${currentLocaleStrings.pageTitle}

${agent.ui_display_info[`card_title_${currentLocaleStrings.lang}`] || agent.ui_display_info.card_title_en}

\n ${agent.ui_display_info.display_type === 'table' ? `\n

\n \n \n \n ${agent.ui_display_info.table_columns.map(col => ``).join('')}\n ${agent.agent_id === 'reference_identifier' ? `` : ''}\n \n \n \n

${col[`header_${currentLocaleStrings.lang}`] \|\| col.header_en}	${currentLocaleStrings.referenceStatusHeader \|\| 'Status'}

\n ` : `

`}\n

`;\n }\n });\n\n const galleryItemsHtml = appConfig.gallery_images.map(img => \n `

$\"${img.alt}\"$

`\n ).join('');\n\n // Note: The CSS () block is largely unchanged from the original,\n // but could also be templatized (e.g., for theme colors). For brevity, it's omitted here but assumed to be present.\n // Ensure all 'data-translate' keys in HTML are present in {{localization_strings_en}} and {{localization_strings_es}}\n return `\n\n \n \n \n ${currentLocaleStrings.pageTitle} \n \n \n \n \n \n\n\n\n \n

\n |\n \n

\n\n

\n \n

\n\n

\n ${resultCardsHtml} {/* Dynamically inserted result cards */}\n

\n\n

${galleryItemsHtml}

\n\n

\n \n \n \n \n

\n\n \n\n \n\n`;\n}\n\n// --- Main Request Handler (Server Code) ---\nexport default async function(req: Request) {\n // --- Dynamic Imports (Unchanged) ---\n const { OpenAI } = await import(\"https://esm.town/v/std/openai\");\n // const { z } = await import(\"npm:zod\"); // Zod might be optional if config is trusted\n const { fetch } = await import(\"https://esm.town/v/std/fetch\");\n const { PDFExtract, PDFExtractOptions } = await import(\"npm:pdf.js-extract\");\n\n // --- CONFIGURATION (These would be replaced by the template variables at generation time) ---\n const APP_CONFIG = `\\{{\\app_config_json}}`; // e.g., { openai_model_name: \"gpt-4o\", text_truncation_length: 25000, ... }\n const ANALYSIS_AGENTS = `\\{\\{analysis_agents_json}}`; // Array of agent objects\n\n // --- Helper: Extract Text using pdf.js-extract (Unchanged) ---\n async function extractPdfTextNative(data: ArrayBuffer, fileName: string, log: LogEntry[]): Promise { /* ... original ... */ }\n\n // --- Helper Function: Call OpenAI API (Uses APP_CONFIG for model) ---\n async function callOpenAI(\n openai: OpenAI,\n systemPrompt: string,\n userMessage: string,\n modelFromConfig = APP_CONFIG.openai_model_name || \"gpt-4o\", // Use configured model\n expectJson = false,\n ): Promise<{ role: \"assistant\" | \"system\"; content: string | object }> {\n /* ... original logic, but use modelFromConfig ... */\n const model = modelFromConfig;\n // ... rest of the original callOpenAI function\n try {\n const response = await openai.chat.completions.create({\n model,\n messages: [{ role: \"system\", content: systemPrompt }, { role: \"user\", content: userMessage }],\n temperature: 0.5,\n ...(expectJson && { response_format: { type: \"json_object\" } }),\n });\n const content = response.choices?.[0]?.message?.content;\n if (!content) throw new Error(\"AI returned empty content.\");\n if (expectJson) {\n try { return { role: \"assistant\", content: JSON.parse(content) }; }\n catch (e) { console.error(\"AI JSON Parse Error:\", e, \"Raw:\", content); throw new Error(`AI response not valid JSON: ${content.substring(0,150)}`); }\n } else { return { role: \"assistant\", content }; }\n } catch (error) { /* ... original error handling ... */ return {role: \"system\", content: \"AI Error\"} }\n }\n\n // --- Helper Function: Traverse References (Optional, could be an agent itself) ---\n // This function is specific. If reference traversal is a common need, keep it.\n // Otherwise, this logic could be part of a dedicated \"Reference Traversal Agent\".\n // For this template, we'll assume it's a potential fixed post-processing step for a reference agent.\n async function traverseReferences(references: Reference[], log: LogEntry[]): Promise {\n const agent = \"Reference Traversal Agent\"; // Fixed name for this specialized step\n // ... original traverseReferences logic ...\n // Ensure it returns Reference[] with updated traversalStatus and error fields\n log.push({ agent, type: \"step\", message: `Traversing ${references.length} potential reference URLs... ` });\n let traversedCount = 0;\n const results: Reference[] = []; // Use Reference[]\n const promises = references.map(async (ref): Promise => { /* ... original ... */ return ref;});\n const settled = await Promise.allSettled(promises);\n /* ... original processing of settled promises ... */\n log.push({ agent: agent, type: \"final\", message: results });\n return results;\n }\n\n // --- Main Agent Flow Logic (DYNAMIC based on ANALYSIS_AGENTS) ---\n async function runAgentFlow(\n input: { documentUrl?: string; documentText?: string; documentFile?: File },\n log: LogEntry[],\n ): Promise {\n const openai = new OpenAI();\n log.push({ agent: \"System\", type: \"step\", message: \"Workflow started.\" });\n // ... initial logging of input type ...\n let docText: string | null = null;\n const ingestAgent = \"Ingestion Agent\";\n\n // --- Ingestion (Largely unchanged, error messages could use localization keys if server logs are localized) ---\n /* ... original ingestion logic ... */\n if (!docText) { /* ... original halt ... */ return log; }\n\n // --- Truncation (Uses APP_CONFIG) ---\n const MAX_LEN = APP_CONFIG.text_truncation_length || 25000;\n let truncText = docText;\n if (docText.length > MAX_LEN) { /* ... original truncation log ... */ }\n\n // --- DYNAMIC AGENT EXECUTION ---\n let previousAgentOutput = null; // For potential chaining, though simple independent execution first\n let extractedReferencesForTraversal: Reference[] | null = null;\n\n for (const agentConfig of ANALYSIS_AGENTS) {\n const currentAgentName = agentConfig.agent_name_en; // Use English name for consistent logging\n log.push({ agent: \"System\", type: \"step\", message: `Executing Agent: ${currentAgentName}...` });\n\n // Prepare user message for the agent (can be enriched with previous output if chaining is implemented)\n // For now, all agents operate on the truncated document text.\n // The system_prompt in agentConfig can use {{document_text}} which we replace here.\n const agentSystemPrompt = agentConfig.system_prompt.replace(\"{{document_text}}\", truncText);\n // If chaining is needed, {{previous_output}} could be another placeholder in prompts.\n\n const agentResult = await callOpenAI(\n openai,\n agentSystemPrompt, // The agent's specific prompt\n truncText, // User message is the doc text itself, or could be empty if prompt is self-contained\n APP_CONFIG.openai_model_name,\n agentConfig.expects_json\n );\n\n if (agentResult.role === \"assistant\" && agentResult.content) {\n log.push({ agent: currentAgentName, type: \"result\", message: agentResult.content });\n previousAgentOutput = agentResult.content; // Store for potential chaining\n\n // Special handling if this agent was the reference extractor\n if (agentConfig.agent_id === 'reference_identifier' && agentConfig.expects_json && \n typeof agentResult.content === 'object' && agentResult.content !== null &&\n Array.isArray(agentResult.content[agentConfig.ui_display_info.source_key || 'references'])) {\n extractedReferencesForTraversal = agentResult.content[agentConfig.ui_display_info.source_key || 'references'];\n }\n\n } else {\n log.push({ agent: currentAgentName, type: \"error\", message: agentResult.content });\n // Decide if workflow should halt on a single agent error, or continue\n // For now, continue with other agents.\n }\n }\n\n // --- Optional Reference Traversal (if references were extracted) ---\n // This part is kept if reference analysis is a common desired feature.\n // It looks for an agent with id 'reference_identifier' having produced output.\n if (extractedReferencesForTraversal && extractedReferencesForTraversal.length > 0) {\n log.push({ agent: \"System\", type: \"step\", message: \"Attempting to verify reference URLs...\" });\n // The traverseReferences function logs its own 'final' results for the 'Reference Traversal Agent'\n await traverseReferences(extractedReferencesForTraversal, log);\n } else {\n log.push({ agent: \"Reference Traversal Agent\", type: \"step\", message: \"Skipped - no valid references for traversal.\" });\n log.push({ agent: \"Reference Traversal Agent\", type: \"final\", message: [] }); // Log empty final state\n }\n\n\n log.push({ agent: \"System\", type: \"final\", message: \"Workflow finished.\" });\n return log;\n }\n\n // --- Request Handling Logic (Largely unchanged, uses APP_CONFIG for source URL) ---\n const url = new URL(req.url);\n const format = url.searchParams.get(\"format\");\n const sourceUrl = APP_CONFIG.footer_powered_by_url || import.meta.url.replace(\"esm.town\", \"val.town\"); // Use configured source URL\n const CORS_HEADERS = { /* ... original ... */ };\n if (req.method === \"OPTIONS\") { /* ... original ... */ }\n\n // Determine current locale strings for HTML shell generation\n // This is a simplified way; proper locale negotiation might be needed if server generates localized HTML.\n // For client-side localization, this mainly affects initial title.\n const initialLocale = APP_CONFIG.default_language === 'es' ? {{localization_strings_es_json}} : {{localization_strings_en_json}};\n initialLocale.lang = APP_CONFIG.default_language || 'en';\n\n\n if (format === \"json\" && req.method === \"POST\") {\n // ... original POST request parsing logic ...\n // ... call runAgentFlow ...\n // ... return JSON response ...\n let inputData: { documentUrl?: string; documentText?: string; documentFile?: File } = {};\n const initialLog: LogEntry[] = [];\n let statusCode = 200;\n try {\n // ... (Original input parsing logic from user's code) ...\n const ct = req.headers.get(\"content-type\");\n if (ct?.includes(\"multipart/form-data\")) { /* ... */ }\n else if (ct?.includes(\"application/json\")) { /* ... */ }\n // ... etc. ...\n if (!inputData.documentFile && !inputData.documentText && !inputData.documentUrl) {\n throw new Error(\"No document input provided in the request.\");\n }\n\n const finalLog = await runAgentFlow(inputData, initialLog);\n const sysHalt = finalLog.some(e => e.agent === \"System\" && e.type === \"error\" && e.message?.includes(\"Halt\"));\n statusCode = sysHalt ? 500 : 200;\n return new Response(JSON.stringify({ log: finalLog }), {\n status: statusCode,\n headers: { ...CORS_HEADERS, \"Content-Type\": \"application/json\" },\n });\n } catch (error) { /* ... original error handling ... */ \n return new Response(\n JSON.stringify({ error: \"Failed to process request.\", details: error.message, log: initialLog }),\n { status: 500, headers: { ...CORS_HEADERS, \"Content-Type\": \"application/json\" } }\n );\n }\n } else { // Serve HTML\n const initialUrlParam = url.searchParams.get(\"documentUrl\") || \"\";\n const initialTextParam = url.searchParams.get(\"documentText\") || \"\";\n // Pass necessary configs to generateHtmlShell\n const html = generateHtmlShell(initialUrlParam, initialTextParam, sourceUrl, APP_CONFIG, ANALYSIS_AGENTS, initialLocale);\n return new Response(html, { status: 200, headers: { \"Content-Type\": \"text/html; charset=utf-8\" } });\n }\n}\n\n/**\n * --- HOW TO USE THIS TEMPLATE ---\n * 1. Define Application Configuration:\n * Fill in the \\`{{app_config_json}}\\` placeholder with general settings for your app\n * (e.g., OpenAI model, max file size, default language).\n *\n * 2. Define Analysis Agents:\n * Fill in the \\`{{analysis_agents_json}}` placeholder. This is an array where each object defines an \"agent\":\n * - `agent_id`: A unique machine-readable ID.\n * - `agent_name_en`/`agent_name_es`: Human-readable names for UI and logs.\n * - `system_prompt`: The OpenAI prompt for this agent. Can use `{{document_text}}`.\n * - `expects_json`: Boolean, if the prompt asks OpenAI for JSON output.\n * - `ui_display_info`: How to render this agent's results:\n * - `card_title_en`/`card_title_es`: Title for the results card.\n * - `display_type`: \"paragraph\", \"list_of_strings\", \"list_of_objects\", \"key_value_pairs\", or \"table\".\n * - `source_key`: (If `expects_json` is true) The key in the agent's JSON output that contains the data to display.\n * - `table_columns`: (If `display_type` is \"table\") Array defining table columns ({header_en, header_es, data_key, is_link}).\n * - `object_properties`: (If `display_type` is \"list_of_objects\") Array defining properties to display for each object.\n *\n * 3. Define Localization Strings:\n * Fill in `{{localization_strings_en_json}}` and `{{localization_strings_es_json}}` with all\n * UI text for English and Spanish respectively. Ensure keys match `data-translate` attributes in the HTML\n * and keys used in the JavaScript `translate()` function.\n * Card titles and table headers defined in `ui_display_info` for agents will also need corresponding\n * entries if you want them translated beyond the direct `card_title_en/es` provided (e.g. for dynamic labels within a key_value_pairs display).\n *\n * 4. PDF Extraction:\n * The template uses 'npm:pdf.js-extract'. Ensure this is available in your environment.\n * The input file type is configurable via `{{app_config.document_format_accepted_mime_type}}` (e.g. \"application/pdf\")\n * and `{{app_config.document_format_accepted_label}}` (e.g. \"PDF\") for UI text.\n *\n * 5. OpenAI API Key:\n * Ensure your environment (e.g., Val Town secrets) has the `OPENAI_API_KEY` (or the appropriate\n * environment variable name for the `OpenAI` library) set.\n *\n * 6. Deployment:\n * This script is designed to be deployed as a single serverless function (e.g., on Val Town).\n * It serves both the HTML interface and the backend API endpoint.\n *\n * --- Customization Notes ---\n * - Agent Chaining: The current template executes agents mostly independently on the main document text.\n * To implement agent chaining (where one agent's output is input to another), you would modify\n * `runAgentFlow` to pass `previousAgentOutput` and update agent prompts to accept/use it (e.g. `{{previous_output}}`).\n * - Advanced UI Rendering: The `renderAgentResult` function provides basic rendering for different data types.\n * For more complex visualizations or interactions, you'd expand this function or add new `display_type` handlers.\n * - Styling: The CSS is extensive. You can customize it directly or by templatizing CSS variables within the `