl' // qwen vl v2 specific config\n ? {\n vl_high_resolution_images: true,\n }\n : {}),\n };\n\n if (isAutoGLM(vlMode)) {\n (commonConfig as unknown as Record<string, number>).top_p = 0.85;\n (commonConfig as unknown as Record<string, number>).frequency_penalty = 0.2;\n }\n\n const {\n config: deepThinkConfig,\n debugMessage,\n warningMessage,\n } = resolveDeepThinkConfig({\n deepThink: options?.deepThink,\n modelFamily,\n });\n if (debugMessage) {\n debugCall(debugMessage);\n }\n if (warningMessage) {\n debugCall(warningMessage);\n console.warn(warningMessage);\n }\n\n try {\n debugCall(\n `sending ${isStreaming ? 'streaming ' : ''}request to ${modelName}`,\n );\n\n if (isStreaming) {\n const stream = (await completion.create(\n {\n model: modelName,\n messages,\n ...commonConfig,\n ...deepThinkConfig,\n },\n {\n stream: true,\n },\n )) as Stream<OpenAI.Chat.Completions.ChatCompletionChunk> & {\n _request_id?: string | null;\n };\n\n for await (const chunk of stream) {\n const content = chunk.choices?.[0]?.delta?.content || '';\n const reasoning_content =\n (chunk.choices?.[0]?.delta as any)?.reasoning_content || '';\n\n // Check for usage info in any chunk (OpenAI provides usage in separate chunks)\n if (chunk.usage) {\n usage = chunk.usage;\n }\n\n if (content || reasoning_content) {\n accumulated += content;\n accumulatedReasoning += reasoning_content;\n const chunkData: CodeGenerationChunk = {\n content,\n reasoning_content,\n accumulated,\n isComplete: false,\n usage: undefined,\n };\n options.onChunk!(chunkData);\n }\n\n // Check if stream is complete\n if (chunk.choices?.[0]?.finish_reason) {\n timeCost = Date.now() - startTime;\n\n // If usage is not available from the stream, provide a basic usage info\n if (!usage) {\n // Estimate token counts based on content length (rough approximation)\n const estimatedTokens = Math.max(\n 1,\n Math.floor(accumulated.length / 4),\n );\n usage = {\n prompt_tokens: estimatedTokens,\n completion_tokens: estimatedTokens,\n total_tokens: estimatedTokens * 2,\n };\n }\n\n // Send final chunk\n const finalChunk: CodeGenerationChunk = {\n content: '',\n accumulated,\n reasoning_content: '',\n isComplete: true,\n usage: buildUsageInfo(usage),\n };\n options.onChunk!(finalChunk);\n break;\n }\n }\n content = accumulated;\n debugProfileStats(\n `streaming model, ${modelName}, mode, ${vlMode || 'default'}, cost-ms, ${timeCost}, temperature, ${temperature ?? ''}`,\n );\n } else {\n // Non-streaming with retry logic\n const retryCount = modelConfig.retryCount ?? 1;\n const retryInterval = modelConfig.retryInterval ?? 2000;\n const maxAttempts = retryCount + 1; // retryCount=1 means 2 total attempts (1 initial + 1 retry)\n\n let lastError: Error | undefined;\n\n for (let attempt = 1; attempt <= maxAttempts; attempt++) {\n try {\n const result = await completion.create({\n model: modelName,\n messages,\n ...commonConfig,\n ...deepThinkConfig,\n } as any);\n\n timeCost = Date.now() - startTime;\n\n debugProfileStats(\n `model, ${modelName}, mode, ${vlMode || 'default'}, ui-tars-version, ${uiTarsVersion}, prompt-tokens, ${result.usage?.prompt_tokens || ''}, completion-tokens, ${result.usage?.completion_tokens || ''}, total-tokens, ${result.usage?.total_tokens || ''}, cost-ms, ${timeCost}, requestId, ${result._request_id || ''}, temperature, ${temperature ?? ''}`,\n );\n\n debugProfileDetail(\n `model usage detail: ${JSON.stringify(result.usage)}`,\n );\n\n if (!result.choices) {\n throw new Error(\n `invalid response from LLM service: ${JSON.stringify(result)}`,\n );\n }\n\n content = result.choices[0].message.content!;\n if (!content) {\n throw new Error('empty content from AI model');\n }\n\n accumulatedReasoning =\n (result.choices[0].message as any)?.reasoning_content || '';\n usage = result.usage;\n break; // Success, exit retry loop\n } catch (error) {\n lastError = error as Error;\n if (attempt < maxAttempts) {\n console.warn(\n `[Midscene] AI call failed (attempt ${attempt}/${maxAttempts}), retrying in ${retryInterval}ms... Error: ${lastError.message}`,\n );\n await new Promise((resolve) => setTimeout(resolve, retryInterval));\n }\n }\n }\n\n if (!content) {\n throw lastError;\n }\n }\n\n debugCall(`response reasoning content: ${accumulatedReasoning}`);\n debugCall(`response content: ${content}`);\n\n // Ensure we always have usage info for streaming responses\n if (isStreaming && !usage) {\n // Estimate token counts based on content length (rough approximation)\n const estimatedTokens = Math.max(\n 1,\n Math.floor((content || '').length / 4),\n );\n usage = {\n prompt_tokens: estimatedTokens,\n completion_tokens: estimatedTokens,\n total_tokens: estimatedTokens * 2,\n } as OpenAI.CompletionUsage;\n }\n\n return {\n content: content || '',\n reasoning_content: accumulatedReasoning || undefined,\n usage: buildUsageInfo(usage),\n isStreamed: !!isStreaming,\n };\n } catch (e: any) {\n console.error(' call AI error', e);\n const newError = new Error(\n `failed to call ${isStreaming ? 'streaming ' : ''}AI model service (${modelName}): ${e.message}\\nTrouble shooting: https://midscenejs.com/model-provider.html`,\n {\n cause: e,\n },\n );\n throw newError;\n }\n}\n\nexport async function callAIWithObjectResponse<T>(\n messages: ChatCompletionMessageParam[],\n modelConfig: IModelConfig,\n options?: {\n deepThink?: DeepThinkOption;\n },\n): Promise<{\n content: T;\n contentString: string;\n usage?: AIUsageInfo;\n reasoning_content?: string;\n}> {\n const response = await callAI(messages, modelConfig, {\n deepThink: options?.deepThink,\n });\n assert(response, 'empty response');\n const vlMode = modelConfig.vlMode;\n const jsonContent = safeParseJson(response.content, vlMode);\n assert(\n typeof jsonContent === 'object',\n `failed to parse json response from model (${modelConfig.modelName}): ${response.content}`,\n );\n return {\n content: jsonContent,\n contentString: response.content,\n usage: response.usage,\n reasoning_content: response.reasoning_content,\n };\n}\n\nexport async function callAIWithStringResponse(\n msgs: AIArgs,\n modelConfig: IModelConfig,\n): Promise<{ content: string; usage?: AIUsageInfo }> {\n const { content, usage } = await callAI(msgs, modelConfig);\n return { content, usage };\n}\n\nexport function extractJSONFromCodeBlock(response: string) {\n try {\n // First, try to match a JSON object directly in the response\n const jsonMatch = response.match(/^\\s*(\\{[\\s\\S]*\\})\\s*$/);\n if (jsonMatch) {\n return jsonMatch[1];\n }\n\n // If no direct JSON object is found, try to extract JSON from a code block\n const codeBlockMatch = response.match(\n /```(?:json)?\\s*(\\{[\\s\\S]*?\\})\\s*```/,\n );\n if (codeBlockMatch) {\n return codeBlockMatch[1];\n }\n\n // If no code block is found, try to find a JSON-like structure in the text\n const jsonLikeMatch = response.match(/\\{[\\s\\S]*\\}/);\n if (jsonLikeMatch) {\n return jsonLikeMatch[0];\n }\n } catch {}\n // If no JSON-like structure is found, return the original response\n return response;\n}\n\nexport function preprocessDoubaoBboxJson(input: string) {\n if (input.includes('bbox')) {\n // when its values like 940 445 969 490, replace all /\\d+\\s+\\d+/g with /$1,$2/g\n while (/\\d+\\s+\\d+/.test(input)) {\n input = input.replace(/(\\d+)\\s+(\\d+)/g, '$1,$2');\n }\n }\n return input;\n}\n\nexport function resolveDeepThinkConfig({\n deepThink,\n modelFamily,\n}: {\n deepThink?: DeepThinkOption;\n modelFamily?: TModelFamily;\n}): {\n config: Record<string, unknown>;\n debugMessage?: string;\n warningMessage?: string;\n} {\n const normalizedDeepThink = deepThink === 'unset' ? undefined : deepThink;\n\n if (normalizedDeepThink === undefined) {\n return { config: {}, debugMessage: undefined };\n }\n\n if (modelFamily === 'qwen3-vl') {\n return {\n config: { enable_thinking: normalizedDeepThink },\n debugMessage: `deepThink mapped to enable_thinking=${normalizedDeepThink} for qwen3-vl`,\n };\n }\n\n if (modelFamily === 'doubao-vision') {\n return {\n config: {\n thinking: { type: normalizedDeepThink ? 'enabled' : 'disabled' },\n },\n debugMessage: `deepThink mapped to thinking.type=${normalizedDeepThink ? 'enabled' : 'disabled'} for doubao-vision`,\n };\n }\n\n if (modelFamily === 'glm-v') {\n return {\n config: {\n thinking: { type: normalizedDeepThink ? 'enabled' : 'disabled' },\n },\n debugMessage: `deepThink mapped to thinking.type=${normalizedDeepThink ? 'enabled' : 'disabled'} for glm-v`,\n };\n }\n\n return {\n config: {},\n debugMessage: `deepThink ignored: unsupported model_family \"${modelFamily ?? 'default'}\"`,\n warningMessage: `The \"deepThink\" option is not supported for model_family \"${modelFamily ?? 'default'}\".`,\n };\n}\n\n/**\n * Normalize a parsed JSON object by trimming whitespace from:\n * 1. All object keys (e.g., \" prompt \" -> \"prompt\")\n * 2. All string values (e.g., \" Tap \" -> \"Tap\")\n * This handles LLM output that may include leading/trailing spaces.\n */\nfunction normalizeJsonObject(obj: any): any {\n // Handle null and undefined\n if (obj === null || obj === undefined) {\n return obj;\n }\n\n // Handle arrays - recursively normalize each element\n if (Array.isArray(obj)) {\n return obj.map((item) => normalizeJsonObject(item));\n }\n\n // Handle objects\n if (typeof obj === 'object') {\n const normalized: any = {};\n\n for (const [key, value] of Object.entries(obj)) {\n // Trim the key to remove leading/trailing spaces\n const trimmedKey = key.trim();\n\n // Recursively normalize the value\n let normalizedValue = normalizeJsonObject(value);\n\n // Trim all string values\n if (typeof normalizedValue === 'string') {\n normalizedValue = normalizedValue.trim();\n }\n\n normalized[trimmedKey] = normalizedValue;\n }\n\n return normalized;\n }\n\n // Handle primitive strings\n if (typeof obj === 'string') {\n return obj.trim();\n }\n\n // Return other primitives as-is\n return obj;\n}\n\nexport function safeParseJson(input: string, vlMode: TVlModeTypes | undefined) {\n const cleanJsonString = extractJSONFromCodeBlock(input);\n // match the point\n if (cleanJsonString?.match(/\\((\\d+),(\\d+)\\)/)) {\n return cleanJsonString\n .match(/\\((\\d+),(\\d+)\\)/)\n ?.slice(1)\n .map(Number);\n }\n\n let parsed: any;\n let lastError: unknown;\n try {\n parsed = JSON.parse(cleanJsonString);\n return normalizeJsonObject(parsed);\n } catch (error) {\n lastError = error;\n }\n try {\n parsed = JSON.parse(jsonrepair(cleanJsonString));\n return normalizeJsonObject(parsed);\n } catch (error) {\n lastError = error;\n }\n\n if (vlMode === 'doubao-vision' || vlMode === 'vlm-ui-tars') {\n const jsonString = preprocessDoubaoBboxJson(cleanJsonString);\n try {\n parsed = JSON.parse(jsonrepair(jsonString));\n return normalizeJsonObject(parsed);\n } catch (error) {\n lastError = error;\n }\n }\n throw Error(\n `failed to parse LLM response into JSON. Error - ${String(\n lastError ?? 'unknown error',\n )}. Response - \\n ${input}`,\n );\n}\n"],"names":["createChatClient","modelConfig","socksProxy","httpProxy","modelName","openaiBaseURL","openaiApiKey","openaiExtraConfig","modelDescription","uiTarsVersion","vlMode","modelFamily","createOpenAIClient","timeout","proxyAgent","debugProxy","getDebug","sanitizeProxyUrl","url","parsed","URL","ifInBrowser","console","moduleName","ProxyAgent","socksDispatcher","proxyUrl","Error","port","Number","protocol","socksType","decodeURIComponent","error","openAIOptions","baseOpenAI","OpenAI","openai","globalConfigManager","MIDSCENE_LANGSMITH_DEBUG","langsmithModule","wrapOpenAI","MIDSCENE_LANGFUSE_DEBUG","langfuseModule","observeOpenAI","wrappedClient","callAI","messages","options","completion","maxTokens","MIDSCENE_MODEL_MAX_TOKENS","OPENAI_MAX_TOKENS","debugCall","debugProfileStats","debugProfileDetail","startTime","Date","temperature","isStreaming","content","accumulated","accumulatedReasoning","usage","timeCost","buildUsageInfo","usageData","cachedInputTokens","commonConfig","isAutoGLM","deepThinkConfig","debugMessage","warningMessage","resolveDeepThinkConfig","stream","chunk","reasoning_content","chunkData","undefined","estimatedTokens","Math","finalChunk","retryCount","retryInterval","maxAttempts","lastError","attempt","result","JSON","Promise","resolve","setTimeout","e","newError","callAIWithObjectResponse","response","assert","jsonContent","safeParseJson","callAIWithStringResponse","msgs","extractJSONFromCodeBlock","jsonMatch","codeBlockMatch","jsonLikeMatch","preprocessDoubaoBboxJson","input","deepThink","normalizedDeepThink","normalizeJsonObject","obj","Array","item","normalized","key","value","Object","trimmedKey","normalizedValue","cleanJsonString","jsonrepair","jsonString","String"],"mappings":";;;;;;AA2BA,eAAeA,iBAAiB,EAC9BC,WAAW,EAGZ;IAQC,MAAM,EACJC,UAAU,EACVC,SAAS,EACTC,SAAS,EACTC,aAAa,EACbC,YAAY,EACZC,iBAAiB,EACjBC,gBAAgB,EAChB,oBAAoBC,aAAa,EACjCC,MAAM,EACNC,WAAW,EACXC,kBAAkB,EAClBC,OAAO,EACR,GAAGZ;IAEJ,IAAIa;IACJ,MAAMC,aAAaC,SAAS;IAI5B,MAAMC,mBAAmB,CAACC;QACxB,IAAI;YACF,MAAMC,SAAS,IAAIC,IAAIF;YACvB,IAAIC,OAAO,QAAQ,EAAE;gBAEnBA,OAAO,QAAQ,GAAG;gBAClB,OAAOA,OAAO,IAAI;YACpB;YACA,OAAOD;QACT,EAAE,OAAM;YAEN,OAAOA;QACT;IACF;IAEA,IAAIf,WAAW;QACbY,WAAW,oBAAoBE,iBAAiBd;QAChD,IAAIkB,aACFC,QAAQ,IAAI,CACV;aAEG;YAEL,MAAMC,aAAa;YACnB,MAAM,EAAEC,UAAU,EAAE,GAAG,MAAM,MAAM,CAACD;YACpCT,aAAa,IAAIU,WAAW;gBAC1B,KAAKrB;YAEP;QACF;IACF,OAAO,IAAID,YAAY;QACrBa,WAAW,qBAAqBE,iBAAiBf;QACjD,IAAImB,aACFC,QAAQ,IAAI,CACV;aAGF,IAAI;YAEF,MAAMC,aAAa;YACnB,MAAM,EAAEE,eAAe,EAAE,GAAG,MAAM,MAAM,CAACF;YAEzC,MAAMG,WAAW,IAAIN,IAAIlB;YAGzB,IAAI,CAACwB,SAAS,QAAQ,EACpB,MAAM,IAAIC,MAAM;YAIlB,MAAMC,OAAOC,OAAO,QAAQ,CAACH,SAAS,IAAI,EAAE;YAC5C,IAAI,CAACA,SAAS,IAAI,IAAIG,OAAO,KAAK,CAACD,OACjC,MAAM,IAAID,MAAM;YAIlB,MAAMG,WAAWJ,SAAS,QAAQ,CAAC,OAAO,CAAC,KAAK;YAChD,MAAMK,YACJD,AAAa,aAAbA,WAAwB,IAAIA,AAAa,aAAbA,WAAwB,IAAI;YAE1DhB,aAAaW,gBAAgB;gBAC3B,MAAMM;gBACN,MAAML,SAAS,QAAQ;gBACvBE;gBACA,GAAIF,SAAS,QAAQ,GACjB;oBACE,QAAQM,mBAAmBN,SAAS,QAAQ;oBAC5C,UAAUM,mBAAmBN,SAAS,QAAQ,IAAI;gBACpD,IACA,CAAC,CAAC;YACR;YACAX,WAAW,uCAAuC;gBAChD,MAAMgB;gBACN,MAAML,SAAS,QAAQ;gBACvB,MAAME;YACR;QACF,EAAE,OAAOK,OAAO;YACdX,QAAQ,KAAK,CAAC,oCAAoCW;YAClD,MAAM,IAAIN,MACR,CAAC,yBAAyB,EAAEzB,WAAW,+GAA+G,CAAC;QAE3J;IAEJ;IAEA,MAAMgC,gBAAgB;QACpB,SAAS7B;QACT,QAAQC;QAGR,GAAIQ,aAAa;YAAE,cAAc;gBAAE,YAAYA;YAAkB;QAAE,IAAI,CAAC,CAAC;QACzE,GAAGP,iBAAiB;QACpB,GAAI,AAAmB,YAAnB,OAAOM,UAAuB;YAAEA;QAAQ,IAAI,CAAC,CAAC;QAClD,yBAAyB;IAC3B;IAEA,MAAMsB,aAAa,IAAIC,SAAOF;IAE9B,IAAIG,SAAiBF;IAGrB,IACEE,UACAC,oBAAoB,qBAAqB,CAACC,2BAC1C;QACA,IAAIlB,aACF,MAAM,IAAIM,MAAM;QAElBL,QAAQ,GAAG,CAAC;QAEZ,MAAMkB,kBAAkB;QACxB,MAAM,EAAEC,UAAU,EAAE,GAAG,MAAM,MAAM,CAACD;QACpCH,SAASI,WAAWJ;IACtB;IAGA,IACEA,UACAC,oBAAoB,qBAAqB,CAACI,0BAC1C;QACA,IAAIrB,aACF,MAAM,IAAIM,MAAM;QAElBL,QAAQ,GAAG,CAAC;QAEZ,MAAMqB,iBAAiB;QACvB,MAAM,EAAEC,aAAa,EAAE,GAAG,MAAM,MAAM,CAACD;QACvCN,SAASO,cAAcP;IACzB;IAEA,IAAIzB,oBAAoB;QACtB,MAAMiC,gBAAgB,MAAMjC,mBAAmBuB,YAAYD;QAE3D,IAAIW,eACFR,SAASQ;IAEb;IAEA,OAAO;QACL,YAAYR,OAAO,IAAI,CAAC,WAAW;QACnCjC;QACAI;QACAC;QACAC;QACAC;IACF;AACF;AAEO,eAAemC,OACpBC,QAAsC,EACtC9C,WAAyB,EACzB+C,OAIC;IAOD,MAAM,EACJC,UAAU,EACV7C,SAAS,EACTI,gBAAgB,EAChBC,aAAa,EACbC,MAAM,EACNC,WAAW,EACZ,GAAG,MAAMX,iBAAiB;QACzBC;IACF;IAEA,MAAMiD,YACJZ,oBAAoB,yBAAyB,CAACa,8BAC9Cb,oBAAoB,yBAAyB,CAACc;IAChD,MAAMC,YAAYrC,SAAS;IAC3B,MAAMsC,oBAAoBtC,SAAS;IACnC,MAAMuC,qBAAqBvC,SAAS;IAEpC,MAAMwC,YAAYC,KAAK,GAAG;IAC1B,MAAMC,cAAczD,YAAY,WAAW,IAAI;IAE/C,MAAM0D,cAAcX,SAAS,UAAUA,SAAS;IAChD,IAAIY;IACJ,IAAIC,cAAc;IAClB,IAAIC,uBAAuB;IAC3B,IAAIC;IACJ,IAAIC;IAEJ,MAAMC,iBAAiB,CAACC;QACtB,IAAI,CAACA,WAAW;QAEhB,MAAMC,oBACJD,WACC,uBAAuB;QAE1B,OAAO;YACL,eAAeA,UAAU,aAAa,IAAI;YAC1C,mBAAmBA,UAAU,iBAAiB,IAAI;YAClD,cAAcA,UAAU,YAAY,IAAI;YACxC,cAAcC,qBAAqB;YACnC,WAAWH,YAAY;YACvB,YAAY5D;YACZ,mBAAmBI;YACnB,QAAQP,YAAY,MAAM;QAC5B;IACF;IAEA,MAAMmE,eAAe;QACnBV;QACA,QAAQ,CAAC,CAACC;QACV,YAAYT;QACZ,GAAIxC,AAAW,iBAAXA,SACA;YACE,2BAA2B;QAC7B,IACA,CAAC,CAAC;IACR;IAEA,IAAI2D,UAAU3D,SAAS;QACpB0D,aAAmD,KAAK,GAAG;QAC3DA,aAAmD,iBAAiB,GAAG;IAC1E;IAEA,MAAM,EACJ,QAAQE,eAAe,EACvBC,YAAY,EACZC,cAAc,EACf,GAAGC,uBAAuB;QACzB,WAAWzB,SAAS;QACpBrC;IACF;IACA,IAAI4D,cACFlB,UAAUkB;IAEZ,IAAIC,gBAAgB;QAClBnB,UAAUmB;QACVlD,QAAQ,IAAI,CAACkD;IACf;IAEA,IAAI;QACFnB,UACE,CAAC,QAAQ,EAAEM,cAAc,eAAe,GAAG,WAAW,EAAEvD,WAAW;QAGrE,IAAIuD,aAAa;YACf,MAAMe,SAAU,MAAMzB,WAAW,MAAM,CACrC;gBACE,OAAO7C;gBACP2C;gBACA,GAAGqB,YAAY;gBACf,GAAGE,eAAe;YACpB,GACA;gBACE,QAAQ;YACV;YAKF,WAAW,MAAMK,SAASD,OAAQ;gBAChC,MAAMd,UAAUe,MAAM,OAAO,EAAE,CAAC,EAAE,EAAE,OAAO,WAAW;gBACtD,MAAMC,oBACHD,MAAM,OAAO,EAAE,CAAC,EAAE,EAAE,OAAe,qBAAqB;gBAG3D,IAAIA,MAAM,KAAK,EACbZ,QAAQY,MAAM,KAAK;gBAGrB,IAAIf,WAAWgB,mBAAmB;oBAChCf,eAAeD;oBACfE,wBAAwBc;oBACxB,MAAMC,YAAiC;wBACrCjB;wBACAgB;wBACAf;wBACA,YAAY;wBACZ,OAAOiB;oBACT;oBACA9B,QAAQ,OAAO,CAAE6B;gBACnB;gBAGA,IAAIF,MAAM,OAAO,EAAE,CAAC,EAAE,EAAE,eAAe;oBACrCX,WAAWP,KAAK,GAAG,KAAKD;oBAGxB,IAAI,CAACO,OAAO;wBAEV,MAAMgB,kBAAkBC,KAAK,GAAG,CAC9B,GACAA,KAAK,KAAK,CAACnB,YAAY,MAAM,GAAG;wBAElCE,QAAQ;4BACN,eAAegB;4BACf,mBAAmBA;4BACnB,cAAcA,AAAkB,IAAlBA;wBAChB;oBACF;oBAGA,MAAME,aAAkC;wBACtC,SAAS;wBACTpB;wBACA,mBAAmB;wBACnB,YAAY;wBACZ,OAAOI,eAAeF;oBACxB;oBACAf,QAAQ,OAAO,CAAEiC;oBACjB;gBACF;YACF;YACArB,UAAUC;YACVP,kBACE,CAAC,iBAAiB,EAAElD,UAAU,QAAQ,EAAEM,UAAU,UAAU,WAAW,EAAEsD,SAAS,eAAe,EAAEN,eAAe,IAAI;QAE1H,OAAO;YAEL,MAAMwB,aAAajF,YAAY,UAAU,IAAI;YAC7C,MAAMkF,gBAAgBlF,YAAY,aAAa,IAAI;YACnD,MAAMmF,cAAcF,aAAa;YAEjC,IAAIG;YAEJ,IAAK,IAAIC,UAAU,GAAGA,WAAWF,aAAaE,UAC5C,IAAI;gBACF,MAAMC,SAAS,MAAMtC,WAAW,MAAM,CAAC;oBACrC,OAAO7C;oBACP2C;oBACA,GAAGqB,YAAY;oBACf,GAAGE,eAAe;gBACpB;gBAEAN,WAAWP,KAAK,GAAG,KAAKD;gBAExBF,kBACE,CAAC,OAAO,EAAElD,UAAU,QAAQ,EAAEM,UAAU,UAAU,mBAAmB,EAAED,cAAc,iBAAiB,EAAE8E,OAAO,KAAK,EAAE,iBAAiB,GAAG,qBAAqB,EAAEA,OAAO,KAAK,EAAE,qBAAqB,GAAG,gBAAgB,EAAEA,OAAO,KAAK,EAAE,gBAAgB,GAAG,WAAW,EAAEvB,SAAS,aAAa,EAAEuB,OAAO,WAAW,IAAI,GAAG,eAAe,EAAE7B,eAAe,IAAI;gBAG9VH,mBACE,CAAC,oBAAoB,EAAEiC,KAAK,SAAS,CAACD,OAAO,KAAK,GAAG;gBAGvD,IAAI,CAACA,OAAO,OAAO,EACjB,MAAM,IAAI5D,MACR,CAAC,mCAAmC,EAAE6D,KAAK,SAAS,CAACD,SAAS;gBAIlE3B,UAAU2B,OAAO,OAAO,CAAC,EAAE,CAAC,OAAO,CAAC,OAAO;gBAC3C,IAAI,CAAC3B,SACH,MAAM,IAAIjC,MAAM;gBAGlBmC,uBACGyB,OAAO,OAAO,CAAC,EAAE,CAAC,OAAO,EAAU,qBAAqB;gBAC3DxB,QAAQwB,OAAO,KAAK;gBACpB;YACF,EAAE,OAAOtD,OAAO;gBACdoD,YAAYpD;gBACZ,IAAIqD,UAAUF,aAAa;oBACzB9D,QAAQ,IAAI,CACV,CAAC,mCAAmC,EAAEgE,QAAQ,CAAC,EAAEF,YAAY,eAAe,EAAED,cAAc,aAAa,EAAEE,UAAU,OAAO,EAAE;oBAEhI,MAAM,IAAII,QAAQ,CAACC,UAAYC,WAAWD,SAASP;gBACrD;YACF;YAGF,IAAI,CAACvB,SACH,MAAMyB;QAEV;QAEAhC,UAAU,CAAC,4BAA4B,EAAES,sBAAsB;QAC/DT,UAAU,CAAC,kBAAkB,EAAEO,SAAS;QAGxC,IAAID,eAAe,CAACI,OAAO;YAEzB,MAAMgB,kBAAkBC,KAAK,GAAG,CAC9B,GACAA,KAAK,KAAK,CAAEpB,AAAAA,CAAAA,WAAW,EAAC,EAAG,MAAM,GAAG;YAEtCG,QAAQ;gBACN,eAAegB;gBACf,mBAAmBA;gBACnB,cAAcA,AAAkB,IAAlBA;YAChB;QACF;QAEA,OAAO;YACL,SAASnB,WAAW;YACpB,mBAAmBE,wBAAwBgB;YAC3C,OAAOb,eAAeF;YACtB,YAAY,CAAC,CAACJ;QAChB;IACF,EAAE,OAAOiC,GAAQ;QACftE,QAAQ,KAAK,CAAC,kBAAkBsE;QAChC,MAAMC,WAAW,IAAIlE,MACnB,CAAC,eAAe,EAAEgC,cAAc,eAAe,GAAG,kBAAkB,EAAEvD,UAAU,GAAG,EAAEwF,EAAE,OAAO,CAAC,8DAA8D,CAAC,EAC9J;YACE,OAAOA;QACT;QAEF,MAAMC;IACR;AACF;AAEO,eAAeC,yBACpB/C,QAAsC,EACtC9C,WAAyB,EACzB+C,OAEC;IAOD,MAAM+C,WAAW,MAAMjD,OAAOC,UAAU9C,aAAa;QACnD,WAAW+C,SAAS;IACtB;IACAgD,OAAOD,UAAU;IACjB,MAAMrF,SAAST,YAAY,MAAM;IACjC,MAAMgG,cAAcC,cAAcH,SAAS,OAAO,EAAErF;IACpDsF,OACE,AAAuB,YAAvB,OAAOC,aACP,CAAC,0CAA0C,EAAEhG,YAAY,SAAS,CAAC,GAAG,EAAE8F,SAAS,OAAO,EAAE;IAE5F,OAAO;QACL,SAASE;QACT,eAAeF,SAAS,OAAO;QAC/B,OAAOA,SAAS,KAAK;QACrB,mBAAmBA,SAAS,iBAAiB;IAC/C;AACF;AAEO,eAAeI,yBACpBC,IAAY,EACZnG,WAAyB;IAEzB,MAAM,EAAE2D,OAAO,EAAEG,KAAK,EAAE,GAAG,MAAMjB,OAAOsD,MAAMnG;IAC9C,OAAO;QAAE2D;QAASG;IAAM;AAC1B;AAEO,SAASsC,yBAAyBN,QAAgB;IACvD,IAAI;QAEF,MAAMO,YAAYP,SAAS,KAAK,CAAC;QACjC,IAAIO,WACF,OAAOA,SAAS,CAAC,EAAE;QAIrB,MAAMC,iBAAiBR,SAAS,KAAK,CACnC;QAEF,IAAIQ,gBACF,OAAOA,cAAc,CAAC,EAAE;QAI1B,MAAMC,gBAAgBT,SAAS,KAAK,CAAC;QACrC,IAAIS,eACF,OAAOA,aAAa,CAAC,EAAE;IAE3B,EAAE,OAAM,CAAC;IAET,OAAOT;AACT;AAEO,SAASU,yBAAyBC,KAAa;IACpD,IAAIA,MAAM,QAAQ,CAAC,SAEjB,MAAO,YAAY,IAAI,CAACA,OACtBA,QAAQA,MAAM,OAAO,CAAC,kBAAkB;IAG5C,OAAOA;AACT;AAEO,SAASjC,uBAAuB,EACrCkC,SAAS,EACThG,WAAW,EAIZ;IAKC,MAAMiG,sBAAsBD,AAAc,YAAdA,YAAwB7B,SAAY6B;IAEhE,IAAIC,AAAwB9B,WAAxB8B,qBACF,OAAO;QAAE,QAAQ,CAAC;QAAG,cAAc9B;IAAU;IAG/C,IAAInE,AAAgB,eAAhBA,aACF,OAAO;QACL,QAAQ;YAAE,iBAAiBiG;QAAoB;QAC/C,cAAc,CAAC,oCAAoC,EAAEA,oBAAoB,aAAa,CAAC;IACzF;IAGF,IAAIjG,AAAgB,oBAAhBA,aACF,OAAO;QACL,QAAQ;YACN,UAAU;gBAAE,MAAMiG,sBAAsB,YAAY;YAAW;QACjE;QACA,cAAc,CAAC,kCAAkC,EAAEA,sBAAsB,YAAY,WAAW,kBAAkB,CAAC;IACrH;IAGF,IAAIjG,AAAgB,YAAhBA,aACF,OAAO;QACL,QAAQ;YACN,UAAU;gBAAE,MAAMiG,sBAAsB,YAAY;YAAW;QACjE;QACA,cAAc,CAAC,kCAAkC,EAAEA,sBAAsB,YAAY,WAAW,UAAU,CAAC;IAC7G;IAGF,OAAO;QACL,QAAQ,CAAC;QACT,cAAc,CAAC,6CAA6C,EAAEjG,eAAe,UAAU,CAAC,CAAC;QACzF,gBAAgB,CAAC,0DAA0D,EAAEA,eAAe,UAAU,EAAE,CAAC;IAC3G;AACF;AAQA,SAASkG,oBAAoBC,GAAQ;IAEnC,IAAIA,QAAAA,KACF,OAAOA;IAIT,IAAIC,MAAM,OAAO,CAACD,MAChB,OAAOA,IAAI,GAAG,CAAC,CAACE,OAASH,oBAAoBG;IAI/C,IAAI,AAAe,YAAf,OAAOF,KAAkB;QAC3B,MAAMG,aAAkB,CAAC;QAEzB,KAAK,MAAM,CAACC,KAAKC,MAAM,IAAIC,OAAO,OAAO,CAACN,KAAM;YAE9C,MAAMO,aAAaH,IAAI,IAAI;YAG3B,IAAII,kBAAkBT,oBAAoBM;YAG1C,IAAI,AAA2B,YAA3B,OAAOG,iBACTA,kBAAkBA,gBAAgB,IAAI;YAGxCL,UAAU,CAACI,WAAW,GAAGC;QAC3B;QAEA,OAAOL;IACT;IAGA,IAAI,AAAe,YAAf,OAAOH,KACT,OAAOA,IAAI,IAAI;IAIjB,OAAOA;AACT;AAEO,SAASZ,cAAcQ,KAAa,EAAEhG,MAAgC;IAC3E,MAAM6G,kBAAkBlB,yBAAyBK;IAEjD,IAAIa,iBAAiB,MAAM,oBACzB,OAAOA,gBACJ,KAAK,CAAC,oBACL,MAAM,GACP,IAAI1F;IAGT,IAAIV;IACJ,IAAIkE;IACJ,IAAI;QACFlE,SAASqE,KAAK,KAAK,CAAC+B;QACpB,OAAOV,oBAAoB1F;IAC7B,EAAE,OAAOc,OAAO;QACdoD,YAAYpD;IACd;IACA,IAAI;QACFd,SAASqE,KAAK,KAAK,CAACgC,WAAWD;QAC/B,OAAOV,oBAAoB1F;IAC7B,EAAE,OAAOc,OAAO;QACdoD,YAAYpD;IACd;IAEA,IAAIvB,AAAW,oBAAXA,UAA8BA,AAAW,kBAAXA,QAA0B;QAC1D,MAAM+G,aAAahB,yBAAyBc;QAC5C,IAAI;YACFpG,SAASqE,KAAK,KAAK,CAACgC,WAAWC;YAC/B,OAAOZ,oBAAoB1F;QAC7B,EAAE,OAAOc,OAAO;YACdoD,YAAYpD;QACd;IACF;IACA,MAAMN,MACJ,CAAC,gDAAgD,EAAE+F,OACjDrC,aAAa,iBACb,gBAAgB,EAAEqB,OAAO;AAE/B"}
|
|
@@ -20,7 +20,7 @@ async function uiTarsPlanning(userInstruction, options) {
|
|
|
20
20
|
let instruction = userInstruction;
|
|
21
21
|
if (actionContext) instruction = `<high_priority_knowledge>${actionContext}</high_priority_knowledge>\n<user_instruction>${userInstruction}</user_instruction>`;
|
|
22
22
|
const systemPrompt = getUiTarsPlanningPrompt() + instruction;
|
|
23
|
-
const screenshotBase64 = context.screenshot.
|
|
23
|
+
const screenshotBase64 = context.screenshot.base64;
|
|
24
24
|
const imagePayload = await resizeImageForUiTars(screenshotBase64, context.size, uiTarsModelVersion);
|
|
25
25
|
conversationHistory.append({
|
|
26
26
|
role: 'user',
|
|
@@ -63,47 +63,50 @@ async function uiTarsPlanning(userInstruction, options) {
|
|
|
63
63
|
if ('click' === actionType) {
|
|
64
64
|
assert(action.action_inputs.start_box, 'start_box is required');
|
|
65
65
|
const point = getPoint(action.action_inputs.start_box, size);
|
|
66
|
+
const locate = {
|
|
67
|
+
prompt: action.thought || '',
|
|
68
|
+
bbox: pointToBbox({
|
|
69
|
+
x: point[0],
|
|
70
|
+
y: point[1]
|
|
71
|
+
}, size.width, size.height)
|
|
72
|
+
};
|
|
66
73
|
transformActions.push({
|
|
67
74
|
type: 'Tap',
|
|
68
75
|
param: {
|
|
69
|
-
locate:
|
|
70
|
-
prompt: action.thought || '',
|
|
71
|
-
bbox: pointToBbox({
|
|
72
|
-
x: point[0],
|
|
73
|
-
y: point[1]
|
|
74
|
-
}, size.width, size.height)
|
|
75
|
-
}
|
|
76
|
+
locate: locate
|
|
76
77
|
}
|
|
77
78
|
});
|
|
78
79
|
} else if ('left_double' === actionType) {
|
|
79
80
|
assert(action.action_inputs.start_box, 'start_box is required');
|
|
80
81
|
const point = getPoint(action.action_inputs.start_box, size);
|
|
82
|
+
const locate = {
|
|
83
|
+
prompt: action.thought || '',
|
|
84
|
+
bbox: pointToBbox({
|
|
85
|
+
x: point[0],
|
|
86
|
+
y: point[1]
|
|
87
|
+
}, size.width, size.height)
|
|
88
|
+
};
|
|
81
89
|
transformActions.push({
|
|
82
90
|
type: 'DoubleClick',
|
|
83
91
|
param: {
|
|
84
|
-
locate:
|
|
85
|
-
prompt: action.thought || '',
|
|
86
|
-
bbox: pointToBbox({
|
|
87
|
-
x: point[0],
|
|
88
|
-
y: point[1]
|
|
89
|
-
}, size.width, size.height)
|
|
90
|
-
}
|
|
92
|
+
locate: locate
|
|
91
93
|
},
|
|
92
94
|
thought: action.thought || ''
|
|
93
95
|
});
|
|
94
96
|
} else if ('right_single' === actionType) {
|
|
95
97
|
assert(action.action_inputs.start_box, 'start_box is required');
|
|
96
98
|
const point = getPoint(action.action_inputs.start_box, size);
|
|
99
|
+
const locate = {
|
|
100
|
+
prompt: action.thought || '',
|
|
101
|
+
bbox: pointToBbox({
|
|
102
|
+
x: point[0],
|
|
103
|
+
y: point[1]
|
|
104
|
+
}, size.width, size.height)
|
|
105
|
+
};
|
|
97
106
|
transformActions.push({
|
|
98
107
|
type: 'RightClick',
|
|
99
108
|
param: {
|
|
100
|
-
locate:
|
|
101
|
-
prompt: action.thought || '',
|
|
102
|
-
bbox: pointToBbox({
|
|
103
|
-
x: point[0],
|
|
104
|
-
y: point[1]
|
|
105
|
-
}, size.width, size.height)
|
|
106
|
-
}
|
|
109
|
+
locate: locate
|
|
107
110
|
},
|
|
108
111
|
thought: action.thought || ''
|
|
109
112
|
});
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"ai-model/ui-tars-planning.mjs","sources":["../../../src/ai-model/ui-tars-planning.ts"],"sourcesContent":["import type {\n PlanningAIResponse,\n PlanningAction,\n Size,\n UIContext,\n} from '@/types';\nimport { type IModelConfig, UITarsModelVersion } from '@midscene/shared/env';\nimport { resizeImgBase64 } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { transformHotkeyInput } from '@midscene/shared/us-keyboard-layout';\nimport { assert } from '@midscene/shared/utils';\nimport { actionParser } from '@ui-tars/action-parser';\nimport type { ConversationHistory } from './conversation-history';\nimport { getSummary, getUiTarsPlanningPrompt } from './prompt/ui-tars-planning';\nimport { callAIWithStringResponse } from './service-caller/index';\ntype ActionType =\n | 'click'\n | 'left_double'\n | 'right_single'\n | 'drag'\n | 'type'\n | 'hotkey'\n | 'finished'\n | 'scroll'\n | 'wait';\n\nconst debug = getDebug('ui-tars-planning');\nconst bboxSize = 10;\nconst pointToBbox = (\n point: { x: number; y: number },\n width: number,\n height: number,\n): [number, number, number, number] => {\n return [\n Math.round(Math.max(point.x - bboxSize / 2, 0)),\n Math.round(Math.max(point.y - bboxSize / 2, 0)),\n Math.round(Math.min(point.x + bboxSize / 2, width)),\n Math.round(Math.min(point.y + bboxSize / 2, height)),\n ];\n};\n\nexport async function uiTarsPlanning(\n userInstruction: string,\n options: {\n conversationHistory: ConversationHistory;\n context: UIContext;\n modelConfig: IModelConfig;\n actionContext?: string;\n },\n): Promise<PlanningAIResponse> {\n const { conversationHistory, context, modelConfig, actionContext } = options;\n const { uiTarsModelVersion } = modelConfig;\n\n let instruction = userInstruction;\n if (actionContext) {\n instruction = `<high_priority_knowledge>${actionContext}</high_priority_knowledge>\\n<user_instruction>${userInstruction}</user_instruction>`;\n }\n\n const systemPrompt = getUiTarsPlanningPrompt() + instruction;\n\n const screenshotBase64 = context.screenshot.getData();\n const imagePayload = await resizeImageForUiTars(\n screenshotBase64,\n context.size,\n uiTarsModelVersion,\n );\n\n conversationHistory.append({\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: {\n url: imagePayload,\n },\n },\n ],\n });\n\n const res = await callAIWithStringResponse(\n [\n {\n role: 'user',\n content: systemPrompt,\n },\n ...conversationHistory.snapshot(),\n ],\n modelConfig,\n );\n const convertedText = convertBboxToCoordinates(res.content);\n\n const { size } = context;\n const { parsed } = actionParser({\n prediction: convertedText,\n factor: [1000, 1000],\n screenContext: {\n width: size.width,\n height: size.height,\n },\n modelVer: uiTarsModelVersion,\n });\n\n debug(\n 'ui-tars modelVer',\n uiTarsModelVersion,\n ', parsed',\n JSON.stringify(parsed),\n );\n\n const transformActions: PlanningAction[] = [];\n const unhandledActions: Array<{ type: string; thought: string }> = [];\n let shouldContinue = true;\n parsed.forEach((action) => {\n const actionType = (action.action_type || '').toLowerCase();\n if (actionType === 'click') {\n assert(action.action_inputs.start_box, 'start_box is required');\n const point = getPoint(action.action_inputs.start_box, size);\n transformActions.push({\n type: 'Tap',\n param: {\n locate: {\n prompt: action.thought || '',\n bbox: pointToBbox(\n { x: point[0], y: point[1] },\n size.width,\n size.height,\n ),\n },\n },\n });\n } else if (actionType === 'left_double') {\n assert(action.action_inputs.start_box, 'start_box is required');\n const point = getPoint(action.action_inputs.start_box, size);\n transformActions.push({\n type: 'DoubleClick',\n param: {\n locate: {\n prompt: action.thought || '',\n bbox: pointToBbox(\n { x: point[0], y: point[1] },\n size.width,\n size.height,\n ),\n },\n },\n thought: action.thought || '',\n });\n } else if (actionType === 'right_single') {\n assert(action.action_inputs.start_box, 'start_box is required');\n const point = getPoint(action.action_inputs.start_box, size);\n transformActions.push({\n type: 'RightClick',\n param: {\n locate: {\n prompt: action.thought || '',\n bbox: pointToBbox(\n { x: point[0], y: point[1] },\n size.width,\n size.height,\n ),\n },\n },\n thought: action.thought || '',\n });\n } else if (actionType === 'drag') {\n assert(action.action_inputs.start_box, 'start_box is required');\n assert(action.action_inputs.end_box, 'end_box is required');\n const startPoint = getPoint(action.action_inputs.start_box, size);\n const endPoint = getPoint(action.action_inputs.end_box, size);\n transformActions.push({\n type: 'DragAndDrop',\n param: {\n from: {\n prompt: action.thought || '',\n bbox: pointToBbox(\n { x: startPoint[0], y: startPoint[1] },\n size.width,\n size.height,\n ),\n },\n to: {\n prompt: action.thought || '',\n bbox: pointToBbox(\n { x: endPoint[0], y: endPoint[1] },\n size.width,\n size.height,\n ),\n },\n },\n thought: action.thought || '',\n });\n } else if (actionType === 'type') {\n transformActions.push({\n type: 'Input',\n param: {\n value: action.action_inputs.content,\n },\n thought: action.thought || '',\n });\n } else if (actionType === 'scroll') {\n transformActions.push({\n type: 'Scroll',\n param: {\n direction: action.action_inputs.direction,\n },\n thought: action.thought || '',\n });\n } else if (actionType === 'finished') {\n shouldContinue = false;\n transformActions.push({\n type: 'Finished',\n param: {},\n thought: action.thought || '',\n });\n } else if (actionType === 'hotkey') {\n if (!action.action_inputs.key) {\n console.warn(\n 'No key found in action: hotkey. Will not perform action.',\n );\n } else {\n const keys = transformHotkeyInput(action.action_inputs.key);\n\n transformActions.push({\n type: 'KeyboardPress',\n param: {\n keyName: keys.join('+'),\n },\n thought: action.thought || '',\n });\n }\n } else if (actionType === 'wait') {\n transformActions.push({\n type: 'Sleep',\n param: {\n timeMs: 1000,\n },\n thought: action.thought || '',\n });\n } else if (actionType) {\n // Track unhandled action types\n unhandledActions.push({\n type: actionType,\n thought: action.thought || '',\n });\n debug('Unhandled action type:', actionType, 'thought:', action.thought);\n }\n });\n\n if (transformActions.length === 0) {\n const errorDetails: string[] = [];\n\n // Check if parsing failed\n if (parsed.length === 0) {\n errorDetails.push('Action parser returned no actions');\n\n // Check if response has Thought but no Action\n if (\n res.content.includes('Thought:') &&\n !res.content.includes('Action:')\n ) {\n errorDetails.push(\n 'Response contains \"Thought:\" but missing \"Action:\" line',\n );\n } else {\n errorDetails.push('Response may be malformed or empty');\n }\n }\n\n // Check if we have unhandled action types\n if (unhandledActions.length > 0) {\n const types = unhandledActions.map((a) => a.type).join(', ');\n errorDetails.push(`Unhandled action types: ${types}`);\n }\n\n const errorMessage = [\n 'No actions found in UI-TARS response.',\n ...errorDetails,\n `\\nRaw response: ${res.content}`,\n ].join('\\n');\n\n throw new Error(errorMessage, {\n cause: {\n prediction: res.content,\n parsed,\n unhandledActions,\n convertedText,\n },\n });\n }\n\n debug('transformActions', JSON.stringify(transformActions, null, 2));\n const log = getSummary(res.content);\n\n conversationHistory.append({\n role: 'assistant',\n content: log,\n });\n\n return {\n actions: transformActions,\n log,\n usage: res.usage,\n rawResponse: JSON.stringify(res.content, undefined, 2),\n shouldContinuePlanning: shouldContinue,\n };\n}\n\n/**\n * Converts bounding box notation to coordinate points\n * @param text - The text containing bbox tags to be converted\n * @returns The text with bbox tags replaced by coordinate points\n */\nfunction convertBboxToCoordinates(text: string): string {\n // Match the four numbers after <bbox>\n const pattern = /<bbox>(\\d+)\\s+(\\d+)\\s+(\\d+)\\s+(\\d+)<\\/bbox>/g;\n\n function replaceMatch(\n match: string,\n x1: string,\n y1: string,\n x2: string,\n y2: string,\n ): string {\n // Convert strings to numbers and calculate center point\n const x1Num = Number.parseInt(x1, 10);\n const y1Num = Number.parseInt(y1, 10);\n const x2Num = Number.parseInt(x2, 10);\n const y2Num = Number.parseInt(y2, 10);\n\n // Use Math.floor to truncate and calculate center point\n const x = Math.floor((x1Num + x2Num) / 2);\n const y = Math.floor((y1Num + y2Num) / 2);\n\n // Return formatted coordinate string\n return `(${x},${y})`;\n }\n\n // Remove [EOS] and replace <bbox> coordinates\n const cleanedText = text.replace(/\\[EOS\\]/g, '');\n return cleanedText.replace(pattern, replaceMatch).trim();\n}\n\nfunction getPoint(startBox: string, size: { width: number; height: number }) {\n const [x, y] = JSON.parse(startBox);\n return [x * size.width, y * size.height];\n}\n\ninterface BaseAction {\n action_type: ActionType;\n action_inputs: Record<string, any>;\n reflection: string | null;\n thought: string | null;\n}\n\ninterface ClickAction extends BaseAction {\n action_type: 'click';\n action_inputs: {\n start_box: string; // JSON string of [x, y] coordinates\n };\n}\n\ninterface DragAction extends BaseAction {\n action_type: 'drag';\n action_inputs: {\n start_box: string; // JSON string of [x, y] coordinates\n end_box: string; // JSON string of [x, y] coordinates\n };\n}\n\ninterface WaitAction extends BaseAction {\n action_type: 'wait';\n action_inputs: {\n time: string; // JSON string of [x, y] coordinates\n };\n}\n\ninterface LeftDoubleAction extends BaseAction {\n action_type: 'left_double';\n action_inputs: {\n start_box: string; // JSON string of [x, y] coordinates\n };\n}\n\ninterface RightSingleAction extends BaseAction {\n action_type: 'right_single';\n action_inputs: {\n start_box: string; // JSON string of [x, y] coordinates\n };\n}\n\ninterface TypeAction extends BaseAction {\n action_type: 'type';\n action_inputs: {\n content: string;\n };\n}\n\ninterface HotkeyAction extends BaseAction {\n action_type: 'hotkey';\n action_inputs: {\n key: string;\n };\n}\n\ninterface ScrollAction extends BaseAction {\n action_type: 'scroll';\n action_inputs: {\n direction: 'up' | 'down';\n };\n}\n\ninterface FinishedAction extends BaseAction {\n action_type: 'finished';\n action_inputs: Record<string, never>;\n}\n\nexport type Action =\n | ClickAction\n | LeftDoubleAction\n | RightSingleAction\n | DragAction\n | TypeAction\n | HotkeyAction\n | ScrollAction\n | FinishedAction\n | WaitAction;\n\nexport async function resizeImageForUiTars(\n imageBase64: string,\n size: Size,\n uiTarsVersion: UITarsModelVersion | undefined,\n) {\n if (uiTarsVersion === UITarsModelVersion.V1_5) {\n debug('ui-tars-v1.5, will check image size', size);\n const currentPixels = size.width * size.height;\n const maxPixels = 16384 * 28 * 28; //\n if (currentPixels > maxPixels) {\n const resizeFactor = Math.sqrt(maxPixels / currentPixels);\n const newWidth = Math.floor(size.width * resizeFactor);\n const newHeight = Math.floor(size.height * resizeFactor);\n debug(\n 'resize image for ui-tars, new width: %s, new height: %s',\n newWidth,\n newHeight,\n );\n const resizedImage = await resizeImgBase64(imageBase64, {\n width: newWidth,\n height: newHeight,\n });\n return resizedImage;\n }\n }\n return imageBase64;\n}\n"],"names":["debug","getDebug","bboxSize","pointToBbox","point","width","height","Math","uiTarsPlanning","userInstruction","options","conversationHistory","context","modelConfig","actionContext","uiTarsModelVersion","instruction","systemPrompt","getUiTarsPlanningPrompt","screenshotBase64","imagePayload","resizeImageForUiTars","res","callAIWithStringResponse","convertedText","convertBboxToCoordinates","size","parsed","actionParser","JSON","transformActions","unhandledActions","shouldContinue","action","actionType","assert","getPoint","startPoint","endPoint","keys","transformHotkeyInput","console","errorDetails","types","a","errorMessage","Error","log","getSummary","undefined","text","pattern","replaceMatch","match","x1","y1","x2","y2","x1Num","Number","y1Num","x2Num","y2Num","x","y","cleanedText","startBox","imageBase64","uiTarsVersion","UITarsModelVersion","currentPixels","maxPixels","resizeFactor","newWidth","newHeight","resizedImage","resizeImgBase64"],"mappings":";;;;;;;;AA0BA,MAAMA,QAAQC,SAAS;AACvB,MAAMC,WAAW;AACjB,MAAMC,cAAc,CAClBC,OACAC,OACAC,SAEO;QACLC,KAAK,KAAK,CAACA,KAAK,GAAG,CAACH,MAAM,CAAC,GAAGF,WAAW,GAAG;QAC5CK,KAAK,KAAK,CAACA,KAAK,GAAG,CAACH,MAAM,CAAC,GAAGF,WAAW,GAAG;QAC5CK,KAAK,KAAK,CAACA,KAAK,GAAG,CAACH,MAAM,CAAC,GAAGF,WAAW,GAAGG;QAC5CE,KAAK,KAAK,CAACA,KAAK,GAAG,CAACH,MAAM,CAAC,GAAGF,WAAW,GAAGI;KAC7C;AAGI,eAAeE,eACpBC,eAAuB,EACvBC,OAKC;IAED,MAAM,EAAEC,mBAAmB,EAAEC,OAAO,EAAEC,WAAW,EAAEC,aAAa,EAAE,GAAGJ;IACrE,MAAM,EAAEK,kBAAkB,EAAE,GAAGF;IAE/B,IAAIG,cAAcP;IAClB,IAAIK,eACFE,cAAc,CAAC,yBAAyB,EAAEF,cAAc,8CAA8C,EAAEL,gBAAgB,mBAAmB,CAAC;IAG9I,MAAMQ,eAAeC,4BAA4BF;IAEjD,MAAMG,mBAAmBP,QAAQ,UAAU,CAAC,OAAO;IACnD,MAAMQ,eAAe,MAAMC,qBACzBF,kBACAP,QAAQ,IAAI,EACZG;IAGFJ,oBAAoB,MAAM,CAAC;QACzB,MAAM;QACN,SAAS;YACP;gBACE,MAAM;gBACN,WAAW;oBACT,KAAKS;gBACP;YACF;SACD;IACH;IAEA,MAAME,MAAM,MAAMC,yBAChB;QACE;YACE,MAAM;YACN,SAASN;QACX;WACGN,oBAAoB,QAAQ;KAChC,EACDE;IAEF,MAAMW,gBAAgBC,yBAAyBH,IAAI,OAAO;IAE1D,MAAM,EAAEI,IAAI,EAAE,GAAGd;IACjB,MAAM,EAAEe,MAAM,EAAE,GAAGC,aAAa;QAC9B,YAAYJ;QACZ,QAAQ;YAAC;YAAM;SAAK;QACpB,eAAe;YACb,OAAOE,KAAK,KAAK;YACjB,QAAQA,KAAK,MAAM;QACrB;QACA,UAAUX;IACZ;IAEAf,MACE,oBACAe,oBACA,YACAc,KAAK,SAAS,CAACF;IAGjB,MAAMG,mBAAqC,EAAE;IAC7C,MAAMC,mBAA6D,EAAE;IACrE,IAAIC,iBAAiB;IACrBL,OAAO,OAAO,CAAC,CAACM;QACd,MAAMC,aAAcD,AAAAA,CAAAA,OAAO,WAAW,IAAI,EAAC,EAAG,WAAW;QACzD,IAAIC,AAAe,YAAfA,YAAwB;YAC1BC,OAAOF,OAAO,aAAa,CAAC,SAAS,EAAE;YACvC,MAAM7B,QAAQgC,SAASH,OAAO,aAAa,CAAC,SAAS,EAAEP;YACvDI,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,QAAQ;wBACN,QAAQG,OAAO,OAAO,IAAI;wBAC1B,MAAM9B,YACJ;4BAAE,GAAGC,KAAK,CAAC,EAAE;4BAAE,GAAGA,KAAK,CAAC,EAAE;wBAAC,GAC3BsB,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;gBACF;YACF;QACF,OAAO,IAAIQ,AAAe,kBAAfA,YAA8B;YACvCC,OAAOF,OAAO,aAAa,CAAC,SAAS,EAAE;YACvC,MAAM7B,QAAQgC,SAASH,OAAO,aAAa,CAAC,SAAS,EAAEP;YACvDI,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,QAAQ;wBACN,QAAQG,OAAO,OAAO,IAAI;wBAC1B,MAAM9B,YACJ;4BAAE,GAAGC,KAAK,CAAC,EAAE;4BAAE,GAAGA,KAAK,CAAC,EAAE;wBAAC,GAC3BsB,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;gBACF;gBACA,SAASO,OAAO,OAAO,IAAI;YAC7B;QACF,OAAO,IAAIC,AAAe,mBAAfA,YAA+B;YACxCC,OAAOF,OAAO,aAAa,CAAC,SAAS,EAAE;YACvC,MAAM7B,QAAQgC,SAASH,OAAO,aAAa,CAAC,SAAS,EAAEP;YACvDI,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,QAAQ;wBACN,QAAQG,OAAO,OAAO,IAAI;wBAC1B,MAAM9B,YACJ;4BAAE,GAAGC,KAAK,CAAC,EAAE;4BAAE,GAAGA,KAAK,CAAC,EAAE;wBAAC,GAC3BsB,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;gBACF;gBACA,SAASO,OAAO,OAAO,IAAI;YAC7B;QACF,OAAO,IAAIC,AAAe,WAAfA,YAAuB;YAChCC,OAAOF,OAAO,aAAa,CAAC,SAAS,EAAE;YACvCE,OAAOF,OAAO,aAAa,CAAC,OAAO,EAAE;YACrC,MAAMI,aAAaD,SAASH,OAAO,aAAa,CAAC,SAAS,EAAEP;YAC5D,MAAMY,WAAWF,SAASH,OAAO,aAAa,CAAC,OAAO,EAAEP;YACxDI,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,MAAM;wBACJ,QAAQG,OAAO,OAAO,IAAI;wBAC1B,MAAM9B,YACJ;4BAAE,GAAGkC,UAAU,CAAC,EAAE;4BAAE,GAAGA,UAAU,CAAC,EAAE;wBAAC,GACrCX,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;oBACA,IAAI;wBACF,QAAQO,OAAO,OAAO,IAAI;wBAC1B,MAAM9B,YACJ;4BAAE,GAAGmC,QAAQ,CAAC,EAAE;4BAAE,GAAGA,QAAQ,CAAC,EAAE;wBAAC,GACjCZ,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;gBACF;gBACA,SAASO,OAAO,OAAO,IAAI;YAC7B;QACF,OAAO,IAAIC,AAAe,WAAfA,YACTJ,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,OAAOG,OAAO,aAAa,CAAC,OAAO;YACrC;YACA,SAASA,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIC,AAAe,aAAfA,YACTJ,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,WAAWG,OAAO,aAAa,CAAC,SAAS;YAC3C;YACA,SAASA,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIC,AAAe,eAAfA,YAA2B;YACpCF,iBAAiB;YACjBF,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO,CAAC;gBACR,SAASG,OAAO,OAAO,IAAI;YAC7B;QACF,OAAO,IAAIC,AAAe,aAAfA,YACT,IAAKD,OAAO,aAAa,CAAC,GAAG,EAItB;YACL,MAAMM,OAAOC,qBAAqBP,OAAO,aAAa,CAAC,GAAG;YAE1DH,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,SAASS,KAAK,IAAI,CAAC;gBACrB;gBACA,SAASN,OAAO,OAAO,IAAI;YAC7B;QACF,OAbEQ,QAAQ,IAAI,CACV;aAaC,IAAIP,AAAe,WAAfA,YACTJ,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,QAAQ;YACV;YACA,SAASG,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIC,YAAY;YAErBH,iBAAiB,IAAI,CAAC;gBACpB,MAAMG;gBACN,SAASD,OAAO,OAAO,IAAI;YAC7B;YACAjC,MAAM,0BAA0BkC,YAAY,YAAYD,OAAO,OAAO;QACxE;IACF;IAEA,IAAIH,AAA4B,MAA5BA,iBAAiB,MAAM,EAAQ;QACjC,MAAMY,eAAyB,EAAE;QAGjC,IAAIf,AAAkB,MAAlBA,OAAO,MAAM,EAAQ;YACvBe,aAAa,IAAI,CAAC;YAGlB,IACEpB,IAAI,OAAO,CAAC,QAAQ,CAAC,eACrB,CAACA,IAAI,OAAO,CAAC,QAAQ,CAAC,YAEtBoB,aAAa,IAAI,CACf;iBAGFA,aAAa,IAAI,CAAC;QAEtB;QAGA,IAAIX,iBAAiB,MAAM,GAAG,GAAG;YAC/B,MAAMY,QAAQZ,iBAAiB,GAAG,CAAC,CAACa,IAAMA,EAAE,IAAI,EAAE,IAAI,CAAC;YACvDF,aAAa,IAAI,CAAC,CAAC,wBAAwB,EAAEC,OAAO;QACtD;QAEA,MAAME,eAAe;YACnB;eACGH;YACH,CAAC,gBAAgB,EAAEpB,IAAI,OAAO,EAAE;SACjC,CAAC,IAAI,CAAC;QAEP,MAAM,IAAIwB,MAAMD,cAAc;YAC5B,OAAO;gBACL,YAAYvB,IAAI,OAAO;gBACvBK;gBACAI;gBACAP;YACF;QACF;IACF;IAEAxB,MAAM,oBAAoB6B,KAAK,SAAS,CAACC,kBAAkB,MAAM;IACjE,MAAMiB,MAAMC,WAAW1B,IAAI,OAAO;IAElCX,oBAAoB,MAAM,CAAC;QACzB,MAAM;QACN,SAASoC;IACX;IAEA,OAAO;QACL,SAASjB;QACTiB;QACA,OAAOzB,IAAI,KAAK;QAChB,aAAaO,KAAK,SAAS,CAACP,IAAI,OAAO,EAAE2B,QAAW;QACpD,wBAAwBjB;IAC1B;AACF;AAOA,SAASP,yBAAyByB,IAAY;IAE5C,MAAMC,UAAU;IAEhB,SAASC,aACPC,KAAa,EACbC,EAAU,EACVC,EAAU,EACVC,EAAU,EACVC,EAAU;QAGV,MAAMC,QAAQC,OAAO,QAAQ,CAACL,IAAI;QAClC,MAAMM,QAAQD,OAAO,QAAQ,CAACJ,IAAI;QAClC,MAAMM,QAAQF,OAAO,QAAQ,CAACH,IAAI;QAClC,MAAMM,QAAQH,OAAO,QAAQ,CAACF,IAAI;QAGlC,MAAMM,IAAIxD,KAAK,KAAK,CAAEmD,AAAAA,CAAAA,QAAQG,KAAI,IAAK;QACvC,MAAMG,IAAIzD,KAAK,KAAK,CAAEqD,AAAAA,CAAAA,QAAQE,KAAI,IAAK;QAGvC,OAAO,CAAC,CAAC,EAAEC,EAAE,CAAC,EAAEC,EAAE,CAAC,CAAC;IACtB;IAGA,MAAMC,cAAcf,KAAK,OAAO,CAAC,YAAY;IAC7C,OAAOe,YAAY,OAAO,CAACd,SAASC,cAAc,IAAI;AACxD;AAEA,SAAShB,SAAS8B,QAAgB,EAAExC,IAAuC;IACzE,MAAM,CAACqC,GAAGC,EAAE,GAAGnC,KAAK,KAAK,CAACqC;IAC1B,OAAO;QAACH,IAAIrC,KAAK,KAAK;QAAEsC,IAAItC,KAAK,MAAM;KAAC;AAC1C;AAkFO,eAAeL,qBACpB8C,WAAmB,EACnBzC,IAAU,EACV0C,aAA6C;IAE7C,IAAIA,kBAAkBC,mBAAmB,IAAI,EAAE;QAC7CrE,MAAM,uCAAuC0B;QAC7C,MAAM4C,gBAAgB5C,KAAK,KAAK,GAAGA,KAAK,MAAM;QAC9C,MAAM6C,YAAY;QAClB,IAAID,gBAAgBC,WAAW;YAC7B,MAAMC,eAAejE,KAAK,IAAI,CAACgE,YAAYD;YAC3C,MAAMG,WAAWlE,KAAK,KAAK,CAACmB,KAAK,KAAK,GAAG8C;YACzC,MAAME,YAAYnE,KAAK,KAAK,CAACmB,KAAK,MAAM,GAAG8C;YAC3CxE,MACE,2DACAyE,UACAC;YAEF,MAAMC,eAAe,MAAMC,gBAAgBT,aAAa;gBACtD,OAAOM;gBACP,QAAQC;YACV;YACA,OAAOC;QACT;IACF;IACA,OAAOR;AACT"}
|
|
1
|
+
{"version":3,"file":"ai-model/ui-tars-planning.mjs","sources":["../../../src/ai-model/ui-tars-planning.ts"],"sourcesContent":["import type {\n PlanningAIResponse,\n PlanningAction,\n Size,\n UIContext,\n} from '@/types';\nimport { type IModelConfig, UITarsModelVersion } from '@midscene/shared/env';\nimport { resizeImgBase64 } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { transformHotkeyInput } from '@midscene/shared/us-keyboard-layout';\nimport { assert } from '@midscene/shared/utils';\nimport { actionParser } from '@ui-tars/action-parser';\nimport type { ConversationHistory } from './conversation-history';\nimport { getSummary, getUiTarsPlanningPrompt } from './prompt/ui-tars-planning';\nimport { callAIWithStringResponse } from './service-caller/index';\n\ntype ActionType =\n | 'click'\n | 'left_double'\n | 'right_single'\n | 'drag'\n | 'type'\n | 'hotkey'\n | 'finished'\n | 'scroll'\n | 'wait';\n\nconst debug = getDebug('ui-tars-planning');\nconst bboxSize = 10;\nconst pointToBbox = (\n point: { x: number; y: number },\n width: number,\n height: number,\n): [number, number, number, number] => {\n return [\n Math.round(Math.max(point.x - bboxSize / 2, 0)),\n Math.round(Math.max(point.y - bboxSize / 2, 0)),\n Math.round(Math.min(point.x + bboxSize / 2, width)),\n Math.round(Math.min(point.y + bboxSize / 2, height)),\n ];\n};\n\nexport async function uiTarsPlanning(\n userInstruction: string,\n options: {\n conversationHistory: ConversationHistory;\n context: UIContext;\n modelConfig: IModelConfig;\n actionContext?: string;\n },\n): Promise<PlanningAIResponse> {\n const { conversationHistory, context, modelConfig, actionContext } = options;\n const { uiTarsModelVersion } = modelConfig;\n\n let instruction = userInstruction;\n if (actionContext) {\n instruction = `<high_priority_knowledge>${actionContext}</high_priority_knowledge>\\n<user_instruction>${userInstruction}</user_instruction>`;\n }\n\n const systemPrompt = getUiTarsPlanningPrompt() + instruction;\n\n const screenshotBase64 = context.screenshot.base64;\n const imagePayload = await resizeImageForUiTars(\n screenshotBase64,\n context.size,\n uiTarsModelVersion,\n );\n\n conversationHistory.append({\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: {\n url: imagePayload,\n },\n },\n ],\n });\n\n const res = await callAIWithStringResponse(\n [\n {\n role: 'user',\n content: systemPrompt,\n },\n ...conversationHistory.snapshot(),\n ],\n modelConfig,\n );\n const convertedText = convertBboxToCoordinates(res.content);\n\n const { size } = context;\n const { parsed } = actionParser({\n prediction: convertedText,\n factor: [1000, 1000],\n screenContext: {\n width: size.width,\n height: size.height,\n },\n modelVer: uiTarsModelVersion,\n });\n\n debug(\n 'ui-tars modelVer',\n uiTarsModelVersion,\n ', parsed',\n JSON.stringify(parsed),\n );\n\n const transformActions: PlanningAction[] = [];\n const unhandledActions: Array<{ type: string; thought: string }> = [];\n let shouldContinue = true;\n parsed.forEach((action) => {\n const actionType = (action.action_type || '').toLowerCase();\n if (actionType === 'click') {\n assert(action.action_inputs.start_box, 'start_box is required');\n const point = getPoint(action.action_inputs.start_box, size);\n\n const locate = {\n prompt: action.thought || '',\n bbox: pointToBbox(\n { x: point[0], y: point[1] },\n size.width,\n size.height,\n ),\n };\n\n transformActions.push({\n type: 'Tap',\n param: {\n locate: locate,\n },\n });\n } else if (actionType === 'left_double') {\n assert(action.action_inputs.start_box, 'start_box is required');\n const point = getPoint(action.action_inputs.start_box, size);\n\n const locate = {\n prompt: action.thought || '',\n bbox: pointToBbox(\n { x: point[0], y: point[1] },\n size.width,\n size.height,\n ),\n };\n\n transformActions.push({\n type: 'DoubleClick',\n param: {\n locate: locate,\n },\n thought: action.thought || '',\n });\n } else if (actionType === 'right_single') {\n assert(action.action_inputs.start_box, 'start_box is required');\n const point = getPoint(action.action_inputs.start_box, size);\n\n const locate = {\n prompt: action.thought || '',\n bbox: pointToBbox(\n { x: point[0], y: point[1] },\n size.width,\n size.height,\n ),\n };\n\n transformActions.push({\n type: 'RightClick',\n param: {\n locate: locate,\n },\n thought: action.thought || '',\n });\n } else if (actionType === 'drag') {\n assert(action.action_inputs.start_box, 'start_box is required');\n assert(action.action_inputs.end_box, 'end_box is required');\n const startPoint = getPoint(action.action_inputs.start_box, size);\n const endPoint = getPoint(action.action_inputs.end_box, size);\n transformActions.push({\n type: 'DragAndDrop',\n param: {\n from: {\n prompt: action.thought || '',\n bbox: pointToBbox(\n { x: startPoint[0], y: startPoint[1] },\n size.width,\n size.height,\n ),\n },\n to: {\n prompt: action.thought || '',\n bbox: pointToBbox(\n { x: endPoint[0], y: endPoint[1] },\n size.width,\n size.height,\n ),\n },\n },\n thought: action.thought || '',\n });\n } else if (actionType === 'type') {\n transformActions.push({\n type: 'Input',\n param: {\n value: action.action_inputs.content,\n },\n thought: action.thought || '',\n });\n } else if (actionType === 'scroll') {\n transformActions.push({\n type: 'Scroll',\n param: {\n direction: action.action_inputs.direction,\n },\n thought: action.thought || '',\n });\n } else if (actionType === 'finished') {\n shouldContinue = false;\n transformActions.push({\n type: 'Finished',\n param: {},\n thought: action.thought || '',\n });\n } else if (actionType === 'hotkey') {\n if (!action.action_inputs.key) {\n console.warn(\n 'No key found in action: hotkey. Will not perform action.',\n );\n } else {\n const keys = transformHotkeyInput(action.action_inputs.key);\n\n transformActions.push({\n type: 'KeyboardPress',\n param: {\n keyName: keys.join('+'),\n },\n thought: action.thought || '',\n });\n }\n } else if (actionType === 'wait') {\n transformActions.push({\n type: 'Sleep',\n param: {\n timeMs: 1000,\n },\n thought: action.thought || '',\n });\n } else if (actionType) {\n // Track unhandled action types\n unhandledActions.push({\n type: actionType,\n thought: action.thought || '',\n });\n debug('Unhandled action type:', actionType, 'thought:', action.thought);\n }\n });\n\n if (transformActions.length === 0) {\n const errorDetails: string[] = [];\n\n // Check if parsing failed\n if (parsed.length === 0) {\n errorDetails.push('Action parser returned no actions');\n\n // Check if response has Thought but no Action\n if (\n res.content.includes('Thought:') &&\n !res.content.includes('Action:')\n ) {\n errorDetails.push(\n 'Response contains \"Thought:\" but missing \"Action:\" line',\n );\n } else {\n errorDetails.push('Response may be malformed or empty');\n }\n }\n\n // Check if we have unhandled action types\n if (unhandledActions.length > 0) {\n const types = unhandledActions.map((a) => a.type).join(', ');\n errorDetails.push(`Unhandled action types: ${types}`);\n }\n\n const errorMessage = [\n 'No actions found in UI-TARS response.',\n ...errorDetails,\n `\\nRaw response: ${res.content}`,\n ].join('\\n');\n\n throw new Error(errorMessage, {\n cause: {\n prediction: res.content,\n parsed,\n unhandledActions,\n convertedText,\n },\n });\n }\n\n debug('transformActions', JSON.stringify(transformActions, null, 2));\n const log = getSummary(res.content);\n\n conversationHistory.append({\n role: 'assistant',\n content: log,\n });\n\n return {\n actions: transformActions,\n log,\n usage: res.usage,\n rawResponse: JSON.stringify(res.content, undefined, 2),\n shouldContinuePlanning: shouldContinue,\n };\n}\n\n/**\n * Converts bounding box notation to coordinate points\n * @param text - The text containing bbox tags to be converted\n * @returns The text with bbox tags replaced by coordinate points\n */\nfunction convertBboxToCoordinates(text: string): string {\n // Match the four numbers after <bbox>\n const pattern = /<bbox>(\\d+)\\s+(\\d+)\\s+(\\d+)\\s+(\\d+)<\\/bbox>/g;\n\n function replaceMatch(\n match: string,\n x1: string,\n y1: string,\n x2: string,\n y2: string,\n ): string {\n // Convert strings to numbers and calculate center point\n const x1Num = Number.parseInt(x1, 10);\n const y1Num = Number.parseInt(y1, 10);\n const x2Num = Number.parseInt(x2, 10);\n const y2Num = Number.parseInt(y2, 10);\n\n // Use Math.floor to truncate and calculate center point\n const x = Math.floor((x1Num + x2Num) / 2);\n const y = Math.floor((y1Num + y2Num) / 2);\n\n // Return formatted coordinate string\n return `(${x},${y})`;\n }\n\n // Remove [EOS] and replace <bbox> coordinates\n const cleanedText = text.replace(/\\[EOS\\]/g, '');\n return cleanedText.replace(pattern, replaceMatch).trim();\n}\n\nfunction getPoint(startBox: string, size: { width: number; height: number }) {\n const [x, y] = JSON.parse(startBox);\n return [x * size.width, y * size.height];\n}\n\ninterface BaseAction {\n action_type: ActionType;\n action_inputs: Record<string, any>;\n reflection: string | null;\n thought: string | null;\n}\n\ninterface ClickAction extends BaseAction {\n action_type: 'click';\n action_inputs: {\n start_box: string; // JSON string of [x, y] coordinates\n };\n}\n\ninterface DragAction extends BaseAction {\n action_type: 'drag';\n action_inputs: {\n start_box: string; // JSON string of [x, y] coordinates\n end_box: string; // JSON string of [x, y] coordinates\n };\n}\n\ninterface WaitAction extends BaseAction {\n action_type: 'wait';\n action_inputs: {\n time: string; // JSON string of [x, y] coordinates\n };\n}\n\ninterface LeftDoubleAction extends BaseAction {\n action_type: 'left_double';\n action_inputs: {\n start_box: string; // JSON string of [x, y] coordinates\n };\n}\n\ninterface RightSingleAction extends BaseAction {\n action_type: 'right_single';\n action_inputs: {\n start_box: string; // JSON string of [x, y] coordinates\n };\n}\n\ninterface TypeAction extends BaseAction {\n action_type: 'type';\n action_inputs: {\n content: string;\n };\n}\n\ninterface HotkeyAction extends BaseAction {\n action_type: 'hotkey';\n action_inputs: {\n key: string;\n };\n}\n\ninterface ScrollAction extends BaseAction {\n action_type: 'scroll';\n action_inputs: {\n direction: 'up' | 'down';\n };\n}\n\ninterface FinishedAction extends BaseAction {\n action_type: 'finished';\n action_inputs: Record<string, never>;\n}\n\nexport type Action =\n | ClickAction\n | LeftDoubleAction\n | RightSingleAction\n | DragAction\n | TypeAction\n | HotkeyAction\n | ScrollAction\n | FinishedAction\n | WaitAction;\n\nexport async function resizeImageForUiTars(\n imageBase64: string,\n size: Size,\n uiTarsVersion: UITarsModelVersion | undefined,\n) {\n if (uiTarsVersion === UITarsModelVersion.V1_5) {\n debug('ui-tars-v1.5, will check image size', size);\n const currentPixels = size.width * size.height;\n const maxPixels = 16384 * 28 * 28; //\n if (currentPixels > maxPixels) {\n const resizeFactor = Math.sqrt(maxPixels / currentPixels);\n const newWidth = Math.floor(size.width * resizeFactor);\n const newHeight = Math.floor(size.height * resizeFactor);\n debug(\n 'resize image for ui-tars, new width: %s, new height: %s',\n newWidth,\n newHeight,\n );\n const resizedImage = await resizeImgBase64(imageBase64, {\n width: newWidth,\n height: newHeight,\n });\n return resizedImage;\n }\n }\n return imageBase64;\n}\n"],"names":["debug","getDebug","bboxSize","pointToBbox","point","width","height","Math","uiTarsPlanning","userInstruction","options","conversationHistory","context","modelConfig","actionContext","uiTarsModelVersion","instruction","systemPrompt","getUiTarsPlanningPrompt","screenshotBase64","imagePayload","resizeImageForUiTars","res","callAIWithStringResponse","convertedText","convertBboxToCoordinates","size","parsed","actionParser","JSON","transformActions","unhandledActions","shouldContinue","action","actionType","assert","getPoint","locate","startPoint","endPoint","keys","transformHotkeyInput","console","errorDetails","types","a","errorMessage","Error","log","getSummary","undefined","text","pattern","replaceMatch","match","x1","y1","x2","y2","x1Num","Number","y1Num","x2Num","y2Num","x","y","cleanedText","startBox","imageBase64","uiTarsVersion","UITarsModelVersion","currentPixels","maxPixels","resizeFactor","newWidth","newHeight","resizedImage","resizeImgBase64"],"mappings":";;;;;;;;AA2BA,MAAMA,QAAQC,SAAS;AACvB,MAAMC,WAAW;AACjB,MAAMC,cAAc,CAClBC,OACAC,OACAC,SAEO;QACLC,KAAK,KAAK,CAACA,KAAK,GAAG,CAACH,MAAM,CAAC,GAAGF,WAAW,GAAG;QAC5CK,KAAK,KAAK,CAACA,KAAK,GAAG,CAACH,MAAM,CAAC,GAAGF,WAAW,GAAG;QAC5CK,KAAK,KAAK,CAACA,KAAK,GAAG,CAACH,MAAM,CAAC,GAAGF,WAAW,GAAGG;QAC5CE,KAAK,KAAK,CAACA,KAAK,GAAG,CAACH,MAAM,CAAC,GAAGF,WAAW,GAAGI;KAC7C;AAGI,eAAeE,eACpBC,eAAuB,EACvBC,OAKC;IAED,MAAM,EAAEC,mBAAmB,EAAEC,OAAO,EAAEC,WAAW,EAAEC,aAAa,EAAE,GAAGJ;IACrE,MAAM,EAAEK,kBAAkB,EAAE,GAAGF;IAE/B,IAAIG,cAAcP;IAClB,IAAIK,eACFE,cAAc,CAAC,yBAAyB,EAAEF,cAAc,8CAA8C,EAAEL,gBAAgB,mBAAmB,CAAC;IAG9I,MAAMQ,eAAeC,4BAA4BF;IAEjD,MAAMG,mBAAmBP,QAAQ,UAAU,CAAC,MAAM;IAClD,MAAMQ,eAAe,MAAMC,qBACzBF,kBACAP,QAAQ,IAAI,EACZG;IAGFJ,oBAAoB,MAAM,CAAC;QACzB,MAAM;QACN,SAAS;YACP;gBACE,MAAM;gBACN,WAAW;oBACT,KAAKS;gBACP;YACF;SACD;IACH;IAEA,MAAME,MAAM,MAAMC,yBAChB;QACE;YACE,MAAM;YACN,SAASN;QACX;WACGN,oBAAoB,QAAQ;KAChC,EACDE;IAEF,MAAMW,gBAAgBC,yBAAyBH,IAAI,OAAO;IAE1D,MAAM,EAAEI,IAAI,EAAE,GAAGd;IACjB,MAAM,EAAEe,MAAM,EAAE,GAAGC,aAAa;QAC9B,YAAYJ;QACZ,QAAQ;YAAC;YAAM;SAAK;QACpB,eAAe;YACb,OAAOE,KAAK,KAAK;YACjB,QAAQA,KAAK,MAAM;QACrB;QACA,UAAUX;IACZ;IAEAf,MACE,oBACAe,oBACA,YACAc,KAAK,SAAS,CAACF;IAGjB,MAAMG,mBAAqC,EAAE;IAC7C,MAAMC,mBAA6D,EAAE;IACrE,IAAIC,iBAAiB;IACrBL,OAAO,OAAO,CAAC,CAACM;QACd,MAAMC,aAAcD,AAAAA,CAAAA,OAAO,WAAW,IAAI,EAAC,EAAG,WAAW;QACzD,IAAIC,AAAe,YAAfA,YAAwB;YAC1BC,OAAOF,OAAO,aAAa,CAAC,SAAS,EAAE;YACvC,MAAM7B,QAAQgC,SAASH,OAAO,aAAa,CAAC,SAAS,EAAEP;YAEvD,MAAMW,SAAS;gBACb,QAAQJ,OAAO,OAAO,IAAI;gBAC1B,MAAM9B,YACJ;oBAAE,GAAGC,KAAK,CAAC,EAAE;oBAAE,GAAGA,KAAK,CAAC,EAAE;gBAAC,GAC3BsB,KAAK,KAAK,EACVA,KAAK,MAAM;YAEf;YAEAI,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,QAAQO;gBACV;YACF;QACF,OAAO,IAAIH,AAAe,kBAAfA,YAA8B;YACvCC,OAAOF,OAAO,aAAa,CAAC,SAAS,EAAE;YACvC,MAAM7B,QAAQgC,SAASH,OAAO,aAAa,CAAC,SAAS,EAAEP;YAEvD,MAAMW,SAAS;gBACb,QAAQJ,OAAO,OAAO,IAAI;gBAC1B,MAAM9B,YACJ;oBAAE,GAAGC,KAAK,CAAC,EAAE;oBAAE,GAAGA,KAAK,CAAC,EAAE;gBAAC,GAC3BsB,KAAK,KAAK,EACVA,KAAK,MAAM;YAEf;YAEAI,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,QAAQO;gBACV;gBACA,SAASJ,OAAO,OAAO,IAAI;YAC7B;QACF,OAAO,IAAIC,AAAe,mBAAfA,YAA+B;YACxCC,OAAOF,OAAO,aAAa,CAAC,SAAS,EAAE;YACvC,MAAM7B,QAAQgC,SAASH,OAAO,aAAa,CAAC,SAAS,EAAEP;YAEvD,MAAMW,SAAS;gBACb,QAAQJ,OAAO,OAAO,IAAI;gBAC1B,MAAM9B,YACJ;oBAAE,GAAGC,KAAK,CAAC,EAAE;oBAAE,GAAGA,KAAK,CAAC,EAAE;gBAAC,GAC3BsB,KAAK,KAAK,EACVA,KAAK,MAAM;YAEf;YAEAI,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,QAAQO;gBACV;gBACA,SAASJ,OAAO,OAAO,IAAI;YAC7B;QACF,OAAO,IAAIC,AAAe,WAAfA,YAAuB;YAChCC,OAAOF,OAAO,aAAa,CAAC,SAAS,EAAE;YACvCE,OAAOF,OAAO,aAAa,CAAC,OAAO,EAAE;YACrC,MAAMK,aAAaF,SAASH,OAAO,aAAa,CAAC,SAAS,EAAEP;YAC5D,MAAMa,WAAWH,SAASH,OAAO,aAAa,CAAC,OAAO,EAAEP;YACxDI,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,MAAM;wBACJ,QAAQG,OAAO,OAAO,IAAI;wBAC1B,MAAM9B,YACJ;4BAAE,GAAGmC,UAAU,CAAC,EAAE;4BAAE,GAAGA,UAAU,CAAC,EAAE;wBAAC,GACrCZ,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;oBACA,IAAI;wBACF,QAAQO,OAAO,OAAO,IAAI;wBAC1B,MAAM9B,YACJ;4BAAE,GAAGoC,QAAQ,CAAC,EAAE;4BAAE,GAAGA,QAAQ,CAAC,EAAE;wBAAC,GACjCb,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;gBACF;gBACA,SAASO,OAAO,OAAO,IAAI;YAC7B;QACF,OAAO,IAAIC,AAAe,WAAfA,YACTJ,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,OAAOG,OAAO,aAAa,CAAC,OAAO;YACrC;YACA,SAASA,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIC,AAAe,aAAfA,YACTJ,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,WAAWG,OAAO,aAAa,CAAC,SAAS;YAC3C;YACA,SAASA,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIC,AAAe,eAAfA,YAA2B;YACpCF,iBAAiB;YACjBF,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO,CAAC;gBACR,SAASG,OAAO,OAAO,IAAI;YAC7B;QACF,OAAO,IAAIC,AAAe,aAAfA,YACT,IAAKD,OAAO,aAAa,CAAC,GAAG,EAItB;YACL,MAAMO,OAAOC,qBAAqBR,OAAO,aAAa,CAAC,GAAG;YAE1DH,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,SAASU,KAAK,IAAI,CAAC;gBACrB;gBACA,SAASP,OAAO,OAAO,IAAI;YAC7B;QACF,OAbES,QAAQ,IAAI,CACV;aAaC,IAAIR,AAAe,WAAfA,YACTJ,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,QAAQ;YACV;YACA,SAASG,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIC,YAAY;YAErBH,iBAAiB,IAAI,CAAC;gBACpB,MAAMG;gBACN,SAASD,OAAO,OAAO,IAAI;YAC7B;YACAjC,MAAM,0BAA0BkC,YAAY,YAAYD,OAAO,OAAO;QACxE;IACF;IAEA,IAAIH,AAA4B,MAA5BA,iBAAiB,MAAM,EAAQ;QACjC,MAAMa,eAAyB,EAAE;QAGjC,IAAIhB,AAAkB,MAAlBA,OAAO,MAAM,EAAQ;YACvBgB,aAAa,IAAI,CAAC;YAGlB,IACErB,IAAI,OAAO,CAAC,QAAQ,CAAC,eACrB,CAACA,IAAI,OAAO,CAAC,QAAQ,CAAC,YAEtBqB,aAAa,IAAI,CACf;iBAGFA,aAAa,IAAI,CAAC;QAEtB;QAGA,IAAIZ,iBAAiB,MAAM,GAAG,GAAG;YAC/B,MAAMa,QAAQb,iBAAiB,GAAG,CAAC,CAACc,IAAMA,EAAE,IAAI,EAAE,IAAI,CAAC;YACvDF,aAAa,IAAI,CAAC,CAAC,wBAAwB,EAAEC,OAAO;QACtD;QAEA,MAAME,eAAe;YACnB;eACGH;YACH,CAAC,gBAAgB,EAAErB,IAAI,OAAO,EAAE;SACjC,CAAC,IAAI,CAAC;QAEP,MAAM,IAAIyB,MAAMD,cAAc;YAC5B,OAAO;gBACL,YAAYxB,IAAI,OAAO;gBACvBK;gBACAI;gBACAP;YACF;QACF;IACF;IAEAxB,MAAM,oBAAoB6B,KAAK,SAAS,CAACC,kBAAkB,MAAM;IACjE,MAAMkB,MAAMC,WAAW3B,IAAI,OAAO;IAElCX,oBAAoB,MAAM,CAAC;QACzB,MAAM;QACN,SAASqC;IACX;IAEA,OAAO;QACL,SAASlB;QACTkB;QACA,OAAO1B,IAAI,KAAK;QAChB,aAAaO,KAAK,SAAS,CAACP,IAAI,OAAO,EAAE4B,QAAW;QACpD,wBAAwBlB;IAC1B;AACF;AAOA,SAASP,yBAAyB0B,IAAY;IAE5C,MAAMC,UAAU;IAEhB,SAASC,aACPC,KAAa,EACbC,EAAU,EACVC,EAAU,EACVC,EAAU,EACVC,EAAU;QAGV,MAAMC,QAAQC,OAAO,QAAQ,CAACL,IAAI;QAClC,MAAMM,QAAQD,OAAO,QAAQ,CAACJ,IAAI;QAClC,MAAMM,QAAQF,OAAO,QAAQ,CAACH,IAAI;QAClC,MAAMM,QAAQH,OAAO,QAAQ,CAACF,IAAI;QAGlC,MAAMM,IAAIzD,KAAK,KAAK,CAAEoD,AAAAA,CAAAA,QAAQG,KAAI,IAAK;QACvC,MAAMG,IAAI1D,KAAK,KAAK,CAAEsD,AAAAA,CAAAA,QAAQE,KAAI,IAAK;QAGvC,OAAO,CAAC,CAAC,EAAEC,EAAE,CAAC,EAAEC,EAAE,CAAC,CAAC;IACtB;IAGA,MAAMC,cAAcf,KAAK,OAAO,CAAC,YAAY;IAC7C,OAAOe,YAAY,OAAO,CAACd,SAASC,cAAc,IAAI;AACxD;AAEA,SAASjB,SAAS+B,QAAgB,EAAEzC,IAAuC;IACzE,MAAM,CAACsC,GAAGC,EAAE,GAAGpC,KAAK,KAAK,CAACsC;IAC1B,OAAO;QAACH,IAAItC,KAAK,KAAK;QAAEuC,IAAIvC,KAAK,MAAM;KAAC;AAC1C;AAkFO,eAAeL,qBACpB+C,WAAmB,EACnB1C,IAAU,EACV2C,aAA6C;IAE7C,IAAIA,kBAAkBC,mBAAmB,IAAI,EAAE;QAC7CtE,MAAM,uCAAuC0B;QAC7C,MAAM6C,gBAAgB7C,KAAK,KAAK,GAAGA,KAAK,MAAM;QAC9C,MAAM8C,YAAY;QAClB,IAAID,gBAAgBC,WAAW;YAC7B,MAAMC,eAAelE,KAAK,IAAI,CAACiE,YAAYD;YAC3C,MAAMG,WAAWnE,KAAK,KAAK,CAACmB,KAAK,KAAK,GAAG+C;YACzC,MAAME,YAAYpE,KAAK,KAAK,CAACmB,KAAK,MAAM,GAAG+C;YAC3CzE,MACE,2DACA0E,UACAC;YAEF,MAAMC,eAAe,MAAMC,gBAAgBT,aAAa;gBACtD,OAAOM;gBACP,QAAQC;YACV;YACA,OAAOC;QACT;IACF;IACA,OAAOR;AACT"}
|
package/dist/es/common.mjs
CHANGED
|
@@ -6,6 +6,19 @@ import { getDebug } from "@midscene/shared/logger";
|
|
|
6
6
|
import { z } from "zod";
|
|
7
7
|
const defaultBboxSize = 20;
|
|
8
8
|
const debugInspectUtils = getDebug('ai:common');
|
|
9
|
+
function pointToBbox(x, y, bboxSize = defaultBboxSize) {
|
|
10
|
+
const halfSize = bboxSize / 2;
|
|
11
|
+
const x1 = Math.max(x - halfSize, 0);
|
|
12
|
+
const y1 = Math.max(y - halfSize, 0);
|
|
13
|
+
const x2 = Math.min(x + halfSize, 1000);
|
|
14
|
+
const y2 = Math.min(y + halfSize, 1000);
|
|
15
|
+
return [
|
|
16
|
+
x1,
|
|
17
|
+
y1,
|
|
18
|
+
x2,
|
|
19
|
+
y2
|
|
20
|
+
];
|
|
21
|
+
}
|
|
9
22
|
function fillBboxParam(locate, width, height, rightLimit, bottomLimit, vlMode) {
|
|
10
23
|
if (locate.bbox_2d && !locate?.bbox) {
|
|
11
24
|
locate.bbox = locate.bbox_2d;
|
|
@@ -86,7 +99,7 @@ function adaptBbox(bbox, width, height, rightLimit, bottomLimit, vlMode) {
|
|
|
86
99
|
0,
|
|
87
100
|
0
|
|
88
101
|
];
|
|
89
|
-
result = 'doubao-vision' === vlMode || 'vlm-ui-tars' === vlMode ? adaptDoubaoBbox(normalizedBbox, width, height) : 'gemini' === vlMode ? adaptGeminiBbox(normalizedBbox, width, height) : '
|
|
102
|
+
result = 'doubao-vision' === vlMode || 'vlm-ui-tars' === vlMode ? adaptDoubaoBbox(normalizedBbox, width, height) : 'gemini' === vlMode ? adaptGeminiBbox(normalizedBbox, width, height) : 'qwen2.5-vl' === vlMode ? adaptQwen2_5Bbox(normalizedBbox) : normalized01000(normalizedBbox, width, height);
|
|
90
103
|
result[2] = Math.min(result[2], rightLimit);
|
|
91
104
|
result[3] = Math.min(result[3], bottomLimit);
|
|
92
105
|
return result;
|
|
@@ -340,6 +353,6 @@ const getReadableTimeString = (format = 'YYYY-MM-DD HH:mm:ss')=>{
|
|
|
340
353
|
const timeString = format.replace('YYYY', String(year)).replace('MM', month).replace('DD', day).replace('HH', hours).replace('mm', minutes).replace('ss', seconds);
|
|
341
354
|
return `${timeString} (${format})`;
|
|
342
355
|
};
|
|
343
|
-
export { PointSchema, RectSchema, SizeSchema, TMultimodalPromptSchema, TUserPromptSchema, adaptBbox, adaptBboxToRect, adaptDoubaoBbox, adaptGeminiBbox, adaptQwen2_5Bbox, buildYamlFlowFromPlans, dumpActionParam, dumpMidsceneLocatorField, expandSearchArea, fillBboxParam, finalizeActionName, findAllMidsceneLocatorField, getMidsceneLocationSchema, getReadableTimeString, ifMidsceneLocatorField, loadActionParam, markupImageForLLM, mergeRects, normalized01000, parseActionParam };
|
|
356
|
+
export { PointSchema, RectSchema, SizeSchema, TMultimodalPromptSchema, TUserPromptSchema, adaptBbox, adaptBboxToRect, adaptDoubaoBbox, adaptGeminiBbox, adaptQwen2_5Bbox, buildYamlFlowFromPlans, dumpActionParam, dumpMidsceneLocatorField, expandSearchArea, fillBboxParam, finalizeActionName, findAllMidsceneLocatorField, getMidsceneLocationSchema, getReadableTimeString, ifMidsceneLocatorField, loadActionParam, markupImageForLLM, mergeRects, normalized01000, parseActionParam, pointToBbox };
|
|
344
357
|
|
|
345
358
|
//# sourceMappingURL=common.mjs.map
|
package/dist/es/common.mjs.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"common.mjs","sources":["../../src/common.ts"],"sourcesContent":["import type {\n BaseElement,\n DeviceAction,\n ElementTreeNode,\n MidsceneYamlFlowItem,\n PlanningAction,\n Rect,\n Size,\n} from '@/types';\nimport { assert, isPlainObject } from '@midscene/shared/utils';\n\nimport type { ChatCompletionMessageParam } from 'openai/resources/index';\n\nimport type { PlanningLocateParam } from '@/types';\nimport { NodeType } from '@midscene/shared/constants';\nimport type { TVlModeTypes } from '@midscene/shared/env';\nimport { treeToList } from '@midscene/shared/extractor';\nimport { compositeElementInfoImg } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { z } from 'zod';\n\nexport type AIArgs = ChatCompletionMessageParam[];\n\nconst defaultBboxSize = 20; // must be even number\nconst debugInspectUtils = getDebug('ai:common');\ntype AdaptBboxInput = number[] | string[] | string | (number[] | string[])[];\n\n// transform the param of locate from qwen mode\nexport function fillBboxParam(\n locate: PlanningLocateParam,\n width: number,\n height: number,\n rightLimit: number,\n bottomLimit: number,\n vlMode: TVlModeTypes | undefined,\n) {\n // The Qwen model might have hallucinations of naming bbox as bbox_2d.\n if ((locate as any).bbox_2d && !locate?.bbox) {\n locate.bbox = (locate as any).bbox_2d;\n // biome-ignore lint/performance/noDelete: <explanation>\n delete (locate as any).bbox_2d;\n }\n\n if (locate?.bbox) {\n locate.bbox = adaptBbox(\n locate.bbox,\n width,\n height,\n rightLimit,\n bottomLimit,\n vlMode,\n );\n }\n\n return locate;\n}\n\nexport function adaptQwen2_5Bbox(\n bbox: number[],\n): [number, number, number, number] {\n if (bbox.length < 2) {\n const msg = `invalid bbox data for qwen-vl mode: ${JSON.stringify(bbox)} `;\n throw new Error(msg);\n }\n\n const result: [number, number, number, number] = [\n Math.round(bbox[0]),\n Math.round(bbox[1]),\n typeof bbox[2] === 'number'\n ? Math.round(bbox[2])\n : Math.round(bbox[0] + defaultBboxSize),\n typeof bbox[3] === 'number'\n ? Math.round(bbox[3])\n : Math.round(bbox[1] + defaultBboxSize),\n ];\n return result;\n}\n\nexport function adaptDoubaoBbox(\n bbox: string[] | number[] | string,\n width: number,\n height: number,\n): [number, number, number, number] {\n assert(\n width > 0 && height > 0,\n 'width and height must be greater than 0 in doubao mode',\n );\n\n if (typeof bbox === 'string') {\n assert(\n /^(\\d+)\\s(\\d+)\\s(\\d+)\\s(\\d+)$/.test(bbox.trim()),\n `invalid bbox data string for doubao-vision mode: ${bbox}`,\n );\n const splitted = bbox.split(' ');\n if (splitted.length === 4) {\n return [\n Math.round((Number(splitted[0]) * width) / 1000),\n Math.round((Number(splitted[1]) * height) / 1000),\n Math.round((Number(splitted[2]) * width) / 1000),\n Math.round((Number(splitted[3]) * height) / 1000),\n ];\n }\n throw new Error(`invalid bbox data string for doubao-vision mode: ${bbox}`);\n }\n\n let bboxList: number[] = [];\n if (Array.isArray(bbox) && typeof bbox[0] === 'string') {\n bbox.forEach((item) => {\n if (typeof item === 'string' && item.includes(',')) {\n const [x, y] = item.split(',');\n bboxList.push(Number(x.trim()), Number(y.trim()));\n } else if (typeof item === 'string' && item.includes(' ')) {\n const [x, y] = item.split(' ');\n bboxList.push(Number(x.trim()), Number(y.trim()));\n } else {\n bboxList.push(Number(item));\n }\n });\n } else {\n bboxList = bbox as any;\n }\n\n if (bboxList.length === 4 || bboxList.length === 5) {\n return [\n Math.round((bboxList[0] * width) / 1000),\n Math.round((bboxList[1] * height) / 1000),\n Math.round((bboxList[2] * width) / 1000),\n Math.round((bboxList[3] * height) / 1000),\n ];\n }\n\n // treat the bbox as a center point\n if (\n bboxList.length === 6 ||\n bboxList.length === 2 ||\n bboxList.length === 3 ||\n bboxList.length === 7\n ) {\n return [\n Math.max(\n 0,\n Math.round((bboxList[0] * width) / 1000) - defaultBboxSize / 2,\n ),\n Math.max(\n 0,\n Math.round((bboxList[1] * height) / 1000) - defaultBboxSize / 2,\n ),\n Math.min(\n width,\n Math.round((bboxList[0] * width) / 1000) + defaultBboxSize / 2,\n ),\n Math.min(\n height,\n Math.round((bboxList[1] * height) / 1000) + defaultBboxSize / 2,\n ),\n ];\n }\n\n if (bbox.length === 8) {\n return [\n Math.round((bboxList[0] * width) / 1000),\n Math.round((bboxList[1] * height) / 1000),\n Math.round((bboxList[4] * width) / 1000),\n Math.round((bboxList[5] * height) / 1000),\n ];\n }\n\n const msg = `invalid bbox data for doubao-vision mode: ${JSON.stringify(bbox)} `;\n throw new Error(msg);\n}\n\nfunction normalizeBboxInput(\n bbox: AdaptBboxInput,\n): number[] | string[] | string {\n if (Array.isArray(bbox)) {\n if (Array.isArray(bbox[0])) {\n return bbox[0] as number[] | string[];\n }\n return bbox as number[] | string[];\n }\n return bbox as string;\n}\n\nexport function adaptBbox(\n bbox: AdaptBboxInput,\n width: number,\n height: number,\n rightLimit: number,\n bottomLimit: number,\n vlMode: TVlModeTypes | undefined,\n): [number, number, number, number] {\n const normalizedBbox = normalizeBboxInput(bbox);\n\n let result: [number, number, number, number] = [0, 0, 0, 0];\n if (vlMode === 'doubao-vision' || vlMode === 'vlm-ui-tars') {\n result = adaptDoubaoBbox(normalizedBbox, width, height);\n } else if (vlMode === 'gemini') {\n result = adaptGeminiBbox(normalizedBbox as number[], width, height);\n } else if (vlMode === 'qwen3-vl' || vlMode === 'glm-v') {\n result = normalized01000(normalizedBbox as number[], width, height);\n } else {\n result = adaptQwen2_5Bbox(normalizedBbox as number[]);\n }\n\n result[2] = Math.min(result[2], rightLimit);\n result[3] = Math.min(result[3], bottomLimit);\n\n return result;\n}\n\n// x1, y1, x2, y2 -> 0-1000\nexport function normalized01000(\n bbox: number[],\n width: number,\n height: number,\n): [number, number, number, number] {\n return [\n Math.round((bbox[0] * width) / 1000),\n Math.round((bbox[1] * height) / 1000),\n Math.round((bbox[2] * width) / 1000),\n Math.round((bbox[3] * height) / 1000),\n ];\n}\n\n// y1, x1, y2, x2 -> 0-1000\nexport function adaptGeminiBbox(\n bbox: number[],\n width: number,\n height: number,\n): [number, number, number, number] {\n const left = Math.round((bbox[1] * width) / 1000);\n const top = Math.round((bbox[0] * height) / 1000);\n const right = Math.round((bbox[3] * width) / 1000);\n const bottom = Math.round((bbox[2] * height) / 1000);\n return [left, top, right, bottom];\n}\n\nexport function adaptBboxToRect(\n bbox: number[],\n width: number,\n height: number,\n offsetX = 0,\n offsetY = 0,\n rightLimit = width,\n bottomLimit = height,\n vlMode?: TVlModeTypes | undefined,\n): Rect {\n debugInspectUtils(\n 'adaptBboxToRect',\n bbox,\n width,\n height,\n 'offset',\n offsetX,\n offsetY,\n 'limit',\n rightLimit,\n bottomLimit,\n 'vlMode',\n vlMode,\n );\n const [left, top, right, bottom] = adaptBbox(\n bbox,\n width,\n height,\n rightLimit,\n bottomLimit,\n vlMode,\n );\n\n // Calculate initial rect dimensions\n const rectLeft = left;\n const rectTop = top;\n let rectWidth = right - left;\n let rectHeight = bottom - top;\n\n // Ensure the rect doesn't exceed image boundaries\n // If right edge exceeds width, adjust the width\n if (rectLeft + rectWidth > width) {\n rectWidth = width - rectLeft;\n }\n\n // If bottom edge exceeds height, adjust the height\n if (rectTop + rectHeight > height) {\n rectHeight = height - rectTop;\n }\n\n // Ensure minimum dimensions (width and height should be at least 1)\n rectWidth = Math.max(1, rectWidth);\n rectHeight = Math.max(1, rectHeight);\n\n const rect = {\n left: rectLeft + offsetX,\n top: rectTop + offsetY,\n width: rectWidth,\n height: rectHeight,\n };\n debugInspectUtils('adaptBboxToRect, result=', rect);\n\n return rect;\n}\n\nexport function mergeRects(rects: Rect[]) {\n const minLeft = Math.min(...rects.map((r) => r.left));\n const minTop = Math.min(...rects.map((r) => r.top));\n const maxRight = Math.max(...rects.map((r) => r.left + r.width));\n const maxBottom = Math.max(...rects.map((r) => r.top + r.height));\n return {\n left: minLeft,\n top: minTop,\n width: maxRight - minLeft,\n height: maxBottom - minTop,\n };\n}\n\n// expand the search area to at least 300 x 300, or add a default padding\nexport function expandSearchArea(\n rect: Rect,\n screenSize: Size,\n vlMode: TVlModeTypes | undefined,\n) {\n let minEdgeSize = 500;\n if (vlMode === 'qwen3-vl') {\n minEdgeSize = 1200;\n }\n const defaultPadding = 160;\n\n // Calculate padding needed to reach minimum edge size\n const paddingSizeHorizontal =\n rect.width < minEdgeSize\n ? Math.ceil((minEdgeSize - rect.width) / 2)\n : defaultPadding;\n const paddingSizeVertical =\n rect.height < minEdgeSize\n ? Math.ceil((minEdgeSize - rect.height) / 2)\n : defaultPadding;\n\n // Calculate new dimensions (ensure minimum edge size)\n let newWidth = Math.max(minEdgeSize, rect.width + paddingSizeHorizontal * 2);\n let newHeight = Math.max(minEdgeSize, rect.height + paddingSizeVertical * 2);\n\n // Calculate initial position with padding\n let newLeft = rect.left - paddingSizeHorizontal;\n let newTop = rect.top - paddingSizeVertical;\n\n // Ensure the rect doesn't exceed screen boundaries by adjusting position\n // If the rect goes beyond the right edge, shift it left\n if (newLeft + newWidth > screenSize.width) {\n newLeft = screenSize.width - newWidth;\n }\n\n // If the rect goes beyond the bottom edge, shift it up\n if (newTop + newHeight > screenSize.height) {\n newTop = screenSize.height - newHeight;\n }\n\n // Ensure the rect doesn't go beyond the left/top edges\n newLeft = Math.max(0, newLeft);\n newTop = Math.max(0, newTop);\n\n // If after position adjustment, the rect still exceeds screen boundaries,\n // clamp the dimensions to fit within screen\n if (newLeft + newWidth > screenSize.width) {\n newWidth = screenSize.width - newLeft;\n }\n if (newTop + newHeight > screenSize.height) {\n newHeight = screenSize.height - newTop;\n }\n\n rect.left = newLeft;\n rect.top = newTop;\n rect.width = newWidth;\n rect.height = newHeight;\n\n return rect;\n}\n\nexport async function markupImageForLLM(\n screenshotBase64: string,\n tree: ElementTreeNode<BaseElement>,\n size: Size,\n) {\n const elementsInfo = treeToList(tree);\n const elementsPositionInfoWithoutText = elementsInfo!.filter(\n (elementInfo) => {\n if (elementInfo.attributes.nodeType === NodeType.TEXT) {\n return false;\n }\n return true;\n },\n );\n\n const imagePayload = await compositeElementInfoImg({\n inputImgBase64: screenshotBase64,\n elementsPositionInfo: elementsPositionInfoWithoutText,\n size,\n });\n return imagePayload;\n}\n\nexport function buildYamlFlowFromPlans(\n plans: PlanningAction[],\n actionSpace: DeviceAction<any>[],\n): MidsceneYamlFlowItem[] {\n const flow: MidsceneYamlFlowItem[] = [];\n\n for (const plan of plans) {\n const verb = plan.type;\n\n const action = actionSpace.find((action) => action.name === verb);\n if (!action) {\n console.warn(\n `Cannot convert action ${verb} to yaml flow. Will ignore it.`,\n );\n continue;\n }\n\n const flowKey = action.interfaceAlias || verb;\n const flowParam = action.paramSchema\n ? dumpActionParam(plan.param || {}, action.paramSchema)\n : {};\n\n const flowItem: MidsceneYamlFlowItem = {\n [flowKey]: '',\n ...flowParam,\n };\n\n flow.push(flowItem);\n }\n\n return flow;\n}\n\n// Zod schemas for shared types\nexport const PointSchema = z.object({\n left: z.number(),\n top: z.number(),\n});\n\nexport const SizeSchema = z.object({\n width: z.number(),\n height: z.number(),\n dpr: z.number().optional(),\n});\n\nexport const RectSchema = PointSchema.and(SizeSchema).and(\n z.object({\n zoom: z.number().optional(),\n }),\n);\n\n// Zod schema for TMultimodalPrompt\nexport const TMultimodalPromptSchema = z.object({\n images: z\n .array(\n z.object({\n name: z.string(),\n url: z.string(),\n }),\n )\n .optional(),\n convertHttpImage2Base64: z.boolean().optional(),\n});\n\n// Zod schema for TUserPrompt\nexport const TUserPromptSchema = z.union([\n z.string(),\n z\n .object({\n prompt: z.string(),\n })\n .and(TMultimodalPromptSchema.partial()),\n]);\n\n// Generate TypeScript types from Zod schemas\nexport type TMultimodalPrompt = z.infer<typeof TMultimodalPromptSchema>;\nexport type TUserPrompt = z.infer<typeof TUserPromptSchema>;\n\nconst locateFieldFlagName = 'midscene_location_field_flag';\n\n// Schema for locator field input (when users provide locate parameters)\nconst MidsceneLocationInput = z\n .object({\n prompt: TUserPromptSchema,\n deepThink: z.boolean().optional(),\n cacheable: z.boolean().optional(),\n xpath: z.union([z.string(), z.boolean()]).optional(),\n })\n .passthrough();\n\n// Schema for locator field result (when AI returns locate results)\nconst MidsceneLocationResult = z\n .object({\n [locateFieldFlagName]: z.literal(true),\n prompt: TUserPromptSchema,\n\n // optional fields\n deepThink: z.boolean().optional(), // only available in vl model\n cacheable: z.boolean().optional(),\n xpath: z.boolean().optional(), // preset result for xpath\n\n // these two fields will only appear in the result\n center: z.tuple([z.number(), z.number()]),\n rect: RectSchema,\n })\n .passthrough();\n\n// Export the result type - this is used for runtime results that include center and rect\nexport type MidsceneLocationResultType = z.infer<typeof MidsceneLocationResult>;\n\n// Export the input type - this is the inferred type from getMidsceneLocationSchema()\nexport type MidsceneLocationInputType = z.infer<typeof MidsceneLocationInput>;\n\n/**\n * Returns the schema for locator fields.\n * This now returns the input schema which is more permissive and suitable for validation.\n */\nexport const getMidsceneLocationSchema = () => {\n return MidsceneLocationInput;\n};\n\nexport const ifMidsceneLocatorField = (field: any): boolean => {\n // Handle optional fields by getting the inner type\n let actualField = field;\n if (actualField._def?.typeName === 'ZodOptional') {\n actualField = actualField._def.innerType;\n }\n\n // Check if this is a ZodObject\n if (actualField._def?.typeName === 'ZodObject') {\n const shape = actualField._def.shape();\n\n // Method 1: Check for the location field flag (for result schema)\n if (locateFieldFlagName in shape) {\n return true;\n }\n\n // Method 2: Check if it's the input schema by checking for 'prompt' field\n // Input schema has 'prompt' as a required field\n if ('prompt' in shape && shape.prompt) {\n return true;\n }\n }\n\n return false;\n};\n\nexport const dumpMidsceneLocatorField = (field: any): string => {\n assert(\n ifMidsceneLocatorField(field),\n 'field is not a midscene locator field',\n );\n\n // If field is a string, return it directly\n if (typeof field === 'string') {\n return field;\n }\n\n // If field is an object with prompt property\n if (field && typeof field === 'object' && field.prompt) {\n // If prompt is a string, return it directly\n if (typeof field.prompt === 'string') {\n return field.prompt;\n }\n // If prompt is a TUserPrompt object, extract the prompt string\n if (typeof field.prompt === 'object' && field.prompt.prompt) {\n return field.prompt.prompt; // TODO: dump images if necessary\n }\n }\n\n // Fallback: try to convert to string\n return String(field);\n};\n\nexport const findAllMidsceneLocatorField = (\n zodType?: z.ZodType<any>,\n requiredOnly?: boolean,\n): string[] => {\n if (!zodType) {\n return [];\n }\n\n // Check if this is a ZodObject by checking if it has a shape property\n const zodObject = zodType as any;\n if (zodObject._def?.typeName === 'ZodObject' && zodObject.shape) {\n const keys = Object.keys(zodObject.shape);\n return keys.filter((key) => {\n const field = zodObject.shape[key];\n if (!ifMidsceneLocatorField(field)) {\n return false;\n }\n\n // If requiredOnly is true, filter out optional fields\n if (requiredOnly) {\n return field._def?.typeName !== 'ZodOptional';\n }\n\n return true;\n });\n }\n\n // For other ZodType instances, we can't extract field names\n return [];\n};\n\nexport const dumpActionParam = (\n jsonObject: Record<string, any>,\n zodSchema: z.ZodType<any>,\n): Record<string, any> => {\n // Prevent spreading strings into {0: 'c', 1: 'o', ...}\n if (!isPlainObject(jsonObject)) {\n return {};\n }\n\n const locatorFields = findAllMidsceneLocatorField(zodSchema);\n const result = { ...jsonObject };\n\n for (const fieldName of locatorFields) {\n const fieldValue = result[fieldName];\n if (fieldValue) {\n // If it's already a string, keep it as is\n if (typeof fieldValue === 'string') {\n result[fieldName] = fieldValue;\n } else if (typeof fieldValue === 'object') {\n // Check if this field is actually a MidsceneLocationType object\n if (fieldValue.prompt) {\n // If prompt is a string, use it directly\n if (typeof fieldValue.prompt === 'string') {\n result[fieldName] = fieldValue.prompt;\n } else if (\n typeof fieldValue.prompt === 'object' &&\n fieldValue.prompt.prompt\n ) {\n // If prompt is a TUserPrompt object, extract the prompt string\n result[fieldName] = fieldValue.prompt.prompt;\n }\n }\n }\n }\n }\n\n return result;\n};\n\nexport const loadActionParam = (\n jsonObject: Record<string, any>,\n zodSchema: z.ZodType<any>,\n): Record<string, any> => {\n const locatorFields = findAllMidsceneLocatorField(zodSchema);\n const result = { ...jsonObject };\n\n for (const fieldName of locatorFields) {\n const fieldValue = result[fieldName];\n if (fieldValue && typeof fieldValue === 'string') {\n result[fieldName] = {\n [locateFieldFlagName]: true,\n prompt: fieldValue,\n };\n }\n }\n\n return result;\n};\n\n/**\n * Parse and validate action parameters using Zod schema.\n * All fields are validated through Zod, EXCEPT locator fields which are skipped.\n * Default values defined in the schema are automatically applied.\n *\n * Locator fields are special business logic fields with complex validation requirements,\n * so they are intentionally excluded from Zod parsing and use existing validation logic.\n */\nexport const parseActionParam = (\n rawParam: Record<string, any> | undefined,\n zodSchema?: z.ZodType<any>,\n): Record<string, any> | undefined => {\n // If no schema is provided, return undefined (action takes no parameters)\n if (!zodSchema) {\n return undefined;\n }\n\n // Handle undefined or null rawParam by providing an empty object\n const param = rawParam ?? {};\n\n // Find all locate fields in the schema\n const locateFields = findAllMidsceneLocatorField(zodSchema);\n\n // If there are no locate fields, just do normal validation\n if (locateFields.length === 0) {\n return zodSchema.parse(param);\n }\n\n // Extract locate field values to restore later\n const locateFieldValues: Record<string, any> = {};\n for (const fieldName of locateFields) {\n if (fieldName in param) {\n locateFieldValues[fieldName] = param[fieldName];\n }\n }\n\n // Build params for validation - skip locate fields and use dummy values\n const paramsForValidation: Record<string, any> = {};\n for (const key in param) {\n if (locateFields.includes(key)) {\n // Use dummy value to satisfy schema validation\n paramsForValidation[key] = { prompt: '_dummy_' };\n } else {\n paramsForValidation[key] = param[key];\n }\n }\n\n // Validate with dummy locate values\n const validated = zodSchema.parse(paramsForValidation);\n\n // Restore the actual locate field values (unvalidated, as per business requirement)\n for (const fieldName in locateFieldValues) {\n validated[fieldName] = locateFieldValues[fieldName];\n }\n\n return validated;\n};\n\nexport const finalizeActionName = 'Finalize';\n\n/**\n * Get a readable time string for the current time\n * @param format - Optional format string. Supports: YYYY, MM, DD, HH, mm, ss. Default: 'YYYY-MM-DD HH:mm:ss'\n * @returns A formatted time string with format label\n */\nexport const getReadableTimeString = (\n format = 'YYYY-MM-DD HH:mm:ss',\n): string => {\n const now = new Date();\n const year = now.getFullYear();\n const month = String(now.getMonth() + 1).padStart(2, '0');\n const day = String(now.getDate()).padStart(2, '0');\n const hours = String(now.getHours()).padStart(2, '0');\n const minutes = String(now.getMinutes()).padStart(2, '0');\n const seconds = String(now.getSeconds()).padStart(2, '0');\n\n const timeString = format\n .replace('YYYY', String(year))\n .replace('MM', month)\n .replace('DD', day)\n .replace('HH', hours)\n .replace('mm', minutes)\n .replace('ss', seconds);\n\n return `${timeString} (${format})`;\n};\n"],"names":["defaultBboxSize","debugInspectUtils","getDebug","fillBboxParam","locate","width","height","rightLimit","bottomLimit","vlMode","adaptBbox","adaptQwen2_5Bbox","bbox","msg","JSON","Error","result","Math","adaptDoubaoBbox","assert","splitted","Number","bboxList","Array","item","x","y","normalizeBboxInput","normalizedBbox","adaptGeminiBbox","normalized01000","left","top","right","bottom","adaptBboxToRect","offsetX","offsetY","rectLeft","rectTop","rectWidth","rectHeight","rect","mergeRects","rects","minLeft","r","minTop","maxRight","maxBottom","expandSearchArea","screenSize","minEdgeSize","defaultPadding","paddingSizeHorizontal","paddingSizeVertical","newWidth","newHeight","newLeft","newTop","markupImageForLLM","screenshotBase64","tree","size","elementsInfo","treeToList","elementsPositionInfoWithoutText","elementInfo","NodeType","imagePayload","compositeElementInfoImg","buildYamlFlowFromPlans","plans","actionSpace","flow","plan","verb","action","console","flowKey","flowParam","dumpActionParam","flowItem","PointSchema","z","SizeSchema","RectSchema","TMultimodalPromptSchema","TUserPromptSchema","locateFieldFlagName","MidsceneLocationInput","getMidsceneLocationSchema","ifMidsceneLocatorField","field","actualField","shape","dumpMidsceneLocatorField","String","findAllMidsceneLocatorField","zodType","requiredOnly","zodObject","keys","Object","key","jsonObject","zodSchema","isPlainObject","locatorFields","fieldName","fieldValue","loadActionParam","parseActionParam","rawParam","param","locateFields","locateFieldValues","paramsForValidation","validated","finalizeActionName","getReadableTimeString","format","now","Date","year","month","day","hours","minutes","seconds","timeString"],"mappings":";;;;;;AAuBA,MAAMA,kBAAkB;AACxB,MAAMC,oBAAoBC,SAAS;AAI5B,SAASC,cACdC,MAA2B,EAC3BC,KAAa,EACbC,MAAc,EACdC,UAAkB,EAClBC,WAAmB,EACnBC,MAAgC;IAGhC,IAAKL,OAAe,OAAO,IAAI,CAACA,QAAQ,MAAM;QAC5CA,OAAO,IAAI,GAAIA,OAAe,OAAO;QAErC,OAAQA,OAAe,OAAO;IAChC;IAEA,IAAIA,QAAQ,MACVA,OAAO,IAAI,GAAGM,UACZN,OAAO,IAAI,EACXC,OACAC,QACAC,YACAC,aACAC;IAIJ,OAAOL;AACT;AAEO,SAASO,iBACdC,IAAc;IAEd,IAAIA,KAAK,MAAM,GAAG,GAAG;QACnB,MAAMC,MAAM,CAAC,oCAAoC,EAAEC,KAAK,SAAS,CAACF,MAAM,CAAC,CAAC;QAC1E,MAAM,IAAIG,MAAMF;IAClB;IAEA,MAAMG,SAA2C;QAC/CC,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE;QAClBK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE;QACC,YAAnB,OAAOA,IAAI,CAAC,EAAE,GACVK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,IAClBK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,GAAGZ;QACN,YAAnB,OAAOY,IAAI,CAAC,EAAE,GACVK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,IAClBK,KAAK,KAAK,CAACL,IAAI,CAAC,EAAE,GAAGZ;KAC1B;IACD,OAAOgB;AACT;AAEO,SAASE,gBACdN,IAAkC,EAClCP,KAAa,EACbC,MAAc;IAEda,OACEd,QAAQ,KAAKC,SAAS,GACtB;IAGF,IAAI,AAAgB,YAAhB,OAAOM,MAAmB;QAC5BO,OACE,+BAA+B,IAAI,CAACP,KAAK,IAAI,KAC7C,CAAC,iDAAiD,EAAEA,MAAM;QAE5D,MAAMQ,WAAWR,KAAK,KAAK,CAAC;QAC5B,IAAIQ,AAAoB,MAApBA,SAAS,MAAM,EACjB,OAAO;YACLH,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAIf,QAAS;YAC3CY,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAId,SAAU;YAC5CW,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAIf,QAAS;YAC3CY,KAAK,KAAK,CAAEI,OAAOD,QAAQ,CAAC,EAAE,IAAId,SAAU;SAC7C;QAEH,MAAM,IAAIS,MAAM,CAAC,iDAAiD,EAAEH,MAAM;IAC5E;IAEA,IAAIU,WAAqB,EAAE;IAC3B,IAAIC,MAAM,OAAO,CAACX,SAAS,AAAmB,YAAnB,OAAOA,IAAI,CAAC,EAAE,EACvCA,KAAK,OAAO,CAAC,CAACY;QACZ,IAAI,AAAgB,YAAhB,OAAOA,QAAqBA,KAAK,QAAQ,CAAC,MAAM;YAClD,MAAM,CAACC,GAAGC,EAAE,GAAGF,KAAK,KAAK,CAAC;YAC1BF,SAAS,IAAI,CAACD,OAAOI,EAAE,IAAI,KAAKJ,OAAOK,EAAE,IAAI;QAC/C,OAAO,IAAI,AAAgB,YAAhB,OAAOF,QAAqBA,KAAK,QAAQ,CAAC,MAAM;YACzD,MAAM,CAACC,GAAGC,EAAE,GAAGF,KAAK,KAAK,CAAC;YAC1BF,SAAS,IAAI,CAACD,OAAOI,EAAE,IAAI,KAAKJ,OAAOK,EAAE,IAAI;QAC/C,OACEJ,SAAS,IAAI,CAACD,OAAOG;IAEzB;SAEAF,WAAWV;IAGb,IAAIU,AAAoB,MAApBA,SAAS,MAAM,IAAUA,AAAoB,MAApBA,SAAS,MAAM,EAC1C,OAAO;QACLL,KAAK,KAAK,CAAEK,QAAQ,CAAC,EAAE,GAAGjB,QAAS;QACnCY,KAAK,KAAK,CAAEK,QAAQ,CAAC,EAAE,GAAGhB,SAAU;QACpCW,KAAK,KAAK,CAAEK,QAAQ,CAAC,EAAE,GAAGjB,QAAS;QACnCY,KAAK,KAAK,CAAEK,QAAQ,CAAC,EAAE,GAAGhB,SAAU;KACrC;IAIH,IACEgB,AAAoB,MAApBA,SAAS,MAAM,IACfA,AAAoB,MAApBA,SAAS,MAAM,IACfA,AAAoB,MAApBA,SAAS,MAAM,IACfA,AAAoB,MAApBA,SAAS,MAAM,EAEf,OAAO;QACLL,KAAK,GAAG,CACN,GACAA,KAAK,KAAK,CAAEK,QAAQ,CAAC,EAAE,GAAGjB,QAAS,QAAQL,kBAAkB;QAE/DiB,KAAK,GAAG,CACN,GACAA,KAAK,KAAK,CAAEK,QAAQ,CAAC,EAAE,GAAGhB,SAAU,QAAQN,kBAAkB;QAEhEiB,KAAK,GAAG,CACNZ,OACAY,KAAK,KAAK,CAAEK,QAAQ,CAAC,EAAE,GAAGjB,QAAS,QAAQL,kBAAkB;QAE/DiB,KAAK,GAAG,CACNX,QACAW,KAAK,KAAK,CAAEK,QAAQ,CAAC,EAAE,GAAGhB,SAAU,QAAQN,kBAAkB;KAEjE;IAGH,IAAIY,AAAgB,MAAhBA,KAAK,MAAM,EACb,OAAO;QACLK,KAAK,KAAK,CAAEK,QAAQ,CAAC,EAAE,GAAGjB,QAAS;QACnCY,KAAK,KAAK,CAAEK,QAAQ,CAAC,EAAE,GAAGhB,SAAU;QACpCW,KAAK,KAAK,CAAEK,QAAQ,CAAC,EAAE,GAAGjB,QAAS;QACnCY,KAAK,KAAK,CAAEK,QAAQ,CAAC,EAAE,GAAGhB,SAAU;KACrC;IAGH,MAAMO,MAAM,CAAC,0CAA0C,EAAEC,KAAK,SAAS,CAACF,MAAM,CAAC,CAAC;IAChF,MAAM,IAAIG,MAAMF;AAClB;AAEA,SAASc,mBACPf,IAAoB;IAEpB,IAAIW,MAAM,OAAO,CAACX,OAChB;QAAA,IAAIW,MAAM,OAAO,CAACX,IAAI,CAAC,EAAE,GACvB,OAAOA,IAAI,CAAC,EAAE;IAChB;IAGF,OAAOA;AACT;AAEO,SAASF,UACdE,IAAoB,EACpBP,KAAa,EACbC,MAAc,EACdC,UAAkB,EAClBC,WAAmB,EACnBC,MAAgC;IAEhC,MAAMmB,iBAAiBD,mBAAmBf;IAE1C,IAAII,SAA2C;QAAC;QAAG;QAAG;QAAG;KAAE;IAEzDA,SADEP,AAAW,oBAAXA,UAA8BA,AAAW,kBAAXA,SACvBS,gBAAgBU,gBAAgBvB,OAAOC,UACvCG,AAAW,aAAXA,SACAoB,gBAAgBD,gBAA4BvB,OAAOC,UACnDG,AAAW,eAAXA,UAAyBA,AAAW,YAAXA,SACzBqB,gBAAgBF,gBAA4BvB,OAAOC,UAEnDK,iBAAiBiB;IAG5BZ,MAAM,CAAC,EAAE,GAAGC,KAAK,GAAG,CAACD,MAAM,CAAC,EAAE,EAAET;IAChCS,MAAM,CAAC,EAAE,GAAGC,KAAK,GAAG,CAACD,MAAM,CAAC,EAAE,EAAER;IAEhC,OAAOQ;AACT;AAGO,SAASc,gBACdlB,IAAc,EACdP,KAAa,EACbC,MAAc;IAEd,OAAO;QACLW,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGP,QAAS;QAC/BY,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGN,SAAU;QAChCW,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGP,QAAS;QAC/BY,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGN,SAAU;KACjC;AACH;AAGO,SAASuB,gBACdjB,IAAc,EACdP,KAAa,EACbC,MAAc;IAEd,MAAMyB,OAAOd,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGP,QAAS;IAC5C,MAAM2B,MAAMf,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGN,SAAU;IAC5C,MAAM2B,QAAQhB,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGP,QAAS;IAC7C,MAAM6B,SAASjB,KAAK,KAAK,CAAEL,IAAI,CAAC,EAAE,GAAGN,SAAU;IAC/C,OAAO;QAACyB;QAAMC;QAAKC;QAAOC;KAAO;AACnC;AAEO,SAASC,gBACdvB,IAAc,EACdP,KAAa,EACbC,MAAc,EACd8B,UAAU,CAAC,EACXC,UAAU,CAAC,EACX9B,aAAaF,KAAK,EAClBG,cAAcF,MAAM,EACpBG,MAAiC;IAEjCR,kBACE,mBACAW,MACAP,OACAC,QACA,UACA8B,SACAC,SACA,SACA9B,YACAC,aACA,UACAC;IAEF,MAAM,CAACsB,MAAMC,KAAKC,OAAOC,OAAO,GAAGxB,UACjCE,MACAP,OACAC,QACAC,YACAC,aACAC;IAIF,MAAM6B,WAAWP;IACjB,MAAMQ,UAAUP;IAChB,IAAIQ,YAAYP,QAAQF;IACxB,IAAIU,aAAaP,SAASF;IAI1B,IAAIM,WAAWE,YAAYnC,OACzBmC,YAAYnC,QAAQiC;IAItB,IAAIC,UAAUE,aAAanC,QACzBmC,aAAanC,SAASiC;IAIxBC,YAAYvB,KAAK,GAAG,CAAC,GAAGuB;IACxBC,aAAaxB,KAAK,GAAG,CAAC,GAAGwB;IAEzB,MAAMC,OAAO;QACX,MAAMJ,WAAWF;QACjB,KAAKG,UAAUF;QACf,OAAOG;QACP,QAAQC;IACV;IACAxC,kBAAkB,4BAA4ByC;IAE9C,OAAOA;AACT;AAEO,SAASC,WAAWC,KAAa;IACtC,MAAMC,UAAU5B,KAAK,GAAG,IAAI2B,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,IAAI;IACnD,MAAMC,SAAS9B,KAAK,GAAG,IAAI2B,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,GAAG;IACjD,MAAME,WAAW/B,KAAK,GAAG,IAAI2B,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,IAAI,GAAGA,EAAE,KAAK;IAC9D,MAAMG,YAAYhC,KAAK,GAAG,IAAI2B,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,GAAG,GAAGA,EAAE,MAAM;IAC/D,OAAO;QACL,MAAMD;QACN,KAAKE;QACL,OAAOC,WAAWH;QAClB,QAAQI,YAAYF;IACtB;AACF;AAGO,SAASG,iBACdR,IAAU,EACVS,UAAgB,EAChB1C,MAAgC;IAEhC,IAAI2C,cAAc;IAClB,IAAI3C,AAAW,eAAXA,QACF2C,cAAc;IAEhB,MAAMC,iBAAiB;IAGvB,MAAMC,wBACJZ,KAAK,KAAK,GAAGU,cACTnC,KAAK,IAAI,CAAEmC,AAAAA,CAAAA,cAAcV,KAAK,KAAI,IAAK,KACvCW;IACN,MAAME,sBACJb,KAAK,MAAM,GAAGU,cACVnC,KAAK,IAAI,CAAEmC,AAAAA,CAAAA,cAAcV,KAAK,MAAK,IAAK,KACxCW;IAGN,IAAIG,WAAWvC,KAAK,GAAG,CAACmC,aAAaV,KAAK,KAAK,GAAGY,AAAwB,IAAxBA;IAClD,IAAIG,YAAYxC,KAAK,GAAG,CAACmC,aAAaV,KAAK,MAAM,GAAGa,AAAsB,IAAtBA;IAGpD,IAAIG,UAAUhB,KAAK,IAAI,GAAGY;IAC1B,IAAIK,SAASjB,KAAK,GAAG,GAAGa;IAIxB,IAAIG,UAAUF,WAAWL,WAAW,KAAK,EACvCO,UAAUP,WAAW,KAAK,GAAGK;IAI/B,IAAIG,SAASF,YAAYN,WAAW,MAAM,EACxCQ,SAASR,WAAW,MAAM,GAAGM;IAI/BC,UAAUzC,KAAK,GAAG,CAAC,GAAGyC;IACtBC,SAAS1C,KAAK,GAAG,CAAC,GAAG0C;IAIrB,IAAID,UAAUF,WAAWL,WAAW,KAAK,EACvCK,WAAWL,WAAW,KAAK,GAAGO;IAEhC,IAAIC,SAASF,YAAYN,WAAW,MAAM,EACxCM,YAAYN,WAAW,MAAM,GAAGQ;IAGlCjB,KAAK,IAAI,GAAGgB;IACZhB,KAAK,GAAG,GAAGiB;IACXjB,KAAK,KAAK,GAAGc;IACbd,KAAK,MAAM,GAAGe;IAEd,OAAOf;AACT;AAEO,eAAekB,kBACpBC,gBAAwB,EACxBC,IAAkC,EAClCC,IAAU;IAEV,MAAMC,eAAeC,WAAWH;IAChC,MAAMI,kCAAkCF,aAAc,MAAM,CAC1D,CAACG;QACC,IAAIA,YAAY,UAAU,CAAC,QAAQ,KAAKC,SAAS,IAAI,EACnD,OAAO;QAET,OAAO;IACT;IAGF,MAAMC,eAAe,MAAMC,wBAAwB;QACjD,gBAAgBT;QAChB,sBAAsBK;QACtBH;IACF;IACA,OAAOM;AACT;AAEO,SAASE,uBACdC,KAAuB,EACvBC,WAAgC;IAEhC,MAAMC,OAA+B,EAAE;IAEvC,KAAK,MAAMC,QAAQH,MAAO;QACxB,MAAMI,OAAOD,KAAK,IAAI;QAEtB,MAAME,SAASJ,YAAY,IAAI,CAAC,CAACI,SAAWA,OAAO,IAAI,KAAKD;QAC5D,IAAI,CAACC,QAAQ;YACXC,QAAQ,IAAI,CACV,CAAC,sBAAsB,EAAEF,KAAK,8BAA8B,CAAC;YAE/D;QACF;QAEA,MAAMG,UAAUF,OAAO,cAAc,IAAID;QACzC,MAAMI,YAAYH,OAAO,WAAW,GAChCI,gBAAgBN,KAAK,KAAK,IAAI,CAAC,GAAGE,OAAO,WAAW,IACpD,CAAC;QAEL,MAAMK,WAAiC;YACrC,CAACH,QAAQ,EAAE;YACX,GAAGC,SAAS;QACd;QAEAN,KAAK,IAAI,CAACQ;IACZ;IAEA,OAAOR;AACT;AAGO,MAAMS,cAAcC,EAAE,MAAM,CAAC;IAClC,MAAMA,EAAE,MAAM;IACd,KAAKA,EAAE,MAAM;AACf;AAEO,MAAMC,aAAaD,EAAE,MAAM,CAAC;IACjC,OAAOA,EAAE,MAAM;IACf,QAAQA,EAAE,MAAM;IAChB,KAAKA,EAAE,MAAM,GAAG,QAAQ;AAC1B;AAEO,MAAME,aAAaH,YAAY,GAAG,CAACE,YAAY,GAAG,CACvDD,EAAE,MAAM,CAAC;IACP,MAAMA,EAAE,MAAM,GAAG,QAAQ;AAC3B;AAIK,MAAMG,0BAA0BH,EAAE,MAAM,CAAC;IAC9C,QAAQA,EAAAA,KACA,CACJA,EAAE,MAAM,CAAC;QACP,MAAMA,EAAE,MAAM;QACd,KAAKA,EAAE,MAAM;IACf,IAED,QAAQ;IACX,yBAAyBA,EAAE,OAAO,GAAG,QAAQ;AAC/C;AAGO,MAAMI,oBAAoBJ,EAAE,KAAK,CAAC;IACvCA,EAAE,MAAM;IACRA,EAAAA,MACS,CAAC;QACN,QAAQA,EAAE,MAAM;IAClB,GACC,GAAG,CAACG,wBAAwB,OAAO;CACvC;AAMD,MAAME,sBAAsB;AAG5B,MAAMC,wBAAwBN,EAAAA,MACrB,CAAC;IACN,QAAQI;IACR,WAAWJ,EAAE,OAAO,GAAG,QAAQ;IAC/B,WAAWA,EAAE,OAAO,GAAG,QAAQ;IAC/B,OAAOA,EAAE,KAAK,CAAC;QAACA,EAAE,MAAM;QAAIA,EAAE,OAAO;KAAG,EAAE,QAAQ;AACpD,GACC,WAAW;AAGiBA,EAAAA,MACtB,CAAC;IACN,CAACK,oBAAoB,EAAEL,EAAE,OAAO,CAAC;IACjC,QAAQI;IAGR,WAAWJ,EAAE,OAAO,GAAG,QAAQ;IAC/B,WAAWA,EAAE,OAAO,GAAG,QAAQ;IAC/B,OAAOA,EAAE,OAAO,GAAG,QAAQ;IAG3B,QAAQA,EAAE,KAAK,CAAC;QAACA,EAAE,MAAM;QAAIA,EAAE,MAAM;KAAG;IACxC,MAAME;AACR,GACC,WAAW;AAYP,MAAMK,4BAA4B,IAChCD;AAGF,MAAME,yBAAyB,CAACC;IAErC,IAAIC,cAAcD;IAClB,IAAIC,YAAY,IAAI,EAAE,aAAa,eACjCA,cAAcA,YAAY,IAAI,CAAC,SAAS;IAI1C,IAAIA,YAAY,IAAI,EAAE,aAAa,aAAa;QAC9C,MAAMC,QAAQD,YAAY,IAAI,CAAC,KAAK;QAGpC,IAAIL,uBAAuBM,OACzB,OAAO;QAKT,IAAI,YAAYA,SAASA,MAAM,MAAM,EACnC,OAAO;IAEX;IAEA,OAAO;AACT;AAEO,MAAMC,2BAA2B,CAACH;IACvC1E,OACEyE,uBAAuBC,QACvB;IAIF,IAAI,AAAiB,YAAjB,OAAOA,OACT,OAAOA;IAIT,IAAIA,SAAS,AAAiB,YAAjB,OAAOA,SAAsBA,MAAM,MAAM,EAAE;QAEtD,IAAI,AAAwB,YAAxB,OAAOA,MAAM,MAAM,EACrB,OAAOA,MAAM,MAAM;QAGrB,IAAI,AAAwB,YAAxB,OAAOA,MAAM,MAAM,IAAiBA,MAAM,MAAM,CAAC,MAAM,EACzD,OAAOA,MAAM,MAAM,CAAC,MAAM;IAE9B;IAGA,OAAOI,OAAOJ;AAChB;AAEO,MAAMK,8BAA8B,CACzCC,SACAC;IAEA,IAAI,CAACD,SACH,OAAO,EAAE;IAIX,MAAME,YAAYF;IAClB,IAAIE,UAAU,IAAI,EAAE,aAAa,eAAeA,UAAU,KAAK,EAAE;QAC/D,MAAMC,OAAOC,OAAO,IAAI,CAACF,UAAU,KAAK;QACxC,OAAOC,KAAK,MAAM,CAAC,CAACE;YAClB,MAAMX,QAAQQ,UAAU,KAAK,CAACG,IAAI;YAClC,IAAI,CAACZ,uBAAuBC,QAC1B,OAAO;YAIT,IAAIO,cACF,OAAOP,MAAM,IAAI,EAAE,aAAa;YAGlC,OAAO;QACT;IACF;IAGA,OAAO,EAAE;AACX;AAEO,MAAMZ,kBAAkB,CAC7BwB,YACAC;IAGA,IAAI,CAACC,cAAcF,aACjB,OAAO,CAAC;IAGV,MAAMG,gBAAgBV,4BAA4BQ;IAClD,MAAM1F,SAAS;QAAE,GAAGyF,UAAU;IAAC;IAE/B,KAAK,MAAMI,aAAaD,cAAe;QACrC,MAAME,aAAa9F,MAAM,CAAC6F,UAAU;QACpC,IAAIC,YAEF;YAAA,IAAI,AAAsB,YAAtB,OAAOA,YACT9F,MAAM,CAAC6F,UAAU,GAAGC;iBACf,IAAI,AAAsB,YAAtB,OAAOA,YAEhB;gBAAA,IAAIA,WAAW,MAAM,EAEnB;oBAAA,IAAI,AAA6B,YAA7B,OAAOA,WAAW,MAAM,EAC1B9F,MAAM,CAAC6F,UAAU,GAAGC,WAAW,MAAM;yBAChC,IACL,AAA6B,YAA7B,OAAOA,WAAW,MAAM,IACxBA,WAAW,MAAM,CAAC,MAAM,EAGxB9F,MAAM,CAAC6F,UAAU,GAAGC,WAAW,MAAM,CAAC,MAAM;gBAC9C;YACF;QACF;IAEJ;IAEA,OAAO9F;AACT;AAEO,MAAM+F,kBAAkB,CAC7BN,YACAC;IAEA,MAAME,gBAAgBV,4BAA4BQ;IAClD,MAAM1F,SAAS;QAAE,GAAGyF,UAAU;IAAC;IAE/B,KAAK,MAAMI,aAAaD,cAAe;QACrC,MAAME,aAAa9F,MAAM,CAAC6F,UAAU;QACpC,IAAIC,cAAc,AAAsB,YAAtB,OAAOA,YACvB9F,MAAM,CAAC6F,UAAU,GAAG;YAClB,CAACpB,oBAAoB,EAAE;YACvB,QAAQqB;QACV;IAEJ;IAEA,OAAO9F;AACT;AAUO,MAAMgG,mBAAmB,CAC9BC,UACAP;IAGA,IAAI,CAACA,WACH;IAIF,MAAMQ,QAAQD,YAAY,CAAC;IAG3B,MAAME,eAAejB,4BAA4BQ;IAGjD,IAAIS,AAAwB,MAAxBA,aAAa,MAAM,EACrB,OAAOT,UAAU,KAAK,CAACQ;IAIzB,MAAME,oBAAyC,CAAC;IAChD,KAAK,MAAMP,aAAaM,aACtB,IAAIN,aAAaK,OACfE,iBAAiB,CAACP,UAAU,GAAGK,KAAK,CAACL,UAAU;IAKnD,MAAMQ,sBAA2C,CAAC;IAClD,IAAK,MAAMb,OAAOU,MAChB,IAAIC,aAAa,QAAQ,CAACX,MAExBa,mBAAmB,CAACb,IAAI,GAAG;QAAE,QAAQ;IAAU;SAE/Ca,mBAAmB,CAACb,IAAI,GAAGU,KAAK,CAACV,IAAI;IAKzC,MAAMc,YAAYZ,UAAU,KAAK,CAACW;IAGlC,IAAK,MAAMR,aAAaO,kBACtBE,SAAS,CAACT,UAAU,GAAGO,iBAAiB,CAACP,UAAU;IAGrD,OAAOS;AACT;AAEO,MAAMC,qBAAqB;AAO3B,MAAMC,wBAAwB,CACnCC,SAAS,qBAAqB;IAE9B,MAAMC,MAAM,IAAIC;IAChB,MAAMC,OAAOF,IAAI,WAAW;IAC5B,MAAMG,QAAQ5B,OAAOyB,IAAI,QAAQ,KAAK,GAAG,QAAQ,CAAC,GAAG;IACrD,MAAMI,MAAM7B,OAAOyB,IAAI,OAAO,IAAI,QAAQ,CAAC,GAAG;IAC9C,MAAMK,QAAQ9B,OAAOyB,IAAI,QAAQ,IAAI,QAAQ,CAAC,GAAG;IACjD,MAAMM,UAAU/B,OAAOyB,IAAI,UAAU,IAAI,QAAQ,CAAC,GAAG;IACrD,MAAMO,UAAUhC,OAAOyB,IAAI,UAAU,IAAI,QAAQ,CAAC,GAAG;IAErD,MAAMQ,aAAaT,OAChB,OAAO,CAAC,QAAQxB,OAAO2B,OACvB,OAAO,CAAC,MAAMC,OACd,OAAO,CAAC,MAAMC,KACd,OAAO,CAAC,MAAMC,OACd,OAAO,CAAC,MAAMC,SACd,OAAO,CAAC,MAAMC;IAEjB,OAAO,GAAGC,WAAW,EAAE,EAAET,OAAO,CAAC,CAAC;AACpC"}
|
|
1
|
+
{"version":3,"file":"common.mjs","sources":["../../src/common.ts"],"sourcesContent":["import type {\n BaseElement,\n DeviceAction,\n ElementTreeNode,\n MidsceneYamlFlowItem,\n PlanningAction,\n Rect,\n Size,\n} from '@/types';\nimport { assert, isPlainObject } from '@midscene/shared/utils';\n\nimport type { ChatCompletionMessageParam } from 'openai/resources/index';\n\nimport type { PlanningLocateParam } from '@/types';\nimport { NodeType } from '@midscene/shared/constants';\nimport type { TVlModeTypes } from '@midscene/shared/env';\nimport { treeToList } from '@midscene/shared/extractor';\nimport { compositeElementInfoImg } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { z } from 'zod';\n\nexport type AIArgs = ChatCompletionMessageParam[];\n\nconst defaultBboxSize = 20; // must be even number\nconst debugInspectUtils = getDebug('ai:common');\ntype AdaptBboxInput = number[] | string[] | string | (number[] | string[])[];\n\n/**\n * Convert a point coordinate [0, 1000] to a small bbox [0, 1000]\n * Creates a small bbox around the center point in the same coordinate space\n *\n * @param x - X coordinate in [0, 1000] range\n * @param y - Y coordinate in [0, 1000] range\n * @param bboxSize - Size of the bbox to create (default: 20)\n * @returns [x1, y1, x2, y2] bbox in [0, 1000] coordinate space\n */\nexport function pointToBbox(\n x: number,\n y: number,\n bboxSize = defaultBboxSize,\n): [number, number, number, number] {\n const halfSize = bboxSize / 2;\n const x1 = Math.max(x - halfSize, 0);\n const y1 = Math.max(y - halfSize, 0);\n const x2 = Math.min(x + halfSize, 1000);\n const y2 = Math.min(y + halfSize, 1000);\n\n return [x1, y1, x2, y2];\n}\n\n// transform the param of locate from qwen mode\nexport function fillBboxParam(\n locate: PlanningLocateParam,\n width: number,\n height: number,\n rightLimit: number,\n bottomLimit: number,\n vlMode: TVlModeTypes | undefined,\n) {\n // The Qwen model might have hallucinations of naming bbox as bbox_2d.\n if ((locate as any).bbox_2d && !locate?.bbox) {\n locate.bbox = (locate as any).bbox_2d;\n // biome-ignore lint/performance/noDelete: <explanation>\n delete (locate as any).bbox_2d;\n }\n\n if (locate?.bbox) {\n locate.bbox = adaptBbox(\n locate.bbox,\n width,\n height,\n rightLimit,\n bottomLimit,\n vlMode,\n );\n }\n\n return locate;\n}\n\nexport function adaptQwen2_5Bbox(\n bbox: number[],\n): [number, number, number, number] {\n if (bbox.length < 2) {\n const msg = `invalid bbox data for qwen-vl mode: ${JSON.stringify(bbox)} `;\n throw new Error(msg);\n }\n\n const result: [number, number, number, number] = [\n Math.round(bbox[0]),\n Math.round(bbox[1]),\n typeof bbox[2] === 'number'\n ? Math.round(bbox[2])\n : Math.round(bbox[0] + defaultBboxSize),\n typeof bbox[3] === 'number'\n ? Math.round(bbox[3])\n : Math.round(bbox[1] + defaultBboxSize),\n ];\n return result;\n}\n\nexport function adaptDoubaoBbox(\n bbox: string[] | number[] | string,\n width: number,\n height: number,\n): [number, number, number, number] {\n assert(\n width > 0 && height > 0,\n 'width and height must be greater than 0 in doubao mode',\n );\n\n if (typeof bbox === 'string') {\n assert(\n /^(\\d+)\\s(\\d+)\\s(\\d+)\\s(\\d+)$/.test(bbox.trim()),\n `invalid bbox data string for doubao-vision mode: ${bbox}`,\n );\n const splitted = bbox.split(' ');\n if (splitted.length === 4) {\n return [\n Math.round((Number(splitted[0]) * width) / 1000),\n Math.round((Number(splitted[1]) * height) / 1000),\n Math.round((Number(splitted[2]) * width) / 1000),\n Math.round((Number(splitted[3]) * height) / 1000),\n ];\n }\n throw new Error(`invalid bbox data string for doubao-vision mode: ${bbox}`);\n }\n\n let bboxList: number[] = [];\n if (Array.isArray(bbox) && typeof bbox[0] === 'string') {\n bbox.forEach((item) => {\n if (typeof item === 'string' && item.includes(',')) {\n const [x, y] = item.split(',');\n bboxList.push(Number(x.trim()), Number(y.trim()));\n } else if (typeof item === 'string' && item.includes(' ')) {\n const [x, y] = item.split(' ');\n bboxList.push(Number(x.trim()), Number(y.trim()));\n } else {\n bboxList.push(Number(item));\n }\n });\n } else {\n bboxList = bbox as any;\n }\n\n if (bboxList.length === 4 || bboxList.length === 5) {\n return [\n Math.round((bboxList[0] * width) / 1000),\n Math.round((bboxList[1] * height) / 1000),\n Math.round((bboxList[2] * width) / 1000),\n Math.round((bboxList[3] * height) / 1000),\n ];\n }\n\n // treat the bbox as a center point\n if (\n bboxList.length === 6 ||\n bboxList.length === 2 ||\n bboxList.length === 3 ||\n bboxList.length === 7\n ) {\n return [\n Math.max(\n 0,\n Math.round((bboxList[0] * width) / 1000) - defaultBboxSize / 2,\n ),\n Math.max(\n 0,\n Math.round((bboxList[1] * height) / 1000) - defaultBboxSize / 2,\n ),\n Math.min(\n width,\n Math.round((bboxList[0] * width) / 1000) + defaultBboxSize / 2,\n ),\n Math.min(\n height,\n Math.round((bboxList[1] * height) / 1000) + defaultBboxSize / 2,\n ),\n ];\n }\n\n if (bbox.length === 8) {\n return [\n Math.round((bboxList[0] * width) / 1000),\n Math.round((bboxList[1] * height) / 1000),\n Math.round((bboxList[4] * width) / 1000),\n Math.round((bboxList[5] * height) / 1000),\n ];\n }\n\n const msg = `invalid bbox data for doubao-vision mode: ${JSON.stringify(bbox)} `;\n throw new Error(msg);\n}\n\nfunction normalizeBboxInput(\n bbox: AdaptBboxInput,\n): number[] | string[] | string {\n if (Array.isArray(bbox)) {\n if (Array.isArray(bbox[0])) {\n return bbox[0] as number[] | string[];\n }\n return bbox as number[] | string[];\n }\n return bbox as string;\n}\n\nexport function adaptBbox(\n bbox: AdaptBboxInput,\n width: number,\n height: number,\n rightLimit: number,\n bottomLimit: number,\n vlMode: TVlModeTypes | undefined,\n): [number, number, number, number] {\n const normalizedBbox = normalizeBboxInput(bbox);\n\n let result: [number, number, number, number] = [0, 0, 0, 0];\n if (vlMode === 'doubao-vision' || vlMode === 'vlm-ui-tars') {\n result = adaptDoubaoBbox(normalizedBbox, width, height);\n } else if (vlMode === 'gemini') {\n result = adaptGeminiBbox(normalizedBbox as number[], width, height);\n } else if (vlMode === 'qwen2.5-vl') {\n result = adaptQwen2_5Bbox(normalizedBbox as number[]);\n } else {\n // Default: normalized 0-1000 coordinate system\n // Includes: qwen3-vl, glm-v, auto-glm, auto-glm-multilingual, and future models\n result = normalized01000(normalizedBbox as number[], width, height);\n }\n\n result[2] = Math.min(result[2], rightLimit);\n result[3] = Math.min(result[3], bottomLimit);\n\n return result;\n}\n\n// x1, y1, x2, y2 -> 0-1000\nexport function normalized01000(\n bbox: number[],\n width: number,\n height: number,\n): [number, number, number, number] {\n return [\n Math.round((bbox[0] * width) / 1000),\n Math.round((bbox[1] * height) / 1000),\n Math.round((bbox[2] * width) / 1000),\n Math.round((bbox[3] * height) / 1000),\n ];\n}\n\n// y1, x1, y2, x2 -> 0-1000\nexport function adaptGeminiBbox(\n bbox: number[],\n width: number,\n height: number,\n): [number, number, number, number] {\n const left = Math.round((bbox[1] * width) / 1000);\n const top = Math.round((bbox[0] * height) / 1000);\n const right = Math.round((bbox[3] * width) / 1000);\n const bottom = Math.round((bbox[2] * height) / 1000);\n return [left, top, right, bottom];\n}\n\nexport function adaptBboxToRect(\n bbox: number[],\n width: number,\n height: number,\n offsetX = 0,\n offsetY = 0,\n rightLimit = width,\n bottomLimit = height,\n vlMode?: TVlModeTypes | undefined,\n): Rect {\n debugInspectUtils(\n 'adaptBboxToRect',\n bbox,\n width,\n height,\n 'offset',\n offsetX,\n offsetY,\n 'limit',\n rightLimit,\n bottomLimit,\n 'vlMode',\n vlMode,\n );\n const [left, top, right, bottom] = adaptBbox(\n bbox,\n width,\n height,\n rightLimit,\n bottomLimit,\n vlMode,\n );\n\n // Calculate initial rect dimensions\n const rectLeft = left;\n const rectTop = top;\n let rectWidth = right - left;\n let rectHeight = bottom - top;\n\n // Ensure the rect doesn't exceed image boundaries\n // If right edge exceeds width, adjust the width\n if (rectLeft + rectWidth > width) {\n rectWidth = width - rectLeft;\n }\n\n // If bottom edge exceeds height, adjust the height\n if (rectTop + rectHeight > height) {\n rectHeight = height - rectTop;\n }\n\n // Ensure minimum dimensions (width and height should be at least 1)\n rectWidth = Math.max(1, rectWidth);\n rectHeight = Math.max(1, rectHeight);\n\n const rect = {\n left: rectLeft + offsetX,\n top: rectTop + offsetY,\n width: rectWidth,\n height: rectHeight,\n };\n debugInspectUtils('adaptBboxToRect, result=', rect);\n\n return rect;\n}\n\nexport function mergeRects(rects: Rect[]) {\n const minLeft = Math.min(...rects.map((r) => r.left));\n const minTop = Math.min(...rects.map((r) => r.top));\n const maxRight = Math.max(...rects.map((r) => r.left + r.width));\n const maxBottom = Math.max(...rects.map((r) => r.top + r.height));\n return {\n left: minLeft,\n top: minTop,\n width: maxRight - minLeft,\n height: maxBottom - minTop,\n };\n}\n\n// expand the search area to at least 300 x 300, or add a default padding\nexport function expandSearchArea(\n rect: Rect,\n screenSize: Size,\n vlMode: TVlModeTypes | undefined,\n) {\n let minEdgeSize = 500;\n if (vlMode === 'qwen3-vl') {\n minEdgeSize = 1200;\n }\n const defaultPadding = 160;\n\n // Calculate padding needed to reach minimum edge size\n const paddingSizeHorizontal =\n rect.width < minEdgeSize\n ? Math.ceil((minEdgeSize - rect.width) / 2)\n : defaultPadding;\n const paddingSizeVertical =\n rect.height < minEdgeSize\n ? Math.ceil((minEdgeSize - rect.height) / 2)\n : defaultPadding;\n\n // Calculate new dimensions (ensure minimum edge size)\n let newWidth = Math.max(minEdgeSize, rect.width + paddingSizeHorizontal * 2);\n let newHeight = Math.max(minEdgeSize, rect.height + paddingSizeVertical * 2);\n\n // Calculate initial position with padding\n let newLeft = rect.left - paddingSizeHorizontal;\n let newTop = rect.top - paddingSizeVertical;\n\n // Ensure the rect doesn't exceed screen boundaries by adjusting position\n // If the rect goes beyond the right edge, shift it left\n if (newLeft + newWidth > screenSize.width) {\n newLeft = screenSize.width - newWidth;\n }\n\n // If the rect goes beyond the bottom edge, shift it up\n if (newTop + newHeight > screenSize.height) {\n newTop = screenSize.height - newHeight;\n }\n\n // Ensure the rect doesn't go beyond the left/top edges\n newLeft = Math.max(0, newLeft);\n newTop = Math.max(0, newTop);\n\n // If after position adjustment, the rect still exceeds screen boundaries,\n // clamp the dimensions to fit within screen\n if (newLeft + newWidth > screenSize.width) {\n newWidth = screenSize.width - newLeft;\n }\n if (newTop + newHeight > screenSize.height) {\n newHeight = screenSize.height - newTop;\n }\n\n rect.left = newLeft;\n rect.top = newTop;\n rect.width = newWidth;\n rect.height = newHeight;\n\n return rect;\n}\n\nexport async function markupImageForLLM(\n screenshotBase64: string,\n tree: ElementTreeNode<BaseElement>,\n size: Size,\n) {\n const elementsInfo = treeToList(tree);\n const elementsPositionInfoWithoutText = elementsInfo!.filter(\n (elementInfo) => {\n if (elementInfo.attributes.nodeType === NodeType.TEXT) {\n return false;\n }\n return true;\n },\n );\n\n const imagePayload = await compositeElementInfoImg({\n inputImgBase64: screenshotBase64,\n elementsPositionInfo: elementsPositionInfoWithoutText,\n size,\n });\n return imagePayload;\n}\n\nexport function buildYamlFlowFromPlans(\n plans: PlanningAction[],\n actionSpace: DeviceAction<any>[],\n): MidsceneYamlFlowItem[] {\n const flow: MidsceneYamlFlowItem[] = [];\n\n for (const plan of plans) {\n const verb = plan.type;\n\n const action = actionSpace.find((action) => action.name === verb);\n if (!action) {\n console.warn(\n `Cannot convert action ${verb} to yaml flow. Will ignore it.`,\n );\n continue;\n }\n\n const flowKey = action.interfaceAlias || verb;\n const flowParam = action.paramSchema\n ? dumpActionParam(plan.param || {}, action.paramSchema)\n : {};\n\n const flowItem: MidsceneYamlFlowItem = {\n [flowKey]: '',\n ...flowParam,\n };\n\n flow.push(flowItem);\n }\n\n return flow;\n}\n\n// Zod schemas for shared types\nexport const PointSchema = z.object({\n left: z.number(),\n top: z.number(),\n});\n\nexport const SizeSchema = z.object({\n width: z.number(),\n height: z.number(),\n dpr: z.number().optional(),\n});\n\nexport const RectSchema = PointSchema.and(SizeSchema).and(\n z.object({\n zoom: z.number().optional(),\n }),\n);\n\n// Zod schema for TMultimodalPrompt\nexport const TMultimodalPromptSchema = z.object({\n images: z\n .array(\n z.object({\n name: z.string(),\n url: z.string(),\n }),\n )\n .optional(),\n convertHttpImage2Base64: z.boolean().optional(),\n});\n\n// Zod schema for TUserPrompt\nexport const TUserPromptSchema = z.union([\n z.string(),\n z\n .object({\n prompt: z.string(),\n })\n .and(TMultimodalPromptSchema.partial()),\n]);\n\n// Generate TypeScript types from Zod schemas\nexport type TMultimodalPrompt = z.infer<typeof TMultimodalPromptSchema>;\nexport type TUserPrompt = z.infer<typeof TUserPromptSchema>;\n\nconst locateFieldFlagName = 'midscene_location_field_flag';\n\n// Schema for locator field input (when users provide locate parameters)\nconst MidsceneLocationInput = z\n .object({\n prompt: TUserPromptSchema,\n deepThink: z.boolean().optional(),\n cacheable: z.boolean().optional(),\n xpath: z.union([z.string(), z.boolean()]).optional(),\n })\n .passthrough();\n\n// Schema for locator field result (when AI returns locate results)\nconst MidsceneLocationResult = z\n .object({\n [locateFieldFlagName]: z.literal(true),\n prompt: TUserPromptSchema,\n\n // optional fields\n deepThink: z.boolean().optional(), // only available in vl model\n cacheable: z.boolean().optional(),\n xpath: z.boolean().optional(), // preset result for xpath\n\n // these two fields will only appear in the result\n center: z.tuple([z.number(), z.number()]),\n rect: RectSchema,\n })\n .passthrough();\n\n// Export the result type - this is used for runtime results that include center and rect\nexport type MidsceneLocationResultType = z.infer<typeof MidsceneLocationResult>;\n\n// Export the input type - this is the inferred type from getMidsceneLocationSchema()\nexport type MidsceneLocationInputType = z.infer<typeof MidsceneLocationInput>;\n\n/**\n * Returns the schema for locator fields.\n * This now returns the input schema which is more permissive and suitable for validation.\n */\nexport const getMidsceneLocationSchema = () => {\n return MidsceneLocationInput;\n};\n\nexport const ifMidsceneLocatorField = (field: any): boolean => {\n // Handle optional fields by getting the inner type\n let actualField = field;\n if (actualField._def?.typeName === 'ZodOptional') {\n actualField = actualField._def.innerType;\n }\n\n // Check if this is a ZodObject\n if (actualField._def?.typeName === 'ZodObject') {\n const shape = actualField._def.shape();\n\n // Method 1: Check for the location field flag (for result schema)\n if (locateFieldFlagName in shape) {\n return true;\n }\n\n // Method 2: Check if it's the input schema by checking for 'prompt' field\n // Input schema has 'prompt' as a required field\n if ('prompt' in shape && shape.prompt) {\n return true;\n }\n }\n\n return false;\n};\n\nexport const dumpMidsceneLocatorField = (field: any): string => {\n assert(\n ifMidsceneLocatorField(field),\n 'field is not a midscene locator field',\n );\n\n // If field is a string, return it directly\n if (typeof field === 'string') {\n return field;\n }\n\n // If field is an object with prompt property\n if (field && typeof field === 'object' && field.prompt) {\n // If prompt is a string, return it directly\n if (typeof field.prompt === 'string') {\n return field.prompt;\n }\n // If prompt is a TUserPrompt object, extract the prompt string\n if (typeof field.prompt === 'object' && field.prompt.prompt) {\n return field.prompt.prompt; // TODO: dump images if necessary\n }\n }\n\n // Fallback: try to convert to string\n return String(field);\n};\n\nexport const findAllMidsceneLocatorField = (\n zodType?: z.ZodType<any>,\n requiredOnly?: boolean,\n): string[] => {\n if (!zodType) {\n return [];\n }\n\n // Check if this is a ZodObject by checking if it has a shape property\n const zodObject = zodType as any;\n if (zodObject._def?.typeName === 'ZodObject' && zodObject.shape) {\n const keys = Object.keys(zodObject.shape);\n return keys.filter((key) => {\n const field = zodObject.shape[key];\n if (!ifMidsceneLocatorField(field)) {\n return false;\n }\n\n // If requiredOnly is true, filter out optional fields\n if (requiredOnly) {\n return field._def?.typeName !== 'ZodOptional';\n }\n\n return true;\n });\n }\n\n // For other ZodType instances, we can't extract field names\n return [];\n};\n\nexport const dumpActionParam = (\n jsonObject: Record<string, any>,\n zodSchema: z.ZodType<any>,\n): Record<string, any> => {\n // Prevent spreading strings into {0: 'c', 1: 'o', ...}\n if (!isPlainObject(jsonObject)) {\n return {};\n }\n\n const locatorFields = findAllMidsceneLocatorField(zodSchema);\n const result = { ...jsonObject };\n\n for (const fieldName of locatorFields) {\n const fieldValue = result[fieldName];\n if (fieldValue) {\n // If it's already a string, keep it as is\n if (typeof fieldValue === 'string') {\n result[fieldName] = fieldValue;\n } else if (typeof fieldValue === 'object') {\n // Check if this field is actually a MidsceneLocationType object\n if (fieldValue.prompt) {\n // If prompt is a string, use it directly\n if (typeof fieldValue.prompt === 'string') {\n result[fieldName] = fieldValue.prompt;\n } else if (\n typeof fieldValue.prompt === 'object' &&\n fieldValue.prompt.prompt\n ) {\n // If prompt is a TUserPrompt object, extract the prompt string\n result[fieldName] = fieldValue.prompt.prompt;\n }\n }\n }\n }\n }\n\n return result;\n};\n\nexport const loadActionParam = (\n jsonObject: Record<string, any>,\n zodSchema: z.ZodType<any>,\n): Record<string, any> => {\n const locatorFields = findAllMidsceneLocatorField(zodSchema);\n const result = { ...jsonObject };\n\n for (const fieldName of locatorFields) {\n const fieldValue = result[fieldName];\n if (fieldValue && typeof fieldValue === 'string') {\n result[fieldName] = {\n [locateFieldFlagName]: true,\n prompt: fieldValue,\n };\n }\n }\n\n return result;\n};\n\n/**\n * Parse and validate action parameters using Zod schema.\n * All fields are validated through Zod, EXCEPT locator fields which are skipped.\n * Default values defined in the schema are automatically applied.\n *\n * Locator fields are special business logic fields with complex validation requirements,\n * so they are intentionally excluded from Zod parsing and use existing validation logic.\n */\nexport const parseActionParam = (\n rawParam: Record<string, any> | undefined,\n zodSchema?: z.ZodType<any>,\n): Record<string, any> | undefined => {\n // If no schema is provided, return undefined (action takes no parameters)\n if (!zodSchema) {\n return undefined;\n }\n\n // Handle undefined or null rawParam by providing an empty object\n const param = rawParam ?? {};\n\n // Find all locate fields in the schema\n const locateFields = findAllMidsceneLocatorField(zodSchema);\n\n // If there are no locate fields, just do normal validation\n if (locateFields.length === 0) {\n return zodSchema.parse(param);\n }\n\n // Extract locate field values to restore later\n const locateFieldValues: Record<string, any> = {};\n for (const fieldName of locateFields) {\n if (fieldName in param) {\n locateFieldValues[fieldName] = param[fieldName];\n }\n }\n\n // Build params for validation - skip locate fields and use dummy values\n const paramsForValidation: Record<string, any> = {};\n for (const key in param) {\n if (locateFields.includes(key)) {\n // Use dummy value to satisfy schema validation\n paramsForValidation[key] = { prompt: '_dummy_' };\n } else {\n paramsForValidation[key] = param[key];\n }\n }\n\n // Validate with dummy locate values\n const validated = zodSchema.parse(paramsForValidation);\n\n // Restore the actual locate field values (unvalidated, as per business requirement)\n for (const fieldName in locateFieldValues) {\n validated[fieldName] = locateFieldValues[fieldName];\n }\n\n return validated;\n};\n\nexport const finalizeActionName = 'Finalize';\n\n/**\n * Get a readable time string for the current time\n * @param format - Optional format string. Supports: YYYY, MM, DD, HH, mm, ss. Default: 'YYYY-MM-DD HH:mm:ss'\n * @returns A formatted time string with format label\n */\nexport const getReadableTimeString = (\n format = 'YYYY-MM-DD HH:mm:ss',\n): string => {\n const now = new Date();\n const year = now.getFullYear();\n const month = String(now.getMonth() + 1).padStart(2, '0');\n const day = String(now.getDate()).padStart(2, '0');\n const hours = String(now.getHours()).padStart(2, '0');\n const minutes = String(now.getMinutes()).padStart(2, '0');\n const seconds = String(now.getSeconds()).padStart(2, '0');\n\n const timeString = format\n .replace('YYYY', String(year))\n .replace('MM', month)\n .replace('DD', day)\n .replace('HH', hours)\n .replace('mm', minutes)\n .replace('ss', seconds);\n\n return `${timeString} (${format})`;\n};\n"],"names":["defaultBboxSize","debugInspectUtils","getDebug","pointToBbox","x","y","bboxSize","halfSize","x1","Math","y1","x2","y2","fillBboxParam","locate","width","height","rightLimit","bottomLimit","vlMode","adaptBbox","adaptQwen2_5Bbox","bbox","msg","JSON","Error","result","adaptDoubaoBbox","assert","splitted","Number","bboxList","Array","item","normalizeBboxInput","normalizedBbox","adaptGeminiBbox","normalized01000","left","top","right","bottom","adaptBboxToRect","offsetX","offsetY","rectLeft","rectTop","rectWidth","rectHeight","rect","mergeRects","rects","minLeft","r","minTop","maxRight","maxBottom","expandSearchArea","screenSize","minEdgeSize","defaultPadding","paddingSizeHorizontal","paddingSizeVertical","newWidth","newHeight","newLeft","newTop","markupImageForLLM","screenshotBase64","tree","size","elementsInfo","treeToList","elementsPositionInfoWithoutText","elementInfo","NodeType","imagePayload","compositeElementInfoImg","buildYamlFlowFromPlans","plans","actionSpace","flow","plan","verb","action","console","flowKey","flowParam","dumpActionParam","flowItem","PointSchema","z","SizeSchema","RectSchema","TMultimodalPromptSchema","TUserPromptSchema","locateFieldFlagName","MidsceneLocationInput","getMidsceneLocationSchema","ifMidsceneLocatorField","field","actualField","shape","dumpMidsceneLocatorField","String","findAllMidsceneLocatorField","zodType","requiredOnly","zodObject","keys","Object","key","jsonObject","zodSchema","isPlainObject","locatorFields","fieldName","fieldValue","loadActionParam","parseActionParam","rawParam","param","locateFields","locateFieldValues","paramsForValidation","validated","finalizeActionName","getReadableTimeString","format","now","Date","year","month","day","hours","minutes","seconds","timeString"],"mappings":";;;;;;AAuBA,MAAMA,kBAAkB;AACxB,MAAMC,oBAAoBC,SAAS;AAY5B,SAASC,YACdC,CAAS,EACTC,CAAS,EACTC,WAAWN,eAAe;IAE1B,MAAMO,WAAWD,WAAW;IAC5B,MAAME,KAAKC,KAAK,GAAG,CAACL,IAAIG,UAAU;IAClC,MAAMG,KAAKD,KAAK,GAAG,CAACJ,IAAIE,UAAU;IAClC,MAAMI,KAAKF,KAAK,GAAG,CAACL,IAAIG,UAAU;IAClC,MAAMK,KAAKH,KAAK,GAAG,CAACJ,IAAIE,UAAU;IAElC,OAAO;QAACC;QAAIE;QAAIC;QAAIC;KAAG;AACzB;AAGO,SAASC,cACdC,MAA2B,EAC3BC,KAAa,EACbC,MAAc,EACdC,UAAkB,EAClBC,WAAmB,EACnBC,MAAgC;IAGhC,IAAKL,OAAe,OAAO,IAAI,CAACA,QAAQ,MAAM;QAC5CA,OAAO,IAAI,GAAIA,OAAe,OAAO;QAErC,OAAQA,OAAe,OAAO;IAChC;IAEA,IAAIA,QAAQ,MACVA,OAAO,IAAI,GAAGM,UACZN,OAAO,IAAI,EACXC,OACAC,QACAC,YACAC,aACAC;IAIJ,OAAOL;AACT;AAEO,SAASO,iBACdC,IAAc;IAEd,IAAIA,KAAK,MAAM,GAAG,GAAG;QACnB,MAAMC,MAAM,CAAC,oCAAoC,EAAEC,KAAK,SAAS,CAACF,MAAM,CAAC,CAAC;QAC1E,MAAM,IAAIG,MAAMF;IAClB;IAEA,MAAMG,SAA2C;QAC/CjB,KAAK,KAAK,CAACa,IAAI,CAAC,EAAE;QAClBb,KAAK,KAAK,CAACa,IAAI,CAAC,EAAE;QACC,YAAnB,OAAOA,IAAI,CAAC,EAAE,GACVb,KAAK,KAAK,CAACa,IAAI,CAAC,EAAE,IAClBb,KAAK,KAAK,CAACa,IAAI,CAAC,EAAE,GAAGtB;QACN,YAAnB,OAAOsB,IAAI,CAAC,EAAE,GACVb,KAAK,KAAK,CAACa,IAAI,CAAC,EAAE,IAClBb,KAAK,KAAK,CAACa,IAAI,CAAC,EAAE,GAAGtB;KAC1B;IACD,OAAO0B;AACT;AAEO,SAASC,gBACdL,IAAkC,EAClCP,KAAa,EACbC,MAAc;IAEdY,OACEb,QAAQ,KAAKC,SAAS,GACtB;IAGF,IAAI,AAAgB,YAAhB,OAAOM,MAAmB;QAC5BM,OACE,+BAA+B,IAAI,CAACN,KAAK,IAAI,KAC7C,CAAC,iDAAiD,EAAEA,MAAM;QAE5D,MAAMO,WAAWP,KAAK,KAAK,CAAC;QAC5B,IAAIO,AAAoB,MAApBA,SAAS,MAAM,EACjB,OAAO;YACLpB,KAAK,KAAK,CAAEqB,OAAOD,QAAQ,CAAC,EAAE,IAAId,QAAS;YAC3CN,KAAK,KAAK,CAAEqB,OAAOD,QAAQ,CAAC,EAAE,IAAIb,SAAU;YAC5CP,KAAK,KAAK,CAAEqB,OAAOD,QAAQ,CAAC,EAAE,IAAId,QAAS;YAC3CN,KAAK,KAAK,CAAEqB,OAAOD,QAAQ,CAAC,EAAE,IAAIb,SAAU;SAC7C;QAEH,MAAM,IAAIS,MAAM,CAAC,iDAAiD,EAAEH,MAAM;IAC5E;IAEA,IAAIS,WAAqB,EAAE;IAC3B,IAAIC,MAAM,OAAO,CAACV,SAAS,AAAmB,YAAnB,OAAOA,IAAI,CAAC,EAAE,EACvCA,KAAK,OAAO,CAAC,CAACW;QACZ,IAAI,AAAgB,YAAhB,OAAOA,QAAqBA,KAAK,QAAQ,CAAC,MAAM;YAClD,MAAM,CAAC7B,GAAGC,EAAE,GAAG4B,KAAK,KAAK,CAAC;YAC1BF,SAAS,IAAI,CAACD,OAAO1B,EAAE,IAAI,KAAK0B,OAAOzB,EAAE,IAAI;QAC/C,OAAO,IAAI,AAAgB,YAAhB,OAAO4B,QAAqBA,KAAK,QAAQ,CAAC,MAAM;YACzD,MAAM,CAAC7B,GAAGC,EAAE,GAAG4B,KAAK,KAAK,CAAC;YAC1BF,SAAS,IAAI,CAACD,OAAO1B,EAAE,IAAI,KAAK0B,OAAOzB,EAAE,IAAI;QAC/C,OACE0B,SAAS,IAAI,CAACD,OAAOG;IAEzB;SAEAF,WAAWT;IAGb,IAAIS,AAAoB,MAApBA,SAAS,MAAM,IAAUA,AAAoB,MAApBA,SAAS,MAAM,EAC1C,OAAO;QACLtB,KAAK,KAAK,CAAEsB,QAAQ,CAAC,EAAE,GAAGhB,QAAS;QACnCN,KAAK,KAAK,CAAEsB,QAAQ,CAAC,EAAE,GAAGf,SAAU;QACpCP,KAAK,KAAK,CAAEsB,QAAQ,CAAC,EAAE,GAAGhB,QAAS;QACnCN,KAAK,KAAK,CAAEsB,QAAQ,CAAC,EAAE,GAAGf,SAAU;KACrC;IAIH,IACEe,AAAoB,MAApBA,SAAS,MAAM,IACfA,AAAoB,MAApBA,SAAS,MAAM,IACfA,AAAoB,MAApBA,SAAS,MAAM,IACfA,AAAoB,MAApBA,SAAS,MAAM,EAEf,OAAO;QACLtB,KAAK,GAAG,CACN,GACAA,KAAK,KAAK,CAAEsB,QAAQ,CAAC,EAAE,GAAGhB,QAAS,QAAQf,kBAAkB;QAE/DS,KAAK,GAAG,CACN,GACAA,KAAK,KAAK,CAAEsB,QAAQ,CAAC,EAAE,GAAGf,SAAU,QAAQhB,kBAAkB;QAEhES,KAAK,GAAG,CACNM,OACAN,KAAK,KAAK,CAAEsB,QAAQ,CAAC,EAAE,GAAGhB,QAAS,QAAQf,kBAAkB;QAE/DS,KAAK,GAAG,CACNO,QACAP,KAAK,KAAK,CAAEsB,QAAQ,CAAC,EAAE,GAAGf,SAAU,QAAQhB,kBAAkB;KAEjE;IAGH,IAAIsB,AAAgB,MAAhBA,KAAK,MAAM,EACb,OAAO;QACLb,KAAK,KAAK,CAAEsB,QAAQ,CAAC,EAAE,GAAGhB,QAAS;QACnCN,KAAK,KAAK,CAAEsB,QAAQ,CAAC,EAAE,GAAGf,SAAU;QACpCP,KAAK,KAAK,CAAEsB,QAAQ,CAAC,EAAE,GAAGhB,QAAS;QACnCN,KAAK,KAAK,CAAEsB,QAAQ,CAAC,EAAE,GAAGf,SAAU;KACrC;IAGH,MAAMO,MAAM,CAAC,0CAA0C,EAAEC,KAAK,SAAS,CAACF,MAAM,CAAC,CAAC;IAChF,MAAM,IAAIG,MAAMF;AAClB;AAEA,SAASW,mBACPZ,IAAoB;IAEpB,IAAIU,MAAM,OAAO,CAACV,OAChB;QAAA,IAAIU,MAAM,OAAO,CAACV,IAAI,CAAC,EAAE,GACvB,OAAOA,IAAI,CAAC,EAAE;IAChB;IAGF,OAAOA;AACT;AAEO,SAASF,UACdE,IAAoB,EACpBP,KAAa,EACbC,MAAc,EACdC,UAAkB,EAClBC,WAAmB,EACnBC,MAAgC;IAEhC,MAAMgB,iBAAiBD,mBAAmBZ;IAE1C,IAAII,SAA2C;QAAC;QAAG;QAAG;QAAG;KAAE;IAEzDA,SADEP,AAAW,oBAAXA,UAA8BA,AAAW,kBAAXA,SACvBQ,gBAAgBQ,gBAAgBpB,OAAOC,UACvCG,AAAW,aAAXA,SACAiB,gBAAgBD,gBAA4BpB,OAAOC,UACnDG,AAAW,iBAAXA,SACAE,iBAAiBc,kBAIjBE,gBAAgBF,gBAA4BpB,OAAOC;IAG9DU,MAAM,CAAC,EAAE,GAAGjB,KAAK,GAAG,CAACiB,MAAM,CAAC,EAAE,EAAET;IAChCS,MAAM,CAAC,EAAE,GAAGjB,KAAK,GAAG,CAACiB,MAAM,CAAC,EAAE,EAAER;IAEhC,OAAOQ;AACT;AAGO,SAASW,gBACdf,IAAc,EACdP,KAAa,EACbC,MAAc;IAEd,OAAO;QACLP,KAAK,KAAK,CAAEa,IAAI,CAAC,EAAE,GAAGP,QAAS;QAC/BN,KAAK,KAAK,CAAEa,IAAI,CAAC,EAAE,GAAGN,SAAU;QAChCP,KAAK,KAAK,CAAEa,IAAI,CAAC,EAAE,GAAGP,QAAS;QAC/BN,KAAK,KAAK,CAAEa,IAAI,CAAC,EAAE,GAAGN,SAAU;KACjC;AACH;AAGO,SAASoB,gBACdd,IAAc,EACdP,KAAa,EACbC,MAAc;IAEd,MAAMsB,OAAO7B,KAAK,KAAK,CAAEa,IAAI,CAAC,EAAE,GAAGP,QAAS;IAC5C,MAAMwB,MAAM9B,KAAK,KAAK,CAAEa,IAAI,CAAC,EAAE,GAAGN,SAAU;IAC5C,MAAMwB,QAAQ/B,KAAK,KAAK,CAAEa,IAAI,CAAC,EAAE,GAAGP,QAAS;IAC7C,MAAM0B,SAAShC,KAAK,KAAK,CAAEa,IAAI,CAAC,EAAE,GAAGN,SAAU;IAC/C,OAAO;QAACsB;QAAMC;QAAKC;QAAOC;KAAO;AACnC;AAEO,SAASC,gBACdpB,IAAc,EACdP,KAAa,EACbC,MAAc,EACd2B,UAAU,CAAC,EACXC,UAAU,CAAC,EACX3B,aAAaF,KAAK,EAClBG,cAAcF,MAAM,EACpBG,MAAiC;IAEjClB,kBACE,mBACAqB,MACAP,OACAC,QACA,UACA2B,SACAC,SACA,SACA3B,YACAC,aACA,UACAC;IAEF,MAAM,CAACmB,MAAMC,KAAKC,OAAOC,OAAO,GAAGrB,UACjCE,MACAP,OACAC,QACAC,YACAC,aACAC;IAIF,MAAM0B,WAAWP;IACjB,MAAMQ,UAAUP;IAChB,IAAIQ,YAAYP,QAAQF;IACxB,IAAIU,aAAaP,SAASF;IAI1B,IAAIM,WAAWE,YAAYhC,OACzBgC,YAAYhC,QAAQ8B;IAItB,IAAIC,UAAUE,aAAahC,QACzBgC,aAAahC,SAAS8B;IAIxBC,YAAYtC,KAAK,GAAG,CAAC,GAAGsC;IACxBC,aAAavC,KAAK,GAAG,CAAC,GAAGuC;IAEzB,MAAMC,OAAO;QACX,MAAMJ,WAAWF;QACjB,KAAKG,UAAUF;QACf,OAAOG;QACP,QAAQC;IACV;IACA/C,kBAAkB,4BAA4BgD;IAE9C,OAAOA;AACT;AAEO,SAASC,WAAWC,KAAa;IACtC,MAAMC,UAAU3C,KAAK,GAAG,IAAI0C,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,IAAI;IACnD,MAAMC,SAAS7C,KAAK,GAAG,IAAI0C,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,GAAG;IACjD,MAAME,WAAW9C,KAAK,GAAG,IAAI0C,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,IAAI,GAAGA,EAAE,KAAK;IAC9D,MAAMG,YAAY/C,KAAK,GAAG,IAAI0C,MAAM,GAAG,CAAC,CAACE,IAAMA,EAAE,GAAG,GAAGA,EAAE,MAAM;IAC/D,OAAO;QACL,MAAMD;QACN,KAAKE;QACL,OAAOC,WAAWH;QAClB,QAAQI,YAAYF;IACtB;AACF;AAGO,SAASG,iBACdR,IAAU,EACVS,UAAgB,EAChBvC,MAAgC;IAEhC,IAAIwC,cAAc;IAClB,IAAIxC,AAAW,eAAXA,QACFwC,cAAc;IAEhB,MAAMC,iBAAiB;IAGvB,MAAMC,wBACJZ,KAAK,KAAK,GAAGU,cACTlD,KAAK,IAAI,CAAEkD,AAAAA,CAAAA,cAAcV,KAAK,KAAI,IAAK,KACvCW;IACN,MAAME,sBACJb,KAAK,MAAM,GAAGU,cACVlD,KAAK,IAAI,CAAEkD,AAAAA,CAAAA,cAAcV,KAAK,MAAK,IAAK,KACxCW;IAGN,IAAIG,WAAWtD,KAAK,GAAG,CAACkD,aAAaV,KAAK,KAAK,GAAGY,AAAwB,IAAxBA;IAClD,IAAIG,YAAYvD,KAAK,GAAG,CAACkD,aAAaV,KAAK,MAAM,GAAGa,AAAsB,IAAtBA;IAGpD,IAAIG,UAAUhB,KAAK,IAAI,GAAGY;IAC1B,IAAIK,SAASjB,KAAK,GAAG,GAAGa;IAIxB,IAAIG,UAAUF,WAAWL,WAAW,KAAK,EACvCO,UAAUP,WAAW,KAAK,GAAGK;IAI/B,IAAIG,SAASF,YAAYN,WAAW,MAAM,EACxCQ,SAASR,WAAW,MAAM,GAAGM;IAI/BC,UAAUxD,KAAK,GAAG,CAAC,GAAGwD;IACtBC,SAASzD,KAAK,GAAG,CAAC,GAAGyD;IAIrB,IAAID,UAAUF,WAAWL,WAAW,KAAK,EACvCK,WAAWL,WAAW,KAAK,GAAGO;IAEhC,IAAIC,SAASF,YAAYN,WAAW,MAAM,EACxCM,YAAYN,WAAW,MAAM,GAAGQ;IAGlCjB,KAAK,IAAI,GAAGgB;IACZhB,KAAK,GAAG,GAAGiB;IACXjB,KAAK,KAAK,GAAGc;IACbd,KAAK,MAAM,GAAGe;IAEd,OAAOf;AACT;AAEO,eAAekB,kBACpBC,gBAAwB,EACxBC,IAAkC,EAClCC,IAAU;IAEV,MAAMC,eAAeC,WAAWH;IAChC,MAAMI,kCAAkCF,aAAc,MAAM,CAC1D,CAACG;QACC,IAAIA,YAAY,UAAU,CAAC,QAAQ,KAAKC,SAAS,IAAI,EACnD,OAAO;QAET,OAAO;IACT;IAGF,MAAMC,eAAe,MAAMC,wBAAwB;QACjD,gBAAgBT;QAChB,sBAAsBK;QACtBH;IACF;IACA,OAAOM;AACT;AAEO,SAASE,uBACdC,KAAuB,EACvBC,WAAgC;IAEhC,MAAMC,OAA+B,EAAE;IAEvC,KAAK,MAAMC,QAAQH,MAAO;QACxB,MAAMI,OAAOD,KAAK,IAAI;QAEtB,MAAME,SAASJ,YAAY,IAAI,CAAC,CAACI,SAAWA,OAAO,IAAI,KAAKD;QAC5D,IAAI,CAACC,QAAQ;YACXC,QAAQ,IAAI,CACV,CAAC,sBAAsB,EAAEF,KAAK,8BAA8B,CAAC;YAE/D;QACF;QAEA,MAAMG,UAAUF,OAAO,cAAc,IAAID;QACzC,MAAMI,YAAYH,OAAO,WAAW,GAChCI,gBAAgBN,KAAK,KAAK,IAAI,CAAC,GAAGE,OAAO,WAAW,IACpD,CAAC;QAEL,MAAMK,WAAiC;YACrC,CAACH,QAAQ,EAAE;YACX,GAAGC,SAAS;QACd;QAEAN,KAAK,IAAI,CAACQ;IACZ;IAEA,OAAOR;AACT;AAGO,MAAMS,cAAcC,EAAE,MAAM,CAAC;IAClC,MAAMA,EAAE,MAAM;IACd,KAAKA,EAAE,MAAM;AACf;AAEO,MAAMC,aAAaD,EAAE,MAAM,CAAC;IACjC,OAAOA,EAAE,MAAM;IACf,QAAQA,EAAE,MAAM;IAChB,KAAKA,EAAE,MAAM,GAAG,QAAQ;AAC1B;AAEO,MAAME,aAAaH,YAAY,GAAG,CAACE,YAAY,GAAG,CACvDD,EAAE,MAAM,CAAC;IACP,MAAMA,EAAE,MAAM,GAAG,QAAQ;AAC3B;AAIK,MAAMG,0BAA0BH,EAAE,MAAM,CAAC;IAC9C,QAAQA,EAAAA,KACA,CACJA,EAAE,MAAM,CAAC;QACP,MAAMA,EAAE,MAAM;QACd,KAAKA,EAAE,MAAM;IACf,IAED,QAAQ;IACX,yBAAyBA,EAAE,OAAO,GAAG,QAAQ;AAC/C;AAGO,MAAMI,oBAAoBJ,EAAE,KAAK,CAAC;IACvCA,EAAE,MAAM;IACRA,EAAAA,MACS,CAAC;QACN,QAAQA,EAAE,MAAM;IAClB,GACC,GAAG,CAACG,wBAAwB,OAAO;CACvC;AAMD,MAAME,sBAAsB;AAG5B,MAAMC,wBAAwBN,EAAAA,MACrB,CAAC;IACN,QAAQI;IACR,WAAWJ,EAAE,OAAO,GAAG,QAAQ;IAC/B,WAAWA,EAAE,OAAO,GAAG,QAAQ;IAC/B,OAAOA,EAAE,KAAK,CAAC;QAACA,EAAE,MAAM;QAAIA,EAAE,OAAO;KAAG,EAAE,QAAQ;AACpD,GACC,WAAW;AAGiBA,EAAAA,MACtB,CAAC;IACN,CAACK,oBAAoB,EAAEL,EAAE,OAAO,CAAC;IACjC,QAAQI;IAGR,WAAWJ,EAAE,OAAO,GAAG,QAAQ;IAC/B,WAAWA,EAAE,OAAO,GAAG,QAAQ;IAC/B,OAAOA,EAAE,OAAO,GAAG,QAAQ;IAG3B,QAAQA,EAAE,KAAK,CAAC;QAACA,EAAE,MAAM;QAAIA,EAAE,MAAM;KAAG;IACxC,MAAME;AACR,GACC,WAAW;AAYP,MAAMK,4BAA4B,IAChCD;AAGF,MAAME,yBAAyB,CAACC;IAErC,IAAIC,cAAcD;IAClB,IAAIC,YAAY,IAAI,EAAE,aAAa,eACjCA,cAAcA,YAAY,IAAI,CAAC,SAAS;IAI1C,IAAIA,YAAY,IAAI,EAAE,aAAa,aAAa;QAC9C,MAAMC,QAAQD,YAAY,IAAI,CAAC,KAAK;QAGpC,IAAIL,uBAAuBM,OACzB,OAAO;QAKT,IAAI,YAAYA,SAASA,MAAM,MAAM,EACnC,OAAO;IAEX;IAEA,OAAO;AACT;AAEO,MAAMC,2BAA2B,CAACH;IACvCxE,OACEuE,uBAAuBC,QACvB;IAIF,IAAI,AAAiB,YAAjB,OAAOA,OACT,OAAOA;IAIT,IAAIA,SAAS,AAAiB,YAAjB,OAAOA,SAAsBA,MAAM,MAAM,EAAE;QAEtD,IAAI,AAAwB,YAAxB,OAAOA,MAAM,MAAM,EACrB,OAAOA,MAAM,MAAM;QAGrB,IAAI,AAAwB,YAAxB,OAAOA,MAAM,MAAM,IAAiBA,MAAM,MAAM,CAAC,MAAM,EACzD,OAAOA,MAAM,MAAM,CAAC,MAAM;IAE9B;IAGA,OAAOI,OAAOJ;AAChB;AAEO,MAAMK,8BAA8B,CACzCC,SACAC;IAEA,IAAI,CAACD,SACH,OAAO,EAAE;IAIX,MAAME,YAAYF;IAClB,IAAIE,UAAU,IAAI,EAAE,aAAa,eAAeA,UAAU,KAAK,EAAE;QAC/D,MAAMC,OAAOC,OAAO,IAAI,CAACF,UAAU,KAAK;QACxC,OAAOC,KAAK,MAAM,CAAC,CAACE;YAClB,MAAMX,QAAQQ,UAAU,KAAK,CAACG,IAAI;YAClC,IAAI,CAACZ,uBAAuBC,QAC1B,OAAO;YAIT,IAAIO,cACF,OAAOP,MAAM,IAAI,EAAE,aAAa;YAGlC,OAAO;QACT;IACF;IAGA,OAAO,EAAE;AACX;AAEO,MAAMZ,kBAAkB,CAC7BwB,YACAC;IAGA,IAAI,CAACC,cAAcF,aACjB,OAAO,CAAC;IAGV,MAAMG,gBAAgBV,4BAA4BQ;IAClD,MAAMvF,SAAS;QAAE,GAAGsF,UAAU;IAAC;IAE/B,KAAK,MAAMI,aAAaD,cAAe;QACrC,MAAME,aAAa3F,MAAM,CAAC0F,UAAU;QACpC,IAAIC,YAEF;YAAA,IAAI,AAAsB,YAAtB,OAAOA,YACT3F,MAAM,CAAC0F,UAAU,GAAGC;iBACf,IAAI,AAAsB,YAAtB,OAAOA,YAEhB;gBAAA,IAAIA,WAAW,MAAM,EAEnB;oBAAA,IAAI,AAA6B,YAA7B,OAAOA,WAAW,MAAM,EAC1B3F,MAAM,CAAC0F,UAAU,GAAGC,WAAW,MAAM;yBAChC,IACL,AAA6B,YAA7B,OAAOA,WAAW,MAAM,IACxBA,WAAW,MAAM,CAAC,MAAM,EAGxB3F,MAAM,CAAC0F,UAAU,GAAGC,WAAW,MAAM,CAAC,MAAM;gBAC9C;YACF;QACF;IAEJ;IAEA,OAAO3F;AACT;AAEO,MAAM4F,kBAAkB,CAC7BN,YACAC;IAEA,MAAME,gBAAgBV,4BAA4BQ;IAClD,MAAMvF,SAAS;QAAE,GAAGsF,UAAU;IAAC;IAE/B,KAAK,MAAMI,aAAaD,cAAe;QACrC,MAAME,aAAa3F,MAAM,CAAC0F,UAAU;QACpC,IAAIC,cAAc,AAAsB,YAAtB,OAAOA,YACvB3F,MAAM,CAAC0F,UAAU,GAAG;YAClB,CAACpB,oBAAoB,EAAE;YACvB,QAAQqB;QACV;IAEJ;IAEA,OAAO3F;AACT;AAUO,MAAM6F,mBAAmB,CAC9BC,UACAP;IAGA,IAAI,CAACA,WACH;IAIF,MAAMQ,QAAQD,YAAY,CAAC;IAG3B,MAAME,eAAejB,4BAA4BQ;IAGjD,IAAIS,AAAwB,MAAxBA,aAAa,MAAM,EACrB,OAAOT,UAAU,KAAK,CAACQ;IAIzB,MAAME,oBAAyC,CAAC;IAChD,KAAK,MAAMP,aAAaM,aACtB,IAAIN,aAAaK,OACfE,iBAAiB,CAACP,UAAU,GAAGK,KAAK,CAACL,UAAU;IAKnD,MAAMQ,sBAA2C,CAAC;IAClD,IAAK,MAAMb,OAAOU,MAChB,IAAIC,aAAa,QAAQ,CAACX,MAExBa,mBAAmB,CAACb,IAAI,GAAG;QAAE,QAAQ;IAAU;SAE/Ca,mBAAmB,CAACb,IAAI,GAAGU,KAAK,CAACV,IAAI;IAKzC,MAAMc,YAAYZ,UAAU,KAAK,CAACW;IAGlC,IAAK,MAAMR,aAAaO,kBACtBE,SAAS,CAACT,UAAU,GAAGO,iBAAiB,CAACP,UAAU;IAGrD,OAAOS;AACT;AAEO,MAAMC,qBAAqB;AAO3B,MAAMC,wBAAwB,CACnCC,SAAS,qBAAqB;IAE9B,MAAMC,MAAM,IAAIC;IAChB,MAAMC,OAAOF,IAAI,WAAW;IAC5B,MAAMG,QAAQ5B,OAAOyB,IAAI,QAAQ,KAAK,GAAG,QAAQ,CAAC,GAAG;IACrD,MAAMI,MAAM7B,OAAOyB,IAAI,OAAO,IAAI,QAAQ,CAAC,GAAG;IAC9C,MAAMK,QAAQ9B,OAAOyB,IAAI,QAAQ,IAAI,QAAQ,CAAC,GAAG;IACjD,MAAMM,UAAU/B,OAAOyB,IAAI,UAAU,IAAI,QAAQ,CAAC,GAAG;IACrD,MAAMO,UAAUhC,OAAOyB,IAAI,UAAU,IAAI,QAAQ,CAAC,GAAG;IAErD,MAAMQ,aAAaT,OAChB,OAAO,CAAC,QAAQxB,OAAO2B,OACvB,OAAO,CAAC,MAAMC,OACd,OAAO,CAAC,MAAMC,KACd,OAAO,CAAC,MAAMC,OACd,OAAO,CAAC,MAAMC,SACd,OAAO,CAAC,MAAMC;IAEjB,OAAO,GAAGC,WAAW,EAAE,EAAET,OAAO,CAAC,CAAC;AACpC"}
|
package/dist/es/index.mjs.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.mjs","sources":["../../src/index.ts"],"sourcesContent":["import { z } from 'zod';\nimport Service from './service/index';\nimport { TaskRunner } from './task-runner';\nimport { getVersion } from './utils';\n\nexport {\n plan,\n describeUserPage,\n AiLocateElement,\n getMidsceneLocationSchema,\n type MidsceneLocationResultType,\n PointSchema,\n SizeSchema,\n RectSchema,\n TMultimodalPromptSchema,\n TUserPromptSchema,\n type TMultimodalPrompt,\n type TUserPrompt,\n} from './ai-model/index';\n\nexport {\n MIDSCENE_MODEL_NAME,\n type CreateOpenAIClientFn,\n} from '@midscene/shared/env';\n\nexport type * from './types';\nexport {\n ServiceError,\n ExecutionDump,\n GroupedActionDump,\n type IExecutionDump,\n type IGroupedActionDump,\n} from './types';\n\nexport { z };\n\nexport default Service;\nexport { TaskRunner, Service, getVersion };\n\nexport type {\n MidsceneYamlScript,\n MidsceneYamlTask,\n MidsceneYamlFlowItem,\n MidsceneYamlConfigResult,\n MidsceneYamlConfig,\n MidsceneYamlScriptWebEnv,\n MidsceneYamlScriptAndroidEnv,\n MidsceneYamlScriptIOSEnv,\n MidsceneYamlScriptEnv,\n LocateOption,\n DetailedLocateParam,\n} from './yaml';\n\nexport { Agent, type AgentOpt, type AiActOptions, createAgent } from './agent';\n\n// Dump utilities\nexport {\n restoreImageReferences,\n escapeContent,\n unescapeContent,\n parseImageScripts,\n parseDumpScript,\n parseDumpScriptAttributes,\n generateImageScriptTag,\n generateDumpScriptTag,\n} from './dump';\n\n// ScreenshotItem\nexport { ScreenshotItem } from './screenshot-item';\n"],"names":["Service"],"mappings":";;;;;;;;;;AAoCA,YAAeA"}
|
|
1
|
+
{"version":3,"file":"index.mjs","sources":["../../src/index.ts"],"sourcesContent":["import { z } from 'zod';\nimport Service from './service/index';\nimport { TaskRunner } from './task-runner';\nimport { getVersion } from './utils';\n\nexport {\n plan,\n describeUserPage,\n AiLocateElement,\n getMidsceneLocationSchema,\n type MidsceneLocationResultType,\n PointSchema,\n SizeSchema,\n RectSchema,\n TMultimodalPromptSchema,\n TUserPromptSchema,\n type TMultimodalPrompt,\n type TUserPrompt,\n} from './ai-model/index';\n\nexport {\n MIDSCENE_MODEL_NAME,\n type CreateOpenAIClientFn,\n} from '@midscene/shared/env';\n\nexport type * from './types';\nexport {\n ServiceError,\n ExecutionDump,\n GroupedActionDump,\n type IExecutionDump,\n type IGroupedActionDump,\n} from './types';\n\nexport { z };\n\nexport default Service;\nexport { TaskRunner, Service, getVersion };\n\nexport type {\n MidsceneYamlScript,\n MidsceneYamlTask,\n MidsceneYamlFlowItem,\n MidsceneYamlConfigResult,\n MidsceneYamlConfig,\n MidsceneYamlScriptWebEnv,\n MidsceneYamlScriptAndroidEnv,\n MidsceneYamlScriptIOSEnv,\n MidsceneYamlScriptEnv,\n LocateOption,\n DetailedLocateParam,\n} from './yaml';\n\nexport { Agent, type AgentOpt, type AiActOptions, createAgent } from './agent';\n\n// Dump utilities\nexport {\n restoreImageReferences,\n escapeContent,\n unescapeContent,\n parseImageScripts,\n parseDumpScript,\n parseDumpScriptAttributes,\n generateImageScriptTag,\n generateDumpScriptTag,\n} from './dump';\n\n// ScreenshotItem\nexport { ScreenshotItem } from './screenshot-item';\nexport type { SerializedScreenshotItem } from './screenshot-item';\n"],"names":["Service"],"mappings":";;;;;;;;;;AAoCA,YAAeA"}
|
|
@@ -12,17 +12,22 @@ class ScreenshotItem {
|
|
|
12
12
|
static create(base64) {
|
|
13
13
|
return new ScreenshotItem(base64);
|
|
14
14
|
}
|
|
15
|
-
|
|
15
|
+
get base64() {
|
|
16
16
|
return this._data;
|
|
17
17
|
}
|
|
18
18
|
toSerializable() {
|
|
19
|
-
return
|
|
19
|
+
return {
|
|
20
|
+
base64: this._data
|
|
21
|
+
};
|
|
22
|
+
}
|
|
23
|
+
toJSON() {
|
|
24
|
+
return this.toSerializable();
|
|
20
25
|
}
|
|
21
26
|
static isSerializedData(value) {
|
|
22
|
-
return '
|
|
27
|
+
return 'object' == typeof value && null !== value && 'base64' in value && 'string' == typeof value.base64;
|
|
23
28
|
}
|
|
24
29
|
static fromSerializedData(data) {
|
|
25
|
-
return new ScreenshotItem(data);
|
|
30
|
+
return new ScreenshotItem(data.base64);
|
|
26
31
|
}
|
|
27
32
|
constructor(data){
|
|
28
33
|
_define_property(this, "_data", void 0);
|