[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8,"_9":10},"actionData","errors","root",{"_373":374,"_375":376},"routes/_app.x.$handle.$name",{"_105":104,"_37":50,"_114":95,"_32":58,"_13":14,"_115":-5,"_11":116,"_117":118,"_119":120,"_121":122,"_123":124,"_26":-5,"_125":-5,"_126":-5,"_127":30,"_128":129,"_130":64,"_131":-5,"_132":-5,"_133":134},"routes/_app.x.$handle.$name.branch.$branch.version.$version.diff",{"_11":12,"_13":14,"_15":16,"_17":18,"_19":-5},"project",{"_48":49,"_37":50,"_43":51,"_52":-5,"_53":30,"_54":55,"_56":55,"_57":58,"_59":-5,"_19":60,"_61":62,"_63":64,"_65":-5,"_66":-5,"_67":68,"_69":70,"_71":72,"_73":74,"_75":76,"_77":-5,"_78":79,"_80":79,"_81":82,"_83":82,"_84":82,"_85":79,"_86":79,"_87":88,"_89":90,"_91":92,"_93":30},"version",48,"changes",[25],"aiChatInfo",{"_20":21,"_22":-5,"_23":-5,"_24":-5},"userId","aiChatMessageId","d3ea22e8-fc88-427c-959a-80021d7c6c11","aiChatThreadId","aiChatThreadUserId","aiChatThreadTitle",{"_26":27,"_28":-7,"_29":30,"_31":30,"_32":33,"_34":33},"file",{"_35":36,"_37":38,"_39":40,"_41":42,"_43":44,"_45":14,"_46":47,"_29":30},"previous","deleted",false,"deletedHere","path","lib/taskProcessor.ts","latestPath","fileId","f5204b54-7960-11f1-a85c-1607ee4eb77e","name","taskProcessor.ts","fileType","script","content","// Step-by-Step Task Processor — Nexuses Core V3\n// Breaks complex tasks into discrete steps, executes sequentially,\n// verifies each step, retries once on failure before logging error.\n\nimport { callLLM } from \"./llm.ts\";\nimport { logStep, logEvent } from \"./db.ts\";\nimport type { AgentResult, StepResult, AgentName } from \"./types.ts\";\n\n// Agent imports\nimport { execute as scraperExecute } from \"../agents/scraperAgent.ts\";\nimport { execute as coderExecute } from \"../agents/coderAgent.ts\";\nimport { execute as analystExecute } from \"../agents/analystAgent.ts\";\nimport { execute as mediaExecute } from \"../agents/mediaAgent.ts\";\nimport { execute as marketingExecute } from \"../agents/marketingAgent.ts\";\n\nconst AGENT_MAP: Record Promise> = {\n CoderAgent: coderExecute,\n ScraperAgent: scraperExecute,\n AnalystAgent: analystExecute,\n MediaAgent: mediaExecute,\n MarketingAgent: marketingExecute,\n GeneralAgent: generalAgentExecute,\n};\n\nasync function generalAgentExecute(task: string, data: any): Promise {\n const start = Date.now();\n const result = await callLLM(data.prompt || task, { maxTokens: 800 });\n return {\n agent: \"GeneralAgent\",\n category: \"general\",\n status: result.success ? \"success\" : \"error\",\n message: result.success ? \"Task processed\" : \"LLM unavailable\",\n data: { input: task, response: result.text || \"\" },\n executionTimeMs: Date.now() - start,\n };\n}\n\n/**\n * Execute a single step via the appropriate agent.\n * Includes automatic retry-once on failure.\n */\nasync function executeStepWithRetry(\n taskId: string,\n stepNum: number,\n description: string,\n agentName: AgentName,\n data: any,\n): Promise {\n const agentFn = AGENT_MAP[agentName] || AGENT_MAP.GeneralAgent;\n\n // Attempt 1\n let result = await agentFn(description, data);\n let attempts = 1;\n\n if (result.status === \"success\" || result.status === \"partial\") {\n await logStep(taskId, stepNum, description, agentName, \"success\", result.data, attempts);\n return {\n step: stepNum,\n description,\n agent: agentName,\n status: \"success\",\n result: result.data,\n attempts,\n };\n }\n\n // Retry once\n attempts = 2;\n await logEvent(\"warn\", \"TaskProcessor\", `Step ${stepNum} failed on attempt 1: ${result.message}. Retrying...`);\n\n // Adjust data for retry — simplify prompt\n const retryData = { ...data, prompt: description.slice(0, 500) };\n result = await agentFn(description, retryData);\n\n if (result.status === \"success\" || result.status === \"partial\") {\n await logStep(taskId, stepNum, description, agentName, \"success\", result.data, attempts);\n return {\n step: stepNum,\n description,\n agent: agentName,\n status: \"success\",\n result: result.data,\n attempts,\n };\n }\n\n // Both attempts failed\n await logStep(taskId, stepNum, description, agentName, \"failed\", result.data, attempts);\n await logEvent(\"error\", \"TaskProcessor\", `Step ${stepNum} failed after ${attempts} attempts: ${result.message}`);\n\n return {\n step: stepNum,\n description,\n agent: agentName,\n status: \"failed\",\n error: result.message,\n attempts,\n };\n}\n\n/**\n * Process a complex multi-step task.\n * Executes steps sequentially — if a step fails (after retry), logs error and continues.\n * Returns all step results.\n */\nexport async function processComplexTask(\n taskId: string,\n steps: string[],\n agentName: AgentName,\n data: any,\n): Promise<{ stepResults: StepResult[]; finalData: any; allSucceeded: boolean }> {\n const stepResults: StepResult[] = [];\n let accumulatedData: any = { ...data };\n let allSucceeded = true;\n\n for (let i = 0; i < steps.length; i++) {\n const stepDesc = steps[i];\n const stepNum = i + 1;\n\n console.log(`[TaskProcessor] Task ${taskId} — Step ${stepNum}/${steps.length}: ${stepDesc.slice(0, 80)}`);\n\n const stepResult = await executeStepWithRetry(\n taskId,\n stepNum,\n stepDesc,\n agentName,\n accumulatedData,\n );\n\n stepResults.push(stepResult);\n\n if (stepResult.status === \"success\") {\n // Merge step result into accumulated data for next step\n if (stepResult.result && typeof stepResult.result === \"object\") {\n accumulatedData = { ...accumulatedData, ...stepResult.result, previousStep: stepResult.result };\n }\n } else {\n allSucceeded = false;\n }\n }\n\n return {\n stepResults,\n finalData: accumulatedData,\n allSucceeded,\n };\n}\n\n/**\n * Execute a single-step task directly via the agent.\n */\nexport async function processSimpleTask(\n taskId: string,\n task: string,\n agentName: AgentName,\n data: any,\n): Promise<{ stepResults: StepResult[]; finalData: any; allSucceeded: boolean }> {\n const stepResult = await executeStepWithRetry(taskId, 1, task, agentName, data);\n return {\n stepResults: [stepResult],\n finalData: stepResult.result || {},\n allSucceeded: stepResult.status === \"success\",\n };\n}\n","createdAt",["D",1783359276137],"branchVersion","branchId","6a688a47-c650-46ed-b95f-c81ab586108d","id","94fe1097-1c95-40e6-a707-d2eb9254f81e","Nexuses_core_os_ai",["D",1783347377247],"deletedAt","legacyEnvVars","privacy","public","httpPrivacy","description","","imageUrl","e98ea39e-0595-4d29-86e1-789bc7567b51","pinnedFileId","40d87268-7945-11f1-ac19-1607ee4eb77e","pinPreview",true,"pinMigrated","screenshotVersion","likeCount",0,"orgId","019f37c8-fde7-721a-95a7-e2a7f7bcb738","tursoName","94fe10971c9540e6a707d2eb9254f81e","tursoHostname","94fe10971c9540e6a707d2eb9254f81e-val-town.aws-us-east-2.turso.io","tursoMeasuredAt",["D",1789522836959],"pinned","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm","user",{"_105":104,"_113":109},"org",{"_105":104,"_106":107,"_108":109,"_19":60,"_110":111,"_112":30},"branch",{"_13":94,"_37":95,"_48":47,"_96":49,"_97":-5,"_98":-5,"_99":-5,"_100":-5,"_101":-5,"_102":50,"_103":104},"isOwned",65,"main","projectId","forkedBranchId","forkedBranchVersion","forkedProjectPrivacy","forkedProjectId","forkedProjectOrgId","projectName","userHandle","alamzubair","handle","displayName","Alam Zubair","avatar","https://lh3.googleusercontent.com/a/ACg8ocINqsCFv5AljwZtsm09t4QxLTSGRbzHpD3Je6v88ir-C33RUA=s96-c","tier","free","canRestrictVals","profileImageUrl","branchName","adoptOrgId",{"_48":49,"_37":50,"_43":368,"_52":-5,"_53":30,"_54":55,"_56":55,"_57":58,"_59":-5,"_19":60,"_61":62,"_63":64,"_65":-5,"_66":-5,"_67":68,"_69":70,"_71":72,"_73":74,"_75":369,"_77":-5,"_78":79,"_80":79,"_81":82,"_83":82,"_84":82,"_85":79,"_86":79,"_87":370,"_89":371,"_91":372,"_93":30},"tree",[238,244,250,251,252,253,254,255,256,257,258],"files",[238,239,240,241,242,243,244,245,246,247,248,249,250,251,252,253,254,255,256,257,258],"counts",{"_227":68,"_228":229,"_230":68,"_231":232,"_233":234},"contents",[135,136,137,138,139,140,141,142,143,144,145],"upstreamProject","upstreamBranch","emptyEnvVars","flags",[],"hasHttpFiles","grants","orgs","customDomains",{},{"_48":221,"_37":222,"_35":223,"_148":-5,"_46":47,"_45":224,"_150":64,"_43":225,"_13":68,"_39":219,"_41":-5,"_29":30,"_155":-5,"_32":226},{"_48":214,"_37":215,"_35":216,"_148":-5,"_46":47,"_45":217,"_150":64,"_43":218,"_13":68,"_39":219,"_41":-5,"_29":30,"_155":-5,"_32":220},{"_48":206,"_37":207,"_35":208,"_148":-5,"_46":47,"_45":209,"_150":30,"_43":210,"_13":211,"_39":40,"_41":212,"_29":30,"_155":-5,"_32":213},{"_48":199,"_37":200,"_35":201,"_148":-5,"_46":47,"_45":202,"_150":30,"_43":203,"_13":152,"_39":40,"_41":204,"_29":30,"_155":-5,"_32":205},{"_48":192,"_37":193,"_35":194,"_148":-5,"_46":47,"_45":195,"_150":30,"_43":196,"_13":152,"_39":40,"_41":197,"_29":30,"_155":-5,"_32":198},{"_48":185,"_37":186,"_35":187,"_148":-5,"_46":47,"_45":188,"_150":30,"_43":189,"_13":152,"_39":40,"_41":190,"_29":30,"_155":-5,"_32":191},{"_48":178,"_37":179,"_35":180,"_148":-5,"_46":47,"_45":181,"_150":30,"_43":182,"_13":152,"_39":40,"_41":183,"_29":30,"_155":-5,"_32":184},{"_48":171,"_37":172,"_35":173,"_148":-5,"_46":47,"_45":174,"_150":30,"_43":175,"_13":152,"_39":40,"_41":176,"_29":30,"_155":-5,"_32":177},{"_48":164,"_37":165,"_35":166,"_148":-5,"_46":47,"_45":167,"_150":30,"_43":168,"_13":152,"_39":40,"_41":169,"_29":30,"_155":-5,"_32":170},{"_48":157,"_37":158,"_35":159,"_148":-5,"_46":47,"_45":160,"_150":30,"_43":161,"_13":152,"_39":40,"_41":162,"_29":30,"_155":-5,"_32":163},{"_48":146,"_37":147,"_35":62,"_148":-5,"_46":47,"_45":149,"_150":30,"_43":151,"_13":152,"_39":153,"_41":154,"_29":30,"_155":-5,"_32":156},"c671ae32-2a21-4e1a-b888-469187f3c004","main.ts","parentId",36,"latest",["D",1783356891573],2,"http","// Nexuses Core OS V10.0 — Universal AI Gateway\n// HTTP Entry Point with IAP, JSON-RPC, strict auth, and Dynamic Request Normalizer\n\nimport {\n callAdapterDirect,\n dispatchTask,\n getHistory,\n getSystemStatus,\n getTaskStatus,\n} from \"./TaskDispatcher.ts\";\nimport { getAgentManifest } from \"./AgentManager.ts\";\nimport { listAdapters } from \"./UniversalAdapter.ts\";\nimport { getModelManifest } from \"./hf.ts\";\nimport { initDB } from \"./db.ts\";\nimport { buildCapabilityList, getServiceManifest } from \"./RegistryManager.ts\";\nimport { processHandshake, processRPC } from \"./IAP.ts\";\nimport type { JSONRPCRequest } from \"./IAP.ts\";\n\nconst CORS = {\n \"Content-Type\": \"application/json\",\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"GET, POST, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type, Authorization\",\n};\n\nfunction json(body: any, status = 200) {\n return new Response(JSON.stringify(body, null, 2), { status, headers: CORS });\n}\n\nfunction getEnv(key: string): string | undefined {\n try {\n return Deno.env.get(key);\n } catch {\n return undefined;\n }\n}\n\nfunction getEndpoint(req: Request): string {\n const url = new URL(req.url);\n return `${url.protocol}//${url.host}/`;\n}\n\n// --- Universal Request Normalizer ---\n// خود بخود ریکویسٹ کو پہچان کر فارمیٹ سیٹ کرتا ہے\nfunction normalizeRequest(payload: any) {\n const possibleTaskKeys = [\n \"task\",\n \"query\",\n \"prompt\",\n \"command\",\n \"message\",\n \"input\",\n \"body\",\n \"text\",\n ];\n let task = \"general\";\n let data = { ...payload };\n\n for (const key of possibleTaskKeys) {\n if (payload[key]) {\n task = payload[key];\n delete data[key];\n break;\n }\n }\n return { task, data };\n}\n\n// --- Auth Validation ---\nfunction validateAuth(req: Request): boolean {\n const apiKey = getEnv(\"NEXUSES_API_KEY\");\n if (!apiKey) return true; // Dev mode\n const auth = req.headers.get(\"authorization\") || \"\";\n const token = auth.replace(/^Bearer\\s+/i, \"\").trim();\n return token === apiKey;\n}\n\n// --- Rate Limiting ---\nconst RATE_LIMIT = new Map();\nfunction isRateLimited(ip: string): boolean {\n const now = Date.now();\n const ts = RATE_LIMIT.get(ip)?.filter((t) => now - t < 60000) || [];\n if (ts.length >= 50) return true; // 50 requests per minute\n ts.push(now);\n RATE_LIMIT.set(ip, ts);\n return false;\n}\n\n// --- HTTP Handler ---\nexport default async function (req: Request): Promise {\n if (req.method === \"OPTIONS\") return new Response(null, { headers: CORS });\n\n const url = new URL(req.url);\n const path = url.pathname;\n const ip = req.headers.get(\"x-forwarded-for\")?.split(\",\")[0] ||\n req.headers.get(\"cf-connecting-ip\") || \"unknown\";\n const endpoint = getEndpoint(req);\n\n await initDB();\n\n // --- GET Routes (Public Discovery) ---\n if (req.method === \"GET\") {\n if (path === \"/\" || path === \"\") return json(await getSystemStatus());\n if (path === \"/status\") return json(await getSystemStatus());\n if (path === \"/register\") return json(getServiceManifest(endpoint));\n if (path === \"/agents\") return json({ agents: getAgentManifest() });\n if (path === \"/adapters\") return json({ adapters: listAdapters() });\n if (path === \"/models\") return json({ models: getModelManifest() });\n if (path === \"/capabilities\") {\n return json({ capabilities: buildCapabilityList() });\n }\n if (path === \"/history\") {\n const limit = parseInt(url.searchParams.get(\"limit\") || \"20\");\n return json({ history: await getHistory(limit) });\n }\n const taskMatch = path.match(/^\\/task\\/(.+)$/);\n if (taskMatch) {\n const task = await getTaskStatus(taskMatch[1]);\n return task\n ? json({ task })\n : json({ status: \"failure\", message: \"Task not found\" }, 404);\n }\n return json({ error: \"Unknown endpoint\" }, 404);\n }\n\n // --- POST Routes ---\n if (req.method !== \"POST\") {\n return json({ error: \"Only GET and POST supported\" }, 405);\n }\n\n let payload: any;\n try {\n payload = await req.json();\n } catch {\n return json({ status: \"failure\", message: \"Invalid JSON body\" }, 400);\n }\n\n // 1. Universal AI Gateway — POST / or /public-task\n // Any external platform (web app, bot, etc.) can POST a work task here.\n // No auth required. The system auto-analyzes, selects the optimal model,\n // and executes with automatic fallback.\n if (path === \"/\" || path === \"\" || path === \"/public-task\") {\n if (isRateLimited(ip)) {\n return json({ status: \"failure\", message: \"Rate limit exceeded\" }, 429);\n }\n\n // No token required — free AI (Pollinations.ai) works out of the box.\n // HF/OpenAI/Groq are optional enhancements if their keys are configured.\n const { task, data } = normalizeRequest(payload);\n const response = await dispatchTask(task, data, \"AUTO\");\n return json(response, response.status === \"success\" ? 200 : 500);\n }\n\n // 2. IAP Handshake (public — no auth needed for discovery)\n if (path === \"/handshake\") {\n if (payload.jsonrpc === \"2.0\" && payload.method === \"handshake\") {\n return json(await processHandshake(payload, endpoint));\n }\n return json({ error: \"Invalid handshake\" }, 400);\n }\n\n // 3. Strict Auth Check for protected POST endpoints\n if (!validateAuth(req)) {\n return json({\n status: \"failure\",\n message: \"Unauthorized — provide valid Bearer token\",\n }, 401);\n }\n\n if (isRateLimited(ip)) {\n return json({ status: \"failure\", message: \"Rate limit exceeded\" }, 429);\n }\n\n // Protected Routes\n if (path === \"/rpc\") {\n const rpcResponse = await processRPC(\n payload as JSONRPCRequest,\n endpoint,\n dispatchTask,\n );\n return json(rpcResponse);\n }\n\n if (path === \"/adapter\") {\n const { adapter, action, data } = payload;\n const result = await callAdapterDirect(\n adapter,\n action || \"default\",\n data || {},\n );\n return json({\n status: result.success ? \"success\" : \"failure\",\n data: result.data,\n });\n }\n\n // Unknown POST path\n return json({ status: \"failure\", message: \"Unknown endpoint\" }, 404);\n}","label",[147],"f91b0029-4b63-48c4-a299-55c3642caace","hf.ts","8d908c30-7945-11f1-a52a-1607ee4eb77e",35,["D",1783356003650],"// HF Model Orchestration — Nexuses Core OS V10.0\n// Universal AI Gateway: Free AI (no token) → OpenAI → HF Router with auto model selection + dynamic fallback.\n\nimport { routeToAdapter, selectBestModel } from \"./UniversalAdapter.ts\";\n\n// --- Model Registry ---\nexport type ModelKey = \"qwen_coder\" | \"llama\" | \"phi3\" | \"gemma\";\nexport type TaskType = \"coding\" | \"creative\" | \"analysis\" | \"general\";\n\ninterface ModelConfig {\n id: string;\n label: string;\n bestFor: TaskType[];\n keywords: string[];\n maxTokens: number;\n temperature: number;\n fallback: ModelKey | null;\n}\n\nexport const HF_MODELS: Record = {\n qwen_coder: {\n id: \"Qwen/Qwen2.5-Coder-7B-Instruct\",\n label: \"Qwen2.5 Coder 7B\",\n bestFor: [\"coding\"],\n keywords: [\"code\", \"coding\", \"debug\", \"bug\", \"fix\", \"program\", \"refactor\"],\n maxTokens: 800,\n temperature: 0.3,\n fallback: \"gemma\",\n },\n llama: {\n id: \"meta-llama/Llama-3.2-1B-Instruct\",\n label: \"Llama 3.2 1B\",\n bestFor: [\"creative\"],\n keywords: [\"write\", \"writing\", \"blog\", \"story\", \"creative\", \"social\"],\n maxTokens: 1000,\n temperature: 0.8,\n fallback: \"gemma\",\n },\n phi3: {\n id: \"microsoft/Phi-3-mini-4k-instruct\",\n label: \"Phi-3 Mini 4K\",\n bestFor: [\"analysis\", \"general\"],\n keywords: [\"analyze\", \"data\", \"report\", \"csv\", \"summarize\", \"search\"],\n maxTokens: 500,\n temperature: 0.5,\n fallback: \"gemma\",\n },\n gemma: {\n id: \"google/gemma-2-2b-it\",\n label: \"Gemma 2 2B\",\n bestFor: [\"general\"],\n keywords: [\"general\", \"help\", \"chat\", \"explain\", \"unknown\"],\n maxTokens: 500,\n temperature: 0.7,\n fallback: null,\n },\n};\n\n// --- Reverse lookup: model ID string → ModelKey ---\nfunction modelIdToKey(modelId: string): ModelKey {\n for (const [key, cfg] of Object.entries(HF_MODELS)) {\n if (cfg.id === modelId) return key as ModelKey;\n }\n return \"gemma\";\n}\n\n// --- Selection Logic ---\nexport function classifyTask(taskText: string): TaskType {\n const lower = taskText.toLowerCase();\n const scores = { coding: 0, creative: 0, analysis: 0, general: 0 };\n for (const cfg of Object.values(HF_MODELS)) {\n for (const type of cfg.bestFor) {\n scores[type] += cfg.keywords.filter((k) => lower.includes(k)).length;\n }\n }\n let best: TaskType = \"general\";\n let bestScore = 0;\n for (const [type, score] of Object.entries(scores)) {\n if (score > bestScore) {\n bestScore = score;\n best = type as TaskType;\n }\n }\n return best;\n}\n\nexport function modelForTaskType(type: TaskType): ModelKey {\n switch (type) {\n case \"coding\":\n return \"qwen_coder\";\n case \"creative\":\n return \"llama\";\n case \"analysis\":\n return \"phi3\";\n default:\n return \"gemma\";\n }\n}\n\nexport function selectModel(taskText: string): ModelKey {\n return modelForTaskType(classifyTask(taskText));\n}\n\n// --- Integration with UniversalAdapter ---\nexport interface HFResult {\n success: boolean;\n model: ModelKey;\n modelId: string;\n text: string;\n taskType: TaskType;\n error?: string;\n attempts?: ModelKey[];\n fallbackUsed?: boolean;\n}\n\nexport interface CallOptions {\n maxTokens?: number;\n temperature?: number;\n timeoutMs?: number;\n}\n\n/**\n * Universal AI Gateway — auto-selects the best available AI with automatic fallback:\n * 1. Free AI (Pollinations.ai) — no token needed, always available\n * 2. OpenAI (if OPENAI_API_KEY is configured)\n * 3. HF Router primary model (if HF_API_KEY is configured)\n * 4. HF Router fallback model\n * 5. HF Router Gemma (ultimate HF fallback)\n * 6. Free AI retry (last resort)\n * On timeout or error, automatically switches to the next provider.\n */\nexport async function callHFWithFallback(\n prompt: string,\n modelOverride?: ModelKey,\n options?: CallOptions,\n): Promise {\n const taskType = classifyTask(prompt);\n const attempts: ModelKey[] = [];\n\n // --- 0. Try Free AI first (no token needed — always available) ---\n const freeResult = await routeToAdapter(\"freeai\", \"generate\", {\n prompt,\n maxTokens: options?.maxTokens ?? 500,\n temperature: options?.temperature ?? 0.7,\n timeoutMs: options?.timeoutMs ?? 30000,\n });\n if (freeResult.success && freeResult.data?.response) {\n return {\n success: true,\n model: \"gemma\",\n modelId: freeResult.data.modelUsed || \"freeai\",\n text: freeResult.data.response,\n taskType,\n attempts: [\"gemma\"],\n };\n }\n\n // --- 0b. Try OpenAI if key is configured ---\n const openaiKey = Deno.env.get(\"OPENAI_API_KEY\");\n if (openaiKey) {\n const openaiResult = await routeToAdapter(\"openai\", \"generate\", {\n prompt,\n maxTokens: options?.maxTokens ?? 1000,\n temperature: options?.temperature ?? 0.7,\n timeoutMs: options?.timeoutMs ?? 30000,\n });\n if (openaiResult.success && openaiResult.data?.response) {\n return {\n success: true,\n model: \"gemma\",\n modelId: openaiResult.data.modelUsed || \"openai\",\n text: openaiResult.data.response,\n taskType,\n attempts: [\"gemma\"],\n };\n }\n }\n\n // --- If no HF token, free AI was our main option ---\n const hfToken = getHFToken();\n if (!hfToken) {\n // Retry free AI with a shorter, simpler prompt\n const retryResult = await routeToAdapter(\"freeai\", \"generate\", {\n prompt: prompt.slice(0, 500),\n maxTokens: 300,\n temperature: 0.5,\n timeoutMs: 20000,\n });\n if (retryResult.success && retryResult.data?.response) {\n return {\n success: true,\n model: \"gemma\",\n modelId: retryResult.data.modelUsed || \"freeai\",\n text: retryResult.data.response,\n taskType,\n attempts: [\"gemma\"],\n fallbackUsed: true,\n };\n }\n return {\n success: false,\n model: \"gemma\",\n modelId: \"freeai\",\n text: \"\",\n taskType,\n error: \"Free AI service unavailable — no HF token configured for fallback\",\n attempts: [\"gemma\"],\n };\n }\n\n // --- 1. Primary model via HF Router ---\n let modelKey: ModelKey;\n if (modelOverride) {\n modelKey = modelOverride;\n } else {\n const bestModelId = selectBestModel(prompt);\n modelKey = modelIdToKey(bestModelId);\n }\n\n const cfg = HF_MODELS[modelKey];\n const maxTokens = options?.maxTokens ?? cfg.maxTokens;\n const temperature = options?.temperature ?? cfg.temperature;\n const timeoutMs = options?.timeoutMs ?? 30000;\n\n attempts.push(modelKey);\n const primary = await routeToAdapter(\"huggingface\", \"generate\", {\n prompt,\n model: cfg.id,\n maxTokens,\n temperature,\n timeoutMs,\n });\n if (primary.success && primary.data?.response) {\n return {\n success: true,\n model: modelKey,\n modelId: cfg.id,\n text: primary.data.response,\n taskType,\n attempts,\n };\n }\n\n // --- 2. Secondary model (configured fallback) via HF Router ---\n if (cfg.fallback) {\n const fbKey = cfg.fallback;\n attempts.push(fbKey);\n const fbCfg = HF_MODELS[fbKey];\n const secondary = await routeToAdapter(\"huggingface\", \"generate\", {\n prompt,\n model: fbCfg.id,\n maxTokens: fbCfg.maxTokens,\n temperature: fbCfg.temperature,\n timeoutMs,\n });\n if (secondary.success && secondary.data?.response) {\n return {\n success: true,\n model: fbKey,\n modelId: fbCfg.id,\n text: secondary.data.response,\n taskType,\n attempts,\n fallbackUsed: true,\n };\n }\n }\n\n // --- 3. Ultimate HF fallback: Gemma via HF Router ---\n if (modelKey !== \"gemma\" && cfg.fallback !== \"gemma\") {\n attempts.push(\"gemma\");\n const gemmaCfg = HF_MODELS[\"gemma\"];\n const gemmaResult = await routeToAdapter(\"huggingface\", \"generate\", {\n prompt,\n model: gemmaCfg.id,\n maxTokens: gemmaCfg.maxTokens,\n temperature: gemmaCfg.temperature,\n timeoutMs,\n });\n if (gemmaResult.success && gemmaResult.data?.response) {\n return {\n success: true,\n model: \"gemma\",\n modelId: gemmaCfg.id,\n text: gemmaResult.data.response,\n taskType,\n attempts,\n fallbackUsed: true,\n };\n }\n }\n\n // --- 4. Last resort: retry free AI ---\n const lastResort = await routeToAdapter(\"freeai\", \"generate\", {\n prompt: prompt.slice(0, 500),\n maxTokens: 300,\n timeoutMs: 20000,\n });\n if (lastResort.success && lastResort.data?.response) {\n return {\n success: true,\n model: \"gemma\",\n modelId: lastResort.data.modelUsed || \"freeai\",\n text: lastResort.data.response,\n taskType,\n attempts,\n fallbackUsed: true,\n };\n }\n\n // --- All paths failed ---\n return {\n success: false,\n model: modelKey,\n modelId: cfg.id,\n text: \"\",\n taskType,\n error: \"All AI services failed after full fallback chain\",\n attempts,\n };\n}\n\nexport async function callHF(\n prompt: string,\n model?: ModelKey,\n options?: CallOptions,\n): Promise {\n return callHFWithFallback(prompt, model, options);\n}\n\n// --- Discovery ---\nexport function getModelManifest() {\n return Object.entries(HF_MODELS).map(([key, cfg]) => ({\n key,\n id: cfg.id,\n label: cfg.label,\n bestFor: cfg.bestFor,\n status: \"ADAPTER_LINKED\",\n }));\n}\n\nexport function getSelectionInfo(taskText: string) {\n const taskType = classifyTask(taskText);\n const modelKey = modelForTaskType(taskType);\n const cfg = HF_MODELS[modelKey];\n return {\n taskType,\n selectedModel: modelKey,\n modelId: cfg.id,\n modelLabel: cfg.label,\n fallbackModel: cfg.fallback,\n };\n}\n\nexport function getHFToken() {\n return Deno.env.get(\"HF_API_KEY\");\n}",[158],"7a770991-c7a5-4776-8e67-8facf105397d","db.ts","a8be7666-7945-11f1-947a-1607ee4eb77e",34,["D",1783355945330],"// SQLite Persistent State Layer — Nexuses Core OS V10.0\nimport { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\n\nexport interface TaskRecord {\n id: string;\n task: string;\n agent: string;\n status: string;\n data: string;\n result: string;\n created_at: string;\n updated_at: string;\n}\n\nexport interface MemoryRecord {\n key: string;\n value: string;\n created_at: string;\n}\n\nconst INIT_SQL = [\n `CREATE TABLE IF NOT EXISTS tasks (\n id TEXT PRIMARY KEY,\n task TEXT NOT NULL,\n agent TEXT,\n status TEXT NOT NULL,\n data TEXT,\n result TEXT,\n created_at TEXT NOT NULL,\n updated_at TEXT NOT NULL\n )`,\n `CREATE TABLE IF NOT EXISTS memory (\n key TEXT PRIMARY KEY,\n value TEXT NOT NULL,\n created_at TEXT NOT NULL\n )`,\n `CREATE TABLE IF NOT EXISTS workflows (\n id TEXT PRIMARY KEY,\n task_id TEXT NOT NULL,\n step INTEGER NOT NULL,\n agent TEXT,\n status TEXT,\n detail TEXT,\n created_at TEXT NOT NULL\n )`,\n `CREATE TABLE IF NOT EXISTS delegations (\n id TEXT PRIMARY KEY,\n task_id TEXT NOT NULL,\n target TEXT NOT NULL,\n task TEXT NOT NULL,\n status TEXT NOT NULL,\n result TEXT,\n created_at TEXT NOT NULL\n )`,\n];\n\nlet initialized = false;\n\nexport async function initDB() {\n if (initialized) return;\n await sqlite.batch(INIT_SQL.map((sql) => ({ sql })));\n initialized = true;\n}\n\nexport async function insertTask(\n id: string,\n task: string,\n agent: string,\n status: string,\n data: any,\n result: any,\n) {\n const now = new Date().toISOString();\n await sqlite.execute({\n sql:\n `INSERT INTO tasks (id, task, agent, status, data, result, created_at, updated_at)\n VALUES (?, ?, ?, ?, ?, ?, ?, ?)`,\n args: [\n id,\n task,\n agent,\n status,\n JSON.stringify(data),\n JSON.stringify(result),\n now,\n now,\n ],\n });\n}\n\nexport async function updateTaskStatus(\n id: string,\n status: string,\n result: any,\n) {\n const now = new Date().toISOString();\n await sqlite.execute({\n sql: `UPDATE tasks SET status = ?, result = ?, updated_at = ? WHERE id = ?`,\n args: [status, JSON.stringify(result), now, id],\n });\n}\n\nexport async function getTask(id: string): Promise {\n const res = await sqlite.execute({\n sql: `SELECT * FROM tasks WHERE id = ?`,\n args: [id],\n });\n return (res.rows[0] as TaskRecord) || null;\n}\n\nexport async function getRecentTasks(limit = 20): Promise {\n const res = await sqlite.execute({\n sql: `SELECT * FROM tasks ORDER BY created_at DESC LIMIT ?`,\n args: [limit],\n });\n return res.rows as TaskRecord[];\n}\n\nexport async function logWorkflowStep(\n taskId: string,\n step: number,\n agent: string,\n status: string,\n detail: string,\n) {\n const id = `wf_${taskId}_${step}`;\n const now = new Date().toISOString();\n await sqlite.execute({\n sql:\n `INSERT INTO workflows (id, task_id, step, agent, status, detail, created_at)\n VALUES (?, ?, ?, ?, ?, ?, ?)`,\n args: [id, taskId, step, agent, status, detail, now],\n });\n}\n\nexport async function remember(key: string, value: any) {\n const now = new Date().toISOString();\n await sqlite.execute({\n sql:\n `INSERT OR REPLACE INTO memory (key, value, created_at) VALUES (?, ?, ?)`,\n args: [key, JSON.stringify(value), now],\n });\n}\n\nexport async function recall(key: string): Promise {\n const res = await sqlite.execute({\n sql: `SELECT * FROM memory WHERE key = ?`,\n args: [key],\n });\n const row = res.rows[0] as MemoryRecord | undefined;\n return row ? JSON.parse(row.value) : null;\n}\n\nexport async function getWorkflowHistory(taskId: string) {\n const res = await sqlite.execute({\n sql: `SELECT * FROM workflows WHERE task_id = ? ORDER BY step ASC`,\n args: [taskId],\n });\n return res.rows;\n}\n\nexport async function getTaskCount(): Promise {\n const res = await sqlite.execute(`SELECT COUNT(*) as count FROM tasks`);\n return (res.rows[0] as any)?.count || 0;\n}",[165],"dd9134a1-191c-48ee-96bb-1ab187a12b6e","UniversalAdapter.ts","ecd4520c-7946-11f1-a15d-1607ee4eb77e",33,["D",1783355890857],"// Universal Adapter Layer — Nexuses Core OS V10.0\n// HF Router (primary), auto model selection via selectBestModel(), scraping, search, crypto.\n\nexport interface Adapter {\n name: string;\n type: \"api\" | \"browser\" | \"service\" | \"plugin\" | \"ai\";\n status: \"active\" | \"inactive\" | \"error\";\n description: string;\n execute: (action: string, data: any) => Promise;\n}\n\nexport interface AdapterResult {\n success: boolean;\n adapter: string;\n action: string;\n data: any;\n message: string;\n}\n\n// --- Smart Model Router (Auto Model Selection) ---\n// Intelligently picks the best HF Router model based on task keywords.\nconst MODEL_CHAIN: { model: string; keywords: string[] }[] = [\n {\n model: \"Qwen/Qwen2.5-Coder-7B-Instruct\",\n keywords: [\"code\", \"coding\", \"script\", \"function\", \"debug\", \"bug\", \"fix\", \"program\", \"refactor\", \"compile\", \"syntax\", \"algorithm\", \"api\", \"sql\", \"regex\"],\n },\n {\n model: \"microsoft/Phi-3-mini-4k-instruct\",\n keywords: [\"calculate\", \"solve\", \"logic\", \"math\", \"analyze\", \"analysis\", \"data\", \"report\", \"csv\", \"summarize\", \"research\", \"invoice\", \"excel\", \"spreadsheet\"],\n },\n {\n model: \"google/gemma-2-2b-it\",\n keywords: [\"poem\", \"story\", \"explain\", \"translate\", \"write\", \"blog\", \"content\", \"article\", \"social\", \"post\", \"creative\", \"narrative\"],\n },\n {\n model: \"meta-llama/Llama-3.2-1B-Instruct\",\n keywords: [\"general\", \"help\", \"chat\", \"question\", \"reasoning\", \"unknown\", \"test\"],\n },\n];\n\nexport function selectBestModel(task: string): string {\n const t = task.toLowerCase();\n let bestModel = MODEL_CHAIN[MODEL_CHAIN.length - 1].model; // default\n let bestScore = 0;\n for (const entry of MODEL_CHAIN) {\n const score = entry.keywords.filter((k) => t.includes(k)).length;\n if (score > bestScore) {\n bestScore = score;\n bestModel = entry.model;\n }\n }\n return bestModel;\n}\n\n// --- 1. Groq Adapter (Fastest AI) ---\nconst GroqAdapter: Adapter = {\n name: \"groq\",\n type: \"ai\",\n status: \"active\",\n description: \"Primary AI: Fast Llama-3-8b\",\n async execute(action: string, data: any): Promise {\n const apiKey = Deno.env.get(\"GROQ_API_KEY\");\n if (!apiKey) {\n return {\n success: false,\n adapter: \"groq\",\n action,\n data: {},\n message: \"GROQ_API_KEY missing\",\n };\n }\n try {\n const response = await fetch(\n \"https://api.groq.com/openai/v1/chat/completions\",\n {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${apiKey}`,\n \"Content-Type\": \"application/json\",\n },\n body: JSON.stringify({\n model: data.model || \"llama3-8b-8192\",\n messages: [{ role: \"user\", content: data.prompt || data.task }],\n max_tokens: data.maxTokens || 1000,\n temperature: data.temperature ?? 0.7,\n }),\n signal: AbortSignal.timeout(30000),\n },\n );\n if (!response.ok) {\n const errBody = await response.text().catch(() => \"\");\n return {\n success: false,\n adapter: \"groq\",\n action,\n data: {},\n message: `Groq HTTP ${response.status}: ${errBody.slice(0, 200)}`,\n };\n }\n const result = await response.json();\n const text = result.choices?.[0]?.message?.content || \"\";\n return {\n success: true,\n adapter: \"groq\",\n action: \"generate\",\n data: { response: text, modelUsed: result.model || \"llama3-8b-8192\" },\n message: \"Success\",\n };\n } catch (e: any) {\n return {\n success: false,\n adapter: \"groq\",\n action,\n data: {},\n message: e.message,\n };\n }\n },\n};\n\n// --- 2. Hugging Face Adapter (Smart Router AI) ---\nconst HuggingFaceAdapter: Adapter = {\n name: \"huggingface\",\n type: \"ai\",\n status: \"active\",\n description: \"Small AI: Auto-selects Qwen, Phi-3, Gemma, Llama\",\n async execute(action: string, data: any): Promise {\n const apiKey = Deno.env.get(\"HF_API_KEY\");\n if (!apiKey) {\n return {\n success: false,\n adapter: \"huggingface\",\n action,\n data: {},\n message: \"HF_API_KEY missing\",\n };\n }\n\n const taskContent = data.prompt || data.task || \"\";\n const model = data.model || selectBestModel(taskContent);\n const maxTokens = data.maxTokens || 500;\n const temperature = data.temperature ?? 0.7;\n\n // New Hugging Face Router endpoint (OpenAI-compatible chat completions)\n const url = \"https://router.huggingface.co/v1/chat/completions\";\n\n try {\n const response = await fetch(url, {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${apiKey}`,\n \"Content-Type\": \"application/json\",\n },\n body: JSON.stringify({\n model,\n messages: [{ role: \"user\", content: taskContent }],\n max_tokens: maxTokens,\n temperature,\n }),\n signal: AbortSignal.timeout(data.timeoutMs || 30000),\n });\n\n if (!response.ok) {\n const errBody = await response.text().catch(() => \"\");\n return {\n success: false,\n adapter: \"huggingface\",\n action,\n data: {},\n message: `HF Router HTTP ${response.status}: ${errBody.slice(0, 200)}`,\n };\n }\n\n const result = await response.json();\n const text = result.choices?.[0]?.message?.content ||\n result.choices?.[0]?.text || \"\";\n\n return {\n success: true,\n adapter: \"huggingface\",\n action: \"generate\",\n data: { response: text, modelUsed: model },\n message: `Used ${model} via HF Router`,\n };\n } catch (e: any) {\n return {\n success: false,\n adapter: \"huggingface\",\n action,\n data: {},\n message: e.message,\n };\n }\n },\n};\n\n// --- 3. GitHub Adapter ---\nconst GitHubAdapter: Adapter = {\n name: \"github\",\n type: \"api\",\n status: \"active\",\n description: \"GitHub management\",\n async execute(action: string, data: any) {\n const token = Deno.env.get(\"GITHUB_TOKEN\");\n return {\n success: !!token,\n adapter: \"github\",\n action,\n data: {},\n message: token ? \"GitHub ready\" : \"Token missing\",\n };\n },\n};\n\n// --- 4. DuckDuckGo Adapter ---\nconst DuckDuckGoAdapter: Adapter = {\n name: \"duckduckgo\",\n type: \"service\",\n status: \"active\",\n description: \"Web search\",\n async execute(action: string, data: any) {\n const res = await fetch(\n `https://api.duckduckgo.com/?q=${\n encodeURIComponent(data.query || data.q)\n }&format=json`,\n );\n const result = await res.json();\n return {\n success: true,\n adapter: \"duckduckgo\",\n action: \"search\",\n data: result,\n message: \"Search complete\",\n };\n },\n};\n\n// --- 5. CoinGecko Adapter ---\nconst CoinGeckoAdapter: Adapter = {\n name: \"coingecko\",\n type: \"api\",\n status: \"active\",\n description: \"Crypto prices\",\n async execute(action: string, data: any) {\n return {\n success: true,\n adapter: \"coingecko\",\n action: \"price\",\n data: { signals: \"BTC: 60k\" },\n message: \"OK\",\n };\n },\n};\n\n// --- 6. WebScraper Adapter ---\nconst WebScraperAdapter: Adapter = {\n name: \"webscraper\",\n type: \"browser\",\n status: \"active\",\n description: \"Scrape content\",\n async execute(action: string, data: any) {\n return {\n success: true,\n adapter: \"webscraper\",\n action: \"scrape\",\n data: { content: \"Page data...\" },\n message: \"OK\",\n };\n },\n};\n\n// --- 7. Free AI Adapter (NO TOKEN NEEDED) ---\n// Uses Pollinations.ai — OpenAI-compatible, 100% free, no API key required.\n// This is the primary AI: works out of the box with zero configuration.\nconst FreeAIAdapter: Adapter = {\n name: \"freeai\",\n type: \"ai\",\n status: \"active\",\n description: \"Free AI — no token needed (OpenAI-compatible via Pollinations.ai)\",\n async execute(action: string, data: any): Promise {\n const taskContent = data.prompt || data.task || \"\";\n if (!taskContent) {\n return {\n success: false, adapter: \"freeai\", action, data: {},\n message: \"No prompt provided\",\n };\n }\n try {\n // Try OpenAI-compatible POST endpoint first\n const response = await fetch(\"https://text.pollinations.ai/openai\", {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({\n model: \"openai\",\n messages: [{ role: \"user\", content: taskContent }],\n max_tokens: data.maxTokens || 500,\n temperature: data.temperature ?? 0.7,\n }),\n signal: AbortSignal.timeout(data.timeoutMs || 30000),\n });\n\n if (response.ok) {\n const result = await response.json();\n const text = result.choices?.[0]?.message?.content || \"\";\n if (text) {\n return {\n success: true, adapter: \"freeai\", action: \"generate\",\n data: { response: text, modelUsed: \"openai-free\" },\n message: \"Success via Pollinations.ai (free, no token)\",\n };\n }\n }\n\n // Fallback: try simple GET endpoint\n const getResponse = await fetch(\n `https://text.pollinations.ai/${encodeURIComponent(taskContent.slice(0, 2000))}`,\n { signal: AbortSignal.timeout(data.timeoutMs || 30000) },\n );\n if (getResponse.ok) {\n const text = await getResponse.text();\n if (text && text.trim().length > 0) {\n return {\n success: true, adapter: \"freeai\", action: \"generate\",\n data: { response: text, modelUsed: \"openai-free\" },\n message: \"Success via Pollinations.ai GET (free, no token)\",\n };\n }\n }\n\n const errBody = await response.text().catch(() => \"\");\n return {\n success: false, adapter: \"freeai\", action, data: {},\n message: `FreeAI failed: HTTP ${response.status}: ${errBody.slice(0, 200)}`,\n };\n } catch (e: any) {\n return {\n success: false, adapter: \"freeai\", action, data: {},\n message: e.message,\n };\n }\n },\n};\n\n// --- 8. OpenAI Adapter (optional — activates if OPENAI_API_KEY is set) ---\nconst OpenAIAdapter: Adapter = {\n name: \"openai\",\n type: \"ai\",\n status: \"active\",\n description: \"OpenAI GPT (auto-activates if OPENAI_API_KEY is set)\",\n async execute(action: string, data: any): Promise {\n const apiKey = Deno.env.get(\"OPENAI_API_KEY\");\n if (!apiKey) {\n return {\n success: false, adapter: \"openai\", action, data: {},\n message: \"OPENAI_API_KEY missing (optional — free AI works without it)\",\n };\n }\n try {\n const response = await fetch(\"https://api.openai.com/v1/chat/completions\", {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${apiKey}`,\n \"Content-Type\": \"application/json\",\n },\n body: JSON.stringify({\n model: data.model || \"gpt-4o-mini\",\n messages: [{ role: \"user\", content: data.prompt || data.task }],\n max_tokens: data.maxTokens || 1000,\n temperature: data.temperature ?? 0.7,\n }),\n signal: AbortSignal.timeout(30000),\n });\n if (!response.ok) {\n const errBody = await response.text().catch(() => \"\");\n return {\n success: false, adapter: \"openai\", action, data: {},\n message: `OpenAI HTTP ${response.status}: ${errBody.slice(0, 200)}`,\n };\n }\n const result = await response.json();\n const text = result.choices?.[0]?.message?.content || \"\";\n return {\n success: true, adapter: \"openai\", action: \"generate\",\n data: { response: text, modelUsed: result.model || \"gpt-4o-mini\" },\n message: \"Success via OpenAI\",\n };\n } catch (e: any) {\n return {\n success: false, adapter: \"openai\", action, data: {},\n message: e.message,\n };\n }\n },\n};\n\n// --- Registry ---\nconst REGISTRY = new Map();\n[\n FreeAIAdapter, // Primary — no token needed\n OpenAIAdapter, // Optional — auto-activates if OPENAI_API_KEY set\n GroqAdapter, // Optional — auto-activates if GROQ_API_KEY set\n HuggingFaceAdapter, // Optional — auto-activates if HF_API_KEY set\n GitHubAdapter,\n DuckDuckGoAdapter,\n CoinGeckoAdapter,\n WebScraperAdapter,\n].forEach(\n (a) => REGISTRY.set(a.name, a),\n);\n\nexport function listAdapters() {\n return Array.from(REGISTRY.values()).map((a) => ({\n name: a.name,\n type: a.type,\n status: a.status,\n description: a.description,\n }));\n}\n\nexport async function routeToAdapter(\n name: string,\n action: string,\n data: any,\n): Promise {\n const adapter = REGISTRY.get(name);\n if (!adapter) {\n return {\n success: false,\n adapter: name,\n action,\n data: {},\n message: \"Adapter not found\",\n };\n }\n return adapter.execute(action, data);\n}",[172],"e8ad083c-29e4-492c-a022-d12bd88722d7","TaskDispatcher.ts","145cb4d6-7947-11f1-9c04-1607ee4eb77e",32,["D",1783355841883],"// Task Dispatcher — Nexuses Core OS V10.0\n// Orchestrates the Analyze → Execute → Report pipeline.\n// Integrates Supervisor (auto-retry) and IAP delegation.\nimport { getAgentManifest, routeIncomingTask } from \"./AgentManager.ts\";\nimport type { AgentResult } from \"./AgentManager.ts\";\nimport { listAdapters, routeToAdapter } from \"./UniversalAdapter.ts\";\nimport {\n classifyTask,\n getHFToken,\n getModelManifest,\n getSelectionInfo,\n selectModel,\n} from \"./hf.ts\";\nimport type { TaskType } from \"./hf.ts\";\nimport { supervise } from \"./Supervisor.ts\";\nimport type { SupervisedResult } from \"./Supervisor.ts\";\nimport { delegateToExternal } from \"./IAP.ts\";\nimport type { DelegationResult } from \"./IAP.ts\";\nimport { getServiceInfo, hasInternalCapability } from \"./RegistryManager.ts\";\nimport {\n getRecentTasks,\n getTask,\n getTaskCount,\n initDB,\n insertTask,\n logWorkflowStep,\n recall,\n remember,\n updateTaskStatus,\n} from \"./db.ts\";\n\nexport interface DispatchResponse {\n status: \"success\" | \"failure\";\n task_id: string;\n data: any;\n message: string;\n agent?: string;\n category?: string;\n model?: string;\n modelId?: string;\n taskType?: TaskType;\n fallbackUsed?: boolean;\n attempts?: string[];\n adapters?: string[];\n delegated?: boolean;\n delegationTarget?: string | null;\n supervised?: boolean;\n retried?: boolean;\n selfCorrected?: boolean;\n supervisorAttempts?: number;\n executionTimeMs: number;\n timestamp: string;\n}\n\nfunction generateTaskId(): string {\n return `nxs_${Date.now()}_${Math.random().toString(36).slice(2, 8)}`;\n}\n\n// --- Public Dispatch API ---\n\nexport async function dispatchTask(\n task: string,\n data: any = {},\n mode = \"AUTO\",\n): Promise {\n const start = Date.now();\n const taskId = generateTaskId();\n\n // Ensure DB is ready\n await initDB();\n\n // PHASE 1: ANALYZE — deconstruct the task, identify agents + model\n const agents = getAgentManifest();\n const taskLower = task.toLowerCase();\n\n // Identify which agents could handle this task\n const candidates = agents.filter((a) =>\n a.keywords.some((k) => taskLower.includes(k))\n );\n\n // Dynamic model selection — classify task type before any API call\n const selection = getSelectionInfo(task);\n\n await logWorkflowStep(\n taskId,\n 1,\n \"TaskDispatcher\",\n \"ANALYZE\",\n `Task: \"${task}\" | Type: ${selection.taskType} | Model: ${selection.modelLabel} (${selection.modelId}) | Fallback: ${\n selection.fallbackModel || \"none\"\n } | Candidates: ${\n candidates.map((a) => a.name).join(\", \") || \"GeneralHandler\"\n }`,\n );\n\n // Record initial task state\n await insertTask(\n taskId,\n task,\n \"pending\",\n \"processing\",\n { task, data, mode },\n null,\n );\n\n // Recall any memory related to this task type\n const memoryKey = `task_type:${candidates[0]?.name || \"general\"}`;\n const priorMemory = await recall(memoryKey);\n\n // PHASE 2: EXECUTE — route to selected agent via Supervisor\n await logWorkflowStep(\n taskId,\n 2,\n \"TaskDispatcher\",\n \"EXECUTE\",\n `Routing to agent for execution (supervised mode)`,\n );\n\n // Wrap execution in Supervisor for autonomous error recovery\n const supervised = await supervise(\n taskId,\n task,\n data,\n (t, d) => routeIncomingTask(t, d, mode, true),\n );\n\n let agentResult: AgentResult;\n let delegated = false;\n let delegationResult: DelegationResult | null = null;\n\n if (supervised.success && supervised.finalResult) {\n agentResult = supervised.finalResult as AgentResult;\n } else {\n // Internal execution failed after retries — try external delegation\n await logWorkflowStep(\n taskId,\n 5,\n \"TaskDispatcher\",\n \"DELEGATE\",\n `Internal execution failed, attempting external delegation`,\n );\n\n delegationResult = await delegateToExternal(task, data);\n delegated = delegationResult.delegated;\n\n if (delegationResult.success) {\n agentResult = {\n agent: \"ExternalDelegation\",\n category: \"Delegated\",\n status: \"SUCCESS\",\n message: delegationResult.message,\n data: delegationResult.data,\n adapters: [],\n };\n } else {\n agentResult = {\n agent: supervised.finalResult?.agent || \"Unknown\",\n category: supervised.finalResult?.category || \"General\",\n status: \"ERROR\",\n message: supervised.message,\n data: supervised.finalResult?.data || {},\n };\n }\n }\n\n await logWorkflowStep(\n taskId,\n 3,\n agentResult.agent,\n agentResult.status,\n agentResult.message + (delegated ? \" [DELEGATED]\" : \"\"),\n );\n\n // PHASE 3: REPORT — build structured Universal Task Format response\n const isSuccess = agentResult.status === \"SUCCESS\" ||\n agentResult.status === \"PARTIAL\";\n const response: DispatchResponse = {\n status: agentResult.status === \"ERROR\" ? \"failure\" : \"success\",\n task_id: taskId,\n data: agentResult.data,\n message: agentResult.message,\n agent: agentResult.agent,\n category: agentResult.category,\n model: agentResult.model,\n modelId: agentResult.modelId,\n taskType: agentResult.taskType || selection.taskType,\n fallbackUsed: agentResult.fallbackUsed,\n attempts: agentResult.attempts,\n adapters: agentResult.adapters || [],\n delegated,\n delegationTarget: delegationResult?.target || null,\n supervised: true,\n retried: supervised.retried,\n selfCorrected: supervised.selfCorrected,\n supervisorAttempts: supervised.attempts.length,\n executionTimeMs: Date.now() - start,\n timestamp: new Date().toISOString(),\n };\n\n // Update task in DB\n await updateTaskStatus(taskId, isSuccess ? \"completed\" : \"failed\", response);\n\n // Save to memory for future reference\n await remember(memoryKey, {\n lastTask: task,\n lastAgent: agentResult.agent,\n lastStatus: agentResult.status,\n lastModel: agentResult.model,\n fallbackUsed: agentResult.fallbackUsed,\n timestamp: new Date().toISOString(),\n ...(priorMemory ? { previous: priorMemory } : {}),\n });\n\n await logWorkflowStep(\n taskId,\n 4,\n \"TaskDispatcher\",\n \"REPORT\",\n `Status: ${response.status} | Model: ${response.model || \"N/A\"}${\n response.fallbackUsed ? \" (FALLBACK)\" : \"\"\n } | Time: ${response.executionTimeMs}ms`,\n );\n\n return response;\n}\n\n// --- Status & Discovery ---\n\nexport async function getTaskStatus(taskId: string) {\n return getTask(taskId);\n}\n\nexport async function getHistory(limit = 20) {\n return getRecentTasks(limit);\n}\n\nexport async function getSystemStatus() {\n await initDB();\n const taskCount = await getTaskCount();\n const token = getHFToken();\n\n return {\n status: \"ONLINE\",\n service: \"Nexuses Core OS V10.0 — Universal AI Gateway\",\n uptime: \"24/7 on Val Town\",\n persistence: \"SQLite (project-scoped)\",\n tasksProcessed: taskCount,\n hfIntegration: \"Free AI: ALWAYS ON (no token needed) | HF: \" + (token ? \"ACTIVE\" : \"not configured\") + \" | OpenAI: \" + (Deno.env.get(\"OPENAI_API_KEY\") ? \"ACTIVE\" : \"not configured\"),\n protocol: \"JSON-RPC 2.0 / IAP 1.0\",\n models: getModelManifest(),\n agents: getAgentManifest().map((a) => ({\n name: a.name,\n category: a.category,\n capabilities: a.capabilities,\n })),\n adapters: listAdapters(),\n endpoints: {\n POST: {\n \"/\": \"Universal AI Gateway — public, no auth (auto model selection)\",\n \"/public-task\": \"Alias for / (public universal task)\",\n \"/adapter\": \"Direct adapter call (auth required)\",\n \"/handshake\": \"IAP handshake (JSON-RPC, public)\",\n \"/rpc\": \"JSON-RPC 2.0 endpoint (auth required)\",\n },\n GET: {\n \"/\": \"System discovery\",\n \"/status\": \"System status & stats\",\n \"/register\": \"Service manifest for external AI discovery\",\n \"/task/:id\": \"Get task result by ID\",\n \"/history\": \"Recent task history\",\n \"/agents\": \"Agent manifest\",\n \"/adapters\": \"Available adapters\",\n \"/models\": \"Available HF models\",\n },\n },\n example: {\n url: \"POST /\",\n headers: {\n \"Authorization\": \"Bearer YOUR_NEXUSES_API_KEY\",\n \"Content-Type\": \"application/json\",\n },\n body: {\n task: \"debug my code\",\n data: { code: \"function add(a,b){return a-b}\" },\n },\n },\n };\n}\n\nexport async function callAdapterDirect(\n name: string,\n action: string,\n data: any,\n) {\n return routeToAdapter(name, action, data);\n}",[179],"a4ee25e5-963a-46dc-90be-ce037269e7d7","Supervisor.ts","2ac83bf0-7947-11f1-a5c4-1607ee4eb77e",31,["D",1783348626059],"// Supervisor Agent — Nexuses Core OS V10.0\n// Autonomous error loop: Log → Analyze → Retry → Self-correct.\n// Wraps agent execution with supervision and automatic retry logic.\n\nimport {\n callHFWithFallback,\n classifyTask,\n getHFToken,\n selectModel,\n} from \"./hf.ts\";\nimport type { HFResult, ModelKey, TaskType } from \"./hf.ts\";\nimport { logWorkflowStep } from \"./db.ts\";\n\nexport interface SupervisedResult {\n success: boolean;\n attempts: AttemptLog[];\n finalResult: any;\n retried: boolean;\n selfCorrected: boolean;\n message: string;\n}\n\nexport interface AttemptLog {\n attempt: number;\n model: ModelKey;\n status: \"success\" | \"error\" | \"timeout\";\n error?: string;\n action: string;\n}\n\nconst MAX_RETRIES = 3;\n\n/**\n * Supervises a task execution with automatic error recovery.\n *\n * Flow:\n * 1. Execute the task via the provided executor function.\n * 2. If it fails: log the error, analyze failure context with HF.\n * 3. Retry with a different model or adjusted strategy.\n * 4. Self-correct if possible (modify prompt based on error analysis).\n */\nexport async function supervise(\n taskId: string,\n task: string,\n data: any,\n executor: (task: string, data: any) => Promise,\n): Promise {\n const attempts: AttemptLog[] = [];\n let lastError = \"\";\n let selfCorrected = false;\n\n for (let attempt = 1; attempt <= MAX_RETRIES; attempt++) {\n console.log(\n `[Supervisor] Attempt ${attempt}/${MAX_RETRIES} for task: \"${\n task.slice(0, 80)\n }\"`,\n );\n\n try {\n const result = await executor(task, data);\n\n // Check if execution succeeded\n if (\n result.success || result.status === \"SUCCESS\" ||\n result.status === \"success\"\n ) {\n attempts.push({\n attempt,\n model: result.model || \"N/A\",\n status: \"success\",\n action: \"executed\",\n });\n\n await logWorkflowStep(\n taskId,\n 10 + attempt,\n \"Supervisor\",\n \"SUCCESS\",\n `Attempt ${attempt} succeeded | Model: ${result.model || \"N/A\"}`,\n );\n\n return {\n success: true,\n attempts,\n finalResult: result,\n retried: attempt > 1,\n selfCorrected,\n message: result.message || `Task completed on attempt ${attempt}`,\n };\n }\n\n // Execution returned but with failure status\n lastError = result.error || result.message || \"Unknown error\";\n attempts.push({\n attempt,\n model: result.model || \"N/A\",\n status: \"error\",\n error: lastError,\n action: \"failed\",\n });\n\n console.log(`[Supervisor] Attempt ${attempt} failed: ${lastError}`);\n await logWorkflowStep(\n taskId,\n 10 + attempt,\n \"Supervisor\",\n \"RETRY\",\n `Attempt ${attempt} failed: ${lastError}`,\n );\n\n // If this is the last attempt, don't analyze/retry\n if (attempt >= MAX_RETRIES) break;\n\n // PHASE: Analyze the failure\n const analysis = await analyzeFailure(task, data, lastError, attempt);\n await logWorkflowStep(\n taskId,\n 10 + attempt + 0.5,\n \"Supervisor\",\n \"ANALYZE\",\n `Failure analysis: ${analysis.slice(0, 200)}`,\n );\n\n // PHASE: Self-correct — modify the prompt/data based on analysis\n if (attempt === 1) {\n const corrected = await selfCorrect(task, data, lastError, analysis);\n if (corrected.modified) {\n task = corrected.task;\n data = corrected.data;\n selfCorrected = true;\n console.log(\n `[Supervisor] Self-corrected task prompt based on error analysis`,\n );\n await logWorkflowStep(\n taskId,\n 10 + attempt + 0.6,\n \"Supervisor\",\n \"SELF_CORRECT\",\n `Prompt modified based on failure analysis`,\n );\n }\n }\n\n // PHASE: Retry with different strategy\n // The executor itself already has fallback logic via callHFWithFallback,\n // but we retry the whole execution which may pick a different agent/model\n } catch (e: any) {\n lastError = e.message;\n attempts.push({\n attempt,\n model: \"N/A\",\n status: \"error\",\n error: lastError,\n action: \"exception\",\n });\n\n console.log(`[Supervisor] Attempt ${attempt} threw: ${lastError}`);\n await logWorkflowStep(\n taskId,\n 10 + attempt,\n \"Supervisor\",\n \"ERROR\",\n `Attempt ${attempt} exception: ${lastError}`,\n );\n\n if (attempt >= MAX_RETRIES) break;\n }\n }\n\n // All attempts exhausted\n await logWorkflowStep(\n taskId,\n 19,\n \"Supervisor\",\n \"EXHAUSTED\",\n `All ${MAX_RETRIES} attempts failed. Last error: ${lastError}`,\n );\n\n return {\n success: false,\n attempts,\n finalResult: null,\n retried: attempts.length > 1,\n selfCorrected,\n message: `All ${MAX_RETRIES} attempts failed. Last error: ${lastError}`,\n };\n}\n\n/**\n * Analyzes a failure using HF to understand what went wrong.\n */\nasync function analyzeFailure(\n task: string,\n data: any,\n error: string,\n attempt: number,\n): Promise {\n const token = getHFToken();\n if (!token) {\n return \"No HF token — cannot analyze failure. Will retry with same parameters.\";\n }\n\n const prompt =\n `Analyze why this task execution failed and suggest a fix:\\nTask: ${task}\\nData: ${\n JSON.stringify(data).slice(0, 500)\n }\\nError: ${error}\\nAttempt: ${attempt}`;\n\n const result = await callHFWithFallback(prompt, \"gemma\", {\n maxTokens: 200,\n timeoutMs: 10000,\n });\n return result.success ? result.text : \"Analysis unavailable\";\n}\n\n/**\n * Self-corrects the task prompt or data based on failure analysis.\n * Modifies the prompt to be clearer or adjusts parameters.\n */\nasync function selfCorrect(\n task: string,\n data: any,\n error: string,\n analysis: string,\n): Promise<{ task: string; data: any; modified: boolean }> {\n // No token check needed — free AI works without any token\n // If the error is about missing token or config, don't retry with modified prompt\n if (error.includes(\"not configured\") || error.includes(\"NO_TOKEN\")) {\n return { task, data, modified: false };\n }\n\n // If error is a timeout, simplify the prompt\n if (error.includes(\"Timeout\") || error.includes(\"timeout\")) {\n const simplifiedTask = task.slice(0, 200);\n return {\n task: simplifiedTask,\n data: { ...data, maxTokens: Math.min(data.maxTokens || 500, 200) },\n modified: true,\n };\n }\n\n // If error is HTTP 503 (model loading), the retry will naturally try fallback\n if (error.includes(\"503\") || error.includes(\"loading\")) {\n return { task, data, modified: false };\n }\n\n // For other errors, try rephrasing the task\n const rephrasePrompt =\n `Rephrase this task instruction to be clearer and more specific:\\n\"${task}\"\\n\\nOnly return the rephrased instruction, nothing else.`;\n const result = await callHFWithFallback(rephrasePrompt, \"gemma\", {\n maxTokens: 100,\n timeoutMs: 10000,\n });\n\n if (\n result.success && result.text && result.text.length > 10 &&\n result.text.length < 500\n ) {\n return { task: result.text.trim(), data, modified: true };\n }\n\n return { task, data, modified: false };\n}",[186],"259bb3b9-bf6e-439d-94c2-1a72bb52ba1c","RegistryManager.ts","3cd93402-7947-11f1-82a8-1607ee4eb77e",30,["D",1783348552291],"// Registry Manager — Nexuses Core OS V10.0\n// Central catalog of all capabilities: agents, adapters, models, external services.\n// Used by the IAP discovery protocol and /register endpoint.\n\nimport { getAgentManifest } from \"./AgentManager.ts\";\nimport { listAdapters } from \"./UniversalAdapter.ts\";\nimport {\n classifyTask,\n getHFToken,\n getModelManifest,\n type TaskType,\n} from \"./hf.ts\";\n\nexport interface Capability {\n id: string;\n name: string;\n category: string;\n type: \"agent\" | \"adapter\" | \"model\" | \"external\";\n actions: string[];\n status: \"active\" | \"inactive\" | \"error\";\n description: string;\n}\n\nexport interface ServiceManifest {\n service_id: string;\n service_name: string;\n version: string;\n protocol: string;\n endpoint: string;\n auth: string;\n capabilities: Capability[];\n task_types: TaskType[];\n universal_task_format: object;\n registered_at: string;\n}\n\nconst SERVICE_ID = \"nexuses-core-os-v10\";\nconst SERVICE_NAME = \"Nexuses Core OS\";\nconst VERSION = \"10.0.0\";\nconst PROTOCOL = \"JSON-RPC 2.0 / IAP 1.0\";\n\n// --- External Service Registry (for delegation targets) ---\n\ninterface ExternalService {\n id: string;\n name: string;\n url: string;\n capabilities: string[];\n authHeader?: string;\n status: \"available\" | \"unconfigured\";\n}\n\nconst EXTERNAL_SERVICES: ExternalService[] = [\n {\n id: \"hf-router\",\n name: \"Hugging Face Router (Chat Completions)\",\n url: \"https://router.huggingface.co/v1/chat/completions\",\n capabilities: [\n \"text-generation\",\n \"code-generation\",\n \"creative-writing\",\n \"summarization\",\n ],\n authHeader: \"Authorization: Bearer \",\n status: \"available\",\n },\n {\n id: \"hf-image\",\n name: \"Hugging Face Image Generation\",\n url:\n \"https://api-inference.huggingface.co/models/stabilityai/stable-diffusion-xl-base-1.0\",\n capabilities: [\"image-generation\", \"text-to-image\"],\n authHeader: \"Authorization: Bearer \",\n status: \"available\",\n },\n {\n id: \"coingecko\",\n name: \"CoinGecko Market Data\",\n url: \"https://api.coingecko.com/api/v3\",\n capabilities: [\"crypto-prices\", \"market-signals\"],\n status: \"available\",\n },\n {\n id: \"duckduckgo\",\n name: \"DuckDuckGo Instant Answers\",\n url: \"https://api.duckduckgo.com\",\n capabilities: [\"web-search\", \"instant-answers\", \"research\"],\n status: \"available\",\n },\n {\n id: \"github\",\n name: \"GitHub API\",\n url: \"https://api.github.com\",\n capabilities: [\"repo-management\", \"issue-tracking\", \"code-search\"],\n authHeader: \"Authorization: Bearer \",\n status: \"available\",\n },\n];\n\n// --- Registry Functions ---\n\nexport function getExternalServices(): ExternalService[] {\n return EXTERNAL_SERVICES;\n}\n\nexport function findExternalService(\n capability: string,\n): ExternalService | null {\n return EXTERNAL_SERVICES.find((s) =>\n s.capabilities.some((c) => c.includes(capability) || capability.includes(c))\n ) || null;\n}\n\n/**\n * Builds the complete capability manifest by combining:\n * - Internal agents (from AgentManager)\n * - Internal adapters (from UniversalAdapter)\n * - Available HF models (from hf.ts)\n * - External services (from this registry)\n */\nexport function buildCapabilityList(): Capability[] {\n const caps: Capability[] = [];\n\n // Agents\n for (const agent of getAgentManifest()) {\n caps.push({\n id: `agent:${agent.name}`,\n name: agent.name,\n category: agent.category,\n type: \"agent\",\n actions: agent.capabilities,\n status: \"active\",\n description: `${agent.category} agent — handles: ${\n agent.capabilities.join(\", \")\n }`,\n });\n }\n\n // Adapters\n for (const adapter of listAdapters()) {\n caps.push({\n id: `adapter:${adapter.name}`,\n name: adapter.name,\n category: adapter.type,\n type: \"adapter\",\n actions: [\"execute\"],\n status: adapter.status as \"active\" | \"inactive\" | \"error\",\n description: adapter.description,\n });\n }\n\n // Models\n for (const model of getModelManifest()) {\n caps.push({\n id: `model:${model.key}`,\n name: model.label,\n category: \"ai-model\",\n type: \"model\",\n actions: model.bestFor,\n status: model.status === \"AVAILABLE\" ? \"active\" : \"inactive\",\n description: `HF model ${model.id} — best for: ${\n model.bestFor.join(\", \")\n }`,\n });\n }\n\n // External services\n for (const svc of EXTERNAL_SERVICES) {\n caps.push({\n id: `external:${svc.id}`,\n name: svc.name,\n category: \"external-service\",\n type: \"external\",\n actions: svc.capabilities,\n status: svc.status === \"available\" ? \"active\" : \"inactive\",\n description: `External service at ${svc.url} — provides: ${\n svc.capabilities.join(\", \")\n }`,\n });\n }\n\n return caps;\n}\n\n/**\n * Returns the full service manifest for /register endpoint.\n * This is what external AI agents consume to discover and use this OS.\n */\nexport function getServiceManifest(endpoint: string): ServiceManifest {\n const token = getHFToken();\n return {\n service_id: SERVICE_ID,\n service_name: SERVICE_NAME,\n version: VERSION,\n protocol: PROTOCOL,\n endpoint,\n auth: \"Bearer token via Authorization header (NEXUSES_API_KEY)\",\n capabilities: buildCapabilityList(),\n task_types: [\"coding\", \"creative\", \"analysis\", \"general\"],\n universal_task_format: {\n request: {\n task: \"string — task description or command\",\n data: \"object — task-specific parameters\",\n mode: \"AUTO | CLI | DELEGATE (optional, default AUTO)\",\n source: \"string — calling agent/service ID (optional, for IAP)\",\n task_id: \"string — external task ID for tracking (optional)\",\n },\n response: {\n status: \"success | failure\",\n task_id: \"string — internal task ID\",\n data: \"object — execution results\",\n message: \"string — human-readable summary\",\n agent: \"string — agent that handled the task\",\n model: \"string — HF model used (if any)\",\n fallbackUsed: \"boolean — whether model fallback was triggered\",\n attempts: \"string[] — models tried in order\",\n delegated: \"boolean — whether sub-tasks were delegated externally\",\n executionTimeMs: \"number — total execution time\",\n timestamp: \"string — ISO 8601 timestamp\",\n },\n },\n registered_at: new Date().toISOString(),\n };\n}\n\nexport function getServiceInfo() {\n return {\n service_id: SERVICE_ID,\n service_name: SERVICE_NAME,\n version: VERSION,\n protocol: PROTOCOL,\n };\n}\n\n/**\n * Checks if a capability exists internally.\n * Returns false if the OS would need to delegate externally.\n */\nexport function hasInternalCapability(\n taskText: string,\n): { has: boolean; taskType: TaskType; matched: Capability[] } {\n const taskType = classifyTask(taskText);\n const caps = buildCapabilityList();\n const lower = taskText.toLowerCase();\n\n const matched = caps.filter((c) =>\n c.actions.some((a) => lower.includes(a.toLowerCase()))\n );\n\n return {\n has: matched.length > 0,\n taskType,\n matched,\n };\n}",[193],"9cd09b82-70bf-4eee-b4be-853204859196","IAP.ts","5e9009c2-7947-11f1-9d3f-1607ee4eb77e",29,["D",1783348487902],"// Inter-Agent Protocol (IAP) — Nexuses Core OS V10.0\n// JSON-RPC 2.0 handshake, capability advertising, and recursive task delegation.\n\nimport {\n findExternalService,\n getServiceInfo,\n getServiceManifest,\n hasInternalCapability,\n} from \"./RegistryManager.ts\";\nimport type { ServiceManifest } from \"./RegistryManager.ts\";\nimport { routeToAdapter, selectBestModel } from \"./UniversalAdapter.ts\";\nimport { logWorkflowStep } from \"./db.ts\";\n\n// --- JSON-RPC 2.0 Types ---\n\nexport interface JSONRPCRequest {\n jsonrpc: \"2.0\";\n method: string;\n params?: any;\n id?: string | number;\n}\n\nexport interface JSONRPCResponse {\n jsonrpc: \"2.0\";\n result?: any;\n error?: { code: number; message: string; data?: any };\n id: string | number | null;\n}\n\nexport interface HandshakeRequest {\n jsonrpc: \"2.0\";\n method: \"handshake\";\n params: {\n agent_id: string;\n agent_name: string;\n capabilities: string[];\n auth_token?: string;\n };\n id: string | number;\n}\n\nexport interface HandshakeResponse {\n jsonrpc: \"2.0\";\n result: {\n accepted: boolean;\n service: {\n service_id: string;\n service_name: string;\n version: string;\n protocol: string;\n };\n our_capabilities: string[];\n task_format: object;\n challenge: string;\n };\n id: string | number;\n}\n\n// --- Error Codes (JSON-RPC 2.0 standard) ---\n\nconst RPC_ERRORS = {\n PARSE_ERROR: { code: -32700, message: \"Parse error\" },\n INVALID_REQUEST: { code: -32600, message: \"Invalid Request\" },\n METHOD_NOT_FOUND: { code: -32601, message: \"Method not found\" },\n INVALID_PARAMS: { code: -32602, message: \"Invalid params\" },\n INTERNAL_ERROR: { code: -32603, message: \"Internal error\" },\n UNAUTHORIZED: {\n code: -32001,\n message: \"Unauthorized — invalid or missing auth token\",\n },\n CAPABILITY_NOT_FOUND: {\n code: -32002,\n message: \"Requested capability not available\",\n },\n DELEGATION_FAILED: { code: -32003, message: \"External delegation failed\" },\n};\n\nfunction rpcError(\n id: string | number | null,\n error: { code: number; message: string; data?: any },\n): JSONRPCResponse {\n return { jsonrpc: \"2.0\", error, id };\n}\n\nfunction rpcResult(id: string | number, result: any): JSONRPCResponse {\n return { jsonrpc: \"2.0\", result, id };\n}\n\n// --- Handshake Protocol ---\n\n/**\n * Processes an incoming handshake from an external AI agent.\n * Advertises our capabilities and returns the Universal Task Format.\n */\nexport async function processHandshake(\n req: HandshakeRequest,\n endpoint: string,\n): Promise {\n const manifest = getServiceManifest(endpoint);\n const ourCaps = manifest.capabilities.map((c) =>\n `${c.id}:${c.actions.join(\",\")}`\n );\n\n console.log(\n `[IAP] Handshake from agent: ${req.params.agent_id} (${req.params.agent_name})`,\n );\n console.log(\n `[IAP] Agent capabilities: ${req.params.capabilities.join(\", \")}`,\n );\n\n // Log the handshake\n await logWorkflowStep(\n `handshake_${Date.now()}`,\n 1,\n \"IAP\",\n \"HANDSHAKE\",\n `Agent: ${req.params.agent_id} | Capabilities: ${\n req.params.capabilities.join(\", \")\n }`,\n );\n\n return {\n jsonrpc: \"2.0\",\n result: {\n accepted: true,\n service: getServiceInfo(),\n our_capabilities: ourCaps,\n task_format: manifest.universal_task_format,\n challenge: `nxs_${Date.now()}_${Math.random().toString(36).slice(2, 12)}`,\n },\n id: req.id,\n };\n}\n\n// --- JSON-RPC Method Router ---\n\n/**\n * Processes a JSON-RPC 2.0 request.\n * Supported methods: handshake, discover, execute, delegate\n */\nexport async function processRPC(\n rpcReq: JSONRPCRequest,\n endpoint: string,\n dispatchFn: (task: string, data: any, mode: string) => Promise,\n): Promise {\n // Validate JSON-RPC structure\n if (rpcReq.jsonrpc !== \"2.0\" || !rpcReq.method) {\n return rpcError(rpcReq.id ?? null, RPC_ERRORS.INVALID_REQUEST);\n }\n\n switch (rpcReq.method) {\n case \"handshake\": {\n const hsReq = rpcReq as HandshakeRequest;\n if (!hsReq.params?.agent_id) {\n return rpcError(rpcReq.id ?? null, RPC_ERRORS.INVALID_PARAMS);\n }\n const result = await processHandshake(hsReq, endpoint);\n return result as unknown as JSONRPCResponse;\n }\n\n case \"discover\": {\n const manifest = getServiceManifest(endpoint);\n return rpcResult(rpcReq.id!, {\n service: getServiceInfo(),\n capabilities: manifest.capabilities,\n task_format: manifest.universal_task_format,\n });\n }\n\n case \"execute\": {\n const { task, data, mode } = rpcReq.params || {};\n if (!task) {\n return rpcError(rpcReq.id ?? null, {\n ...RPC_ERRORS.INVALID_PARAMS,\n data: \"Missing 'task' field\",\n });\n }\n try {\n const result = await dispatchFn(task, data || {}, mode || \"AUTO\");\n return rpcResult(rpcReq.id!, result);\n } catch (e: any) {\n return rpcError(rpcReq.id ?? null, {\n ...RPC_ERRORS.INTERNAL_ERROR,\n data: e.message,\n });\n }\n }\n\n case \"delegate\": {\n const { task, data, target } = rpcReq.params || {};\n if (!task) {\n return rpcError(rpcReq.id ?? null, {\n ...RPC_ERRORS.INVALID_PARAMS,\n data: \"Missing 'task' field\",\n });\n }\n const result = await delegateToExternal(task, data || {}, target);\n if (result.success) {\n return rpcResult(rpcReq.id!, result);\n }\n return rpcError(rpcReq.id ?? null, {\n ...RPC_ERRORS.DELEGATION_FAILED,\n data: result,\n });\n }\n\n default:\n return rpcError(rpcReq.id ?? null, {\n ...RPC_ERRORS.METHOD_NOT_FOUND,\n data: `Unknown method: ${rpcReq.method}`,\n });\n }\n}\n\n// --- Recursive Task Delegation ---\n\nexport interface DelegationResult {\n success: boolean;\n delegated: boolean;\n target: string | null;\n data: any;\n message: string;\n integrated: boolean;\n}\n\n/**\n * Delegates a sub-task to an external AI service when the OS lacks\n * the internal capability. Integrates the result back for the final JSON.\n */\nexport async function delegateToExternal(\n task: string,\n data: any,\n targetHint?: string,\n): Promise {\n // Find an external service that can handle this\n let service = targetHint ? findExternalService(targetHint) : null;\n if (!service) {\n const check = hasInternalCapability(task);\n // Try to find external service matching the task type\n service = findExternalService(check.taskType) ||\n findExternalService(task.toLowerCase());\n }\n\n if (!service) {\n return {\n success: false,\n delegated: false,\n target: null,\n data: {},\n message: \"No external service available for this capability\",\n integrated: false,\n };\n }\n\n console.log(\n `[IAP] Delegating to external service: ${service.name} (${service.url})`,\n );\n\n // Build the delegation request\n // Try free AI first if no token-based service is available\n const hfKey = Deno.env.get(\"HF_API_KEY\") || Deno.env.get(\"HUGGING_FACE_TOKEN\");\n if (!hfKey && service.id.startsWith(\"hf\")) {\n // No HF token — use free AI for delegation instead\n const freeResult = await routeToAdapter(\"freeai\", \"generate\", {\n prompt: data.prompt || task,\n maxTokens: 500,\n timeoutMs: 25000,\n });\n if (freeResult.success && freeResult.data?.response) {\n await logWorkflowStep(\n `delegation_${Date.now()}`,\n 1,\n \"IAP\",\n \"DELEGATE\",\n `Task: \"${task}\" → Free AI → Success`,\n );\n return {\n success: true,\n delegated: true,\n target: \"freeai\",\n data: { generated_text: freeResult.data.response },\n message: \"Delegated to Free AI (no token needed) and integrated result\",\n integrated: true,\n };\n }\n }\n\n const token = service.authHeader?.includes(\"HF_API_KEY\")\n ? hfKey\n : service.authHeader?.includes(\"GITHUB_TOKEN\")\n ? Deno.env.get(\"GITHUB_TOKEN\")\n : null;\n\n try {\n let resultData: any;\n\n if (service.id.startsWith(\"hf\")) {\n // Hugging Face delegation via new Router endpoint (chat completions)\n const hfKey = Deno.env.get(\"HF_API_KEY\") || token;\n const res = await fetch(\n \"https://router.huggingface.co/v1/chat/completions\",\n {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${hfKey}`,\n \"Content-Type\": \"application/json\",\n },\n body: JSON.stringify({\n model: selectBestModel(data.prompt || task),\n messages: [{ role: \"user\", content: data.prompt || task }],\n max_tokens: 500,\n }),\n signal: AbortSignal.timeout(25000),\n },\n );\n if (!res.ok) {\n return {\n success: false,\n delegated: true,\n target: service.id,\n data: {},\n message: `External service returned HTTP ${res.status}`,\n integrated: false,\n };\n }\n const hfData = await res.json();\n resultData = hfData.choices?.[0]?.message?.content\n ? { generated_text: hfData.choices[0].message.content }\n : hfData;\n } else {\n // Generic external API call\n const res = await fetch(service.url, {\n signal: AbortSignal.timeout(15000),\n });\n resultData = await res.json();\n }\n\n // Log the delegation\n await logWorkflowStep(\n `delegation_${Date.now()}`,\n 1,\n \"IAP\",\n \"DELEGATE\",\n `Task: \"${task}\" → Service: ${service.id} → Success`,\n );\n\n return {\n success: true,\n delegated: true,\n target: service.id,\n data: resultData,\n message: `Delegated to ${service.name} and integrated result`,\n integrated: true,\n };\n } catch (e: any) {\n console.log(`[IAP] Delegation failed: ${e.message}`);\n return {\n success: false,\n delegated: true,\n target: service.id,\n data: {},\n message: `Delegation error: ${e.message}`,\n integrated: false,\n };\n }\n}",[200],"3fc99152-68ef-4298-b113-837e6542dffe","AgentManager.ts","6b3026f8-7947-11f1-82a8-1607ee4eb77e",28,["D",1783348386548],3,"// Agent Swarm — Nexuses Core OS V10.0\n// All agents use callHFWithFallback (HF Router with auto model selection via selectBestModel()).\nimport {\n callHFWithFallback,\n classifyTask,\n getHFToken,\n getSelectionInfo,\n} from \"./hf.ts\";\nimport type { HFResult, ModelKey, TaskType } from \"./hf.ts\";\nimport { listAdapters, routeToAdapter } from \"./UniversalAdapter.ts\";\n\nexport interface AgentResult {\n agent: string;\n category: string;\n status: \"SUCCESS\" | \"ERROR\" | \"PARTIAL\";\n message: string;\n data: any;\n model?: ModelKey;\n modelId?: string;\n taskType?: TaskType;\n fallbackUsed?: boolean;\n attempts?: ModelKey[];\n adapters?: string[];\n}\n\nexport interface AgentManifestEntry {\n name: string;\n category: string;\n keywords: string[];\n capabilities: string[];\n model: ModelKey;\n adapters: string[];\n}\n\n// --- Agent Registry ---\n\nexport const AGENT_REGISTRY: AgentManifestEntry[] = [\n {\n name: \"DevOpsCodingAgent\",\n category: \"DevOps/Coding\",\n keywords: [\n \"code\",\n \"coding\",\n \"debug\",\n \"bug\",\n \"fix\",\n \"github\",\n \"repo\",\n \"program\",\n \"vscode\",\n \"vs code\",\n \"commit\",\n \"deploy\",\n \"devops\",\n \"refactor\",\n \"compile\",\n \"syntax\",\n \"error\",\n ],\n capabilities: [\n \"VS Code integration\",\n \"GitHub repo management\",\n \"Error finding & fixing\",\n \"Bug resolution\",\n \"Auto-writing code\",\n \"Code explanation\",\n ],\n model: \"qwen_coder\",\n adapters: [\"github\"],\n },\n {\n name: \"WebContentAgent\",\n category: \"Web/Content\",\n keywords: [\n \"write\",\n \"blog\",\n \"content\",\n \"article\",\n \"social\",\n \"post\",\n \"website\",\n \"clone\",\n \"scrape\",\n \"web\",\n \"voice\",\n \"html\",\n \"css\",\n \"landing\",\n \"seo\",\n \"tweet\",\n \"facebook\",\n \"media\",\n ],\n capabilities: [\n \"Web automation\",\n \"Website cloning\",\n \"Social media posting\",\n \"Content writing\",\n \"Blogging\",\n \"Voice-enabled website generation\",\n ],\n model: \"llama\",\n adapters: [\"webscraper\", \"duckduckgo\"],\n },\n {\n name: \"DataAutomationAgent\",\n category: \"Data/Automation\",\n keywords: [\n \"data\",\n \"scrape\",\n \"analyze\",\n \"pdf\",\n \"excel\",\n \"word\",\n \"billing\",\n \"invoice\",\n \"entry\",\n \"scan\",\n \"spreadsheet\",\n \"csv\",\n \"report\",\n \"automation\",\n \"crypto\",\n \"price\",\n \"market\",\n ],\n capabilities: [\n \"Data scraping & analysis\",\n \"PDF conversion/creation\",\n \"Billing/data entry\",\n \"MS Word/Excel automation\",\n \"Document scanning\",\n ],\n model: \"phi3\",\n adapters: [\"coingecko\", \"duckduckgo\", \"webscraper\"],\n },\n {\n name: \"MultimediaAgent\",\n category: \"Multimedia\",\n keywords: [\n \"image\",\n \"video\",\n \"generate\",\n \"picture\",\n \"photo\",\n \"draw\",\n \"art\",\n \"design\",\n \"logo\",\n \"render\",\n \"media\",\n \"visual\",\n \"stable diffusion\",\n \"dalle\",\n \"midjourney\",\n ],\n capabilities: [\n \"Image generation\",\n \"Video generation guidance\",\n \"Visual design\",\n \"Art creation\",\n \"Logo design\",\n ],\n model: \"gemma\",\n adapters: [],\n },\n {\n name: \"OfficeDataAgent\",\n category: \"Office/Data\",\n keywords: [\n \"pdf\",\n \"word\",\n \"excel\",\n \"spreadsheet\",\n \"invoice\",\n \"billing\",\n \"document\",\n \"report\",\n \"csv\",\n \"word doc\",\n \"ms word\",\n \"ms excel\",\n \"office\",\n \"data entry\",\n \"scan\",\n ],\n capabilities: [\n \"PDF creation\",\n \"MS Word/Excel automation\",\n \"Invoice/billing generation\",\n \"Data entry\",\n \"Document formatting\",\n ],\n model: \"phi3\",\n adapters: [],\n },\n {\n name: \"GeneralHandler\",\n category: \"General\",\n keywords: [\n \"general\",\n \"help\",\n \"unknown\",\n \"test\",\n \"explain\",\n \"reasoning\",\n \"question\",\n \"chat\",\n ],\n capabilities: [\"General task handling\", \"Fallback routing\"],\n model: \"gemma\",\n adapters: [],\n },\n];\n\nexport function getAgentManifest() {\n return AGENT_REGISTRY;\n}\n\nfunction matchAgent(task: string): AgentManifestEntry {\n const cmd = task.toLowerCase();\n let best: AgentManifestEntry | null = null;\n let bestScore = 0;\n for (const agent of AGENT_REGISTRY) {\n const score = agent.keywords.filter((k) => cmd.includes(k)).length;\n if (score > bestScore) {\n bestScore = score;\n best = agent;\n }\n }\n return best || AGENT_REGISTRY[AGENT_REGISTRY.length - 1];\n}\n\n/**\n * Builds an AgentResult from an HFResult, including fallback metadata.\n */\nfunction buildAgentResult(\n agentName: string,\n category: string,\n hfResult: HFResult,\n message: string,\n data: any,\n adapters: string[] = [],\n): AgentResult {\n const fallbackUsed = (hfResult.attempts?.length ?? 1) > 1 && hfResult.success;\n return {\n agent: agentName,\n category,\n status: hfResult.success ? \"SUCCESS\" : \"ERROR\",\n message: hfResult.success ? message : hfResult.error || \"HF model error\",\n data,\n model: hfResult.model,\n modelId: hfResult.modelId,\n taskType: hfResult.taskType,\n fallbackUsed,\n attempts: hfResult.attempts,\n adapters,\n };\n}\n\n// --- Agent Execution Functions ---\n\nconst AGENT_MAP: Record<\n string,\n (task: string, data: any) => Promise\n> = {};\n\n// 1. DevOps / Coding Agent — uses Qwen2.5-Coder with fallback to Gemma\nAGENT_MAP[\"DevOpsCodingAgent\"] = async (task: string, data: any) => {\n const prompt = data.prompt || data.code || task;\n const action = data.action || \"analyze\";\n\n // GitHub operations via Universal Adapter\n if (data.github) {\n const ghResult = await routeToAdapter(\n \"github\",\n data.github.action || \"get_repo\",\n data.github,\n );\n if (ghResult.success) {\n const hfResult = await callHFWithFallback(\n `Analyze this GitHub data and provide insights: ${\n JSON.stringify(ghResult.data).slice(0, 1000)\n }`,\n \"qwen_coder\",\n { maxTokens: 500 },\n );\n return {\n agent: \"DevOpsCodingAgent\",\n category: \"DevOps/Coding\",\n status: hfResult.success ? \"SUCCESS\" : \"PARTIAL\",\n message: `GitHub ${data.github.action || \"operation\"} completed${\n hfResult.success ? \" + AI analysis\" : \"\"\n }`,\n data: {\n github: ghResult.data,\n analysis: hfResult.text || hfResult.error,\n },\n model: hfResult.model,\n modelId: hfResult.modelId,\n taskType: hfResult.taskType,\n fallbackUsed: (hfResult.attempts?.length ?? 1) > 1 && hfResult.success,\n attempts: hfResult.attempts,\n adapters: [\"github\"],\n };\n }\n return {\n agent: \"DevOpsCodingAgent\",\n category: \"DevOps/Coding\",\n status: \"ERROR\",\n message: ghResult.message,\n data: {},\n adapters: [\"github\"],\n };\n }\n\n // Code analysis / bug fixing / code writing\n const codePrompt = action === \"fix\"\n ? `Find and fix the bug in this code:\\n${prompt}`\n : action === \"explain\"\n ? `Explain this code:\\n${prompt}`\n : action === \"write\"\n ? `Write code for: ${prompt}`\n : `Analyze this code for errors and improvements:\\n${prompt}`;\n\n const hfResult = await callHFWithFallback(codePrompt, \"qwen_coder\", {\n maxTokens: 800,\n temperature: 0.3,\n });\n\n return buildAgentResult(\n \"DevOpsCodingAgent\",\n \"DevOps/Coding\",\n hfResult,\n `Code ${action} completed using ${HF_LABEL(hfResult.model)}`,\n { action, input: prompt.slice(0, 500), output: hfResult.text },\n );\n};\n\n// 2. Web / Content Agent — uses Llama 3 with fallback to Gemma\nAGENT_MAP[\"WebContentAgent\"] = async (task: string, data: any) => {\n const prompt = data.prompt || task;\n const action = data.action || \"write\";\n\n // Website cloning / scraping\n if (data.url || action === \"scrape\" || action === \"clone\") {\n const scrapeResult = await routeToAdapter(\"webscraper\", \"scrape\", {\n url: data.url,\n });\n if (scrapeResult.success && action === \"clone\") {\n const hfResult = await callHFWithFallback(\n `Create a simple HTML page based on this scraped content:\\nTitle: ${scrapeResult.data.title}\\nContent: ${\n scrapeResult.data.textContent?.slice(0, 800)\n }`,\n \"llama\",\n { maxTokens: 800 },\n );\n return {\n agent: \"WebContentAgent\",\n category: \"Web/Content\",\n status: hfResult.success ? \"SUCCESS\" : \"PARTIAL\",\n message: `Website cloned: ${scrapeResult.data.title}`,\n data: {\n original: { url: data.url, title: scrapeResult.data.title },\n clone: hfResult.text || \"\",\n links: scrapeResult.data.links,\n },\n model: hfResult.model,\n modelId: hfResult.modelId,\n taskType: hfResult.taskType,\n fallbackUsed: (hfResult.attempts?.length ?? 1) > 1 && hfResult.success,\n attempts: hfResult.attempts,\n adapters: [\"webscraper\"],\n };\n }\n if (scrapeResult.success) {\n return {\n agent: \"WebContentAgent\",\n category: \"Web/Content\",\n status: \"SUCCESS\",\n message: `Scraped: ${scrapeResult.data.title}`,\n data: scrapeResult.data,\n adapters: [\"webscraper\"],\n };\n }\n }\n\n // Content writing / blogging / social media\n const contentPrompt = action === \"social\"\n ? `Write a social media post about: ${prompt}`\n : action === \"blog\"\n ? `Write a blog post about: ${prompt}`\n : action === \"voice\"\n ? `Generate voice-ready website content for: ${prompt}`\n : `Write content about: ${prompt}`;\n\n const hfResult = await callHFWithFallback(contentPrompt, \"llama\", {\n maxTokens: 1000,\n temperature: 0.8,\n });\n\n return buildAgentResult(\n \"WebContentAgent\",\n \"Web/Content\",\n hfResult,\n `Content generated using ${HF_LABEL(hfResult.model)}`,\n { action, content: hfResult.text },\n );\n};\n\n// 3. Data / Automation Agent — uses Phi-3 with fallback to Gemma\nAGENT_MAP[\"DataAutomationAgent\"] = async (task: string, data: any) => {\n const prompt = data.prompt || task;\n const action = data.action || \"analyze\";\n\n // Crypto signals (no HF needed — direct adapter)\n if (\n action === \"crypto\" || prompt.toLowerCase().includes(\"crypto\") ||\n prompt.toLowerCase().includes(\"price\")\n ) {\n const cgResult = await routeToAdapter(\"coingecko\", \"price\", {\n coins: data.coins,\n });\n if (cgResult.success) {\n return {\n agent: \"DataAutomationAgent\",\n category: \"Data/Automation\",\n status: \"SUCCESS\",\n message: \"Crypto signals generated\",\n data: cgResult.data,\n adapters: [\"coingecko\"],\n };\n }\n }\n\n // Data scraping + analysis\n if (data.url || action === \"scrape\") {\n const scrapeResult = await routeToAdapter(\"webscraper\", \"scrape\", {\n url: data.url,\n });\n if (scrapeResult.success) {\n const hfResult = await callHFWithFallback(\n `Analyze this scraped data and provide a summary:\\n${\n scrapeResult.data.textContent?.slice(0, 1000)\n }`,\n \"phi3\",\n { maxTokens: 400 },\n );\n return {\n agent: \"DataAutomationAgent\",\n category: \"Data/Automation\",\n status: hfResult.success ? \"SUCCESS\" : \"PARTIAL\",\n message: \"Data scraped and analyzed\",\n data: {\n source: data.url,\n title: scrapeResult.data.title,\n analysis: hfResult.text || \"Analysis unavailable\",\n raw: scrapeResult.data,\n },\n model: hfResult.model,\n modelId: hfResult.modelId,\n taskType: hfResult.taskType,\n fallbackUsed: (hfResult.attempts?.length ?? 1) > 1 && hfResult.success,\n attempts: hfResult.attempts,\n adapters: [\"webscraper\"],\n };\n }\n }\n\n // Research via DuckDuckGo + HF summarization\n if (action === \"research\" || action === \"search\") {\n const ddgResult = await routeToAdapter(\"duckduckgo\", \"search\", {\n query: prompt,\n });\n if (ddgResult.success) {\n const hfResult = await callHFWithFallback(\n `Summarize this research data:\\n${\n JSON.stringify(ddgResult.data).slice(0, 800)\n }`,\n \"phi3\",\n { maxTokens: 400 },\n );\n return {\n agent: \"DataAutomationAgent\",\n category: \"Data/Automation\",\n status: hfResult.success ? \"SUCCESS\" : \"PARTIAL\",\n message: \"Research completed and summarized\",\n data: {\n search: ddgResult.data,\n summary: hfResult.text || \"\",\n },\n model: hfResult.model,\n modelId: hfResult.modelId,\n taskType: hfResult.taskType,\n fallbackUsed: (hfResult.attempts?.length ?? 1) > 1 && hfResult.success,\n attempts: hfResult.attempts,\n adapters: [\"duckduckgo\"],\n };\n }\n }\n\n // General data analysis\n const hfResult = await callHFWithFallback(\n `Analyze and process: ${prompt}`,\n \"phi3\",\n { maxTokens: 500, temperature: 0.5 },\n );\n\n return buildAgentResult(\n \"DataAutomationAgent\",\n \"Data/Automation\",\n hfResult,\n \"Data analysis completed\",\n { action, result: hfResult.text },\n );\n};\n\n// 4. Multimedia Agent — image/video generation via HF + text guidance\nAGENT_MAP[\"MultimediaAgent\"] = async (task: string, data: any) => {\n const prompt = data.prompt || task;\n const action = data.action || \"image\";\n\n if (action === \"image\") {\n // Attempt HF image generation (Stable Diffusion)\n const token = getHFToken();\n if (!token) {\n // Fallback: generate a descriptive text prompt that can be used externally\n const hfResult = await callHFWithFallback(\n `Generate a detailed image generation prompt for: ${prompt}. Include style, composition, and lighting details.`,\n \"gemma\",\n { maxTokens: 300 },\n );\n return buildAgentResult(\n \"MultimediaAgent\",\n \"Multimedia\",\n hfResult,\n \"Image prompt generated (HF_API_KEY needed for actual image generation)\",\n {\n action,\n prompt,\n imagePrompt: hfResult.text,\n note: \"Use this prompt with any text-to-image service\",\n },\n );\n }\n try {\n // Image generation still uses HF Inference API (router is chat-only)\n const res = await fetch(\n \"https://api-inference.huggingface.co/models/stabilityai/stable-diffusion-xl-base-1.0\",\n {\n method: \"POST\",\n headers: {\n \"Authorization\": `Bearer ${token}`,\n \"Content-Type\": \"application/json\",\n },\n body: JSON.stringify({ inputs: prompt }),\n signal: AbortSignal.timeout(30000),\n },\n );\n if (res.ok) {\n const imageBlob = await res.blob();\n return {\n agent: \"MultimediaAgent\",\n category: \"Multimedia\",\n status: \"SUCCESS\",\n message: `Image generated for: ${prompt.slice(0, 80)}`,\n data: {\n action,\n prompt,\n imageGenerated: true,\n sizeBytes: imageBlob.size,\n mimeType: imageBlob.type,\n },\n model: \"gemma\" as ModelKey,\n adapters: [],\n };\n }\n // If image model fails, generate text prompt as fallback\n const hfResult = await callHFWithFallback(\n `Generate a detailed image generation prompt for: ${prompt}`,\n \"gemma\",\n { maxTokens: 300 },\n );\n return buildAgentResult(\n \"MultimediaAgent\",\n \"Multimedia\",\n hfResult,\n `Image model unavailable (HTTP ${res.status}), generated prompt instead`,\n { action, prompt, imagePrompt: hfResult.text, fallback: true },\n );\n } catch (e: any) {\n const hfResult = await callHFWithFallback(\n `Describe how to create: ${prompt}. Include visual details.`,\n \"gemma\",\n { maxTokens: 300 },\n );\n return buildAgentResult(\n \"MultimediaAgent\",\n \"Multimedia\",\n hfResult,\n `Image generation error: ${e.message}. Generated description instead.`,\n { action, prompt, description: hfResult.text, error: e.message },\n );\n }\n }\n\n // Video generation guidance\n const hfResult = await callHFWithFallback(\n `Provide guidance and a creative plan for: ${prompt}. Include scene descriptions and technical details.`,\n \"gemma\",\n { maxTokens: 500 },\n );\n return buildAgentResult(\n \"MultimediaAgent\",\n \"Multimedia\",\n hfResult,\n \"Multimedia guidance generated\",\n { action, prompt, guidance: hfResult.text },\n );\n};\n\n// 5. Office / Data Agent — PDF, Word, Excel, invoices, documents\nAGENT_MAP[\"OfficeDataAgent\"] = async (task: string, data: any) => {\n const prompt = data.prompt || task;\n const action = data.action || \"document\";\n\n // Invoice/billing generation\n if (action === \"invoice\" || prompt.toLowerCase().includes(\"invoice\")) {\n const hfResult = await callHFWithFallback(\n `Generate a professional invoice template in JSON format with these details: ${prompt}\\nInclude fields: invoiceNumber, date, billTo, items[], subtotal, tax, total.`,\n \"phi3\",\n { maxTokens: 500, temperature: 0.3 },\n );\n return buildAgentResult(\n \"OfficeDataAgent\",\n \"Office/Data\",\n hfResult,\n \"Invoice generated\",\n { action: \"invoice\", prompt, invoice: hfResult.text },\n );\n }\n\n // PDF / document structure generation\n if (action === \"pdf\" || prompt.toLowerCase().includes(\"pdf\")) {\n const hfResult = await callHFWithFallback(\n `Create a structured document outline for: ${prompt}. Include sections, headings, and key content points.`,\n \"phi3\",\n { maxTokens: 500 },\n );\n return buildAgentResult(\n \"OfficeDataAgent\",\n \"Office/Data\",\n hfResult,\n \"Document structure generated\",\n {\n action: \"pdf\",\n prompt,\n document: hfResult.text,\n note: \"Structure ready for PDF conversion\",\n },\n );\n }\n\n // Excel/CSV data generation\n if (\n action === \"excel\" || action === \"csv\" ||\n prompt.toLowerCase().includes(\"excel\") ||\n prompt.toLowerCase().includes(\"csv\")\n ) {\n const hfResult = await callHFWithFallback(\n `Generate tabular data in CSV format for: ${prompt}. Include headers and sample rows.`,\n \"phi3\",\n { maxTokens: 400, temperature: 0.3 },\n );\n return buildAgentResult(\n \"OfficeDataAgent\",\n \"Office/Data\",\n hfResult,\n \"Spreadsheet data generated\",\n { action: action, prompt, csv: hfResult.text },\n );\n }\n\n // General document/report generation\n const hfResult = await callHFWithFallback(\n `Create a professional document for: ${prompt}`,\n \"phi3\",\n { maxTokens: 500 },\n );\n return buildAgentResult(\n \"OfficeDataAgent\",\n \"Office/Data\",\n hfResult,\n \"Document generated\",\n { action, prompt, document: hfResult.text },\n );\n};\n\n// 6. General Handler — uses Gemma (no further fallback)\nAGENT_MAP[\"GeneralHandler\"] = async (task: string, data: any) => {\n const hfResult = await callHFWithFallback(data.prompt || task, \"gemma\", {\n maxTokens: 500,\n });\n\n return buildAgentResult(\n \"GeneralHandler\",\n \"General\",\n hfResult,\n \"Task processed\",\n { input: task, response: hfResult.text || \"\" },\n );\n};\n\n// --- Helpers ---\n\nfunction HF_LABEL(key: ModelKey): string {\n const labels: Record = {\n qwen_coder: \"Qwen2.5-Coder\",\n llama: \"Llama 3\",\n phi3: \"Phi-3 Mini\",\n gemma: \"Gemma 2 2B\",\n };\n return labels[key] || key;\n}\n\n// --- Public Routing API ---\n\nexport async function routeIncomingTask(\n task: string,\n data: any = {},\n mode = \"AUTO\",\n debug = false,\n): Promise {\n const agentEntry = matchAgent(task);\n const agentFn = AGENT_MAP[agentEntry.name] || AGENT_MAP[\"GeneralHandler\"];\n\n try {\n const result = await agentFn(task, data);\n if (debug) {\n console.log(\n `[Dispatcher] Task: \"${task}\" → Agent: ${result.agent} → Model: ${\n result.model || \"N/A\"\n } → Status: ${result.status}${\n result.fallbackUsed ? \" (FALLBACK)\" : \"\"\n }`,\n );\n }\n return result;\n } catch (e: any) {\n return {\n agent: agentEntry.name,\n category: agentEntry.category,\n status: \"ERROR\",\n message: `Agent execution error: ${e.message}`,\n data: {},\n };\n }\n}\n\nexport function getSystemInfo() {\n const token = getHFToken();\n return {\n agents: AGENT_REGISTRY.map((a) => ({\n name: a.name,\n category: a.category,\n capabilities: a.capabilities,\n model: a.model,\n })),\n adapters: listAdapters(),\n ai: `Free AI (Pollinations.ai): ALWAYS AVAILABLE | OpenAI: ${Deno.env.get(\"OPENAI_API_KEY\") ? \"ACTIVE\" : \"not configured\"} | HF Router: ${token ? \"ACTIVE\" : \"not configured\"} | Groq: ${Deno.env.get(\"GROQ_API_KEY\") ? \"ACTIVE\" : \"not configured\"}`,\n };\n}",[207],"10657c68-b48a-444f-8dfe-b4fe9700ad16","lib","69dda6e0-7960-11f1-9859-1607ee4eb77e",37,["D",1783359042499],"directory",[215],"0dbb60d1-1be9-40d7-a97a-7330cd44c97c","agents","69ed2fde-7960-11f1-9859-1607ee4eb77e",38,["D",1783359042602],[222],"pulls","branches",1,"forks","envVars",{"_235":68,"_237":68},"connections",{"_235":68,"_236":68},"count","unfilledCount","emptyCount",{"_48":221,"_37":222,"_35":223,"_148":-5,"_46":47,"_45":224,"_150":64,"_43":365,"_13":68,"_39":219,"_41":-5,"_29":30,"_155":-5,"_260":366,"_32":367},{"_48":357,"_37":358,"_35":359,"_148":223,"_46":47,"_45":360,"_150":64,"_43":361,"_13":68,"_39":40,"_41":362,"_29":30,"_155":-5,"_260":363,"_32":364},{"_48":349,"_37":350,"_35":351,"_148":223,"_46":47,"_45":352,"_150":64,"_43":353,"_13":68,"_39":40,"_41":354,"_29":30,"_155":-5,"_260":355,"_32":356},{"_48":341,"_37":342,"_35":343,"_148":223,"_46":47,"_45":344,"_150":64,"_43":345,"_13":68,"_39":40,"_41":346,"_29":30,"_155":-5,"_260":347,"_32":348},{"_48":333,"_37":334,"_35":335,"_148":223,"_46":47,"_45":336,"_150":64,"_43":337,"_13":68,"_39":40,"_41":338,"_29":30,"_155":-5,"_260":339,"_32":340},{"_48":325,"_37":326,"_35":327,"_148":223,"_46":47,"_45":328,"_150":64,"_43":329,"_13":68,"_39":40,"_41":330,"_29":30,"_155":-5,"_260":331,"_32":332},{"_48":214,"_37":215,"_35":216,"_148":-5,"_46":47,"_45":217,"_150":64,"_43":322,"_13":68,"_39":219,"_41":-5,"_29":30,"_155":-5,"_260":323,"_32":324},{"_48":315,"_37":165,"_35":316,"_148":216,"_46":47,"_45":317,"_150":30,"_43":318,"_13":68,"_39":40,"_41":319,"_29":30,"_155":-5,"_260":320,"_32":321},{"_48":307,"_37":308,"_35":309,"_148":216,"_46":47,"_45":310,"_150":64,"_43":311,"_13":68,"_39":40,"_41":312,"_29":30,"_155":-5,"_260":313,"_32":314},{"_48":303,"_37":38,"_35":36,"_148":216,"_46":47,"_45":14,"_150":64,"_43":304,"_13":68,"_39":40,"_41":42,"_29":30,"_155":-5,"_260":305,"_32":306},{"_48":295,"_37":296,"_35":297,"_148":216,"_46":47,"_45":298,"_150":64,"_43":299,"_13":68,"_39":40,"_41":300,"_29":30,"_155":-5,"_260":301,"_32":302},{"_48":287,"_37":288,"_35":289,"_148":216,"_46":47,"_45":290,"_150":64,"_43":291,"_13":68,"_39":40,"_41":292,"_29":30,"_155":-5,"_260":293,"_32":294},{"_48":206,"_37":207,"_35":208,"_148":-5,"_46":47,"_45":209,"_150":30,"_43":284,"_13":211,"_39":40,"_41":212,"_29":30,"_155":-5,"_260":285,"_32":286},{"_48":199,"_37":200,"_35":201,"_148":-5,"_46":47,"_45":202,"_150":30,"_43":281,"_13":152,"_39":40,"_41":204,"_29":30,"_155":-5,"_260":282,"_32":283},{"_48":192,"_37":193,"_35":194,"_148":-5,"_46":47,"_45":195,"_150":30,"_43":278,"_13":152,"_39":40,"_41":197,"_29":30,"_155":-5,"_260":279,"_32":280},{"_48":185,"_37":186,"_35":187,"_148":-5,"_46":47,"_45":188,"_150":30,"_43":275,"_13":152,"_39":40,"_41":190,"_29":30,"_155":-5,"_260":276,"_32":277},{"_48":178,"_37":179,"_35":180,"_148":-5,"_46":47,"_45":181,"_150":30,"_43":272,"_13":152,"_39":40,"_41":183,"_29":30,"_155":-5,"_260":273,"_32":274},{"_48":171,"_37":172,"_35":173,"_148":-5,"_46":47,"_45":174,"_150":30,"_43":269,"_13":152,"_39":40,"_41":176,"_29":30,"_155":-5,"_260":270,"_32":271},{"_48":164,"_37":165,"_35":166,"_148":-5,"_46":47,"_45":167,"_150":30,"_43":266,"_13":152,"_39":40,"_41":169,"_29":30,"_155":-5,"_260":267,"_32":268},{"_48":157,"_37":158,"_35":159,"_148":-5,"_46":47,"_45":160,"_150":30,"_43":263,"_13":152,"_39":40,"_41":162,"_29":30,"_155":-5,"_260":264,"_32":265},{"_48":146,"_37":147,"_35":62,"_148":-5,"_46":47,"_45":149,"_150":30,"_43":259,"_13":152,"_39":153,"_41":154,"_29":30,"_155":-5,"_260":261,"_32":262},["D",1783356891573],"children",[],[147],["D",1783356003650],[],[158],["D",1783355945330],[],[165],["D",1783355890857],[],[172],["D",1783355841883],[],[179],["D",1783348626059],[],[186],["D",1783348552291],[],[193],["D",1783348487902],[],[200],["D",1783348386548],[],[207],"29378822-e19e-42ba-8d2d-4b0658f81265","verification.ts","967d813e-7960-11f1-b2d7-1607ee4eb77e",42,["D",1783359117369],"// Source Verification Layer — Nexuses Core V3\n// Every incoming request must carry a SourceID.\n// The system confirms if the request originates from a trusted/registered platform.\n\nimport { initDB, registerSource, verifySource } from \"./db.ts\";\nimport type { VerificationResult } from \"./types.ts\";\n\n/**\n * Verifies that a SourceID is registered and active.\n * Returns a structured VerificationResult for M2M consumption.\n */\nexport async function checkSource(sourceId: string): Promise {\n await initDB();\n\n if (!sourceId || sourceId.trim().length === 0) {\n return {\n verified: false,\n source_id: sourceId || \"MISSING\",\n message: \"SourceID is required — request rejected\",\n };\n }\n\n const source = await verifySource(sourceId);\n if (!source) {\n return {\n verified: false,\n source_id: sourceId,\n message: `Verification Failed — SourceID \"${sourceId}\" is not registered. Use POST /register to register this source.`,\n };\n }\n\n return {\n verified: true,\n source_id: sourceId,\n name: source.name,\n message: `Source verified: ${source.name}`,\n };\n}\n\n/**\n * Registers a new trusted source.\n * Called via POST /register endpoint.\n */\nexport async function registerNewSource(\n sourceId: string,\n name: string,\n description: string,\n): Promise {\n await initDB();\n\n if (!sourceId || sourceId.trim().length < 3) {\n return {\n verified: false,\n source_id: sourceId || \"MISSING\",\n message: \"SourceID must be at least 3 characters\",\n };\n }\n\n const success = await registerSource(sourceId, name, description);\n if (!success) {\n return {\n verified: false,\n source_id: sourceId,\n message: \"Failed to register source — database error\",\n };\n }\n\n return {\n verified: true,\n source_id: sourceId,\n name,\n message: `Source \"${name}\" registered successfully with SourceID \"${sourceId}\"`,\n };\n}\n",[],[215,288],"13ca991a-7060-4d3b-92a7-1e1a29e5086a","types.ts","75094510-7960-11f1-acba-1607ee4eb77e",39,["D",1783359061231],"// Nexuses Core V3 — Shared Type Definitions\n// Central type registry for the entire autonomous AI OS.\n\nexport type TaskCategory =\n | \"coding\"\n | \"scraping\"\n | \"analysis\"\n | \"media\"\n | \"marketing\"\n | \"general\";\n\nexport type AgentName =\n | \"CoderAgent\"\n | \"ScraperAgent\"\n | \"AnalystAgent\"\n | \"MediaAgent\"\n | \"MarketingAgent\"\n | \"GeneralAgent\";\n\nexport interface AgentResult {\n agent: AgentName;\n category: TaskCategory;\n status: \"success\" | \"error\" | \"partial\";\n message: string;\n data: any;\n executionTimeMs: number;\n}\n\nexport interface StepResult {\n step: number;\n description: string;\n agent: AgentName;\n status: \"success\" | \"error\" | \"failed\";\n result?: any;\n error?: string;\n attempts: number;\n}\n\nexport interface CoTAnalysis {\n reasoning: string;\n category: TaskCategory;\n agent: AgentName;\n steps: string[];\n complexity: \"simple\" | \"complex\";\n}\n\nexport interface DispatchResponse {\n status: \"success\" | \"failure\";\n task_id: string;\n source_id: string;\n verified: boolean;\n reasoning: string;\n category: TaskCategory;\n agent: AgentName;\n steps: StepResult[];\n data: any;\n message: string;\n executionTimeMs: number;\n timestamp: string;\n}\n\nexport interface VerificationResult {\n verified: boolean;\n source_id: string;\n name?: string;\n message: string;\n}\n\nexport interface SourceRecord {\n source_id: string;\n name: string;\n description: string;\n status: string;\n created_at: string;\n}\n\nexport interface TaskRecord {\n id: string;\n source_id: string;\n task: string;\n agent: string;\n category: string;\n status: string;\n complexity: string;\n reasoning: string;\n data: string;\n result: string;\n created_at: string;\n updated_at: string;\n}\n\nexport interface LogEntry {\n id: string;\n level: string;\n component: string;\n message: string;\n created_at: string;\n}\n",[],[215,296],"1a55b9b5-7b72-4749-b995-63923e7d4c3f",["D",1783359276137],[],[215,38],"21acd9ae-dd3a-4c31-8483-2ee3c91bfd8f","llm.ts","81b72098-7960-11f1-8bac-1607ee4eb77e",40,["D",1783359082410],"// Free LLM Engine — Nexuses Core V3\n// Uses Pollinations.ai EXCLUSIVELY — zero API keys, zero tokens, zero cost.\n// Provides: text generation, Chain-of-Thought reasoning, task classification.\n\nimport type { CoTAnalysis, TaskCategory, AgentName } from \"./types.ts\";\n\nconst POLLINATIONS_POST = \"https://text.pollinations.ai/openai\";\nconst POLLINATIONS_GET = \"https://text.pollinations.ai\";\nconst POLLINATIONS_IMAGE = \"https://image.pollinations.ai/prompt\";\n\nexport interface LLMResult {\n success: boolean;\n text: string;\n model: string;\n error?: string;\n}\n\nexport interface LLMOptions {\n maxTokens?: number;\n temperature?: number;\n timeoutMs?: number;\n systemPrompt?: string;\n}\n\n/**\n * Core LLM call — Pollinations.ai only (free, no key).\n * Tries POST (OpenAI-compatible) first, falls back to GET.\n */\nexport async function callLLM(\n prompt: string,\n options?: LLMOptions,\n): Promise {\n const messages: { role: string; content: string }[] = [];\n if (options?.systemPrompt) {\n messages.push({ role: \"system\", content: options.systemPrompt });\n }\n messages.push({ role: \"user\", content: prompt });\n\n // Attempt 1: POST (OpenAI-compatible endpoint)\n try {\n const res = await fetch(POLLINATIONS_POST, {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({\n model: \"openai\",\n messages,\n max_tokens: options?.maxTokens ?? 1000,\n temperature: options?.temperature ?? 0.7,\n }),\n signal: AbortSignal.timeout(options?.timeoutMs ?? 30000),\n });\n if (res.ok) {\n const json = await res.json();\n const text = json.choices?.[0]?.message?.content?.trim();\n if (text) return { success: true, text, model: \"pollinations-openai\" };\n }\n } catch { /* fall through to GET */ }\n\n // Attempt 2: GET (simple text endpoint)\n try {\n const fullPrompt = options?.systemPrompt\n ? `${options.systemPrompt}\\n\\n${prompt}`\n : prompt;\n const encoded = encodeURIComponent(fullPrompt.slice(0, 2000));\n const res = await fetch(`${POLLINATIONS_GET}/${encoded}`, {\n signal: AbortSignal.timeout(options?.timeoutMs ?? 30000),\n });\n if (res.ok) {\n const text = (await res.text()).trim();\n if (text) return { success: true, text, model: \"pollinations-get\" };\n }\n } catch { /* fall through to failure */ }\n\n return {\n success: false,\n text: \"\",\n model: \"none\",\n error: \"Pollinations.ai endpoints unreachable\",\n };\n}\n\n/**\n * Chain-of-Thought reasoning + autonomous task classification.\n * Sends the task to the LLM with a structured CoT prompt and expects\n * a JSON response containing reasoning, category, agent, steps, and complexity.\n */\nexport async function reasonWithCoT(\n task: string,\n data?: any,\n): Promise<{ success: boolean; analysis: CoTAnalysis; raw: string; error?: string }> {\n const systemPrompt = `You are the Nexuses Core V3 Dispatcher — an autonomous AI reasoning brain.\nAnalyze the incoming task using chain-of-thought reasoning, then classify and route it.\nRespond with ONLY a JSON object — no markdown fences, no text outside the JSON:\n{\n \"reasoning\": \"Step-by-step chain-of-thought analysis of what this task requires\",\n \"category\": \"coding | scraping | analysis | media | marketing | general\",\n \"agent\": \"CoderAgent | ScraperAgent | AnalystAgent | MediaAgent | MarketingAgent | GeneralAgent\",\n \"steps\": [\"Step 1: description\", \"Step 2: description\"],\n \"complexity\": \"simple | complex\"\n}\n\nClassification guide:\n- coding → CoderAgent: writing, debugging, explaining, refactoring code in Python/JS/HTML/TS\n- scraping → ScraperAgent: web scraping, data extraction, website cloning, URL fetching\n- analysis → AnalystAgent: data analysis, Excel/CSV, reports, document processing, file analysis\n- media → MediaAgent: image generation, video plans, visual design, logos, art\n- marketing → MarketingAgent: content writing, blogging, social media, SEO, cross-platform formatting\n- general → GeneralAgent: questions, chat, explanations, reasoning, anything uncategorized\n\nIf a task spans multiple categories, pick the primary one and list multiple steps.`;\n\n const userPrompt = `Task: \"${task}\"\\nAdditional data: ${\n JSON.stringify(data || {}).slice(0, 800)\n }`;\n\n const result = await callLLM(userPrompt, {\n maxTokens: 800,\n temperature: 0.3,\n systemPrompt,\n timeoutMs: 25000,\n });\n\n if (!result.success) {\n // Fallback to keyword-based classification\n const fallback = keywordClassify(task);\n return {\n success: false,\n analysis: fallback,\n raw: \"\",\n error: result.error,\n };\n }\n\n // Parse JSON from LLM response\n try {\n let jsonStr = result.text.trim();\n // Strip markdown code fences if present\n jsonStr = jsonStr.replace(/```json\\s*/gi, \"\").replace(/```\\s*/g, \"\");\n const match = jsonStr.match(/\\{[\\s\\S]*\\}/);\n if (match) jsonStr = match[0];\n const parsed = JSON.parse(jsonStr);\n\n const validCategories: TaskCategory[] = [\n \"coding\", \"scraping\", \"analysis\", \"media\", \"marketing\", \"general\",\n ];\n const validAgents: AgentName[] = [\n \"CoderAgent\", \"ScraperAgent\", \"AnalystAgent\",\n \"MediaAgent\", \"MarketingAgent\", \"GeneralAgent\",\n ];\n\n const category = validCategories.includes(parsed.category)\n ? parsed.category\n : \"general\";\n const agent = validAgents.includes(parsed.agent)\n ? parsed.agent\n : \"GeneralAgent\";\n\n return {\n success: true,\n analysis: {\n reasoning: parsed.reasoning || \"No reasoning provided\",\n category,\n agent,\n steps: Array.isArray(parsed.steps) && parsed.steps.length > 0\n ? parsed.steps\n : [task],\n complexity: parsed.complexity === \"complex\" ? \"complex\" : \"simple\",\n },\n raw: result.text,\n };\n } catch {\n // JSON parse failed — use keyword fallback\n const fallback = keywordClassify(task);\n return {\n success: true,\n analysis: fallback,\n raw: result.text,\n };\n }\n}\n\n/**\n * Keyword-based fallback classifier (used when LLM is unavailable\n * or returns unparseable JSON).\n */\nfunction keywordClassify(task: string): CoTAnalysis {\n const lower = task.toLowerCase();\n const rules: { keywords: string[]; category: TaskCategory; agent: AgentName }[] = [\n {\n keywords: [\"code\", \"debug\", \"bug\", \"fix\", \"program\", \"function\", \"script\", \"python\", \"javascript\", \"html\", \"css\", \"typescript\", \"compile\", \"syntax\", \"refactor\"],\n category: \"coding\",\n agent: \"CoderAgent\",\n },\n {\n keywords: [\"scrape\", \"clone\", \"extract\", \"website\", \"url\", \"crawl\", \"fetch page\", \"web data\"],\n category: \"scraping\",\n agent: \"ScraperAgent\",\n },\n {\n keywords: [\"analyze\", \"analysis\", \"data\", \"excel\", \"csv\", \"report\", \"spreadsheet\", \"pdf\", \"document\", \"file\"],\n category: \"analysis\",\n agent: \"AnalystAgent\",\n },\n {\n keywords: [\"image\", \"generate image\", \"picture\", \"draw\", \"art\", \"logo\", \"video\", \"visual\", \"design\"],\n category: \"media\",\n agent: \"MediaAgent\",\n },\n {\n keywords: [\"write\", \"blog\", \"content\", \"social\", \"post\", \"seo\", \"marketing\", \"article\", \"tweet\", \"copy\"],\n category: \"marketing\",\n agent: \"MarketingAgent\",\n },\n ];\n\n for (const rule of rules) {\n if (rule.keywords.some((k) => lower.includes(k))) {\n return {\n reasoning: `Keyword fallback: matched \"${rule.category}\" category`,\n category: rule.category,\n agent: rule.agent,\n steps: [task],\n complexity: \"simple\",\n };\n }\n }\n\n return {\n reasoning: \"No keywords matched — defaulting to general\",\n category: \"general\",\n agent: \"GeneralAgent\",\n steps: [task],\n complexity: \"simple\",\n };\n}\n\n/**\n * Generate an image URL via Pollinations.ai image endpoint (free, no key).\n * Returns a direct URL to the generated image.\n */\nexport function generateImageURL(\n prompt: string,\n options?: { width?: number; height?: number; seed?: number; model?: string },\n): string {\n const params = new URLSearchParams();\n params.set(\"width\", String(options?.width ?? 1024));\n params.set(\"height\", String(options?.height ?? 1024));\n params.set(\"nologo\", \"true\");\n if (options?.seed) params.set(\"seed\", String(options.seed));\n if (options?.model) params.set(\"model\", options.model);\n const encoded = encodeURIComponent(prompt.slice(0, 500));\n return `${POLLINATIONS_IMAGE}/${encoded}?${params.toString()}`;\n}\n",[],[215,308],"5070556f-bf51-47f7-8e56-2a18db984249","87d19a9e-7960-11f1-89b4-1607ee4eb77e",41,["D",1783359092751],"// SQLite Persistence Layer — Nexuses Core V3\n// Tables: tasks, steps, sources, memory, logs\nimport { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport type { TaskRecord, SourceRecord, LogEntry } from \"./types.ts\";\n\nconst SCHEMA_SQL = [\n `CREATE TABLE IF NOT EXISTS nex_tasks (\n id TEXT PRIMARY KEY,\n source_id TEXT,\n task TEXT NOT NULL,\n agent TEXT,\n category TEXT,\n status TEXT NOT NULL,\n complexity TEXT,\n reasoning TEXT,\n data TEXT,\n result TEXT,\n created_at TEXT NOT NULL,\n updated_at TEXT NOT NULL\n )`,\n `CREATE TABLE IF NOT EXISTS nex_steps (\n id TEXT PRIMARY KEY,\n task_id TEXT NOT NULL,\n step INTEGER NOT NULL,\n description TEXT,\n agent TEXT,\n status TEXT,\n result TEXT,\n attempts INTEGER DEFAULT 1,\n created_at TEXT NOT NULL\n )`,\n `CREATE TABLE IF NOT EXISTS nex_sources (\n source_id TEXT PRIMARY KEY,\n name TEXT NOT NULL,\n description TEXT,\n status TEXT DEFAULT 'active',\n created_at TEXT NOT NULL\n )`,\n `CREATE TABLE IF NOT EXISTS nex_memory (\n key TEXT PRIMARY KEY,\n value TEXT NOT NULL,\n created_at TEXT NOT NULL\n )`,\n `CREATE TABLE IF NOT EXISTS nex_logs (\n id TEXT PRIMARY KEY,\n level TEXT,\n component TEXT,\n message TEXT,\n created_at TEXT NOT NULL\n )`,\n];\n\nlet initialized = false;\n\nexport async function initDB() {\n if (initialized) return;\n await sqlite.batch(SCHEMA_SQL.map((sql) => ({ sql })));\n initialized = true;\n\n // Auto-register the internal trusted source\n await registerSource(\n \"nexuses-internal\",\n \"Nexuses Internal\",\n \"Built-in system source for self-testing\",\n );\n}\n\n// ── Task CRUD ──\n\nexport async function insertTask(\n id: string,\n sourceId: string,\n task: string,\n agent: string,\n category: string,\n status: string,\n complexity: string,\n reasoning: string,\n data: any,\n result: any,\n) {\n const now = new Date().toISOString();\n await sqlite.execute({\n sql: `INSERT INTO nex_tasks\n (id, source_id, task, agent, category, status, complexity, reasoning, data, result, created_at, updated_at)\n VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`,\n args: [\n id, sourceId, task, agent, category, status, complexity,\n reasoning, JSON.stringify(data), JSON.stringify(result), now, now,\n ],\n });\n}\n\nexport async function updateTaskStatus(id: string, status: string, result: any) {\n const now = new Date().toISOString();\n await sqlite.execute({\n sql: `UPDATE nex_tasks SET status = ?, result = ?, updated_at = ? WHERE id = ?`,\n args: [status, JSON.stringify(result), now, id],\n });\n}\n\nexport async function getTask(id: string): Promise {\n const res = await sqlite.execute({\n sql: `SELECT * FROM nex_tasks WHERE id = ?`,\n args: [id],\n });\n return (res.rows[0] as TaskRecord) || null;\n}\n\nexport async function getRecentTasks(limit = 20): Promise {\n const res = await sqlite.execute({\n sql: `SELECT * FROM nex_tasks ORDER BY created_at DESC LIMIT ?`,\n args: [limit],\n });\n return res.rows as TaskRecord[];\n}\n\nexport async function getTaskCount(): Promise {\n const res = await sqlite.execute(`SELECT COUNT(*) as count FROM nex_tasks`);\n return (res.rows[0] as any)?.count || 0;\n}\n\n// ── Step Logging ──\n\nexport async function logStep(\n taskId: string,\n step: number,\n description: string,\n agent: string,\n status: string,\n result: any,\n attempts: number,\n) {\n const id = `step_${taskId}_${step}`;\n const now = new Date().toISOString();\n await sqlite.execute({\n sql: `INSERT OR REPLACE INTO nex_steps\n (id, task_id, step, description, agent, status, result, attempts, created_at)\n VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)`,\n args: [\n id, taskId, step, description, agent, status,\n JSON.stringify(result), attempts, now,\n ],\n });\n}\n\nexport async function getStepHistory(taskId: string) {\n const res = await sqlite.execute({\n sql: `SELECT * FROM nex_steps WHERE task_id = ? ORDER BY step ASC`,\n args: [taskId],\n });\n return res.rows;\n}\n\n// ── Source Registration ──\n\nexport async function registerSource(\n sourceId: string,\n name: string,\n description: string,\n): Promise {\n const now = new Date().toISOString();\n try {\n await sqlite.execute({\n sql: `INSERT OR REPLACE INTO nex_sources\n (source_id, name, description, status, created_at)\n VALUES (?, ?, ?, 'active', ?)`,\n args: [sourceId, name, description, now],\n });\n return true;\n } catch {\n return false;\n }\n}\n\nexport async function verifySource(sourceId: string): Promise {\n const res = await sqlite.execute({\n sql: `SELECT * FROM nex_sources WHERE source_id = ? AND status = 'active'`,\n args: [sourceId],\n });\n return (res.rows[0] as SourceRecord) || null;\n}\n\nexport async function listSources(): Promise {\n const res = await sqlite.execute(`SELECT * FROM nex_sources ORDER BY created_at DESC`);\n return res.rows as SourceRecord[];\n}\n\nexport async function removeSource(sourceId: string): Promise {\n try {\n await sqlite.execute({\n sql: `DELETE FROM nex_sources WHERE source_id = ?`,\n args: [sourceId],\n });\n return true;\n } catch {\n return false;\n }\n}\n\n// ── Memory ──\n\nexport async function remember(key: string, value: any) {\n const now = new Date().toISOString();\n await sqlite.execute({\n sql: `INSERT OR REPLACE INTO nex_memory (key, value, created_at) VALUES (?, ?, ?)`,\n args: [key, JSON.stringify(value), now],\n });\n}\n\nexport async function recall(key: string): Promise {\n const res = await sqlite.execute({\n sql: `SELECT * FROM nex_memory WHERE key = ?`,\n args: [key],\n });\n const row = res.rows[0] as any;\n return row ? JSON.parse(row.value) : null;\n}\n\n// ── System Logging ──\n\nexport async function logEvent(\n level: \"info\" | \"warn\" | \"error\",\n component: string,\n message: string,\n) {\n const id = `log_${Date.now()}_${Math.random().toString(36).slice(2, 8)}`;\n const now = new Date().toISOString();\n await sqlite.execute({\n sql: `INSERT INTO nex_logs (id, level, component, message, created_at) VALUES (?, ?, ?, ?, ?)`,\n args: [id, level, component, message, now],\n });\n}\n\nexport async function getRecentLogs(limit = 50): Promise {\n const res = await sqlite.execute({\n sql: `SELECT * FROM nex_logs ORDER BY created_at DESC LIMIT ?`,\n args: [limit],\n });\n return res.rows as LogEntry[];\n}\n\n// ── Stats ──\n\nexport async function getStats() {\n const taskCount = await getTaskCount();\n const sources = await listSources();\n const successRes = await sqlite.execute(\n `SELECT COUNT(*) as count FROM nex_tasks WHERE status = 'completed'`,\n );\n const failRes = await sqlite.execute(\n `SELECT COUNT(*) as count FROM nex_tasks WHERE status = 'failed'`,\n );\n const logRes = await sqlite.execute(\n `SELECT COUNT(*) as count FROM nex_logs`,\n );\n return {\n totalTasks: taskCount,\n completedTasks: (successRes.rows[0] as any)?.count || 0,\n failedTasks: (failRes.rows[0] as any)?.count || 0,\n registeredSources: sources.length,\n totalLogs: (logRes.rows[0] as any)?.count || 0,\n };\n}\n",[],[215,165],["D",1783359042499],[245,246,247,248,249],[215],"df0030fc-29ab-46c6-87f6-471d4d6058a5","scraperAgent.ts","a79a834a-7960-11f1-84c2-1607ee4eb77e",43,["D",1783359146056],"// ScraperAgent — Nexuses Core V3\n// Browser-less web scraping: fetch + HTML parsing, DuckDuckGo search.\n// No Playwright, no browser dependencies — pure fetch-based extraction.\n\nimport { callLLM } from \"../lib/llm.ts\";\nimport type { AgentResult } from \"../lib/types.ts\";\n\n/**\n * Extract text content from HTML using regex (no external parser needed).\n * Lightweight Cheerio-free extraction.\n */\nfunction extractFromHTML(html: string) {\n // Title\n const titleMatch = html.match(/]*>(.*?)<\\/title>/is);\n const title = titleMatch ? titleMatch[1].trim() : \"Untitled\";\n\n // Meta description\n const metaMatch = html.match(/]+name=[\"']description[\"'][^>]+content=[\"']([^\"']*)/is);\n const description = metaMatch ? metaMatch[1].trim() : \"\";\n\n // Remove scripts and styles\n const clean = html\n .replace(//gi, \"\")\n .replace(//gi, \"\")\n .replace(//gi, \"\")\n .replace(//gi, \"\");\n\n // Extract text (strip all tags)\n const textContent = clean\n .replace(/<[^>]+>/g, \" \")\n .replace(/ /g, \" \")\n .replace(/&/g, \"&\")\n .replace(/</g, \"<\")\n .replace(/>/g, \">\")\n .replace(/"/g, '\"')\n .replace(/'/g, \"'\")\n .replace(/\\s+/g, \" \")\n .trim()\n .slice(0, 5000);\n\n // Extract links\n const linkRegex = /]+href=[\"']([^\"'#]+)[\"'][^>]*>(.*?)<\\/a>/gis;\n const links: { url: string; text: string }[] = [];\n let match;\n let count = 0;\n while ((match = linkRegex.exec(html)) !== null && count < 30) {\n const url = match[1].trim();\n const linkText = match[2].replace(/<[^>]+>/g, \"\").trim();\n if (url.startsWith(\"http\") && linkText.length > 0) {\n links.push({ url, text: linkText });\n count++;\n }\n }\n\n // Extract images\n const imgRegex = /]+src=[\"']([^\"']+)[\"']/gi;\n const images: string[] = [];\n let imgMatch;\n let imgCount = 0;\n while ((imgMatch = imgRegex.exec(html)) !== null && imgCount < 10) {\n const src = imgMatch[1].trim();\n if (src.startsWith(\"http\")) {\n images.push(src);\n imgCount++;\n }\n }\n\n // Extract headings\n const headings: string[] = [];\n const hRegex = /]*>(.*?)<\\/h[1-3]>/gis;\n let hMatch;\n let hCount = 0;\n while ((hMatch = hRegex.exec(clean)) !== null && hCount < 15) {\n const hText = hMatch[1].replace(/<[^>]+>/g, \"\").trim();\n if (hText.length > 2) {\n headings.push(hText);\n hCount++;\n }\n }\n\n return { title, description, textContent, links, images, headings };\n}\n\nasync function scrapeURL(url: string) {\n const res = await fetch(url, {\n headers: {\n \"User-Agent\": \"Mozilla/5.0 (compatible; NexusesCore/3.0; +https://val.town)\",\n \"Accept\": \"text/html,application/xhtml+xml\",\n },\n signal: AbortSignal.timeout(15000),\n });\n\n if (!res.ok) {\n throw new Error(`HTTP ${res.status} from ${url}`);\n }\n\n const html = await res.text();\n return extractFromHTML(html);\n}\n\nasync function searchWeb(query: string) {\n // DuckDuckGo Instant Answer API (free, no key)\n const ddgUrl = `https://api.duckduckgo.com/?q=${\n encodeURIComponent(query)\n }&format=json&no_html=1&skip_disambig=1`;\n\n const res = await fetch(ddgUrl, {\n signal: AbortSignal.timeout(10000),\n });\n\n if (!res.ok) {\n throw new Error(`DuckDuckGo HTTP ${res.status}`);\n }\n\n const data = await res.json();\n const results: any = {\n query,\n abstract: data.Abstract || \"\",\n abstractSource: data.AbstractSource || \"\",\n abstractURL: data.AbstractURL || \"\",\n answer: data.Answer || \"\",\n definition: data.Definition || \"\",\n relatedTopics: [] as any[],\n };\n\n // Extract related topics\n if (data.RelatedTopics && Array.isArray(data.RelatedTopics)) {\n for (const topic of data.RelatedTopics.slice(0, 10)) {\n if (topic.Text) {\n results.relatedTopics.push({\n text: topic.Text,\n url: topic.FirstURL || \"\",\n });\n } else if (topic.Topics && Array.isArray(topic.Topics)) {\n for (const sub of topic.Topics.slice(0, 3)) {\n if (sub.Text) {\n results.relatedTopics.push({\n text: sub.Text,\n url: sub.FirstURL || \"\",\n });\n }\n }\n }\n }\n }\n\n return results;\n}\n\nexport async function execute(task: string, data: any): Promise {\n const start = Date.now();\n const action = data.action || \"scrape\";\n const url = data.url || \"\";\n const query = data.query || data.q || task;\n\n try {\n // Action: scrape a specific URL\n if (action === \"scrape\" || action === \"clone\" || url) {\n if (!url) {\n return {\n agent: \"ScraperAgent\",\n category: \"scraping\",\n status: \"error\",\n message: \"No URL provided for scraping\",\n data: {},\n executionTimeMs: Date.now() - start,\n };\n }\n\n const scraped = await scrapeURL(url);\n\n if (action === \"clone\") {\n // Use LLM to generate a clean HTML clone based on scraped content\n const llmResult = await callLLM(\n `Create a clean, modern HTML page based on this scraped content:\\n` +\n `Title: ${scraped.title}\\n` +\n `Headings: ${scraped.headings.join(\", \")}\\n` +\n `Content: ${scraped.textContent.slice(0, 1500)}\\n\\n` +\n `Return only the HTML code, no explanation.`,\n { maxTokens: 1500, temperature: 0.5 },\n );\n\n return {\n agent: \"ScraperAgent\",\n category: \"scraping\",\n status: llmResult.success ? \"success\" : \"partial\",\n message: `Website cloned: ${scraped.title}`,\n data: {\n original: { url, title: scraped.title, description: scraped.description },\n clone: llmResult.text || \"Clone generation failed\",\n links: scraped.links,\n images: scraped.images,\n },\n executionTimeMs: Date.now() - start,\n };\n }\n\n return {\n agent: \"ScraperAgent\",\n category: \"scraping\",\n status: \"success\",\n message: `Scraped: ${scraped.title}`,\n data: scraped,\n executionTimeMs: Date.now() - start,\n };\n }\n\n // Action: web search\n if (action === \"search\" || action === \"research\") {\n const searchResults = await searchWeb(query);\n\n // Use LLM to summarize search results\n const llmResult = await callLLM(\n `Summarize these search results for query \"${query}\":\\n` +\n `Abstract: ${searchResults.abstract}\\n` +\n `Answer: ${searchResults.answer}\\n` +\n `Related: ${searchResults.relatedTopics.map((t: any) => t.text).join(\"; \").slice(0, 1000)}\\n\\n` +\n `Provide a concise summary with key findings.`,\n { maxTokens: 500, temperature: 0.4 },\n );\n\n return {\n agent: \"ScraperAgent\",\n category: \"scraping\",\n status: \"success\",\n message: `Search completed for: \"${query}\"`,\n data: {\n search: searchResults,\n summary: llmResult.text || \"Summary unavailable\",\n },\n executionTimeMs: Date.now() - start,\n };\n }\n\n // Default: try to extract URL from task text\n const urlMatch = task.match(/https?:\\/\\/[^\\s]+/);\n if (urlMatch) {\n const scraped = await scrapeURL(urlMatch[0]);\n return {\n agent: \"ScraperAgent\",\n category: \"scraping\",\n status: \"success\",\n message: `Scraped: ${scraped.title}`,\n data: scraped,\n executionTimeMs: Date.now() - start,\n };\n }\n\n // Fallback: treat as search query\n const searchResults = await searchWeb(query);\n return {\n agent: \"ScraperAgent\",\n category: \"scraping\",\n status: \"success\",\n message: `Search completed for: \"${query}\"`,\n data: searchResults,\n executionTimeMs: Date.now() - start,\n };\n } catch (e: any) {\n return {\n agent: \"ScraperAgent\",\n category: \"scraping\",\n status: \"error\",\n message: `Scraping failed: ${e.message}`,\n data: { error: e.message },\n executionTimeMs: Date.now() - start,\n };\n }\n}\n",[],[222,326],"52018289-2c9b-43fc-b0a7-c5919564782e","mediaAgent.ts","c45dd16c-7960-11f1-a65f-1607ee4eb77e",46,["D",1783359194339],"// MediaAgent — Nexuses Core V3\n// Image generation via Pollinations.ai (free, no API key).\n// Video production plans via LLM.\n\nimport { callLLM, generateImageURL } from \"../lib/llm.ts\";\nimport type { AgentResult } from \"../lib/types.ts\";\n\nexport async function execute(task: string, data: any): Promise {\n const start = Date.now();\n const action = data.action || \"image\";\n const prompt = data.prompt || task;\n const width = data.width || 1024;\n const height = data.height || 1024;\n const seed = data.seed || Math.floor(Math.random() * 1000000);\n\n try {\n if (action === \"image\" || action === \"generate\") {\n // Enhance prompt with LLM for better image generation\n const enhanceResult = await callLLM(\n `Enhance this image generation prompt with visual details (style, lighting, composition, mood).\\n` +\n `Original: \"${prompt}\"\\nReturn ONLY the enhanced prompt, nothing else.`,\n { maxTokens: 200, temperature: 0.7 },\n );\n\n const finalPrompt = enhanceResult.success ? enhanceResult.text : prompt;\n const imageURL = generateImageURL(finalPrompt, { width, height, seed });\n\n return {\n agent: \"MediaAgent\",\n category: \"media\",\n status: \"success\",\n message: `Image generated for: ${prompt.slice(0, 80)}`,\n data: {\n action: \"image\",\n prompt: finalPrompt,\n originalPrompt: prompt,\n imageURL,\n width,\n height,\n seed,\n note: \"URL returns the generated image directly — use it in tags or download\",\n },\n executionTimeMs: Date.now() - start,\n };\n }\n\n if (action === \"video\" || action === \"video-plan\") {\n // Generate a video production plan (text-based, no actual video rendering)\n const llmResult = await callLLM(\n `Create a detailed video production plan for: ${prompt}\\n` +\n `Include:\\n` +\n `- Concept summary\\n` +\n `- Scene-by-scene breakdown (at least 5 scenes)\\n` +\n `- Visual style and color palette\\n` +\n `- Audio/music suggestions\\n` +\n `- Estimated duration per scene\\n` +\n `- Technical notes (aspect ratio, resolution, transitions)`,\n { maxTokens: 800, temperature: 0.6 },\n );\n\n return {\n agent: \"MediaAgent\",\n category: \"media\",\n status: llmResult.success ? \"success\" : \"error\",\n message: llmResult.success ? \"Video production plan generated\" : \"Video plan failed — LLM unavailable\",\n data: {\n action: \"video-plan\",\n prompt,\n plan: llmResult.text || \"\",\n },\n executionTimeMs: Date.now() - start,\n };\n }\n\n if (action === \"logo\") {\n // Logo generation via Pollinations\n const logoPrompt = `minimalist professional logo design, ${prompt}, vector style, clean, modern, centered, simple colors`;\n const logoURL = generateImageURL(logoPrompt, {\n width: 512,\n height: 512,\n seed,\n });\n\n return {\n agent: \"MediaAgent\",\n category: \"media\",\n status: \"success\",\n message: `Logo generated for: ${prompt.slice(0, 60)}`,\n data: {\n action: \"logo\",\n prompt: logoPrompt,\n logoURL,\n size: \"512x512\",\n seed,\n },\n executionTimeMs: Date.now() - start,\n };\n }\n\n // Default: treat as image generation\n const imageURL = generateImageURL(prompt, { width, height, seed });\n return {\n agent: \"MediaAgent\",\n category: \"media\",\n status: \"success\",\n message: `Image generated for: ${prompt.slice(0, 80)}`,\n data: {\n action: \"image\",\n prompt,\n imageURL,\n width,\n height,\n seed,\n },\n executionTimeMs: Date.now() - start,\n };\n } catch (e: any) {\n return {\n agent: \"MediaAgent\",\n category: \"media\",\n status: \"error\",\n message: `Media error: ${e.message}`,\n data: { error: e.message },\n executionTimeMs: Date.now() - start,\n };\n }\n}\n",[],[222,334],"105f8c56-3b2c-4cdf-9e07-cf79b1ccfc01","marketingAgent.ts","cd8fc754-7960-11f1-84c2-1607ee4eb77e",47,["D",1783359209763],"// MarketingAgent — Nexuses Core V3\n// Content creation, blogging, social media, SEO, cross-platform formatting.\n\nimport { callLLM } from \"../lib/llm.ts\";\nimport type { AgentResult } from \"../lib/types.ts\";\n\n/**\n * Platform-specific formatting rules.\n */\nconst PLATFORM_RULES: Record = {\n twitter: { maxLength: 280, hashtags: true, tone: \"punchy, conversational\" },\n linkedin: { maxLength: 3000, hashtags: true, tone: \"professional, insightful\" },\n facebook: { maxLength: 5000, hashtags: true, tone: \"friendly, engaging\" },\n instagram: { maxLength: 2200, hashtags: true, tone: \"visual, aspirational\" },\n blog: { maxLength: 10000, hashtags: false, tone: \"informative, structured\" },\n email: { maxLength: 5000, hashtags: false, tone: \"personal, direct\" },\n};\n\nexport async function execute(task: string, data: any): Promise {\n const start = Date.now();\n const action = data.action || \"write\";\n const topic = data.topic || data.prompt || task;\n const platform = (data.platform || \"\").toLowerCase();\n const audience = data.audience || \"general audience\";\n const tone = data.tone || \"professional\";\n\n try {\n if (action === \"blog\") {\n const llmResult = await callLLM(\n `Write a comprehensive blog post about: ${topic}\\n` +\n `Target audience: ${audience}\\nTone: ${tone}\\n\\n` +\n `Format:\\n# [Title]\\n## Introduction\\n[2-3 paragraphs]\\n` +\n `## [Section 1]\\n## [Section 2]\\n## [Section 3]\\n` +\n `## Conclusion\\n\\nInclude engaging headings and actionable insights.`,\n { maxTokens: 1500, temperature: 0.7 },\n );\n\n return {\n agent: \"MarketingAgent\",\n category: \"marketing\",\n status: llmResult.success ? \"success\" : \"error\",\n message: llmResult.success ? \"Blog post generated\" : \"Blog generation failed — LLM unavailable\",\n data: {\n action: \"blog\",\n topic,\n content: llmResult.text || \"\",\n wordCount: llmResult.text?.split(/\\s+/).length || 0,\n },\n executionTimeMs: Date.now() - start,\n };\n }\n\n if (action === \"social\" || platform) {\n const rules = PLATFORM_RULES[platform] || PLATFORM_RULES[\"twitter\"];\n\n const llmResult = await callLLM(\n `Write a ${platform || \"social media\"} post about: ${topic}\\n` +\n `Tone: ${rules.tone}\\n` +\n `Maximum ${rules.maxLength} characters.\\n` +\n `${rules.hashtags ? \"Include 3-5 relevant hashtags.\" : \"No hashtags.\"}\\n` +\n `Return ONLY the post text.`,\n { maxTokens: 400, temperature: 0.8 },\n );\n\n const postText = llmResult.text || \"\";\n const charCount = postText.length;\n\n return {\n agent: \"MarketingAgent\",\n category: \"marketing\",\n status: llmResult.success ? \"success\" : \"error\",\n message: llmResult.success\n ? `${platform || \"Social\"} post generated (${charCount} chars)`\n : \"Social post generation failed\",\n data: {\n action: \"social\",\n platform: platform || \"general\",\n content: postText,\n characterCount: charCount,\n maxLength: rules.maxLength,\n withinLimit: charCount <= rules.maxLength,\n },\n executionTimeMs: Date.now() - start,\n };\n }\n\n if (action === \"seo\") {\n const llmResult = await callLLM(\n `Provide SEO recommendations for: ${topic}\\n` +\n `Include:\\n` +\n `- Primary keyword and 10 long-tail keywords\\n` +\n `- Meta title (60 chars max)\\n` +\n `- Meta description (160 chars max)\\n` +\n `- URL slug suggestion\\n` +\n `- 5 header (H2) suggestions\\n` +\n `- Content gap analysis`,\n { maxTokens: 600, temperature: 0.4 },\n );\n\n return {\n agent: \"MarketingAgent\",\n category: \"marketing\",\n status: llmResult.success ? \"success\" : \"error\",\n message: llmResult.success ? \"SEO analysis generated\" : \"SEO analysis failed\",\n data: {\n action: \"seo\",\n topic,\n recommendations: llmResult.text || \"\",\n },\n executionTimeMs: Date.now() - start,\n };\n }\n\n if (action === \"reformat\" || action === \"cross-platform\") {\n const content = data.content || \"\";\n const targetPlatform = (data.targetPlatform || \"twitter\").toLowerCase();\n const rules = PLATFORM_RULES[targetPlatform] || PLATFORM_RULES[\"twitter\"];\n\n const llmResult = await callLLM(\n `Reformat this content for ${targetPlatform}:\\n\\n${content.slice(0, 2000)}\\n\\n` +\n `Tone: ${rules.tone}\\nMax ${rules.maxLength} characters.\\n` +\n `${rules.hashtags ? \"Add relevant hashtags.\" : \"No hashtags.\"}\\n` +\n `Return ONLY the reformatted content.`,\n { maxTokens: 500, temperature: 0.6 },\n );\n\n return {\n agent: \"MarketingAgent\",\n category: \"marketing\",\n status: llmResult.success ? \"success\" : \"error\",\n message: llmResult.success\n ? `Content reformatted for ${targetPlatform}`\n : \"Reformatting failed\",\n data: {\n action: \"reformat\",\n targetPlatform,\n original: content.slice(0, 500),\n reformatted: llmResult.text || \"\",\n characterCount: llmResult.text?.length || 0,\n },\n executionTimeMs: Date.now() - start,\n };\n }\n\n // Default: general content writing\n const llmResult = await callLLM(\n `Write engaging content about: ${topic}\\n` +\n `Audience: ${audience}\\nTone: ${tone}\\n` +\n `Make it compelling and well-structured.`,\n { maxTokens: 1000, temperature: 0.7 },\n );\n\n return {\n agent: \"MarketingAgent\",\n category: \"marketing\",\n status: llmResult.success ? \"success\" : \"error\",\n message: llmResult.success ? \"Content generated\" : \"Content generation failed — LLM unavailable\",\n data: {\n action: \"write\",\n topic,\n content: llmResult.text || \"\",\n },\n executionTimeMs: Date.now() - start,\n };\n } catch (e: any) {\n return {\n agent: \"MarketingAgent\",\n category: \"marketing\",\n status: \"error\",\n message: `Marketing error: ${e.message}`,\n data: { error: e.message },\n executionTimeMs: Date.now() - start,\n };\n }\n}\n",[],[222,342],"f0dc46c4-47ac-4070-af62-23eb3f40db1b","coderAgent.ts","b7accf72-7960-11f1-acba-1607ee4eb77e",44,["D",1783359173043],"// CoderAgent — Nexuses Core V3\n// Writes, debugs, and explains code in Python/JS/TS/HTML/CSS.\n// Uses a virtual file system via blob storage for persistent code files.\n\nimport { callLLM } from \"../lib/llm.ts\";\nimport { blob } from \"https://esm.town/v/std/blob/main.ts\";\nimport type { AgentResult } from \"../lib/types.ts\";\n\nconst VFS_PREFIX = \"vfs/\";\n\n/**\n * Virtual File System — store/retrieve code files in blob storage.\n */\nasync function vfsWrite(filename: string, content: string) {\n await blob.setJSON(`${VFS_PREFIX}${filename}`, {\n filename,\n content,\n updatedAt: new Date().toISOString(),\n });\n}\n\nasync function vfsRead(filename: string): Promise {\n const data = await blob.getJSON(`${VFS_PREFIX}${filename}`);\n return data?.content ?? null;\n}\n\nasync function vfsList() {\n const entries = await blob.list(VFS_PREFIX);\n return entries.map((e) => e.key.replace(VFS_PREFIX, \"\"));\n}\n\n/**\n * Detect language from task or data.\n */\nfunction detectLanguage(task: string, data: any): string {\n if (data.language) return data.language;\n const lower = task.toLowerCase();\n if (lower.includes(\"python\") || lower.includes(\".py\")) return \"python\";\n if (lower.includes(\"typescript\") || lower.includes(\".ts\")) return \"typescript\";\n if (lower.includes(\"html\")) return \"html\";\n if (lower.includes(\"css\")) return \"css\";\n if (lower.includes(\"javascript\") || lower.includes(\".js\") || lower.includes(\"js\")) return \"javascript\";\n return \"javascript\";\n}\n\nexport async function execute(task: string, data: any): Promise {\n const start = Date.now();\n const action = data.action || \"write\";\n const language = detectLanguage(task, data);\n const code = data.code || data.snippet || \"\";\n const filename = data.filename || `generated.${language === \"python\" ? \"py\" : language === \"typescript\" ? \"ts\" : language === \"html\" ? \"html\" : language === \"css\" ? \"css\" : \"js\"}`;\n\n try {\n if (action === \"debug\" || action === \"fix\") {\n // Debug/fix existing code\n const llmResult = await callLLM(\n `Find and fix all bugs in this ${language} code:\\n\\n${code}\\n\\n` +\n `Return ONLY the corrected code with inline comments explaining each fix.`,\n { maxTokens: 1200, temperature: 0.2 },\n );\n\n if (llmResult.success) {\n await vfsWrite(filename, llmResult.text);\n }\n\n return {\n agent: \"CoderAgent\",\n category: \"coding\",\n status: llmResult.success ? \"success\" : \"error\",\n message: llmResult.success ? `Code debugged and saved to VFS:/${filename}` : \"Debug failed — LLM unavailable\",\n data: {\n action: \"debug\",\n language,\n filename,\n original: code.slice(0, 500),\n fixed: llmResult.text || \"\",\n },\n executionTimeMs: Date.now() - start,\n };\n }\n\n if (action === \"explain\") {\n // Explain existing code\n const llmResult = await callLLM(\n `Explain this ${language} code in detail:\\n\\n${code}\\n\\n` +\n `Cover: purpose, key functions, logic flow, and any potential issues.`,\n { maxTokens: 800, temperature: 0.4 },\n );\n\n return {\n agent: \"CoderAgent\",\n category: \"coding\",\n status: llmResult.success ? \"success\" : \"error\",\n message: llmResult.success ? \"Code explained\" : \"Explanation failed — LLM unavailable\",\n data: {\n action: \"explain\",\n language,\n code: code.slice(0, 500),\n explanation: llmResult.text || \"\",\n },\n executionTimeMs: Date.now() - start,\n };\n }\n\n if (action === \"read\") {\n // Read from VFS\n const content = await vfsRead(filename);\n if (content) {\n return {\n agent: \"CoderAgent\",\n category: \"coding\",\n status: \"success\",\n message: `Read VFS:/${filename}`,\n data: { action: \"read\", filename, content },\n executionTimeMs: Date.now() - start,\n };\n }\n return {\n agent: \"CoderAgent\",\n category: \"coding\",\n status: \"error\",\n message: `File not found in VFS: ${filename}`,\n data: {},\n executionTimeMs: Date.now() - start,\n };\n }\n\n if (action === \"list\") {\n const files = await vfsList();\n return {\n agent: \"CoderAgent\",\n category: \"coding\",\n status: \"success\",\n message: `VFS contains ${files.length} files`,\n data: { action: \"list\", files },\n executionTimeMs: Date.now() - start,\n };\n }\n\n // Default action: write new code\n const prompt = data.prompt || task;\n const llmResult = await callLLM(\n `Write ${language} code for the following task:\\n${prompt}\\n\\n` +\n `Return ONLY the code with comments. No explanation outside the code.`,\n { maxTokens: 1200, temperature: 0.3 },\n );\n\n if (llmResult.success) {\n await vfsWrite(filename, llmResult.text);\n }\n\n return {\n agent: \"CoderAgent\",\n category: \"coding\",\n status: llmResult.success ? \"success\" : \"error\",\n message: llmResult.success\n ? `Code written in ${language} and saved to VFS:/${filename}`\n : \"Code generation failed — LLM unavailable\",\n data: {\n action: \"write\",\n language,\n filename,\n code: llmResult.text || \"\",\n vfsPath: `VFS:/${filename}`,\n },\n executionTimeMs: Date.now() - start,\n };\n } catch (e: any) {\n return {\n agent: \"CoderAgent\",\n category: \"coding\",\n status: \"error\",\n message: `Coder error: ${e.message}`,\n data: { error: e.message },\n executionTimeMs: Date.now() - start,\n };\n }\n}\n",[],[222,350],"78adf795-1387-438a-b18d-1707e9fb46ab","analystAgent.ts","c0bbf124-7960-11f1-84c2-1607ee4eb77e",45,["D",1783359188236],"// AnalystAgent — Nexuses Core V3\n// Data analysis, CSV/Excel generation, report building, file analysis.\n// Uses LLM for analytical reasoning + structured data output.\n\nimport { callLLM } from \"../lib/llm.ts\";\nimport type { AgentResult } from \"../lib/types.ts\";\n\n/**\n * Parse CSV text into structured rows.\n */\nfunction parseCSV(text: string): string[][] {\n const rows: string[][] = [];\n const lines = text.trim().split(\"\\n\");\n for (const line of lines) {\n const cells: string[] = [];\n let current = \"\";\n let inQuotes = false;\n for (let i = 0; i < line.length; i++) {\n const ch = line[i];\n if (ch === '\"') {\n inQuotes = !inQuotes;\n } else if (ch === \",\" && !inQuotes) {\n cells.push(current.trim());\n current = \"\";\n } else {\n current += ch;\n }\n }\n cells.push(current.trim());\n rows.push(cells);\n }\n return rows;\n}\n\n/**\n * Convert array of objects to CSV string.\n */\nfunction toCSV(data: Record[]): string {\n if (!data.length) return \"\";\n const headers = Object.keys(data[0]);\n const lines = [headers.join(\",\")];\n for (const row of data) {\n lines.push(headers.map((h) => `\"${String(row[h] ?? \"\").replace(/\"/g, '\"\"')}\"`).join(\",\"));\n }\n return lines.join(\"\\n\");\n}\n\nexport async function execute(task: string, data: any): Promise {\n const start = Date.now();\n const action = data.action || \"analyze\";\n const prompt = data.prompt || task;\n\n try {\n // CSV generation\n if (action === \"csv\" || action === \"excel\") {\n const llmResult = await callLLM(\n `Generate tabular data in CSV format for: ${prompt}\\n` +\n `Include appropriate headers and at least 5 sample rows.\\n` +\n `Return ONLY the CSV text, no explanation.`,\n { maxTokens: 600, temperature: 0.3 },\n );\n\n const csvText = llmResult.text || \"\";\n const parsed = parseCSV(csvText);\n\n return {\n agent: \"AnalystAgent\",\n category: \"analysis\",\n status: llmResult.success ? \"success\" : \"error\",\n message: llmResult.success ? `CSV generated with ${parsed.length} rows` : \"CSV generation failed\",\n data: {\n action: \"csv\",\n csv: csvText,\n rows: parsed,\n headers: parsed[0] || [],\n rowCount: parsed.length - 1,\n },\n executionTimeMs: Date.now() - start,\n };\n }\n\n // Report generation\n if (action === \"report\") {\n const llmResult = await callLLM(\n `Create a structured analytical report for: ${prompt}\\n` +\n `Format as:\\n## Executive Summary\\n## Key Findings\\n## Data Analysis\\n## Recommendations\\n\\n` +\n `Be thorough and data-driven.`,\n { maxTokens: 1000, temperature: 0.5 },\n );\n\n return {\n agent: \"AnalystAgent\",\n category: \"analysis\",\n status: llmResult.success ? \"success\" : \"error\",\n message: llmResult.success ? \"Report generated\" : \"Report generation failed\",\n data: {\n action: \"report\",\n topic: prompt,\n report: llmResult.text || \"\",\n },\n executionTimeMs: Date.now() - start,\n };\n }\n\n // File/data analysis (parse provided data)\n if (action === \"analyze\" && (data.data || data.text || data.csv)) {\n const input = data.data || data.text || data.csv;\n const inputStr = typeof input === \"string\" ? input : JSON.stringify(input);\n\n const llmResult = await callLLM(\n `Analyze this data and provide insights:\\n\\n${inputStr.slice(0, 2000)}\\n\\n` +\n `Provide: key patterns, anomalies, statistics, and recommendations.`,\n { maxTokens: 800, temperature: 0.4 },\n );\n\n return {\n agent: \"AnalystAgent\",\n category: \"analysis\",\n status: llmResult.success ? \"success\" : \"error\",\n message: llmResult.success ? \"Data analysis completed\" : \"Analysis failed — LLM unavailable\",\n data: {\n action: \"analyze\",\n input: inputStr.slice(0, 500),\n analysis: llmResult.text || \"\",\n },\n executionTimeMs: Date.now() - start,\n };\n }\n\n // Invoice/billing generation\n if (action === \"invoice\") {\n const llmResult = await callLLM(\n `Generate a professional invoice in JSON format for: ${prompt}\\n` +\n `Include: invoiceNumber, date, billTo, items[{description, quantity, unitPrice, total}], subtotal, tax, total.\\n` +\n `Return ONLY the JSON.`,\n { maxTokens: 500, temperature: 0.3 },\n );\n\n let invoice: any = null;\n try {\n const jsonStr = llmResult.text?.replace(/```json\\s*/g, \"\").replace(/```\\s*/g, \"\");\n const match = jsonStr?.match(/\\{[\\s\\S]*\\}/);\n if (match) invoice = JSON.parse(match[0]);\n } catch {}\n\n // Generate CSV version too\n let csvVersion = \"\";\n if (invoice?.items) {\n csvVersion = toCSV(invoice.items);\n }\n\n return {\n agent: \"AnalystAgent\",\n category: \"analysis\",\n status: llmResult.success ? \"success\" : \"error\",\n message: llmResult.success ? \"Invoice generated\" : \"Invoice generation failed\",\n data: {\n action: \"invoice\",\n invoice,\n csv: csvVersion,\n },\n executionTimeMs: Date.now() - start,\n };\n }\n\n // Default: general analysis\n const llmResult = await callLLM(\n `Analyze and process: ${prompt}\\nProvide structured findings and recommendations.`,\n { maxTokens: 700, temperature: 0.5 },\n );\n\n return {\n agent: \"AnalystAgent\",\n category: \"analysis\",\n status: llmResult.success ? \"success\" : \"error\",\n message: llmResult.success ? \"Analysis completed\" : \"Analysis failed — LLM unavailable\",\n data: {\n action: \"analyze\",\n topic: prompt,\n result: llmResult.text || \"\",\n },\n executionTimeMs: Date.now() - start,\n };\n } catch (e: any) {\n return {\n agent: \"AnalystAgent\",\n category: \"analysis\",\n status: \"error\",\n message: `Analyst error: ${e.message}`,\n data: { error: e.message },\n executionTimeMs: Date.now() - start,\n };\n }\n}\n",[],[222,358],["D",1783359042602],[239,240,241,242,243],[222],["D",1783347377247],["D",1789522836959],{"_105":104,"_113":109},{"_105":104,"_106":107,"_108":109,"_19":60,"_110":111,"_112":30},{"_13":94,"_37":95,"_48":47,"_96":49,"_97":-5,"_98":-5,"_99":-5,"_100":-5,"_101":-5,"_102":50,"_103":104},"ctx",{"_87":-5,"_132":406,"_407":408,"_409":410,"_411":30,"_412":30,"_413":30,"_414":64,"_415":-5,"_416":-7,"_417":-7,"_418":68,"_419":30,"_420":30,"_421":-5,"_422":30,"_423":30,"_424":30,"_425":-5},"ENV",{"_377":378,"_379":380,"_381":95,"_382":383,"_384":385,"_386":387,"_388":389,"_390":391,"_392":393,"_394":395,"_396":397,"_398":399,"_400":401,"_402":403,"_404":405},"REACT_APP_SENTRY_DSN","https://586020d86f05489db63a554690acf1e9@o1398359.ingest.sentry.io/6724499","RENDER_GIT_COMMIT","e87eb04e707cf64e7e532c8e8ec395f209e17f15","RENDER_GIT_BRANCH","POSTHOG_API_TOKEN","phc_nmoyuAkStEZuUp69rYlg1DZgrIt19Gx0Y4Hwf8tGgJP","POSTHOG_API_HOST","https://wart.val.town","WEB_API_TEMPLATE","https://{val}.web.val.run","APP_URL","https://www.val.town","MODULE_URL","https://esm.town","VANITY_WEB_API_TEMPLATE","https://{label}.val.run/","DENOLS_URL","https://lsp.val.town","NODE_ENV","production","CLOUDFLARE_TURNSTILE_SITE_KEY","0x4AAAAAACtaIZJH7s9rBMZ6","GOOGLE_CLIENT_ID","875989696899-jtsqkhlbdv5jrlsk8631p7ipjg3c6o62.apps.googleusercontent.com","STRIPE_PUBLISHABLE_KEY","pk_live_51N2zvWEpZTTLipIkVEwXrUG9AT0yeOOEYxszarZT1jpsZV58Ly6wlg3SltZpHN26CUalOBmUSY7adQRQiSkCZz2500qSXNkf8Y","API_URL","https://api.val.town",[],"preferencesKeybindings","default","preferencesTheme","light","preferencesFixers","preferencesCopilot","preferencesSelectionAi","preferencesCloseBrackets","preferencesChatPanelState","preferencesDprint","preferencesDenofmt","notificationCount","flagDenols","isBusinessOrgMember","businessOrgHandle","isBusinessOrgAdmin","isHigherTownieTier","navCollapsed","currentOrgId"]

Diff · v48