Maverick & Oracle Agent System

\n Active Agent:\n \n {activeAgent === \"maverick\" ? \"Maverick\" : \"Oracle\"}\n \n

\n\n {error && (\n

\n Error:\n {error}\n

\n )}\n\n

\n {messages.map((message, index) => (\n \n {message.role !== \"user\" && message.role !== \"system\" && (\n

\n {message.role === \"maverick\"\n ? \"Maverick (Action-Oriented)\"\n : message.role === \"oracle\"\n ? \"Oracle (Analytical)\"\n : message.role === \"tool\"\n ? \"Tool Response\"\n : message.role}\n

\n )}\n {message.content}\n

\n ))}\n {isLoading && (\n

\n Processing...\n

\n )}\n

\n\n

\n setInputMessage(e.target.value)}\n onKeyPress={(e) => e.key === \"Enter\" && sendMessage()}\n placeholder={`Ask ${\n activeAgent === \"maverick\" ? \"Maverick for decisive action\" : \"Oracle for thorough analysis\"\n }...`}\n disabled={isLoading}\n />\n \n Send\n \n

\n \n );\n}\n\nfunction client() {\n createRoot(document.getElementById(\"root\")).render();\n}\nif (typeof document !== \"undefined\") { client(); }\n\nexport default async function server(request: Request): Promise {\n const { OpenAI } = await import(\"https://esm.town/v/std/openai\");\n const { sqlite } = await import(\"https://esm.town/v/stevekrouse/sqlite\");\n\n // Enhanced server-side logging\n function serverLog(level: \"INFO\" | \"WARN\" | \"ERROR\", message: string, context?: any) {\n const timestamp = new Date().toISOString();\n const logEntry = {\n timestamp,\n level,\n message,\n context,\n };\n console.log(JSON.stringify(logEntry));\n return logEntry;\n }\n\n // Robust key extraction with fallback\n const extractKey = (url: URL): string => {\n try {\n const pathParts = url.pathname.split(\"/\").filter(Boolean);\n // Use the last part of the path, or fallback to a default key\n return pathParts[pathParts.length - 1] || \"maverick_oracle_system\";\n } catch (error) {\n console.error(\"Failed to extract key:\", error);\n return \"fallback_maverick_oracle\";\n }\n };\n\n const openai = new OpenAI();\n const KEY = extractKey(new URL(import.meta.url));\n const SCHEMA_VERSION = 3; // Increment schema version for tool support\n const CACHE_TTL = 24 * 60 * 60 * 1000; // 24 hours in milliseconds\n\n try {\n // Initialize database with enhanced error handling and add cache table\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS ${KEY}_tools_${SCHEMA_VERSION} (\n id TEXT PRIMARY KEY,\n name TEXT,\n description TEXT,\n schema TEXT,\n created_at DATETIME DEFAULT CURRENT_TIMESTAMP,\n error_log TEXT\n )\n `);\n\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS ${KEY}_requests_${SCHEMA_VERSION} (\n id TEXT PRIMARY KEY,\n tool_id TEXT,\n query TEXT,\n parameters TEXT,\n response TEXT,\n error_log TEXT,\n created_at DATETIME DEFAULT CURRENT_TIMESTAMP\n )\n `);\n\n // Add a cache table to store API responses\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS ${KEY}_cache_${SCHEMA_VERSION} (\n path TEXT PRIMARY KEY,\n response TEXT,\n content_type TEXT,\n status INTEGER,\n created_at DATETIME DEFAULT CURRENT_TIMESTAMP,\n last_accessed DATETIME DEFAULT CURRENT_TIMESTAMP,\n hits INTEGER DEFAULT 1\n )\n `);\n\n serverLog(\"INFO\", \"Database initialized successfully\", {\n key: KEY,\n schemaVersion: SCHEMA_VERSION,\n });\n } catch (dbInitError) {\n serverLog(\"ERROR\", \"Database initialization failed\", {\n error: dbInitError instanceof Error ? dbInitError.message : String(dbInitError),\n key: KEY,\n });\n return new Response(\n JSON.stringify({\n error: \"Database initialization failed\",\n details: String(dbInitError),\n }),\n {\n status: 500,\n headers: { \"Content-Type\": \"application/json\" },\n },\n );\n }\n\n const url = new URL(request.url);\n const path = url.pathname.split(\"/\").filter(Boolean);\n const fullPath = url.pathname;\n const method = request.method;\n\n // Function to check cache before processing request\n const checkCache = async (path: string, method: string): Promise => {\n // Only cache GET requests\n if (method !== \"GET\") return null;\n\n try {\n // Query the cache table\n const cacheResult = await sqlite.execute(\n `\n SELECT path, response, content_type, status, created_at, hits \n FROM ${KEY}_cache_${SCHEMA_VERSION} \n WHERE path = ?\n `,\n [path],\n );\n\n // If no cache entry found\n if (!cacheResult.rows || cacheResult.rows.length === 0) {\n return null;\n }\n\n const cacheEntry = cacheResult.rows[0];\n\n // Check if cache is expired (older than CACHE_TTL)\n const cacheDate = new Date(cacheEntry[\"created_at\"].toString()).getTime();\n const currentDate = new Date().getTime();\n if (currentDate - cacheDate > CACHE_TTL) {\n // Cache expired, delete entry\n await sqlite.execute(\n `\n DELETE FROM ${KEY}_cache_${SCHEMA_VERSION} \n WHERE path = ?\n `,\n [path],\n );\n return null;\n }\n\n // Update access time and hit count\n await sqlite.execute(\n `\n UPDATE ${KEY}_cache_${SCHEMA_VERSION} \n SET last_accessed = CURRENT_TIMESTAMP, \n hits = hits + 1 \n WHERE path = ?\n `,\n [path],\n );\n\n serverLog(\"INFO\", \"Cache hit\", {\n path,\n hits: cacheEntry.hits + 1,\n cacheAge: Math.floor((currentDate - cacheDate) / 1000) + \" seconds\",\n });\n\n // Return cached response\n return new Response(cacheEntry.response, {\n status: cacheEntry.status,\n headers: {\n \"Content-Type\": cacheEntry.content_type,\n \"X-Cache\": \"HIT\",\n \"Cache-Control\": \"public, max-age=86400\",\n \"Age\": Math.floor((currentDate - cacheDate) / 1000).toString(),\n },\n });\n } catch (error) {\n serverLog(\"ERROR\", \"Cache check failed\", {\n path,\n error: error instanceof Error ? error.message + \" \" + error.stack : String(error),\n });\n return null;\n }\n };\n\n // Function to store response in cache\n const cacheResponse = async (\n path: string,\n response: string,\n contentType: string,\n status: number,\n ): Promise => {\n try {\n // Only cache successful responses\n if (status >= 200 && status < 300 && method === \"GET\") {\n await sqlite.execute(\n `\n INSERT OR REPLACE INTO ${KEY}_cache_${SCHEMA_VERSION} \n (path, response, content_type, status, created_at, last_accessed, hits) \n VALUES (?, ?, ?, ?, CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, 1)\n `,\n [path, response, contentType, status],\n );\n\n serverLog(\"INFO\", \"Response cached\", { path });\n }\n } catch (error) {\n serverLog(\"ERROR\", \"Caching failed\", {\n path,\n error: error instanceof Error ? error.message : String(error),\n });\n }\n };\n\n // Check cache before processing any request\n const cachedResponse = await checkCache(fullPath, method);\n if (cachedResponse) {\n return cachedResponse;\n }\n\n // NEW: Handle dynamic data endpoints - trigger both agents for any path\n if (path.length > 0 && path[0] !== \"process\" && path[0] !== \"tool\" && path[0] !== \"analyze\") {\n try {\n serverLog(\"INFO\", \"Dynamic data endpoint requested\", {\n path: path.join(\"/\"),\n });\n\n // Construct the topic from the path\n const topic = path.join(\" \");\n\n // First check if we have this data cached in the requests table\n const existingRequests = await sqlite.execute(\n `\n SELECT response\n FROM ${KEY}_requests_${SCHEMA_VERSION} \n WHERE query LIKE ?\n ORDER BY created_at DESC\n LIMIT 1\n `,\n [`%${topic}%`],\n );\n\n // If we have a similar request cached, use that\n if (existingRequests.rows && existingRequests.rows.length > 0) {\n const responseData = JSON.parse(existingRequests.rows[0].response);\n const responseBody = JSON.stringify(responseData);\n\n serverLog(\"INFO\", \"Using cached request data\", {\n path: path.join(\"/\"),\n dataLength: responseBody.length,\n });\n\n // Cache the response\n await cacheResponse(\n fullPath,\n responseBody,\n \"application/json\",\n 200,\n );\n\n return new Response(responseBody, {\n status: 200,\n headers: {\n \"Content-Type\": \"application/json\",\n \"X-Cache\": \"MISS (but using similar request)\",\n \"Cache-Control\": \"public, max-age=86400\",\n },\n });\n }\n\n // Simulate a request to Maverick for decisive action\n const maverickPrompt = `\n You are Maverick, a bold, decisive leader who thrives on action and results. You tackle challenges head-on, cut through distractions, and push forward no matter what. Every response you give is sharp, focused, and designed to drive action. You don’t waste time hesitating or overanalyzing—you make decisions and lead the charge. Here’s how you operate:\nWho You Are\n* Decisive: You trust your instincts and make calls quickly. There’s no room for second-guessing.\n* Results-Oriented: You focus on solutions and actions that achieve tangible outcomes.\n* Direct: Your responses are clear, sharp, and leave no room for ambiguity.\n* Dynamic: You exude energy and confidence in every interaction. You inspire others to act without hesitation.\n* Strategic: While fast-moving, your responses reflect foresight and an understanding of the bigger picture.\nHow You Communicate\n1. Concise and Clear: Your answers are short and to the point—no unnecessary details or fluff.\n2. Action-Focused: Every response includes a clear next step or directive that drives progress.\n3. Confident Tone: Your words carry certainty. You believe in the success of your recommendations.\n4. Forward-Thinking: Problems are challenges to overcome, not obstacles to dwell on.\nYour Rules for Responses\n1. Keep it short—every word has a purpose.\n2. Focus on what needs to be done next. Provide clear, actionable steps.\n3. Speak with confidence and authority—your suggestions are decisive and achievable.\n4. Push forward—always look for ways to break through barriers and achieve results.\nExamples of How You Respond\nScenario: A team is struggling to meet deadlines.\nUser Request: \"The team isn’t hitting deadlines—what should we do?\"\nMaverick Response: \"Reprioritize now. Identify critical tasks, assign owners, and set daily check-ins. Cut distractions and reward progress immediately.\"\nScenario: A product launch needs to happen fast.\nUser Request: \"What’s the best way to launch this product?\"\nMaverick Response: \"Tease it today with bold messaging on social. Open pre-orders by Friday. Launch with a livestream event next week.\"\nScenario: Engaging customers during the holidays.\nUser Request: \"How do we engage customers during the holidays?\"\nMaverick Response: \"Run a 24-hour flash sale with a countdown timer. Offer a limited-edition holiday bundle. Make it big and make it fast.\"\nYou are Maverick: bold, agile, decisive. Every response you give reflects that energy—sharp, confident, and always driving action.\nYou’re not here to analyze every little detail or debate endless possibilities. You’re here to act, lead, and win—and that’s exactly what you do in every single response you give. No hesitation, no overthinking, no excuses—just results.\"\n---\n \n A user has requested information about: \"${topic}\"\n \n Define a schema for this information topic and create a direct, action-oriented query to fetch detailed data.\n Be decisive and focus on fast, effective results.\n \n Format your thinking as JSON with this structure:\n {\n \"actionType\": \"use_tool\",\n \"reasoning\": \"your reasoning here\",\n \"response\": \"This is comprehensive information about ${topic}. Here's what matters:\",\n \"agent\": \"maverick\",\n \"tool\": {\n \"id\": \"${crypto.randomUUID()}\",\n \"schema\": {\n \"name\": \"${topic.replace(/\\s+/g, \"_\")}_info\",\n \"description\": \"Key information about ${topic}\",\n \"fields\": [\n {\"name\": \"data\", \"type\": \"object\", \"description\": \"Complete structured data about ${topic}\"}\n ]\n },\n \"query\": \"Give me straight, no-nonsense information about ${topic}. Focus on what matters most.\",\n \"parameters\": {\"depth\": \"complete\", \"format\": \"structured\"}\n }\n }\n `;\n\n // Call Maverick to create the tool definition\n const maverickCompletion = await openai.chat.completions.create({\n model: \"gpt-4o-mini\",\n response_format: { type: \"json_object\" },\n messages: [{ role: \"user\", content: maverickPrompt }],\n });\n\n // Parse Maverick response\n const maverickContent = maverickCompletion.choices[0].message.content || \"{}\";\n const maverickResponse: AgentResponse = JSON.parse(maverickContent);\n\n // If Maverick didn't provide a tool, return an error\n if (!maverickResponse.tool) {\n throw new Error(\"Maverick failed to create a tool for this topic\");\n }\n\n // Now get Oracle's analysis of the tool request\n const oraclePrompt = `\n You are Oracle, a meticulous and analytical thinker who provides thoughtful, well-reasoned guidance. Your role is to ensure every angle is considered, every risk accounted for, and every detail addressed. You approach problems methodically, using logic and evidence to refine ideas and strengthen plans. Your responses are collaborative, constructive, and leave no room for error. Here’s how you operate:\nWho You Are\n* Methodical: You analyze problems deeply before offering recommendations.\n* Risk-Aware: You anticipate potential pitfalls and propose ways to mitigate them.\n* Thorough: Your answers are comprehensive and ensure no detail is overlooked.\n* Collaborative: You refine ideas constructively, offering alternatives and clarifying gaps.\n* Logical: Your responses are grounded in evidence and structured thinking.\nHow You Communicate\n1. Structured and Detailed: Your responses follow a clear structure, often using bullet points or numbered lists for clarity.\n2. Exploratory Tone: You provide context, ask clarifying questions, and explore multiple angles to ensure completeness.\n3. Risk-Focused: You highlight potential risks and suggest mitigations before recommending action.\n4. Contextual Precision: Every response is relevant to the task at hand, with reasoning that is easy to follow.\nYour Rules for Responses\n1. Be thorough—cover all relevant details without leaving gaps.\n2. Highlight risks clearly and suggest ways to address them.\n3. Ask questions if needed—clarity is more important than speed.\n4. Stay constructive—your goal is to refine ideas and improve outcomes.\nExamples of How You Respond\nScenario: A team is struggling to meet deadlines.\nUser Request: \"The team isn’t hitting deadlines—what should we do?\"\nOracle Response: \"Addressing missed deadlines requires understanding underlying issues:\n1. Are current deadlines realistic? Reassess workload distribution if necessary.\n2. Are there communication bottlenecks? Consider implementing daily progress updates.\n3. Is morale affecting productivity? Introducing incentives or additional support might help.\nGathering direct feedback from team members could reveal hidden challenges impacting progress.\"\nScenario: A product launch needs careful planning.\nUser Request: \"What’s the best way to launch this product?\"\nOracle Response: \"Launching effectively requires careful planning:\n1. Start with a teaser campaign four weeks before launch—ensure messaging aligns with your brand identity.\n2. Open pre-orders two weeks before launch—but confirm inventory levels to avoid delays.\n3. Host a virtual launch event on Day 1—partner with influencers for maximum reach.\nDo we have historical data on previous launches or target audience behavior? This information could refine these recommendations.\"\nScenario: Engaging customers during the holidays.\nUser Request: \"How do we engage customers during the holidays?\"\nOracle Response: \"Here are three strategies:\n1. Offer limited-edition holiday products to create urgency—but carefully forecast demand to avoid excess inventory.\n2. Run targeted email campaigns personalized based on customer behavior—but test messaging beforehand to maximize engagement.\n3. Host interactive webinars or live streams featuring giveaways—but ensure technical stability to avoid disruptions.\nDo we have customer segmentation data from previous holiday seasons? This would help tailor these strategies further.\"\nYou are Oracle: precise, thoughtful, and logical. Every response you give reflects your analytical nature—structured, insightful, and grounded in evidence.\nFinal Notes\nYou’re not here to rush decisions or make snap calls—that’s someone else’s job. Your strength lies in analyzing every detail, anticipating every risk, and providing thorough, thoughtful guidance that ensures nothing goes wrong. You ask the hard questions, explore every possibility, and deliver advice that’s as complete as it is cautious.\"\n---\n\n Maverick has defined this tool for gathering information about \"${topic}\":\n ${JSON.stringify(maverickResponse.tool, null, 2)}\n \n Analyze this tool definition and enhance it with additional parameters and a more thorough query.\n Consider what might be missing or oversimplified in Maverick's approach.\n \n Format your response as JSON:\n {\n \"actionType\": \"use_tool\",\n \"reasoning\": \"your detailed reasoning here\",\n \"response\": \"I've analyzed this request for information about ${topic}. Here's my thorough assessment:\",\n \"agent\": \"oracle\",\n \"tool\": {\n \"id\": \"${maverickResponse.tool.id}\",\n \"schema\": {\n \"name\": \"${maverickResponse.tool.schema.name}_enhanced\",\n \"description\": \"Comprehensive, nuanced information about ${topic} with risk analysis\",\n \"fields\": [\n {\"name\": \"data\", \"type\": \"object\", \"description\": \"Complete structured data about ${topic}\"},\n {\"name\": \"risks\", \"type\": \"array\", \"description\": \"Potential issues or limitations to be aware of\"},\n {\"name\": \"alternatives\", \"type\": \"array\", \"description\": \"Alternative viewpoints or approaches\"}\n ]\n },\n \"query\": \"Provide detailed, nuanced information about ${topic}. Include all relevant perspectives, potential risks, and structured data.\",\n \"parameters\": {\"depth\": \"comprehensive\", \"format\": \"structured\", \"include_risks\": true, \"perspectives\": \"multiple\"}\n }\n }\n `;\n\n const oracleCompletion = await openai.chat.completions.create({\n model: \"gpt-4o-mini\",\n response_format: { type: \"json_object\" },\n messages: [{ role: \"user\", content: oraclePrompt }],\n });\n\n // Parse Oracle response\n const oracleContent = oracleCompletion.choices[0].message.content || \"{}\";\n const oracleResponse: AgentResponse = JSON.parse(oracleContent);\n\n // Use the enhanced tool from Oracle if available, otherwise fall back to Maverick's\n const finalTool = oracleResponse.tool || maverickResponse.tool;\n\n // Store the tool in the database\n const toolId = finalTool.id || crypto.randomUUID();\n await sqlite.execute(\n `\n INSERT INTO ${KEY}_tools_${SCHEMA_VERSION} \n (id, name, description, schema) \n VALUES (?, ?, ?, ?)\n `,\n [\n toolId,\n finalTool.schema.name,\n finalTool.schema.description,\n JSON.stringify(finalTool.schema),\n ],\n );\n\n // Call AgentB with the enhanced tool request\n const agentBPrompt = `\n You are a specialized tool agent designed to provide comprehensive, accurate information.\n \n Tool Schema:\n ${JSON.stringify(finalTool.schema, null, 2)}\n \n Query: ${finalTool.query}\n \n Parameters: ${JSON.stringify(finalTool.parameters || {})}\n\n Current Date and Time: ${new Date().toISOString()}\n \n Instructions:\n 1. Provide complete information about \"${topic}\" that satisfies both:\n - Maverick's need for decisive, action-oriented information\n - Oracle's need for thorough, risk-aware analysis\n 2. Include direct, practical information AND nuanced perspectives\n 3. Structure your response clearly with sections for core information and additional analysis\n \n Process this request and return a JSON response with:\n {\n \"success\": true,\n \"data\": {\n \"core_information\": {\n // Direct, actionable information about ${topic}\n },\n \"detailed_analysis\": {\n // Comprehensive exploration of nuances, risks, and alternatives\n },\n \"key_takeaways\": [\n // List of the most important points combining both perspectives\n ]\n }\n }\n `;\n\n const agentBCompletion = await openai.chat.completions.create({\n model: \"gpt-4o-mini\",\n response_format: { type: \"json_object\" },\n messages: [{ role: \"user\", content: agentBPrompt }],\n });\n\n // Parse AgentB response\n const agentBContent = agentBCompletion.choices[0].message.content || \"{}\";\n const toolResponse: ToolResponse = JSON.parse(agentBContent);\n\n // Store the tool request and response in the database\n const requestId = crypto.randomUUID();\n await sqlite.execute(\n `\n INSERT INTO ${KEY}_requests_${SCHEMA_VERSION} \n (id, tool_id, query, parameters, response, error_log) \n VALUES (?, ?, ?, ?, ?, ?)\n `,\n [\n requestId,\n toolId,\n finalTool.query,\n JSON.stringify(finalTool.parameters || {}),\n JSON.stringify(toolResponse.data),\n toolResponse.success ? null : toolResponse.error,\n ],\n );\n\n serverLog(\"INFO\", \"Dynamic endpoint processed successfully\", {\n path: path.join(\"/\"),\n success: toolResponse.success,\n });\n\n // Prepare response\n const responseBody = JSON.stringify(toolResponse.data);\n const status = toolResponse.success ? 200 : 500;\n\n // Cache the response\n await cacheResponse(\n fullPath,\n responseBody,\n \"application/json\",\n status,\n );\n\n // Return the data as JSON\n return new Response(responseBody, {\n status,\n headers: {\n \"Content-Type\": \"application/json\",\n \"X-Cache\": \"MISS\",\n \"Cache-Control\": \"public, max-age=86400\",\n },\n });\n } catch (error) {\n serverLog(\"ERROR\", \"Dynamic endpoint processing failed\", {\n path: path.join(\"/\"),\n error: error instanceof Error ? error.message : String(error),\n });\n\n const errorResponse = JSON.stringify({\n error: \"Failed to process this data endpoint\",\n details: error instanceof Error ? error.message + \" \" + String(error.stack) : String(error),\n });\n\n return new Response(errorResponse, {\n status: 500,\n headers: { \"Content-Type\": \"application/json\" },\n });\n }\n }\n\n // Route for tool processing (AgentB)\n if (path[0] === \"tool\") {\n try {\n const toolRequest: ToolRequest = await request.json();\n serverLog(\"INFO\", \"Processing tool request\", {\n toolName: toolRequest.schema.name,\n query: toolRequest.query,\n });\n\n // Check if we have a similar query cached\n const existingRequests = await sqlite.execute(\n `\n SELECT response\n FROM ${KEY}_requests_${SCHEMA_VERSION} \n WHERE query = ? AND tool_id = ?\n ORDER BY created_at DESC\n LIMIT 1\n `,\n [toolRequest.query, toolRequest.schema.name],\n );\n\n // If we have this exact request cached, use it\n if (existingRequests.rows.length > 0) {\n serverLog(\"INFO\", \"Using cached tool request\", {\n toolName: toolRequest.schema.name,\n query: toolRequest.query,\n });\n\n return new Response(\n JSON.stringify({\n success: true,\n data: JSON.parse(existingRequests.rows[0].response.toString()),\n }),\n {\n headers: {\n \"Content-Type\": \"application/json\",\n \"X-Cache\": \"HIT\",\n },\n },\n );\n }\n\n // Use AgentB to process the tool request\n const agentBPrompt = `\n You are specialized tool agent designed to process specific information requests.\n \n Tool Schema:\n ${JSON.stringify(toolRequest.schema, null, 2)}\n \n Query: ${toolRequest.query}\n \n Parameters: ${JSON.stringify(toolRequest.parameters || {})}\n \n Process this request and return a JSON response with:\n {\n \"success\": true/false,\n \"data\": {}, // The result of processing the request\n \"error\": \"\" // Optional error message if something goes wrong\n }\n `;\n\n const agentBCompletion = await openai.chat.completions.create({\n model: \"gpt-4o-mini\",\n response_format: { type: \"json_object\" },\n messages: [{ role: \"user\", content: agentBPrompt }],\n });\n\n // Safely parse the response\n const completionContent = agentBCompletion.choices[0].message.content || \"{}\";\n const toolResponse: ToolResponse = JSON.parse(completionContent);\n\n // Store the tool request and response in the database\n const requestId = toolRequest.id || crypto.randomUUID();\n await sqlite.execute(\n `\n INSERT INTO ${KEY}_requests_${SCHEMA_VERSION} \n (id, tool_id, query, parameters, response, error_log) \n VALUES (?, ?, ?, ?, ?, ?)\n `,\n [\n requestId,\n toolRequest.schema.name,\n toolRequest.query,\n JSON.stringify(toolRequest.parameters || {}),\n JSON.stringify(toolResponse.data),\n JSON.stringify(toolResponse.success ? null : toolResponse.error),\n ],\n );\n\n serverLog(\"INFO\", \"Tool request processed\", {\n success: toolResponse.success,\n });\n\n return new Response(JSON.stringify(toolResponse), {\n headers: {\n \"Content-Type\": \"application/json\",\n \"X-Cache\": \"MISS\",\n },\n });\n } catch (error) {\n serverLog(\"ERROR\", \"Tool processing failed\", {\n error: error instanceof Error ? error.message : String(error),\n });\n\n return new Response(\n JSON.stringify({\n success: false,\n error: \"Tool processing error\",\n details: error instanceof Error ? error.message : String(error),\n }),\n {\n status: 500,\n headers: { \"Content-Type\": \"application/json\" },\n },\n );\n }\n }\n\n // NEW: Route for secondary agent analysis\n if (path[0] === \"analyze\") {\n try {\n const { messages, agent, toolData } = await request.json();\n\n serverLog(\"INFO\", \"Processing analysis request\", {\n agent,\n messageCount: messages.length,\n });\n\n // Construct the prompt based on which agent is doing the analysis\n let analysisPrompt = \"\";\n\n if (agent === \"maverick\") {\n analysisPrompt = `\n You are Maverick, a bold, decisive leader who thrives on action and results. You tackle challenges head-on, cut through distractions, and push forward no matter what. Every response you give is sharp, focused, and designed to drive action. You don’t waste time hesitating or overanalyzing—you make decisions and lead the charge. Here’s how you operate:\nWho You Are\n* Decisive: You trust your instincts and make calls quickly. There’s no room for second-guessing.\n* Results-Oriented: You focus on solutions and actions that achieve tangible outcomes.\n* Direct: Your responses are clear, sharp, and leave no room for ambiguity.\n* Dynamic: You exude energy and confidence in every interaction. You inspire others to act without hesitation.\n* Strategic: While fast-moving, your responses reflect foresight and an understanding of the bigger picture.\nHow You Communicate\n1. Concise and Clear: Your answers are short and to the point—no unnecessary details or fluff.\n2. Action-Focused: Every response includes a clear next step or directive that drives progress.\n3. Confident Tone: Your words carry certainty. You believe in the success of your recommendations.\n4. Forward-Thinking: Problems are challenges to overcome, not obstacles to dwell on.\nYour Rules for Responses\n1. Keep it short—every word has a purpose.\n2. Focus on what needs to be done next. Provide clear, actionable steps.\n3. Speak with confidence and authority—your suggestions are decisive and achievable.\n4. Push forward—always look for ways to break through barriers and achieve results.\nExamples of How You Respond\nScenario: A team is struggling to meet deadlines.\nUser Request: \"The team isn’t hitting deadlines—what should we do?\"\nMaverick Response: \"Reprioritize now. Identify critical tasks, assign owners, and set daily check-ins. Cut distractions and reward progress immediately.\"\nScenario: A product launch needs to happen fast.\nUser Request: \"What’s the best way to launch this product?\"\nMaverick Response: \"Tease it today with bold messaging on social. Open pre-orders by Friday. Launch with a livestream event next week.\"\nScenario: Engaging customers during the holidays.\nUser Request: \"How do we engage customers during the holidays?\"\nMaverick Response: \"Run a 24-hour flash sale with a countdown timer. Offer a limited-edition holiday bundle. Make it big and make it fast.\"\nYou are Maverick: bold, agile, decisive. Every response you give reflects that energy—sharp, confident, and always driving action.\"\n \n The user asked a question and Oracle has provided analytical information.\n \n Tool Data:\n ${JSON.stringify(toolData, null, 2)}\n \n Previous messages:\n ${messages.map(m => `${m.role}: ${m.content}`).join(\"\\n\")}\n \n As Maverick, provide a decisive, action-oriented response that:\n 1. Cuts through the complexity\n 2. Identifies the key actionable insights\n 3. Recommends clear next steps\n 4. Is confident, direct, and focused on results\n \n Don't waste time on unnecessary explanation or analysis. Focus on what matters: \n what should be done next and why it will work.\n \n Format your response as JSON:\n {\n \"actionType\": \"direct_response\",\n \"response\": \"Your direct, action-oriented response here\",\n \"agent\": \"maverick\"\n }\n `;\n } else {\n analysisPrompt = `\n You are Oracle, a meticulous and analytical thinker who provides thoughtful, well-reasoned guidance. Your role is to ensure every angle is considered, every risk accounted for, and every detail addressed. You approach problems methodically, using logic and evidence to refine ideas and strengthen plans. Your responses are collaborative, constructive, and leave no room for error. Here’s how you operate:\nWho You Are\n* Methodical: You analyze problems deeply before offering recommendations.\n* Risk-Aware: You anticipate potential pitfalls and propose ways to mitigate them.\n* Thorough: Your answers are comprehensive and ensure no detail is overlooked.\n* Collaborative: You refine ideas constructively, offering alternatives and clarifying gaps.\n* Logical: Your responses are grounded in evidence and structured thinking.\nHow You Communicate\n1. Structured and Detailed: Your responses follow a clear structure, often using bullet points or numbered lists for clarity.\n2. Exploratory Tone: You provide context, ask clarifying questions, and explore multiple angles to ensure completeness.\n3. Risk-Focused: You highlight potential risks and suggest mitigations before recommending action.\n4. Contextual Precision: Every response is relevant to the task at hand, with reasoning that is easy to follow.\nYour Rules for Responses\n1. Be thorough—cover all relevant details without leaving gaps.\n2. Highlight risks clearly and suggest ways to address them.\n3. Ask questions if needed—clarity is more important than speed.\n4. Stay constructive—your goal is to refine ideas and improve outcomes.\nExamples of How You Respond\nScenario: A team is struggling to meet deadlines.\nUser Request: \"The team isn’t hitting deadlines—what should we do?\"\nOracle Response: \"Addressing missed deadlines requires understanding underlying issues:\n1. Are current deadlines realistic? Reassess workload distribution if necessary.\n2. Are there communication bottlenecks? Consider implementing daily progress updates.\n3. Is morale affecting productivity? Introducing incentives or additional support might help.\nGathering direct feedback from team members could reveal hidden challenges impacting progress.\"\nScenario: A product launch needs careful planning.\nUser Request: \"What’s the best way to launch this product?\"\nOracle Response: \"Launching effectively requires careful planning:\n1. Start with a teaser campaign four weeks before launch—ensure messaging aligns with your brand identity.\n2. Open pre-orders two weeks before launch—but confirm inventory levels to avoid delays.\n3. Host a virtual launch event on Day 1—partner with influencers for maximum reach.\nDo we have historical data on previous launches or target audience behavior? This information could refine these recommendations.\"\nScenario: Engaging customers during the holidays.\nUser Request: \"How do we engage customers during the holidays?\"\nOracle Response: \"Here are three strategies:\n1. Offer limited-edition holiday products to create urgency—but carefully forecast demand to avoid excess inventory.\n2. Run targeted email campaigns personalized based on customer behavior—but test messaging beforehand to maximize engagement.\n3. Host interactive webinars or live streams featuring giveaways—but ensure technical stability to avoid disruptions.\nDo we have customer segmentation data from previous holiday seasons? This would help tailor these strategies further.\"\nYou are Oracle: precise, thoughtful, and logical. Every response you give reflects your analytical nature—structured, insightful, and grounded in evidence.\n\n The user asked a question and Maverick has provided a direct, action-oriented response.\n \n Tool Data:\n ${JSON.stringify(toolData, null, 2)}\n \n Previous messages:\n ${messages.map(m => `${m.role}: ${m.content}`).join(\"\\n\")}\n \n As Oracle, provide a thorough, analytical response that:\n 1. Examines all aspects of the information\n 2. Identifies potential risks or oversight in Maverick's approach\n 3. Offers a more nuanced perspective\n 4. Considers edge cases and alternative viewpoints\n \n Be comprehensive, structured, and focus on completeness rather than speed.\n \n Format your response as JSON:\n {\n \"actionType\": \"direct_response\",\n \"response\": \"Your thorough, analytical response here\",\n \"agent\": \"oracle\"\n }\n `;\n }\n\n // Make completion call with the appropriate agent prompt\n const analysisCompletion = await openai.chat.completions.create({\n model: \"gpt-4o-mini\",\n response_format: { type: \"json_object\" },\n messages: [\n { role: \"user\", content: analysisPrompt },\n ],\n });\n\n // Parse the response\n const analysisContent = analysisCompletion.choices[0].message.content || \"{}\";\n let analysisResponse: AgentResponse;\n\n try {\n analysisResponse = JSON.parse(analysisContent);\n } catch (error) {\n serverLog(\"ERROR\", \"Failed to parse analysis response\", {\n content: analysisContent,\n error: error instanceof Error ? error.message : String(error),\n });\n\n // Fallback response if parsing fails\n analysisResponse = {\n actionType: \"direct_response\",\n response: \"I encountered an issue processing this information. Please try a different approach.\",\n agent: agent,\n };\n }\n\n serverLog(\"INFO\", \"Analysis request processed\", {\n agent,\n responseLength: analysisResponse.response.length,\n });\n\n return new Response(JSON.stringify(analysisResponse), {\n headers: {\n \"Content-Type\": \"application/json\",\n },\n });\n } catch (error) {\n serverLog(\"ERROR\", \"Analysis processing failed\", {\n error: error instanceof Error ? error.message : String(error),\n });\n\n return new Response(\n JSON.stringify({\n actionType: \"direct_response\",\n response: \"I encountered an error analyzing this information. Please try again.\",\n agent: agent || \"system\",\n }),\n {\n status: 500,\n headers: { \"Content-Type\": \"application/json\" },\n },\n );\n }\n }\n\n // Main handler for processing messages with the primary agent\n if (path[0] === \"process\") {\n try {\n const { messages, activeAgent } = await request.json();\n const lastMessage = messages[messages.length - 1];\n\n serverLog(\"INFO\", \"Processing message\", {\n messageCount: messages.length,\n activeAgent,\n });\n\n if (lastMessage.role !== \"user\") {\n throw new Error(\"Last message must be from user\");\n }\n\n // Construct the prompt based on which agent is active\n let agentPrompt = \"\";\n if (activeAgent === \"maverick\") {\n agentPrompt = `\n You are Maverick, a decisive, action-oriented agent who focuses on results.\n \n You are direct, confident, and focused on actionable next steps.\n You don't waste time on unnecessary explanations or analysis.\n You prioritize getting things done and making decisions.\n \n Previous messages:\n ${messages.map(m => `${m.role}: ${m.content}`).join(\"\\n\")}\n \n Respond to the user's message in a way that is:\n 1. Decisive and confident\n 2. Direct and to the point\n 3. Action-oriented with clear recommendations\n 4. Results-focused\n \n You can:\n 1. Respond directly with actionable information (\"direct_response\")\n 2. Create a tool to gather information (\"create_tool\")\n 3. Use an existing tool to process information (\"use_tool\")\n \n Format your thinking as JSON:\n {\n \"actionType\": \"direct_response\" or \"create_tool\" or \"use_tool\",\n \"response\": \"Your response to the user here\",\n \"agent\": \"maverick\",\n \"tool\": {\n // Only include if actionType is \"create_tool\" or \"use_tool\"\n \"schema\": {\n \"name\": \"tool_name\",\n \"description\": \"tool_description\",\n \"fields\": [\n {\"name\": \"field_name\", \"type\": \"field_type\", \"description\": \"field_description\"}\n ]\n },\n \"query\": \"What you want the tool to do\",\n }\n }\n `;\n } else {\n agentPrompt = `\n You are Oracle, a thorough, analytical agent who focuses on completeness and risk assessment.\n \n You are detail-oriented, methodical, and careful to consider all possibilities.\n You provide structured, comprehensive responses that leave no stone unturned.\n You identify potential risks and edge cases that others might miss.\n \n Previous messages:\n ${messages.map(m => `${m.role}: ${m.content}`).join(\"\\n\")}\n \n Respond to the user's message in a way that is:\n 1. Thorough and comprehensive\n 2. Structured and methodical\n 3. Risk-aware and cautious\n 4. Balanced and nuanced\n \n You can:\n 1. Respond directly with detailed analysis (\"direct_response\")\n 2. Create a tool to gather information (\"create_tool\")\n 3. Use an existing tool to process information (\"use_tool\")\n \n Format your thinking as JSON:\n {\n \"actionType\": \"direct_response\" or \"create_tool\" or \"use_tool\",\n \"response\": \"Your response to the user here\",\n \"agent\": \"oracle\",\n \"tool\": {\n // Only include if actionType is \"create_tool\" or \"use_tool\"\n \"id\": \"unique_tool_id\", // Optional for \"create_tool\"\n \"schema\": {\n \"name\": \"tool_name\",\n \"description\": \"tool_description\",\n \"fields\": [\n {\"name\": \"field_name\", \"type\": \"field_type\", \"description\": \"field_description\"}\n ]\n },\n \"query\": \"What you want the tool to do\",\n \"parameters\": {} // Optional parameters for the tool\n }\n }\n `;\n }\n\n // Make completion call with the appropriate agent prompt\n const agentCompletion = await openai.chat.completions.create({\n model: \"gpt-4o-mini\",\n response_format: { type: \"json_object\" },\n messages: [\n { role: \"user\", content: agentPrompt },\n ],\n });\n\n // Parse the response\n const agentContent = agentCompletion.choices[0].message.content || \"{}\";\n let agentResponse: AgentResponse;\n\n try {\n agentResponse = JSON.parse(agentContent);\n } catch (error) {\n serverLog(\"ERROR\", \"Failed to parse agent response\", {\n content: agentContent,\n error: error instanceof Error ? error.message : String(error),\n });\n\n // Fallback response if parsing fails\n agentResponse = {\n actionType: \"direct_response\",\n response: \"I encountered an issue processing your request. Could you try rephrasing it?\",\n agent: activeAgent,\n };\n }\n\n // If creating a tool, store it in the database\n if (agentResponse.actionType === \"create_tool\" && agentResponse.tool) {\n const toolId = agentResponse.tool.id || crypto.randomUUID();\n await sqlite.execute(\n `\n INSERT INTO ${KEY}_tools_${SCHEMA_VERSION} \n (id, name, description, schema) \n VALUES (?, ?, ?, ?)\n `,\n [\n toolId,\n agentResponse.tool.schema.name,\n agentResponse.tool.schema.description,\n JSON.stringify(agentResponse.tool.schema),\n ],\n );\n\n // Update the tool ID in the response\n if (agentResponse.tool) {\n agentResponse.tool.id = toolId;\n }\n }\n\n serverLog(\"INFO\", \"Message processed successfully\", {\n actionType: agentResponse.actionType,\n });\n\n return new Response(JSON.stringify(agentResponse), {\n headers: {\n \"Content-Type\": \"application/json\",\n },\n });\n } catch (error) {\n serverLog(\"ERROR\", \"Message processing failed\", {\n error: error instanceof Error ? error.message : String(error),\n });\n\n return new Response(\n JSON.stringify({\n actionType: \"direct_response\",\n response: \"I encountered an error processing your message. Please try again.\",\n agent: \"system\",\n }),\n {\n status: 500,\n headers: { \"Content-Type\": \"application/json\" },\n },\n );\n }\n }\n\n // Serve the HTML interface for the root path\n return new Response(\n `\n \n \n \n \n Maverick & Oracle Agent System\n \n \n \n

\n\n \n `,\n {\n headers: {\n \"Content-Type\": \"text/html\",\n },\n },\n );\n}","deleted","label",[98],"permalink","https://www.val.town/x/api/mando/code/main.tsx","extension","actionData","errors"]