AIHive · Agent World

[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8,"_9":10},"actionData","errors","root",{"_1903":1904,"_1905":1906},"routes/_app.x.$handle.$name",{"_101":100,"_32":46,"_109":91,"_27":110,"_13":14,"_111":-5,"_11":112,"_113":114,"_115":116,"_117":118,"_119":120,"_20":-5,"_121":-5,"_122":-5,"_123":25,"_124":125,"_126":60,"_127":-5,"_128":-5,"_129":130},"routes/_app.x.$handle.$name.branch.$branch.version.$version.diff",{"_11":12,"_13":14,"_15":16,"_17":-5,"_18":-5},"project",{"_44":45,"_32":46,"_39":47,"_48":-5,"_49":25,"_50":51,"_52":51,"_53":54,"_55":-5,"_18":56,"_57":58,"_59":60,"_61":-5,"_62":-5,"_63":64,"_65":66,"_67":68,"_69":70,"_71":72,"_73":-5,"_74":75,"_76":75,"_77":78,"_79":78,"_80":78,"_81":75,"_82":75,"_83":84,"_85":86,"_87":88,"_89":25},"version",1072,"changes",[19],"aiChatInfo","userId",{"_20":21,"_22":23,"_24":25,"_26":25,"_27":28,"_29":28},"file",{"_30":31,"_32":33,"_34":35,"_36":38,"_39":40,"_41":14,"_42":43,"_24":25},"previous",{"_30":31,"_32":33,"_34":35,"_36":37},"deleted",false,"deletedHere","path","api/chat.ts","latestPath","fileId","31ca5a90-56ca-11f1-a61d-ee650bb23af1","name","chat.ts","fileType","script","content","import { sqlite, currentTenantId } from \"../db.ts\";\nimport { getLLMConfig, callLLM, callLLMStream } from \"./llm.ts\";\nimport { getFunnelSummary, getCorrelations } from \"./intelligence.ts\";\nimport { listAgents, createAgent, getAgent, deleteAgent, startAgentRun, finishAgentRun } from \"./agents.ts\";\nimport { getCredentials, handleIntegrationProxy } from \"./integrations.ts\";\nimport { enqueueAction } from \"./actions.ts\";\nimport { createTask, listTasks, updateTask, deleteTask } from \"./tasks.ts\";\nimport { createCalendarEntry } from \"./calendar_unified.ts\";\nimport { listObjectives, createObjective } from \"./okrs.ts\";\nimport { createCampaign, deleteCampaign, listCampaigns, getCampaign, getCampaignAgents } from \"./campaigns.ts\";\nimport { createExperiment } from \"./experiments.ts\";\nimport { getCompanyContextString } from \"./company.ts\";\nimport { getMemoryContextString, captureMemoryFromConversation } from \"./memory.ts\";\nimport { loadEnabledKnowledgeContext } from \"./knowledge.ts\";\nimport { getSkillsIndex, applySkillContext, getSkillReference, getSkillReferenceList } from \"./skills.ts\";\nimport type { ChatMessage } from \"../types.ts\";\n\n// ─── Constants ────────────────────────────────────────────────────────────────\n\nconst CORS = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"GET, POST, PUT, DELETE, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type, Authorization\",\n};\n\nconst DEFAULT_SYSTEM_PROMPT =\n \"You are an AI assistant for AIHive Command Center, a marketing and sales platform. \" +\n \"You can analyze data, create agents, and execute tasks across integrations.\";\n\n// Grounds the model in its REAL execution capabilities so it actually uses the tools instead of\n// hallucinating that it is \"blocked\" or \"needs credentials\" (the platform already holds them).\nconst EXECUTION_GUIDE =\n \"EXECUTION CAPABILITIES — IMPORTANT: You can actually DO work via tools. NEVER claim you are \" +\n \"'blocked', that you 'need credentials/site details', or that there is a 'security boundary you cannot bypass' — \" +\n \"the platform already stores all integration credentials. \" +\n \"PUBLIC/IRREVERSIBLE actions QUEUE for one-click human approval in the Approvals inbox — call the matching tool: \" +\n \"prepare_wordpress_post (publish to WordPress), prepare_email (send via Gmail), prepare_outlook_email (send via Outlook), prepare_linkedin_post (post to LinkedIn). \" +\n \"INTERNAL/REVERSIBLE actions run IMMEDIATELY (no approval): create_campaign (start a campaign/project), create_task and schedule_calendar_event (create work), \" +\n \"create_drive_folder / create_google_doc / create_google_sheet (create files in Google Drive/Docs/Sheets). \" +\n \"Use get_integration_data to read live data. None of these require the user to provide credentials. \" +\n \"When the user asks you to publish or send something, call the appropriate prepare_* tool with the FULL content, then tell them it is queued in Approvals. \" +\n \"When the user asks to create a campaign/project or a file, call the matching create_* tool and confirm it was created. \" +\n \"RUNNING THE WORK: when the user asks you to RUN / DO / EXECUTE / START the project's work (e.g. 'run a funnel analysis', 'do it', 'kick it off'), \" +\n \"you MUST call run_campaign — do NOT just narrate or fabricate the work yourself. run_campaign hands off to the member agents, who actually do the work and queue the deliverable to Approvals. \" +\n \"TEAM MANAGEMENT: you can also REMOVE work. Use list_campaigns / list_agents to find ids, then delete_campaign (removes the campaign AND its member agents) or delete_agent (removes one standalone agent). \" +\n \"Deletes are permanent — when the user clearly asks to delete/remove something, just do it and confirm; if they are vague about which one, list them first and ask. \" +\n \"Do NOT fabricate step-by-step 'execution logs', and do NOT ask the user for credentials.\";\n\n// ─── Helpers ──────────────────────────────────────────────────────────────────\n\nfunction json(data: any, status = 200): Response {\n return new Response(JSON.stringify(data), {\n status,\n headers: { \"Content-Type\": \"application/json\", ...CORS },\n });\n}\n\n// ─── Active project (campaign) context ─────────────────────────────────────────\n// When the Director chat is scoped to a campaign (frontend sends body.campaign = campaign id),\n// load that campaign + its parent OKR objective + its member agents and render a concise\n// \"ACTIVE PROJECT\" block. Injected into the system prompt so the Director directs THAT project and\n// treats the linked objective as the goal. Also returns campaign_id + objective_id so tasks the\n// chat creates can be linked back to the project. Fully guarded — never throws.\ntype CampaignScope = { campaign_id?: string; objective_id?: string };\n\nasync function buildCampaignContext(\n campaignId?: string\n): Promise<{ context: string; scope: CampaignScope }> {\n if (!campaignId || typeof campaignId !== \"string\" || !campaignId.trim()) {\n return { context: \"\", scope: {} };\n }\n try {\n const campaign = await getCampaign(campaignId.trim());\n if (!campaign) return { context: \"\", scope: {} };\n\n const scope: CampaignScope = { campaign_id: String(campaign.id) };\n\n // Parent objective (the FINAL goal). listObjectives has no get-by-id → list + filter.\n let objLine = \"\";\n if (campaign.objective_id != null) {\n scope.objective_id = String(campaign.objective_id);\n try {\n const objs = await listObjectives({});\n const obj = objs.find((o: any) => String(o.id) === String(campaign.objective_id));\n if (obj) {\n const krs = (obj.key_results || []).slice(0, 3).map((k: any) =>\n ` • ${k.title} — ${k.progress}% (now ${k.current_value}${k.unit ? \" \" + k.unit : \"\"} → target ${k.target_value}${k.unit ? \" \" + k.unit : \"\"})`\n ).join(\"\\n\");\n objLine =\n `\\n- Parent objective (the FINAL goal): \"${obj.title}\" [${obj.progress}% complete]` +\n (krs ? \"\\n Key results:\\n\" + krs : \"\");\n }\n } catch { /* objective unavailable — skip */ }\n }\n\n // Member agents the Director orchestrates.\n let agentsLine = \"\";\n try {\n const members = (campaign.agents && campaign.agents.length\n ? campaign.agents\n : await getCampaignAgents(String(campaign.id))) as any[];\n const names = members\n .map((m) => {\n const nm = m?.agent?.name || m?.role || \"agent\";\n const role = m?.role && m?.agent?.name ? ` (${m.role})` : \"\";\n return `${nm}${role}`;\n })\n .filter(Boolean);\n if (names.length) agentsLine = `\\n- Member agents you direct: ${names.join(\", \")}`;\n } catch { /* members unavailable — skip */ }\n\n const context = (\n \"## ACTIVE PROJECT — you are directing THIS campaign\\n\" +\n \"You are the DIRECTOR of the project below. Direct your member agents and steer every action toward this project's goal and its parent objective. \" +\n \"When you create tasks, campaigns or content, scope them to THIS project.\\n\" +\n \"When the user asks you to RUN / DO / EXECUTE / START this project's work (e.g. 'run a funnel analysis', 'do it', 'go'), CALL run_campaign (it defaults to this campaign) — the member agents then actually do the work and the deliverable lands in Approvals. Do NOT just narrate or pretend to do the work yourself.\\n\" +\n `- Project (campaign): \"${campaign.name}\"` +\n (campaign.goal ? `\\n- Goal: ${campaign.goal}` : \"\") +\n (campaign.channel ? `\\n- Channel: ${campaign.channel}` : \"\") +\n (campaign.audience ? `\\n- Audience: ${campaign.audience}` : \"\") +\n (campaign.workspace ? `\\n- Workspace: ${campaign.workspace}` : \"\") +\n objLine +\n agentsLine\n ).slice(0, 2000);\n\n return { context, scope };\n } catch {\n return { context: \"\", scope: {} };\n }\n}\n\nfunction rowToConversation(\n columns: string[],\n row: any[]\n): { id: string; messages: ChatMessage[]; agent_id?: string; workspace?: string } {\n const obj: Record = {};\n for (let i = 0; i < columns.length; i++) {\n obj[columns[i]] = row[i];\n }\n\n let messages: ChatMessage[] = [];\n if (obj.messages) {\n try {\n messages = JSON.parse(obj.messages);\n } catch {\n messages = [];\n }\n }\n\n return {\n id: obj.id,\n messages,\n agent_id: obj.agent_id ?? undefined,\n workspace: obj.workspace ?? undefined,\n };\n}\n\n// ─── Conversation DB ops ──────────────────────────────────────────────────────\n\nasync function loadConversation(\n conversationId: string\n): Promise<{ id: string; messages: ChatMessage[]; agent_id?: string; workspace?: string } | null> {\n const result = await sqlite.execute(\n \"SELECT id, messages, agent_id, workspace FROM conversations WHERE id = ? AND tenant_id = ?\",\n [conversationId, currentTenantId()]\n );\n\n if (!result.rows || result.rows.length === 0) {\n return null;\n }\n\n return rowToConversation(result.columns, result.rows[0]);\n}\n\nasync function saveConversation(\n id: string,\n messages: ChatMessage[],\n agentId?: string,\n workspace?: string\n): Promise {\n const now = new Date().toISOString();\n const messagesJson = JSON.stringify(messages);\n\n const existing = await sqlite.execute(\n \"SELECT id FROM conversations WHERE id = ? AND tenant_id = ?\",\n [id, currentTenantId()]\n );\n\n const ws = workspace ?? \"general\";\n if (existing.rows && existing.rows.length > 0) {\n await sqlite.execute(\n \"UPDATE conversations SET messages = ?, agent_id = ?, workspace = ?, updated_at = ? WHERE id = ? AND tenant_id = ?\",\n [messagesJson, agentId ?? null, ws, now, id, currentTenantId()]\n );\n } else {\n await sqlite.execute(\n `INSERT INTO conversations (id, messages, agent_id, workspace, created_at, updated_at, tenant_id)\n VALUES (?, ?, ?, ?, ?, ?, ?)`,\n [id, messagesJson, agentId ?? null, ws, now, now, currentTenantId()]\n );\n }\n}\n\n// ─── Tool definitions ─────────────────────────────────────────────────────────\n\nexport function buildTools(): any[] {\n return [\n {\n type: \"function\",\n function: {\n name: \"get_integration_data\",\n description:\n \"Fetch data from ANY connected integration. Marketing/SEO: ahrefs, ga4, gsc, youtube, wordpress, producthunt, linkedin. \" +\n \"Microsoft 365 (all share one connected account): outlook (inbox/calendar/contacts), onedrive (files), \" +\n \"teams (teams/channels/messages/chats), excel (workbooks/worksheets/range). Google Workspace: gmail, drive, sheets, docs, calendar.\",\n parameters: {\n type: \"object\",\n properties: {\n name: {\n type: \"string\",\n description:\n \"Integration name: ahrefs, ga4, gsc, linkedin, outlook, onedrive, teams, excel, wordpress, youtube, producthunt, gmail, drive, sheets, docs, calendar\",\n },\n endpoint: {\n type: \"string\",\n description:\n \"The endpoint within the integration, e.g. onedrive 'recent'/'files'/'search'; teams 'teams'/'channels'/'messages'/'chats'; excel 'workbooks'/'worksheets'/'range'; outlook 'inbox'/'calendar'/'contacts'; ahrefs 'site-explorer'/'keywords'; gsc 'search-analytics'; linkedin 'posts'.\",\n },\n params: {\n type: \"object\",\n description: \"Additional query parameters to forward to the API\",\n additionalProperties: { type: \"string\" },\n },\n },\n required: [\"name\", \"endpoint\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"create_agent\",\n description:\n \"Create a new AI agent in AIHive Command Center with a specific role, workspace, and toolset.\",\n parameters: {\n type: \"object\",\n properties: {\n name: {\n type: \"string\",\n description: \"Name for the new agent\",\n },\n workspace: {\n type: \"string\",\n enum: [\"marketing\", \"sales\"],\n description: \"The workspace this agent belongs to\",\n },\n description: {\n type: \"string\",\n description: \"Short description of what this agent does\",\n },\n system_prompt: {\n type: \"string\",\n description: \"System prompt that defines the agent's behavior\",\n },\n tools: {\n type: \"array\",\n items: { type: \"string\" },\n description:\n \"List of tool names the agent has access to (e.g. ['get_integration_data', 'compose_content'])\",\n },\n },\n required: [\"name\", \"workspace\", \"description\", \"system_prompt\", \"tools\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"list_agents\",\n description: \"List all AI agents, optionally filtered by workspace.\",\n parameters: {\n type: \"object\",\n properties: {\n workspace: {\n type: \"string\",\n enum: [\"marketing\", \"sales\"],\n description: \"Filter agents by workspace (optional)\",\n },\n },\n required: [],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"list_objectives\",\n description: \"List OKR objectives (id + title). Use before create_campaign to link a campaign to an existing objective.\",\n parameters: {\n type: \"object\",\n properties: {\n workspace: { type: \"string\", enum: [\"marketing\", \"sales\"], description: \"Filter by workspace (optional)\" },\n },\n required: [],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"list_campaigns\",\n description: \"List campaigns (id, name, status, mode, workspace). Use this to resolve a campaign's id before delete_campaign, or to report campaign status.\",\n parameters: {\n type: \"object\",\n properties: {\n workspace: { type: \"string\", enum: [\"marketing\", \"sales\"], description: \"Filter by workspace (optional)\" },\n },\n required: [],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"delete_campaign\",\n description: \"Permanently delete a campaign AND all of its member agents. Provide campaign_id (preferred) or name. This is irreversible — confirm with the user first unless they were explicit about deleting it.\",\n parameters: {\n type: \"object\",\n properties: {\n campaign_id: { type: \"string\", description: \"The campaign id to delete (preferred)\" },\n name: { type: \"string\", description: \"The campaign name — used to resolve the id when campaign_id is omitted\" },\n },\n required: [],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"delete_agent\",\n description: \"Permanently delete a standalone AI agent by agent_id (preferred) or name. Irreversible — confirm first unless the user was explicit. To remove a campaign's member agents, delete the campaign instead.\",\n parameters: {\n type: \"object\",\n properties: {\n agent_id: { type: \"string\", description: \"The agent id to delete (preferred)\" },\n name: { type: \"string\", description: \"The agent name — used to resolve the id when agent_id is omitted\" },\n },\n required: [],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"compose_content\",\n description:\n \"Generate marketing or sales content such as blog posts, LinkedIn updates, emails, or YouTube descriptions.\",\n parameters: {\n type: \"object\",\n properties: {\n type: {\n type: \"string\",\n enum: [\"blog\", \"linkedin\", \"email\", \"youtube_description\"],\n description: \"The type of content to generate\",\n },\n topic: {\n type: \"string\",\n description: \"The topic or subject of the content\",\n },\n tone: {\n type: \"string\",\n description:\n \"The tone of voice for the content (e.g. 'professional', 'casual', 'persuasive')\",\n },\n },\n required: [\"type\", \"topic\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"prepare_email\",\n description:\n \"Draft an email and QUEUE it for human approval (Gmail). It is NOT sent until the user approves it in Approvals. Use this to actually act on email outreach.\",\n parameters: {\n type: \"object\",\n properties: {\n to: { type: \"string\", description: \"Recipient email address\" },\n subject: { type: \"string\", description: \"Email subject\" },\n body: { type: \"string\", description: \"Email body (plain text)\" },\n cc: { type: \"string\", description: \"Optional CC address\" },\n task_id: { type: \"string\", description: \"Optional task id to mark done once sent\" },\n },\n required: [\"to\", \"subject\", \"body\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"prepare_linkedin_post\",\n description:\n \"Draft a LinkedIn post and QUEUE it for human approval. NOT posted until the user approves.\",\n parameters: {\n type: \"object\",\n properties: {\n text: { type: \"string\", description: \"The post text\" },\n task_id: { type: \"string\", description: \"Optional task id to mark done once posted\" },\n },\n required: [\"text\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"prepare_wordpress_post\",\n description:\n \"Draft a WordPress blog post and QUEUE it for human approval. NOT published until the user approves.\",\n parameters: {\n type: \"object\",\n properties: {\n title: { type: \"string\", description: \"Post title\" },\n content: { type: \"string\", description: \"Post HTML/markdown content\" },\n status: { type: \"string\", enum: [\"publish\", \"draft\"], description: \"publish (default) or draft\" },\n task_id: { type: \"string\", description: \"Optional task id to mark done once published\" },\n },\n required: [\"title\", \"content\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"create_task\",\n description:\n \"Create a task in the task board. Runs immediately (internal, no approval). Optionally link it to an objective, key result, or experiment.\",\n parameters: {\n type: \"object\",\n properties: {\n title: { type: \"string\" },\n description: { type: \"string\" },\n workspace: { type: \"string\", enum: [\"marketing\", \"sales\"] },\n priority: { type: \"string\", enum: [\"low\", \"medium\", \"high\", \"urgent\"] },\n due_date: { type: \"string\", description: \"YYYY-MM-DD\" },\n objective_id: { type: \"string\" },\n key_result_id: { type: \"string\" },\n experiment_id: { type: \"string\" },\n },\n required: [\"title\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"list_tasks\",\n description: \"List tasks (id, title, status, priority, due_date, campaign_id). Use to find a task's id before update_task / delete_task, or to report the board.\",\n parameters: {\n type: \"object\",\n properties: {\n workspace: { type: \"string\", enum: [\"marketing\", \"sales\"] },\n status: { type: \"string\", enum: [\"todo\", \"doing\", \"done\", \"blocked\"] },\n campaign_id: { type: \"string\" },\n },\n required: [],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"update_task\",\n description: \"Update / modify an existing task — change its title, description, status (todo|doing|done|blocked), priority, due_date, assignee, or its objective/key-result/campaign links. Provide task_id (use list_tasks to find it).\",\n parameters: {\n type: \"object\",\n properties: {\n task_id: { type: \"string\", description: \"The task id to update\" },\n title: { type: \"string\" },\n description: { type: \"string\" },\n status: { type: \"string\", enum: [\"todo\", \"doing\", \"done\", \"blocked\"] },\n priority: { type: \"string\", enum: [\"low\", \"medium\", \"high\", \"urgent\"] },\n due_date: { type: \"string\", description: \"YYYY-MM-DD\" },\n assignee: { type: \"string\" },\n objective_id: { type: \"string\" },\n key_result_id: { type: \"string\" },\n campaign_id: { type: \"string\" },\n },\n required: [\"task_id\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"delete_task\",\n description: \"Permanently delete a task by task_id (use list_tasks to find it). Irreversible — when the user clearly asks to delete it, just do it; if vague about which, list_tasks first.\",\n parameters: {\n type: \"object\",\n properties: { task_id: { type: \"string\", description: \"The task id to delete\" } },\n required: [\"task_id\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"schedule_calendar_event\",\n description:\n \"Add an event to the unified calendar. Runs immediately (internal, no approval).\",\n parameters: {\n type: \"object\",\n properties: {\n title: { type: \"string\" },\n event_date: { type: \"string\", description: \"YYYY-MM-DD\" },\n event_time: { type: \"string\", description: \"HH:MM (optional)\" },\n workspace: { type: \"string\", enum: [\"marketing\", \"sales\"] },\n },\n required: [\"title\", \"event_date\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"create_campaign\",\n description:\n \"Create a REAL multi-agent campaign: a DIRECTOR orchestrates several member agents that cooperate to achieve the goal. \" +\n \"Runs immediately (internal, no approval). \" +\n \"By default the campaign gets a 3-agent research -> strategy -> compose pipeline; supply a custom 'agents' array to define your own team. \" +\n \"Set 'mode' to 'sequential' (each agent hands off to the next, default) or 'parallel' (agents work independently and the director merges).\",\n parameters: {\n type: \"object\",\n properties: {\n name: { type: \"string\", description: \"Campaign name\" },\n workspace: { type: \"string\", enum: [\"marketing\", \"sales\"] },\n channel: { type: \"string\", description: \"linkedin | email | content | blog | seo | social | paid | mixed\" },\n goal: { type: \"string\", description: \"What the campaign aims to achieve\" },\n audience: { type: \"string\", description: \"Target audience (optional)\" },\n objective_id: { type: \"string\", description: \"Optional OKR objective id this campaign serves (the FINAL goal). Resolve it first via list_objectives, then pass its id so a completed campaign run advances that objective.\" },\n mode: { type: \"string\", enum: [\"sequential\", \"parallel\"], description: \"How the member agents cooperate. 'sequential' (default): each agent hands its output to the next. 'parallel': agents work independently and the director merges results.\" },\n agents: {\n type: \"array\",\n description: \"Optional custom team of member agents the director orchestrates. Omit to use the default research -> strategy -> compose pipeline.\",\n items: {\n type: \"object\",\n properties: {\n name: { type: \"string\", description: \"Member agent name, e.g. 'Research Agent'\" },\n role: { type: \"string\", description: \"Short role label, e.g. research | strategy | compose\" },\n system_prompt: { type: \"string\", description: \"Instructions defining what this member agent does\" },\n },\n required: [\"name\"],\n },\n },\n },\n required: [\"name\", \"channel\", \"goal\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"run_campaign\",\n description:\n \"Execute the active campaign — the member agents actually DO the work (research/strategy/compose/etc.) and the result is queued to Approvals. \" +\n \"Call this when the user asks you to run / do / execute / start the project's work (e.g. 'run a funnel analysis').\",\n parameters: {\n type: \"object\",\n properties: {\n campaign_id: {\n type: \"string\",\n description: \"The campaign id to execute. Optional — defaults to the campaign-scoped chat's campaign.\",\n },\n },\n required: [],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"create_drive_folder\",\n description: \"Create a folder in Google Drive. Runs immediately (internal, no approval).\",\n parameters: {\n type: \"object\",\n properties: { name: { type: \"string\", description: \"Folder name\" } },\n required: [\"name\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"create_google_doc\",\n description: \"Create a Google Doc, optionally with initial content. Runs immediately (internal, no approval).\",\n parameters: {\n type: \"object\",\n properties: {\n title: { type: \"string\", description: \"Document title\" },\n content: { type: \"string\", description: \"Optional initial body text\" },\n },\n required: [\"title\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"create_google_sheet\",\n description: \"Create a Google Sheet. Runs immediately (internal, no approval).\",\n parameters: {\n type: \"object\",\n properties: { title: { type: \"string\", description: \"Spreadsheet title\" } },\n required: [\"title\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"prepare_outlook_email\",\n description:\n \"Draft an email and QUEUE it for human approval (Outlook/Microsoft 365). NOT sent until the user approves it in Approvals.\",\n parameters: {\n type: \"object\",\n properties: {\n to: { type: \"string\", description: \"Recipient email address\" },\n subject: { type: \"string\", description: \"Email subject\" },\n body: { type: \"string\", description: \"Email body (plain text)\" },\n cc: { type: \"string\", description: \"Optional CC address\" },\n task_id: { type: \"string\", description: \"Optional task id to mark done once sent\" },\n },\n required: [\"to\", \"subject\", \"body\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"get_file_content\",\n description:\n \"Read the full content of a single file or message from a connected integration by its id. \" +\n \"Use for drive/onedrive files or outlook/gmail messages — pass the integration name and the item's id \" +\n \"(get the id first via get_integration_data, e.g. listing drive 'files' or gmail 'inbox').\",\n parameters: {\n type: \"object\",\n properties: {\n integration: {\n type: \"string\",\n description: \"Integration name: drive, onedrive, outlook, gmail\",\n },\n id: {\n type: \"string\",\n description: \"The file id (drive/onedrive) or message id (outlook/gmail) to read\",\n },\n },\n required: [\"integration\", \"id\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"apply_skill\",\n description:\n \"Load a marketing/sales skill playbook into context. Call this with the skill name (from the \" +\n \"MARKETING SKILLS INDEX) when a request matches one, then follow the returned playbook.\",\n parameters: {\n type: \"object\",\n properties: {\n skill_name: { type: \"string\", description: \"The skill name to apply (from the skills index)\" },\n },\n required: [\"skill_name\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"get_skill_reference\",\n description:\n \"List a skill's reference documents (omit reference) or fetch one reference doc's content \" +\n \"(pass reference). Use after apply_skill when the playbook points to a named reference.\",\n parameters: {\n type: \"object\",\n properties: {\n skill_name: { type: \"string\", description: \"The skill name\" },\n reference: { type: \"string\", description: \"Optional reference doc name to fetch; omit to list available references\" },\n },\n required: [\"skill_name\"],\n },\n },\n },\n ];\n}\n\n// ─── Tool executor ────────────────────────────────────────────────────────────\n\n// Run a WRITE/mutate endpoint on an integration immediately, POSTing through the same proxy\n// the approval queue uses. Returns parsed JSON (or {error}).\nasync function runIntegrationWrite(integration: string, endpoint: string, payload: any): Promise {\n const creds = await getCredentials(integration);\n if (!creds) return { error: `Integration '${integration}' is not connected. Add it in Settings first.` };\n const params = new URLSearchParams({ endpoint });\n const req = new Request(\n `https://aihive.internal/api/integrations/${integration}/data?endpoint=${encodeURIComponent(endpoint)}`,\n { method: \"POST\", headers: { \"Content-Type\": \"application/json\" }, body: JSON.stringify(payload ?? {}) },\n );\n try {\n const res = await handleIntegrationProxy(req, integration, params, creds);\n return await res.json().catch(() => ({}));\n } catch (err: any) {\n return { error: err?.message ?? \"Integration call failed\" };\n }\n}\n\nexport async function executeTool(\n toolName: string,\n toolArgs: Record,\n llmConfig: Awaited>,\n // When the chat is scoped to a campaign/project, this carries its campaign_id (+ objective_id)\n // so created tasks link back to the project. Optional — preserves all existing callers.\n scope?: CampaignScope\n): Promise {\n switch (toolName) {\n case \"get_integration_data\": {\n const { name, endpoint, params = {} } = toolArgs as {\n name: string;\n endpoint: string;\n params?: Record;\n };\n\n const creds = await getCredentials(name);\n if (!creds) {\n return JSON.stringify({\n error: `Integration '${name}' is not configured. Please add credentials first.`,\n });\n }\n\n // Build a synthetic URLSearchParams from args\n const searchParams = new URLSearchParams({ endpoint });\n for (const [k, v] of Object.entries(params)) {\n searchParams.set(k, String(v));\n }\n\n // Build a minimal Request so handlers can read body if needed\n const syntheticReq = new Request(\"https://aihive.internal/proxy\", {\n method: \"GET\",\n });\n\n try {\n const proxyRes = await handleIntegrationProxy(\n syntheticReq,\n name,\n searchParams,\n creds\n );\n const data = await proxyRes.json();\n return JSON.stringify(data);\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Proxy request failed\" });\n }\n }\n\n case \"create_agent\": {\n const { name, workspace, description, system_prompt, tools } = toolArgs as {\n name: string;\n workspace: \"marketing\" | \"sales\";\n description: string;\n system_prompt: string;\n tools: string[];\n };\n\n try {\n const agent = await createAgent({\n name,\n workspace,\n description,\n system_prompt,\n tools,\n });\n return JSON.stringify({ success: true, agent });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to create agent\" });\n }\n }\n\n case \"list_agents\": {\n const { workspace } = toolArgs as { workspace?: string };\n try {\n const agents = await listAgents(workspace);\n return JSON.stringify({ agents });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to list agents\" });\n }\n }\n\n case \"compose_content\": {\n const { type, topic, tone = \"professional\" } = toolArgs as {\n type: \"blog\" | \"linkedin\" | \"email\" | \"youtube_description\";\n topic: string;\n tone?: string;\n };\n\n const contentPrompts: Record = {\n blog: `Write a comprehensive, SEO-friendly blog post about: \"${topic}\". Tone: ${tone}. Include an introduction, 3-5 sections with headers, and a conclusion.`,\n linkedin: `Write an engaging LinkedIn post about: \"${topic}\". Tone: ${tone}. Keep it under 300 words, include relevant hashtags, and a call-to-action.`,\n email: `Write a professional email about: \"${topic}\". Tone: ${tone}. Include subject line, greeting, body paragraphs, and a clear call-to-action.`,\n youtube_description: `Write an optimized YouTube video description for a video about: \"${topic}\". Tone: ${tone}. Include a hook, timestamps placeholder, links section, and relevant tags.`,\n };\n\n const prompt = contentPrompts[type] ?? `Write content about: \"${topic}\". Tone: ${tone}.`;\n\n // W1 — prepend unified company context so generated content stays on-brand.\n const companyCtx = await getCompanyContextString(1500).catch(() => \"\");\n\n try {\n const contentMessages: ChatMessage[] = [\n { role: \"user\", content: prompt },\n ];\n\n const result = await callLLM(llmConfig, contentMessages, {\n systemPrompt:\n (companyCtx ? companyCtx + \"\\n\\n---\\n\\n\" : \"\") +\n \"You are an expert content writer for marketing and sales teams. Produce high-quality, engaging content tailored to the specified format and tone.\",\n maxTokens: 2048,\n });\n\n // Persist the composed output as a REVIEW-ONLY deliverable so it doesn't evaporate.\n // integration:\"content\" is internal — approveAction simply ACCEPTS it (no external send).\n // Guarded so a queue failure never breaks the compose tool result.\n let action_id: number | undefined;\n try {\n const title = `Content: ${type} — ${String(topic).slice(0, 80)}`;\n const action = await enqueueAction({\n integration: \"content\",\n endpoint: \"review\",\n payload: { type, topic, tone, content: result.content },\n title,\n preview: String(result.content ?? \"\").slice(0, 600),\n source_type: \"chat\",\n });\n action_id = action?.id;\n } catch { /* non-critical — still return the composed content */ }\n\n return JSON.stringify({\n type,\n topic,\n tone,\n content: result.content,\n action_id,\n queued_for_review: action_id != null,\n note: action_id != null\n ? \"Composed and saved to Approvals as a review-only deliverable.\"\n : undefined,\n });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to compose content\" });\n }\n }\n\n case \"prepare_email\": {\n const { to, subject, body, cc, task_id } = toolArgs as any;\n if (!to || !subject) return JSON.stringify({ error: \"to + subject required\" });\n const action = await enqueueAction({\n integration: \"gmail\", endpoint: \"send\",\n payload: { to, subject, body: body ?? \"\", cc },\n title: `Email: ${subject}`,\n preview: `To: ${to}\\nSubject: ${subject}\\n\\n${String(body ?? \"\").slice(0, 600)}`,\n task_id, source_type: \"chat\",\n });\n return JSON.stringify({ queued: true, status: \"prepared\", action_id: action?.id, note: \"Email drafted and QUEUED FOR APPROVAL — it will NOT send until the user approves it in Approvals.\" });\n }\n\n case \"prepare_linkedin_post\": {\n const { text, task_id } = toolArgs as any;\n if (!text) return JSON.stringify({ error: \"text required\" });\n const action = await enqueueAction({\n integration: \"linkedin\", endpoint: \"create-post\",\n payload: { text },\n title: \"LinkedIn post\",\n preview: String(text).slice(0, 600),\n task_id, source_type: \"chat\", workspace: \"sales\",\n });\n return JSON.stringify({ queued: true, status: \"prepared\", action_id: action?.id, note: \"LinkedIn post drafted and QUEUED FOR APPROVAL.\" });\n }\n\n case \"prepare_wordpress_post\": {\n const { title, content, status, task_id } = toolArgs as any;\n if (!title || !content) return JSON.stringify({ error: \"title + content required\" });\n const action = await enqueueAction({\n integration: \"wordpress\", endpoint: \"publish-post\",\n payload: { title, content, status: status === \"draft\" ? \"draft\" : \"publish\" },\n title: `WordPress: ${title}`,\n preview: `${title}\\n\\n${String(content).slice(0, 600)}`,\n task_id, source_type: \"chat\",\n });\n return JSON.stringify({ queued: true, status: \"prepared\", action_id: action?.id, note: \"WordPress post drafted and QUEUED FOR APPROVAL.\" });\n }\n\n case \"create_task\": {\n const a = toolArgs as any;\n if (!a.title) return JSON.stringify({ error: \"title required\" });\n try {\n // Active-project scope: when this chat is directing a campaign, link the task to that\n // campaign (and its parent objective) unless the model explicitly supplied its own.\n const campaign_id = a.campaign_id ?? scope?.campaign_id;\n const objective_id = a.objective_id ?? scope?.objective_id;\n const task = await createTask({\n title: a.title, description: a.description ?? null,\n priority: a.priority, workspace: a.workspace ?? \"marketing\",\n due_date: a.due_date, objective_id,\n key_result_id: a.key_result_id, experiment_id: a.experiment_id,\n campaign_id,\n } as any);\n return JSON.stringify({ success: true, task });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to create task\" });\n }\n }\n\n case \"list_tasks\": {\n const a = toolArgs as { workspace?: string; status?: string; campaign_id?: string };\n try {\n const filters: any = {};\n if (a.workspace) filters.workspace = a.workspace;\n if (a.status) filters.status = a.status;\n if (a.campaign_id ?? scope?.campaign_id) filters.campaign_id = a.campaign_id ?? scope?.campaign_id;\n const tasks = await listTasks(filters);\n return JSON.stringify({ tasks: tasks.map((t: any) => ({ id: t.id, title: t.title, status: t.status, priority: t.priority, due_date: t.due_date, campaign_id: t.campaign_id })) });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to list tasks\" });\n }\n }\n\n case \"update_task\": {\n const a = toolArgs as any;\n if (!a.task_id) return JSON.stringify({ error: \"task_id required (use list_tasks to find it)\" });\n try {\n const patch: any = {};\n for (const k of [\"title\", \"description\", \"status\", \"priority\", \"due_date\", \"assignee\", \"objective_id\", \"key_result_id\", \"campaign_id\"]) {\n if (a[k] !== undefined) patch[k] = a[k];\n }\n const task = await updateTask(String(a.task_id), patch);\n return JSON.stringify(task ? { success: true, task } : { error: `Task ${a.task_id} not found.` });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to update task\" });\n }\n }\n\n case \"delete_task\": {\n const a = toolArgs as { task_id?: string };\n if (!a.task_id) return JSON.stringify({ error: \"task_id required (use list_tasks to find it)\" });\n try {\n const ok = await deleteTask(String(a.task_id));\n return JSON.stringify(ok ? { success: true, deleted_task: a.task_id, note: \"Task permanently deleted.\" } : { error: `Task ${a.task_id} not found or already deleted.` });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to delete task\" });\n }\n }\n\n case \"schedule_calendar_event\": {\n const a = toolArgs as any;\n if (!a.title || !a.event_date) return JSON.stringify({ error: \"title + event_date (YYYY-MM-DD) required\" });\n try {\n const entry = await createCalendarEntry({\n title: a.title, event_date: a.event_date, event_time: a.event_time ?? null,\n workspace: a.workspace ?? \"marketing\", event_type: \"custom\",\n } as any);\n return JSON.stringify({ success: true, entry });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to schedule event\" });\n }\n }\n\n case \"list_objectives\": {\n const { workspace } = toolArgs as { workspace?: string };\n try {\n const objectives = await listObjectives(workspace ? { workspace } : {});\n return JSON.stringify({ objectives: objectives.map((o: any) => ({ id: o.id, title: o.title, status: o.status, progress: o.progress })) });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to list objectives\" });\n }\n }\n\n case \"create_campaign\": {\n const a = toolArgs as any;\n if (!a.name) return JSON.stringify({ error: \"name required\" });\n const workspace = a.workspace === \"sales\" ? \"sales\" : \"marketing\";\n try {\n // A campaign is a REAL multi-agent system: a DIRECTOR orchestrates ordered member agents\n // that cooperate. Use the model-supplied team if given, else createCampaign falls back to its\n // default research -> strategy -> compose pipeline.\n const agents = Array.isArray(a.agents) && a.agents.length ? a.agents : undefined;\n // Link the campaign to an OKR objective when the model resolved one via list_objectives.\n // The tool delivers objective_id as a string; createCampaign expects a number — coerce + guard.\n const objNum = a.objective_id != null ? Number(a.objective_id) : NaN;\n const objective_id = Number.isFinite(objNum) ? objNum : undefined;\n const campaign = await createCampaign({\n name: a.name,\n workspace,\n goal: a.goal ?? a.description ?? undefined,\n channel: a.channel ?? undefined,\n audience: a.audience ?? undefined,\n mode: a.mode,\n objective_id,\n agents,\n });\n const members = (campaign?.agents || []) as any[];\n const memberNames = members\n .map((m) => m?.agent?.name || m?.role || \"agent\")\n .filter(Boolean);\n const mode = campaign?.mode || \"sequential\";\n const summary =\n `Created campaign \"${campaign?.name ?? a.name}\" with a director orchestrating ` +\n `${memberNames.length} member agent${memberNames.length === 1 ? \"\" : \"s\"} (${mode}): ` +\n `${memberNames.join(\" -> \")}.`;\n return JSON.stringify({ success: true, campaign, message: summary });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to create campaign\" });\n }\n }\n\n case \"run_campaign\": {\n // EXECUTE the active campaign — the member agents actually do the work and the deliverable\n // lands in Approvals. CRITICAL: we do NOT run the ~77s orchestration inline here — that would\n // block the request / time out the SSE stream. Instead we resolve the campaign id and return a\n // short \"kicking off\" message. The BROWSER fires the real run: the streamed {tool} frame carries\n // name:\"run_campaign\" + the resolved campaign_id (in args/result), and the world entry POSTs to\n // /api/campaigns/:id/runs so the user watches the team work in the World.\n const a = toolArgs as { campaign_id?: string };\n const id = a.campaign_id ?? scope?.campaign_id;\n if (!id) {\n return JSON.stringify({\n error:\n \"No campaign to run — open a project (campaign-scoped chat) or pass campaign_id. Use list_campaigns to find one.\",\n });\n }\n return JSON.stringify({\n kicking_off: true,\n campaign_id: String(id),\n message:\n \"Kicking off the team now — the agents are executing the campaign; watch them work in the World and the deliverable will land in Approvals.\",\n });\n }\n\n case \"list_campaigns\": {\n const { workspace } = toolArgs as { workspace?: string };\n try {\n const campaigns = await listCampaigns(workspace);\n return JSON.stringify({ campaigns: campaigns.map((c: any) => ({ id: c.id, name: c.name, status: c.status, mode: c.mode, workspace: c.workspace })) });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to list campaigns\" });\n }\n }\n\n case \"delete_campaign\": {\n const a = toolArgs as { campaign_id?: string; name?: string };\n try {\n let id = a.campaign_id;\n if (!id && a.name) {\n const all = await listCampaigns();\n const q = String(a.name).toLowerCase();\n const match = all.find((c: any) => String(c.name).toLowerCase() === q)\n || all.find((c: any) => String(c.name).toLowerCase().includes(q));\n id = match?.id;\n }\n if (!id) return JSON.stringify({ error: `No campaign found matching ${a.campaign_id ?? a.name ?? \"(none)\"}. Use list_campaigns to see ids.` });\n const members = await getCampaignAgents(id).catch(() => []);\n const ok = await deleteCampaign(id);\n return JSON.stringify(ok\n ? { success: true, deleted_campaign: id, deleted_member_agents: members.length, note: \"Campaign and its member agents were permanently deleted.\" }\n : { error: `Campaign ${id} not found or already deleted.` });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to delete campaign\" });\n }\n }\n\n case \"delete_agent\": {\n const a = toolArgs as { agent_id?: string; name?: string };\n try {\n let id = a.agent_id;\n let agent = id ? await getAgent(id) : null;\n if (!agent && a.name) {\n const all = await listAgents();\n const q = String(a.name).toLowerCase();\n agent = all.find((g: any) => String(g.name).toLowerCase() === q)\n || all.find((g: any) => String(g.name).toLowerCase().includes(q)) || null;\n id = agent?.id;\n }\n if (!id) return JSON.stringify({ error: `No agent found matching ${a.agent_id ?? a.name ?? \"(none)\"}. Use list_agents to see ids.` });\n const ok = await deleteAgent(id);\n return JSON.stringify(ok\n ? { success: true, deleted_agent: id, name: agent?.name ?? null, note: \"Agent permanently deleted.\" }\n : { error: `Agent ${id} not found or already deleted.` });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to delete agent\" });\n }\n }\n\n case \"create_drive_folder\": {\n const a = toolArgs as any;\n if (!a.name) return JSON.stringify({ error: \"name required\" });\n const data = await runIntegrationWrite(\"drive\", \"createFolder\", { name: a.name });\n return JSON.stringify(data?.error ? data : { success: true, file: data });\n }\n\n case \"create_google_doc\": {\n const a = toolArgs as any;\n if (!a.title) return JSON.stringify({ error: \"title required\" });\n const data = await runIntegrationWrite(\"docs\", \"create\", { title: a.title, content: a.content });\n return JSON.stringify(data?.error ? data : { success: true, doc: data });\n }\n\n case \"create_google_sheet\": {\n const a = toolArgs as any;\n if (!a.title) return JSON.stringify({ error: \"title required\" });\n const data = await runIntegrationWrite(\"sheets\", \"create\", { title: a.title });\n return JSON.stringify(data?.error ? data : { success: true, sheet: data });\n }\n\n case \"prepare_outlook_email\": {\n const a = toolArgs as any;\n if (!a.to || !a.subject) return JSON.stringify({ error: \"to + subject required\" });\n const action = await enqueueAction({\n integration: \"outlook\", endpoint: \"send\",\n payload: {\n message: {\n subject: a.subject,\n body: { contentType: \"Text\", content: a.body ?? \"\" },\n toRecipients: [{ emailAddress: { address: a.to } }],\n ...(a.cc ? { ccRecipients: [{ emailAddress: { address: a.cc } }] } : {}),\n },\n saveToSentItems: true,\n },\n title: `Outlook email: ${a.subject}`,\n preview: `To: ${a.to}\\nSubject: ${a.subject}\\n\\n${String(a.body ?? \"\").slice(0, 600)}`,\n task_id: a.task_id, source_type: \"chat\",\n });\n return JSON.stringify({ queued: true, status: \"prepared\", action_id: action?.id, note: \"Outlook email drafted and QUEUED FOR APPROVAL.\" });\n }\n\n case \"get_file_content\": {\n // Read a single file/message from an integration via the SAME read path as get_integration_data.\n // drive/onedrive use endpoint 'read'; outlook/gmail use endpoint 'message'. The id is passed as ?id=.\n const { integration, id } = toolArgs as { integration: string; id: string };\n if (!integration || !id) return JSON.stringify({ error: \"integration + id required\" });\n\n const creds = await getCredentials(integration);\n if (!creds) {\n return JSON.stringify({\n error: `Integration '${integration}' is not configured. Please add credentials first.`,\n });\n }\n\n const endpoint =\n integration === \"outlook\" || integration === \"gmail\" ? \"message\" : \"read\";\n\n const searchParams = new URLSearchParams({ endpoint });\n searchParams.set(\"id\", String(id));\n\n const syntheticReq = new Request(\"https://aihive.internal/proxy\", {\n method: \"GET\",\n });\n\n try {\n const proxyRes = await handleIntegrationProxy(\n syntheticReq,\n integration,\n searchParams,\n creds\n );\n const data = await proxyRes.json();\n return JSON.stringify(data);\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Proxy request failed\" });\n }\n }\n\n case \"apply_skill\": {\n const ctx = await applySkillContext(toolArgs.skill_name);\n return ctx || JSON.stringify({ error: \"No such skill\" });\n }\n\n case \"get_skill_reference\": {\n if (!toolArgs.reference) {\n return JSON.stringify({ references: getSkillReferenceList(toolArgs.skill_name) });\n }\n const r = getSkillReference(toolArgs.skill_name, toolArgs.reference);\n return r || JSON.stringify({ error: \"No such reference\", available: getSkillReferenceList(toolArgs.skill_name) });\n }\n\n default:\n return JSON.stringify({ error: `Unknown tool: ${toolName}` });\n }\n}\n\n// ─── Main chat handler ────────────────────────────────────────────────────────\n\nexport async function handleChat(req: Request): Promise {\n if (req.method === \"OPTIONS\") {\n return new Response(null, { status: 204, headers: CORS });\n }\n\n if (req.method !== \"POST\") {\n return json({ error: \"Method not allowed\" }, 405);\n }\n\n let body: any;\n try {\n body = await req.json();\n } catch {\n return json({ error: \"Invalid JSON body\" }, 400);\n }\n\n const {\n message,\n agent_id,\n conversation_id: incomingConversationId,\n workspace,\n record_run,\n campaign: campaignId,\n } = body as {\n message?: string;\n agent_id?: string;\n conversation_id?: string;\n workspace?: string;\n record_run?: boolean;\n campaign?: string;\n };\n\n if (!message || typeof message !== \"string\" || message.trim() === \"\") {\n return json({ error: \"message field is required and must be a non-empty string\" }, 400);\n }\n\n // Part 14.1/14.2 — own the agent-run record + capture a structured trace.\n // wantRun is only true when an agent is attached AND the caller asked to record the run.\n const wantRun = !!(agent_id && record_run);\n let runId: string | null = null;\n if (wantRun) {\n runId = await startAgentRun(agent_id as string, \"manual\", message.trim()).catch(() => null);\n }\n // Trace steps accumulated as the tool loop runs (round/tool/message shapes).\n const trace: any[] = [];\n\n // Load LLM config\n const llmConfig = await getLLMConfig();\n\n // Load agent if specified\n let agent = null;\n if (agent_id) {\n agent = await getAgent(agent_id);\n }\n\n // Determine conversation ID\n const conversationId = incomingConversationId ?? crypto.randomUUID();\n\n // Load existing conversation history or start fresh\n let conversationHistory: ChatMessage[] = [];\n if (incomingConversationId) {\n const existing = await loadConversation(incomingConversationId);\n if (existing) {\n conversationHistory = existing.messages;\n }\n }\n\n // Add the new user message\n conversationHistory.push({ role: \"user\", content: message.trim() });\n\n // Build funnel-aware system prompt: inject live funnel state + top correlations\n // so the LLM always answers in causal/positional terms (TOFU/MOFU/BOFU).\n // Falls back silently if intelligence layer is unavailable.\n let funnelContext = \"\";\n try {\n const [summary, correlations] = await Promise.all([getFunnelSummary(), getCorrelations()]);\n const stageLines = summary.stages.map((s) =>\n ` ${s.stage} (${s.label}): health ${s.health_score}/100 · ${s.metrics.length} metrics · ${s.diagnosis}`\n ).join(\"\\n\");\n const bottleneck = summary.bottleneck ? `Bottleneck: ${summary.bottleneck.stage} — ${summary.bottleneck.reason}` : \"No bottleneck detected.\";\n const corrLines = correlations.slice(0, 5).map((c) =>\n ` - ${c.narrative} [${c.funnel_stage_from}→${c.funnel_stage_to}, ${c.strength}, lag ${c.lag_days}d, conf ${Math.round(c.confidence*100)}%]`\n ).join(\"\\n\");\n funnelContext = `\n\n## Current Funnel State (live, last 30d)\n${stageLines}\n${bottleneck}\nIntegrations connected: ${summary.total_integrations_connected} · Metrics observed: ${summary.total_metrics_observed}\n\n## Active Signal Correlations\n${corrLines || \" (no correlations yet — connect more integrations)\"}\n\n## Your Role\nYou are **Nexus** — a senior marketing & sales intelligence analyst embedded in the dashboard.\n- When a metric is mentioned, explain its funnel position (TOFU/MOFU/BOFU) and causal relationships\n- Proactively surface the highest-priority bottleneck in every response\n- Suggest specific actions tied to the integration data above\n- Use academic terminology: CAC, LTV, MQL, SQL, attribution, conversion velocity, message-market fit\n- End responses with a concrete next action when applicable\n`.trim();\n } catch { /* intelligence layer unavailable — proceed without context */ }\n\n // I3: agent role prompts. When agent.role is one of listener/topic/creator/analyst, prepend\n // the specialized prefix so the model behaves consistently with the role badge shown in the UI.\n const AGENT_ROLE_PROMPTS: Record = {\n listener: `You are a Listener Agent. Your job:\n1. Monitor every incoming signal from Gmail, Outlook, and LinkedIn for buyer intent.\n2. Flag mentions of: pain points, competitor references, pricing questions, budget signals.\n3. Surface weekly patterns (\"3 leads this week mentioned X\").\n4. Err on the side of flagging — never miss a commercially-relevant signal.`,\n topic: `You are a Topic Agent. Your job:\n1. Take buyer pain points + intent signals (from the Listener Agent or Signals workspace).\n2. Generate content ideas: blog titles, LinkedIn angles, email subject lines.\n3. Score each idea by search-volume potential (Ahrefs), funnel-stage fit, estimated ICE impact.\n4. Output a prioritized content backlog formatted as: Title | Funnel Stage | AARRR Lever | Est. ICE Score.\n5. Link every output to an active OKR Key Result when possible.`,\n creator: `You are a Creator Agent. Your job:\n1. Take approved topic ideas and draft complete marketing assets.\n2. Match brand voice: direct, data-backed, B2B SaaS tone.\n3. Blog posts: H1/H2 structure, 800–1200 words, CTA at end.\n4. LinkedIn posts: hook in first line, insight in body, question as CTA, max 250 words.\n5. Email sequences: subject + 3-email sequence (awareness → consideration → CTA).\n6. Always cite one data point from the connected integrations as social proof.`,\n analyst: `You are an Analyst Agent. Your job:\n1. Analyze the funnel summary across all 14 integrations.\n2. Identify the single biggest bottleneck in the current TOFU/MOFU/BOFU funnel.\n3. Propose the top 3 ICE-scored experiments to address it.\n4. Track experiment outcomes and update affected OKR Key Results.\n5. Produce a weekly growth digest: wins, losses, signals, recommended experiments.`,\n };\n const rolePrompt = agent?.role && AGENT_ROLE_PROMPTS[agent.role] ? AGENT_ROLE_PROMPTS[agent.role] : \"\";\n\n // Part 15.4 / W1 — CONTEXT STACK. Each source is fetched in its own try/catch (default \"\")\n // so a failure in any one never breaks chat. Non-empty blocks are appended additively to the\n // system prompt below, separated by blank lines — existing prompt content is left untouched.\n let companyContext = \"\";\n try { companyContext = await getCompanyContextString(2000); } catch { companyContext = \"\"; }\n let memoryContext = \"\";\n try { memoryContext = await getMemoryContextString(2000); } catch { memoryContext = \"\"; }\n let knowledgeContext = \"\";\n try { knowledgeContext = await loadEnabledKnowledgeContext({ target: \"chat\", maxChars: 4000 }); } catch { knowledgeContext = \"\"; }\n let skillsIndex = \"\";\n try { skillsIndex = await getSkillsIndex({ maxChars: 6000 }); } catch { skillsIndex = \"\"; }\n\n // GOAL AWARENESS — the FINAL GOAL is the OKR. Inject a concise summary of the ACTIVE objective(s)\n // + their key results (title · progress · target) so every answer is steered toward the objective.\n // Short (a few hundred chars) and fully guarded — never breaks chat if OKRs are unavailable.\n let goalContext = \"\";\n try {\n const objs = await listObjectives(workspace ? { workspace } : {});\n const active = objs.filter((o: any) => o.status === \"active\").slice(0, 3);\n if (active.length) {\n const lines = active.map((o: any) => {\n const krs = (o.key_results || []).slice(0, 3).map((k: any) =>\n ` • ${k.title} — ${k.progress}% (now ${k.current_value}${k.unit ? \" \" + k.unit : \"\"} → target ${k.target_value}${k.unit ? \" \" + k.unit : \"\"})`\n ).join(\"\\n\");\n return ` - ${o.title} [${o.progress}% complete]${krs ? \"\\n\" + krs : \"\"}`;\n }).join(\"\\n\");\n goalContext = (\n \"## FINAL GOAL — Active OKRs (the priority)\\n\" +\n \"Everything you do should advance these objectives + key results. Tie tasks, campaigns and recommendations back to them.\\n\" +\n lines\n ).slice(0, 1200);\n }\n } catch { goalContext = \"\"; }\n\n // ACTIVE PROJECT — when scoped to a campaign, load it + its objective + member agents and prime\n // the Director with it. scope carries campaign_id/objective_id so created tasks link to the project.\n const { context: campaignContext, scope: campaignScope } = await buildCampaignContext(campaignId);\n\n const baseSystemPrompt: string =\n agent?.system_prompt?.trim()\n ? agent.system_prompt.trim()\n : DEFAULT_SYSTEM_PROMPT;\n const systemPrompt = [\n rolePrompt, baseSystemPrompt, EXECUTION_GUIDE, campaignContext, goalContext, funnelContext,\n companyContext, memoryContext, knowledgeContext, skillsIndex,\n ]\n .filter((s) => s && s.length > 0)\n .join(\"\\n\\n\");\n\n // Determine tools — always give all tools; agents narrow by integration name via get_integration_data\n // Agent.tools stores integration names (e.g. [\"ga4\",\"gsc\"]), not tool function names,\n // so filtering by function name would always produce an empty list. Give all tools instead.\n const tools = buildTools();\n\n // Read temperature + frequency_penalty from app_config so Settings sliders are real, not cosmetic.\n let temperature: number | undefined = undefined;\n let frequencyPenalty: number | undefined = undefined;\n try {\n const cfgRows = await sqlite.execute(\"SELECT key, value FROM app_config WHERE key IN ('temperature','frequency_penalty') AND tenant_id = ?\", [currentTenantId()]);\n for (const row of cfgRows.rows) {\n const [k, v] = row as any[];\n const num = parseFloat(String(v).replace(/^\"|\"$/g, \"\"));\n if (!isNaN(num)) {\n if (k === \"temperature\") temperature = num;\n if (k === \"frequency_penalty\") frequencyPenalty = num;\n }\n }\n } catch { /* fall back to provider defaults */ }\n\n // Part 14.2 — wrap the LLM/tool work so we can finalize the agent-run record + trace on both\n // success and failure. Behavior is otherwise unchanged from the v551 baseline.\n try {\n // First LLM call\n let llmResult = await callLLM(llmConfig, conversationHistory, {\n systemPrompt,\n tools,\n maxTokens: 4096,\n temperature,\n frequencyPenalty,\n });\n\n const allToolCalls: any[] = [];\n\n // Handle tool use — bounded multi-round loop (was a single round, which truncated\n // any plan needing chained tool calls). Cap rounds to avoid runaway cost/latency.\n let toolRounds = 0;\n const MAX_TOOL_ROUNDS = 5;\n while (llmResult.tool_calls && llmResult.tool_calls.length > 0 && toolRounds < MAX_TOOL_ROUNDS) {\n toolRounds++;\n // Trace: mark the start of this tool round.\n trace.push({ t: \"round\", n: toolRounds, ts: new Date().toISOString() });\n // Record the assistant message with tool calls.\n // reasoning_content must be echoed back for DeepSeek thinking-mode multi-turn.\n const assistantMsg: any = {\n role: \"assistant\",\n content: llmResult.content || \"\",\n tool_calls: llmResult.tool_calls,\n };\n if ((llmResult as any).reasoning_content) {\n assistantMsg.reasoning_content = (llmResult as any).reasoning_content;\n }\n conversationHistory.push(assistantMsg);\n\n // Execute each tool call and collect results\n const toolResults: Array<{ tool_call_id: string; role: \"tool\"; content: string }> = [];\n\n for (const toolCall of llmResult.tool_calls) {\n allToolCalls.push(toolCall);\n\n let args: Record = {};\n try {\n args =\n typeof toolCall.function.arguments === \"string\"\n ? JSON.parse(toolCall.function.arguments)\n : toolCall.function.arguments ?? {};\n } catch {\n args = {};\n }\n\n // Measure wall-clock elapsed around the existing executeTool call (no double-call).\n const _start = performance.now();\n const toolResult = await executeTool(toolCall.function.name, args, llmConfig, campaignScope);\n const _ms = Math.round(performance.now() - _start);\n\n // Trace: record this tool call with short arg/result previews.\n trace.push({\n t: \"tool\",\n name: toolCall.function.name,\n args_preview: JSON.stringify(args).slice(0, 300),\n result_preview: String(toolResult).slice(0, 300),\n ok: true,\n ms: _ms,\n ts: new Date().toISOString(),\n });\n\n toolResults.push({\n tool_call_id: toolCall.id,\n role: \"tool\",\n content: toolResult,\n });\n }\n\n // Add tool results to history\n for (const tr of toolResults) {\n conversationHistory.push(tr as any);\n }\n\n // Second LLM call with tool results\n llmResult = await callLLM(llmConfig, conversationHistory, {\n systemPrompt,\n tools,\n maxTokens: 4096,\n temperature,\n frequencyPenalty,\n });\n }\n\n // Add final assistant response to history\n conversationHistory.push({\n role: \"assistant\",\n content: llmResult.content,\n });\n\n // Trace: record the final assistant message.\n trace.push({\n t: \"message\",\n role: \"assistant\",\n preview: String(llmResult.content).slice(0, 300),\n ts: new Date().toISOString(),\n });\n\n // Persist conversation\n await saveConversation(\n conversationId,\n conversationHistory,\n agent_id,\n workspace\n );\n\n // W1 — fire-and-forget: distill durable memory from this conversation. Never blocks the response.\n captureMemoryFromConversation(conversationHistory as any).catch(() => {});\n\n // Part 14.1/14.2 — finalize the agent-run record + persist the trace on success.\n if (wantRun && runId) {\n await finishAgentRun(runId, \"completed\", llmResult.content).catch(() => {});\n await sqlite.execute(\"UPDATE agent_runs SET trace=? WHERE id=? AND tenant_id = ?\", [JSON.stringify(trace), runId, currentTenantId()]).catch(() => {});\n }\n\n return json({\n response: llmResult.content,\n conversation_id: conversationId,\n tool_calls: allToolCalls.length > 0 ? allToolCalls : undefined,\n agent_id: agent_id ?? undefined,\n trace,\n });\n } catch (err: any) {\n // Part 14.1/14.2 — mark the run failed and persist whatever partial trace we captured.\n if (wantRun && runId) {\n await finishAgentRun(runId, \"failed\", undefined, String(err)).catch(() => {});\n await sqlite.execute(\"UPDATE agent_runs SET trace=? WHERE id=? AND tenant_id = ?\", [JSON.stringify(trace), runId, currentTenantId()]).catch(() => {});\n }\n // Preserve prior behavior: surface the error as a 500 JSON response.\n return json({ error: err?.message ?? \"Chat failed\" }, 500);\n }\n}\n\n// ─── Streaming chat handler (SSE) ────────────────────────────────────────────\n\nexport async function handleChatStream(req: Request): Promise {\n if (req.method === \"OPTIONS\") {\n return new Response(null, { status: 204, headers: CORS });\n }\n if (req.method !== \"POST\") {\n return new Response(JSON.stringify({ error: \"Method not allowed\" }), { status: 405, headers: { \"Content-Type\": \"application/json\", ...CORS } });\n }\n\n let body: any;\n try { body = await req.json(); } catch {\n return new Response(JSON.stringify({ error: \"Invalid JSON body\" }), { status: 400, headers: { \"Content-Type\": \"application/json\", ...CORS } });\n }\n\n const { message, agent_id, conversation_id: incomingConversationId, workspace, campaign: campaignId } = body as {\n message?: string; agent_id?: string; conversation_id?: string; workspace?: string; campaign?: string;\n };\n\n if (!message || typeof message !== \"string\" || message.trim() === \"\") {\n return new Response(JSON.stringify({ error: \"message field is required\" }), { status: 400, headers: { \"Content-Type\": \"application/json\", ...CORS } });\n }\n\n const llmConfig = await getLLMConfig();\n let agent = null;\n if (agent_id) { agent = await getAgent(agent_id); }\n\n const conversationId = incomingConversationId ?? crypto.randomUUID();\n let conversationHistory: ChatMessage[] = [];\n if (incomingConversationId) {\n const existing = await loadConversation(incomingConversationId);\n if (existing) conversationHistory = existing.messages;\n }\n conversationHistory.push({ role: \"user\", content: message.trim() });\n\n const baseSystemPrompt: string = agent?.system_prompt?.trim() ? agent.system_prompt.trim() :\n \"You are an AI assistant for AIHive Command Center, a marketing and sales platform.\";\n\n // W1 — inject unified company context (cheap single app_config read; safe on failure).\n const companyCtx = await getCompanyContextString(2000).catch(() => \"\");\n // ACTIVE PROJECT — the /world Director chat sends body.campaign (campaign id). Load that project\n // + its parent objective + member agents so the Director answers WITH project context instead of\n // generically. scope carries campaign_id/objective_id so created tasks link back to the project.\n const { context: campaignContext, scope: campaignScope } = await buildCampaignContext(campaignId);\n // Ground the streaming chat in its REAL tool capabilities (same guide as handleChat) so it\n // actually runs create_agent/create_campaign/create_task/prepare_* instead of narrating.\n const streamSystemPrompt = [baseSystemPrompt, EXECUTION_GUIDE, campaignContext, companyCtx]\n .filter((s) => s && s.trim()).join(\"\\n\\n\");\n\n // The streaming chat gets the SAME toolset as handleChat so it can DO work mid-stream.\n const tools = buildTools();\n\n const encoder = new TextEncoder();\n const stream = new ReadableStream({\n async start(controller) {\n const send = (data: object) => {\n controller.enqueue(encoder.encode(`data: ${JSON.stringify(data)}\\n\\n`));\n };\n\n let fullResponse = \"\";\n try {\n // ── TOOL LOOP (mirrors handleChat's executeTool loop, but over SSE) ──────────\n // Phase 1 (tool rounds): call the model NON-streaming so we can detect tool_calls.\n // For each requested tool we emit a {tool} event, execute it (create_agent /\n // create_campaign / create_task / prepare_* …), append the result, and loop —\n // exactly like handleChat. Any assistant prose accompanying a tool round is emitted\n // as a {delta} so no text is lost.\n // Phase 2 (final answer): once the model stops requesting tools, we STREAM the\n // closing assistant turn token-by-token via callLLMStream — preserving the existing\n // text streaming behaviour for the user-visible answer.\n const MAX_TOOL_ROUNDS = 5;\n let toolRounds = 0;\n\n while (toolRounds < MAX_TOOL_ROUNDS) {\n const llmResult = await callLLM(llmConfig, conversationHistory, {\n systemPrompt: streamSystemPrompt, tools, maxTokens: 4096,\n });\n\n // No tools requested → the model is ready to answer. Break to the streaming pass,\n // which re-prompts once and streams the final text token-by-token.\n if (!llmResult.tool_calls || llmResult.tool_calls.length === 0) break;\n\n // Tool round: surface any assistant prose, then run + report each tool call.\n toolRounds++;\n if (llmResult.content) {\n fullResponse += llmResult.content;\n send({ delta: llmResult.content, done: false });\n }\n const assistantMsg: any = {\n role: \"assistant\",\n content: llmResult.content || \"\",\n tool_calls: llmResult.tool_calls,\n };\n if ((llmResult as any).reasoning_content) {\n assistantMsg.reasoning_content = (llmResult as any).reasoning_content;\n }\n conversationHistory.push(assistantMsg);\n\n for (const toolCall of llmResult.tool_calls) {\n let args: Record = {};\n try {\n args = typeof toolCall.function.arguments === \"string\"\n ? JSON.parse(toolCall.function.arguments)\n : toolCall.function.arguments ?? {};\n } catch { args = {}; }\n\n // run_campaign defaults to the campaign-scoped chat's campaign — make sure the streamed\n // frame ALWAYS carries that resolved campaign_id (in args) so the world entry can fire the\n // real run even when the model didn't pass campaign_id explicitly.\n if (toolCall.function.name === \"run_campaign\" && args.campaign_id == null && campaignScope?.campaign_id) {\n args = { ...args, campaign_id: campaignScope.campaign_id };\n }\n\n // Tell the client a tool is running (name + arg preview) before we execute it.\n send({ tool: { name: toolCall.function.name, args, status: \"running\" } });\n\n const toolResult = await executeTool(toolCall.function.name, args, llmConfig, campaignScope);\n\n // Report completion + a short result preview. For run_campaign also lift the resolved\n // campaign_id onto the {tool} frame itself so the world entry can act on it (fire the run).\n const doneFrame: any = { name: toolCall.function.name, status: \"done\", result_preview: String(toolResult).slice(0, 300) };\n if (toolCall.function.name === \"run_campaign\") {\n let cid: string | undefined;\n try { cid = JSON.parse(toolResult)?.campaign_id; } catch { /* keep undefined */ }\n cid = cid ?? args.campaign_id ?? campaignScope?.campaign_id;\n if (cid != null) { doneFrame.campaign_id = String(cid); doneFrame.args = { campaign_id: String(cid) }; }\n }\n send({ tool: doneFrame });\n\n conversationHistory.push({\n tool_call_id: toolCall.id,\n role: \"tool\",\n content: toolResult,\n } as any);\n }\n // loop again — the model now sees the tool results\n }\n\n // Phase 2 — STREAM the final assistant turn token-by-token (existing text streaming).\n let streamed = \"\";\n const gen = callLLMStream(llmConfig, conversationHistory, {\n systemPrompt: streamSystemPrompt, maxTokens: 4096,\n });\n for await (const token of gen) {\n if (token) { streamed += token; fullResponse += token; send({ delta: token, done: false }); }\n }\n conversationHistory.push({ role: \"assistant\", content: streamed });\n\n await saveConversation(conversationId, conversationHistory, agent_id, workspace);\n captureMemoryFromConversation(conversationHistory as any).catch(() => {});\n send({ delta: \"\", done: true, response: fullResponse, conversation_id: conversationId });\n } catch (err: any) {\n send({ delta: \"\", done: true, error: err?.message ?? \"Stream error\", conversation_id: conversationId });\n } finally {\n controller.close();\n }\n },\n });\n\n return new Response(stream, {\n headers: {\n \"Content-Type\": \"text/event-stream\",\n \"Cache-Control\": \"no-cache\",\n \"Connection\": \"keep-alive\",\n ...CORS,\n },\n });\n}\n\n// ─── Conversation history CRUD ────────────────────────────────────────────────\n\nexport async function listConversations(): Promise {\n try {\n const result = await sqlite.execute(\n `SELECT id, workspace, agent_id, created_at, updated_at, messages\n FROM conversations WHERE tenant_id = ? ORDER BY updated_at DESC LIMIT 50`,\n [currentTenantId()]\n );\n return result.rows.map((row) => {\n const obj: Record = {};\n for (let i = 0; i < result.columns.length; i++) obj[result.columns[i]] = row[i];\n // Extract first user message as title\n let firstMsg = \"\";\n try {\n const msgs: ChatMessage[] = JSON.parse(String(obj.messages || \"[]\"));\n const first = msgs.find((m) => m.role === \"user\");\n firstMsg = first ? String(first.content).slice(0, 60) : \"\";\n } catch {}\n return {\n id: obj.id,\n workspace: obj.workspace,\n agent_id: obj.agent_id,\n created_at: obj.created_at,\n updated_at: obj.updated_at,\n title: firstMsg || \"(no messages)\",\n };\n });\n } catch { return []; }\n}\n\nexport async function getConversationById(id: string): Promise {\n try {\n const result = await sqlite.execute(\n \"SELECT id, workspace, agent_id, created_at, updated_at, messages FROM conversations WHERE id = ? AND tenant_id = ?\",\n [id, currentTenantId()]\n );\n if (!result.rows?.length) return null;\n const obj: Record = {};\n for (let i = 0; i < result.columns.length; i++) obj[result.columns[i]] = result.rows[0][i];\n let messages: ChatMessage[] = [];\n try { messages = JSON.parse(String(obj.messages || \"[]\")); } catch {}\n return { id: obj.id, workspace: obj.workspace, agent_id: obj.agent_id,\n created_at: obj.created_at, updated_at: obj.updated_at, messages };\n } catch { return null; }\n}\n\nexport async function deleteConversation(id: string): Promise {\n try {\n await sqlite.execute(\"DELETE FROM conversations WHERE id = ? AND tenant_id = ?\", [id, currentTenantId()]);\n return true;\n } catch { return false; }\n}\n","import { sqlite, currentTenantId } from \"../db.ts\";\nimport { getLLMConfig, callLLM, callLLMStream } from \"./llm.ts\";\nimport { getFunnelSummary, getCorrelations } from \"./intelligence.ts\";\nimport { listAgents, createAgent, getAgent, deleteAgent, startAgentRun, finishAgentRun } from \"./agents.ts\";\nimport { getCredentials, handleIntegrationProxy } from \"./integrations.ts\";\nimport { enqueueAction } from \"./actions.ts\";\nimport { createTask, listTasks, updateTask, deleteTask } from \"./tasks.ts\";\nimport { createCalendarEntry } from \"./calendar_unified.ts\";\nimport { listObjectives, createObjective } from \"./okrs.ts\";\nimport { createCampaign, deleteCampaign, listCampaigns, getCampaign, getCampaignAgents } from \"./campaigns.ts\";\nimport { createExperiment } from \"./experiments.ts\";\nimport { getCompanyContextString } from \"./company.ts\";\nimport { getMemoryContextString, captureMemoryFromConversation } from \"./memory.ts\";\nimport { loadEnabledKnowledgeContext } from \"./knowledge.ts\";\nimport { getSkillsIndex, applySkillContext, getSkillReference, getSkillReferenceList } from \"./skills.ts\";\nimport type { ChatMessage } from \"../types.ts\";\n\n// ─── Constants ────────────────────────────────────────────────────────────────\n\nconst CORS = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"GET, POST, PUT, DELETE, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type, Authorization\",\n};\n\nconst DEFAULT_SYSTEM_PROMPT =\n \"You are an AI assistant for this marketing and sales platform. \" +\n \"You can analyze data, create agents, and execute tasks across integrations.\";\n\n// Grounds the model in its REAL execution capabilities so it actually uses the tools instead of\n// hallucinating that it is \"blocked\" or \"needs credentials\" (the platform already holds them).\nconst EXECUTION_GUIDE =\n \"EXECUTION CAPABILITIES — IMPORTANT: You can actually DO work via tools. NEVER claim you are \" +\n \"'blocked', that you 'need credentials/site details', or that there is a 'security boundary you cannot bypass' — \" +\n \"the platform already stores all integration credentials. \" +\n \"PUBLIC/IRREVERSIBLE actions QUEUE for one-click human approval in the Approvals inbox — call the matching tool: \" +\n \"prepare_wordpress_post (publish to WordPress), prepare_email (send via Gmail), prepare_outlook_email (send via Outlook), prepare_linkedin_post (post to LinkedIn). \" +\n \"INTERNAL/REVERSIBLE actions run IMMEDIATELY (no approval): create_campaign (start a campaign/project), create_task and schedule_calendar_event (create work), \" +\n \"create_drive_folder / create_google_doc / create_google_sheet (create files in Google Drive/Docs/Sheets). \" +\n \"Use get_integration_data to read live data. None of these require the user to provide credentials. \" +\n \"When the user asks you to publish or send something, call the appropriate prepare_* tool with the FULL content, then tell them it is queued in Approvals. \" +\n \"When the user asks to create a campaign/project or a file, call the matching create_* tool and confirm it was created. \" +\n \"RUNNING THE WORK: when the user asks you to RUN / DO / EXECUTE / START the project's work (e.g. 'run a funnel analysis', 'do it', 'kick it off'), \" +\n \"you MUST call run_campaign — do NOT just narrate or fabricate the work yourself. run_campaign hands off to the member agents, who actually do the work and queue the deliverable to Approvals. \" +\n \"TEAM MANAGEMENT: you can also REMOVE work. Use list_campaigns / list_agents to find ids, then delete_campaign (removes the campaign AND its member agents) or delete_agent (removes one standalone agent). \" +\n \"Deletes are permanent — when the user clearly asks to delete/remove something, just do it and confirm; if they are vague about which one, list them first and ask. \" +\n \"Do NOT fabricate step-by-step 'execution logs', and do NOT ask the user for credentials.\";\n\n// ─── Helpers ──────────────────────────────────────────────────────────────────\n\nfunction json(data: any, status = 200): Response {\n return new Response(JSON.stringify(data), {\n status,\n headers: { \"Content-Type\": \"application/json\", ...CORS },\n });\n}\n\n// ─── Active project (campaign) context ─────────────────────────────────────────\n// When the Director chat is scoped to a campaign (frontend sends body.campaign = campaign id),\n// load that campaign + its parent OKR objective + its member agents and render a concise\n// \"ACTIVE PROJECT\" block. Injected into the system prompt so the Director directs THAT project and\n// treats the linked objective as the goal. Also returns campaign_id + objective_id so tasks the\n// chat creates can be linked back to the project. Fully guarded — never throws.\ntype CampaignScope = { campaign_id?: string; objective_id?: string };\n\nasync function buildCampaignContext(\n campaignId?: string\n): Promise<{ context: string; scope: CampaignScope }> {\n if (!campaignId || typeof campaignId !== \"string\" || !campaignId.trim()) {\n return { context: \"\", scope: {} };\n }\n try {\n const campaign = await getCampaign(campaignId.trim());\n if (!campaign) return { context: \"\", scope: {} };\n\n const scope: CampaignScope = { campaign_id: String(campaign.id) };\n\n // Parent objective (the FINAL goal). listObjectives has no get-by-id → list + filter.\n let objLine = \"\";\n if (campaign.objective_id != null) {\n scope.objective_id = String(campaign.objective_id);\n try {\n const objs = await listObjectives({});\n const obj = objs.find((o: any) => String(o.id) === String(campaign.objective_id));\n if (obj) {\n const krs = (obj.key_results || []).slice(0, 3).map((k: any) =>\n ` • ${k.title} — ${k.progress}% (now ${k.current_value}${k.unit ? \" \" + k.unit : \"\"} → target ${k.target_value}${k.unit ? \" \" + k.unit : \"\"})`\n ).join(\"\\n\");\n objLine =\n `\\n- Parent objective (the FINAL goal): \"${obj.title}\" [${obj.progress}% complete]` +\n (krs ? \"\\n Key results:\\n\" + krs : \"\");\n }\n } catch { /* objective unavailable — skip */ }\n }\n\n // Member agents the Director orchestrates.\n let agentsLine = \"\";\n try {\n const members = (campaign.agents && campaign.agents.length\n ? campaign.agents\n : await getCampaignAgents(String(campaign.id))) as any[];\n const names = members\n .map((m) => {\n const nm = m?.agent?.name || m?.role || \"agent\";\n const role = m?.role && m?.agent?.name ? ` (${m.role})` : \"\";\n return `${nm}${role}`;\n })\n .filter(Boolean);\n if (names.length) agentsLine = `\\n- Member agents you direct: ${names.join(\", \")}`;\n } catch { /* members unavailable — skip */ }\n\n const context = (\n \"## ACTIVE PROJECT — you are directing THIS campaign\\n\" +\n \"You are the DIRECTOR of the project below. Direct your member agents and steer every action toward this project's goal and its parent objective. \" +\n \"When you create tasks, campaigns or content, scope them to THIS project.\\n\" +\n \"When the user asks you to RUN / DO / EXECUTE / START this project's work (e.g. 'run a funnel analysis', 'do it', 'go'), CALL run_campaign (it defaults to this campaign) — the member agents then actually do the work and the deliverable lands in Approvals. Do NOT just narrate or pretend to do the work yourself.\\n\" +\n `- Project (campaign): \"${campaign.name}\"` +\n (campaign.goal ? `\\n- Goal: ${campaign.goal}` : \"\") +\n (campaign.channel ? `\\n- Channel: ${campaign.channel}` : \"\") +\n (campaign.audience ? `\\n- Audience: ${campaign.audience}` : \"\") +\n (campaign.workspace ? `\\n- Workspace: ${campaign.workspace}` : \"\") +\n objLine +\n agentsLine\n ).slice(0, 2000);\n\n return { context, scope };\n } catch {\n return { context: \"\", scope: {} };\n }\n}\n\nfunction rowToConversation(\n columns: string[],\n row: any[]\n): { id: string; messages: ChatMessage[]; agent_id?: string; workspace?: string } {\n const obj: Record = {};\n for (let i = 0; i < columns.length; i++) {\n obj[columns[i]] = row[i];\n }\n\n let messages: ChatMessage[] = [];\n if (obj.messages) {\n try {\n messages = JSON.parse(obj.messages);\n } catch {\n messages = [];\n }\n }\n\n return {\n id: obj.id,\n messages,\n agent_id: obj.agent_id ?? undefined,\n workspace: obj.workspace ?? undefined,\n };\n}\n\n// ─── Conversation DB ops ──────────────────────────────────────────────────────\n\nasync function loadConversation(\n conversationId: string\n): Promise<{ id: string; messages: ChatMessage[]; agent_id?: string; workspace?: string } | null> {\n const result = await sqlite.execute(\n \"SELECT id, messages, agent_id, workspace FROM conversations WHERE id = ? AND tenant_id = ?\",\n [conversationId, currentTenantId()]\n );\n\n if (!result.rows || result.rows.length === 0) {\n return null;\n }\n\n return rowToConversation(result.columns, result.rows[0]);\n}\n\nasync function saveConversation(\n id: string,\n messages: ChatMessage[],\n agentId?: string,\n workspace?: string\n): Promise {\n const now = new Date().toISOString();\n const messagesJson = JSON.stringify(messages);\n\n const existing = await sqlite.execute(\n \"SELECT id FROM conversations WHERE id = ? AND tenant_id = ?\",\n [id, currentTenantId()]\n );\n\n const ws = workspace ?? \"general\";\n if (existing.rows && existing.rows.length > 0) {\n await sqlite.execute(\n \"UPDATE conversations SET messages = ?, agent_id = ?, workspace = ?, updated_at = ? WHERE id = ? AND tenant_id = ?\",\n [messagesJson, agentId ?? null, ws, now, id, currentTenantId()]\n );\n } else {\n await sqlite.execute(\n `INSERT INTO conversations (id, messages, agent_id, workspace, created_at, updated_at, tenant_id)\n VALUES (?, ?, ?, ?, ?, ?, ?)`,\n [id, messagesJson, agentId ?? null, ws, now, now, currentTenantId()]\n );\n }\n}\n\n// ─── Tool definitions ─────────────────────────────────────────────────────────\n\nexport function buildTools(): any[] {\n return [\n {\n type: \"function\",\n function: {\n name: \"get_integration_data\",\n description:\n \"Fetch data from ANY connected integration. Marketing/SEO: ahrefs, ga4, gsc, youtube, wordpress, producthunt, linkedin. \" +\n \"Microsoft 365 (all share one connected account): outlook (inbox/calendar/contacts), onedrive (files), \" +\n \"teams (teams/channels/messages/chats), excel (workbooks/worksheets/range). Google Workspace: gmail, drive, sheets, docs, calendar.\",\n parameters: {\n type: \"object\",\n properties: {\n name: {\n type: \"string\",\n description:\n \"Integration name: ahrefs, ga4, gsc, linkedin, outlook, onedrive, teams, excel, wordpress, youtube, producthunt, gmail, drive, sheets, docs, calendar\",\n },\n endpoint: {\n type: \"string\",\n description:\n \"The endpoint within the integration, e.g. onedrive 'recent'/'files'/'search'; teams 'teams'/'channels'/'messages'/'chats'; excel 'workbooks'/'worksheets'/'range'; outlook 'inbox'/'calendar'/'contacts'; ahrefs 'site-explorer'/'keywords'; gsc 'search-analytics'; linkedin 'posts'.\",\n },\n params: {\n type: \"object\",\n description: \"Additional query parameters to forward to the API\",\n additionalProperties: { type: \"string\" },\n },\n },\n required: [\"name\", \"endpoint\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"create_agent\",\n description:\n \"Create a new AI agent in the platform with a specific role, workspace, and toolset.\",\n parameters: {\n type: \"object\",\n properties: {\n name: {\n type: \"string\",\n description: \"Name for the new agent\",\n },\n workspace: {\n type: \"string\",\n enum: [\"marketing\", \"sales\"],\n description: \"The workspace this agent belongs to\",\n },\n description: {\n type: \"string\",\n description: \"Short description of what this agent does\",\n },\n system_prompt: {\n type: \"string\",\n description: \"System prompt that defines the agent's behavior\",\n },\n tools: {\n type: \"array\",\n items: { type: \"string\" },\n description:\n \"List of tool names the agent has access to (e.g. ['get_integration_data', 'compose_content'])\",\n },\n },\n required: [\"name\", \"workspace\", \"description\", \"system_prompt\", \"tools\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"list_agents\",\n description: \"List all AI agents, optionally filtered by workspace.\",\n parameters: {\n type: \"object\",\n properties: {\n workspace: {\n type: \"string\",\n enum: [\"marketing\", \"sales\"],\n description: \"Filter agents by workspace (optional)\",\n },\n },\n required: [],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"list_objectives\",\n description: \"List OKR objectives (id + title). Use before create_campaign to link a campaign to an existing objective.\",\n parameters: {\n type: \"object\",\n properties: {\n workspace: { type: \"string\", enum: [\"marketing\", \"sales\"], description: \"Filter by workspace (optional)\" },\n },\n required: [],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"list_campaigns\",\n description: \"List campaigns (id, name, status, mode, workspace). Use this to resolve a campaign's id before delete_campaign, or to report campaign status.\",\n parameters: {\n type: \"object\",\n properties: {\n workspace: { type: \"string\", enum: [\"marketing\", \"sales\"], description: \"Filter by workspace (optional)\" },\n },\n required: [],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"delete_campaign\",\n description: \"Permanently delete a campaign AND all of its member agents. Provide campaign_id (preferred) or name. This is irreversible — confirm with the user first unless they were explicit about deleting it.\",\n parameters: {\n type: \"object\",\n properties: {\n campaign_id: { type: \"string\", description: \"The campaign id to delete (preferred)\" },\n name: { type: \"string\", description: \"The campaign name — used to resolve the id when campaign_id is omitted\" },\n },\n required: [],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"delete_agent\",\n description: \"Permanently delete a standalone AI agent by agent_id (preferred) or name. Irreversible — confirm first unless the user was explicit. To remove a campaign's member agents, delete the campaign instead.\",\n parameters: {\n type: \"object\",\n properties: {\n agent_id: { type: \"string\", description: \"The agent id to delete (preferred)\" },\n name: { type: \"string\", description: \"The agent name — used to resolve the id when agent_id is omitted\" },\n },\n required: [],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"compose_content\",\n description:\n \"Generate marketing or sales content such as blog posts, LinkedIn updates, emails, or YouTube descriptions.\",\n parameters: {\n type: \"object\",\n properties: {\n type: {\n type: \"string\",\n enum: [\"blog\", \"linkedin\", \"email\", \"youtube_description\"],\n description: \"The type of content to generate\",\n },\n topic: {\n type: \"string\",\n description: \"The topic or subject of the content\",\n },\n tone: {\n type: \"string\",\n description:\n \"The tone of voice for the content (e.g. 'professional', 'casual', 'persuasive')\",\n },\n },\n required: [\"type\", \"topic\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"prepare_email\",\n description:\n \"Draft an email and QUEUE it for human approval (Gmail). It is NOT sent until the user approves it in Approvals. Use this to actually act on email outreach.\",\n parameters: {\n type: \"object\",\n properties: {\n to: { type: \"string\", description: \"Recipient email address\" },\n subject: { type: \"string\", description: \"Email subject\" },\n body: { type: \"string\", description: \"Email body (plain text)\" },\n cc: { type: \"string\", description: \"Optional CC address\" },\n task_id: { type: \"string\", description: \"Optional task id to mark done once sent\" },\n },\n required: [\"to\", \"subject\", \"body\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"prepare_linkedin_post\",\n description:\n \"Draft a LinkedIn post and QUEUE it for human approval. NOT posted until the user approves.\",\n parameters: {\n type: \"object\",\n properties: {\n text: { type: \"string\", description: \"The post text\" },\n task_id: { type: \"string\", description: \"Optional task id to mark done once posted\" },\n },\n required: [\"text\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"prepare_wordpress_post\",\n description:\n \"Draft a WordPress blog post and QUEUE it for human approval. NOT published until the user approves.\",\n parameters: {\n type: \"object\",\n properties: {\n title: { type: \"string\", description: \"Post title\" },\n content: { type: \"string\", description: \"Post HTML/markdown content\" },\n status: { type: \"string\", enum: [\"publish\", \"draft\"], description: \"publish (default) or draft\" },\n task_id: { type: \"string\", description: \"Optional task id to mark done once published\" },\n },\n required: [\"title\", \"content\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"create_task\",\n description:\n \"Create a task in the task board. Runs immediately (internal, no approval). Optionally link it to an objective, key result, or experiment.\",\n parameters: {\n type: \"object\",\n properties: {\n title: { type: \"string\" },\n description: { type: \"string\" },\n workspace: { type: \"string\", enum: [\"marketing\", \"sales\"] },\n priority: { type: \"string\", enum: [\"low\", \"medium\", \"high\", \"urgent\"] },\n due_date: { type: \"string\", description: \"YYYY-MM-DD\" },\n objective_id: { type: \"string\" },\n key_result_id: { type: \"string\" },\n experiment_id: { type: \"string\" },\n },\n required: [\"title\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"list_tasks\",\n description: \"List tasks (id, title, status, priority, due_date, campaign_id). Use to find a task's id before update_task / delete_task, or to report the board.\",\n parameters: {\n type: \"object\",\n properties: {\n workspace: { type: \"string\", enum: [\"marketing\", \"sales\"] },\n status: { type: \"string\", enum: [\"todo\", \"doing\", \"done\", \"blocked\"] },\n campaign_id: { type: \"string\" },\n },\n required: [],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"update_task\",\n description: \"Update / modify an existing task — change its title, description, status (todo|doing|done|blocked), priority, due_date, assignee, or its objective/key-result/campaign links. Provide task_id (use list_tasks to find it).\",\n parameters: {\n type: \"object\",\n properties: {\n task_id: { type: \"string\", description: \"The task id to update\" },\n title: { type: \"string\" },\n description: { type: \"string\" },\n status: { type: \"string\", enum: [\"todo\", \"doing\", \"done\", \"blocked\"] },\n priority: { type: \"string\", enum: [\"low\", \"medium\", \"high\", \"urgent\"] },\n due_date: { type: \"string\", description: \"YYYY-MM-DD\" },\n assignee: { type: \"string\" },\n objective_id: { type: \"string\" },\n key_result_id: { type: \"string\" },\n campaign_id: { type: \"string\" },\n },\n required: [\"task_id\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"delete_task\",\n description: \"Permanently delete a task by task_id (use list_tasks to find it). Irreversible — when the user clearly asks to delete it, just do it; if vague about which, list_tasks first.\",\n parameters: {\n type: \"object\",\n properties: { task_id: { type: \"string\", description: \"The task id to delete\" } },\n required: [\"task_id\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"schedule_calendar_event\",\n description:\n \"Add an event to the unified calendar. Runs immediately (internal, no approval).\",\n parameters: {\n type: \"object\",\n properties: {\n title: { type: \"string\" },\n event_date: { type: \"string\", description: \"YYYY-MM-DD\" },\n event_time: { type: \"string\", description: \"HH:MM (optional)\" },\n workspace: { type: \"string\", enum: [\"marketing\", \"sales\"] },\n },\n required: [\"title\", \"event_date\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"create_campaign\",\n description:\n \"Create a REAL multi-agent campaign: a DIRECTOR orchestrates several member agents that cooperate to achieve the goal. \" +\n \"Runs immediately (internal, no approval). \" +\n \"By default the campaign gets a 3-agent research -> strategy -> compose pipeline; supply a custom 'agents' array to define your own team. \" +\n \"Set 'mode' to 'sequential' (each agent hands off to the next, default) or 'parallel' (agents work independently and the director merges).\",\n parameters: {\n type: \"object\",\n properties: {\n name: { type: \"string\", description: \"Campaign name\" },\n workspace: { type: \"string\", enum: [\"marketing\", \"sales\"] },\n channel: { type: \"string\", description: \"linkedin | email | content | blog | seo | social | paid | mixed\" },\n goal: { type: \"string\", description: \"What the campaign aims to achieve\" },\n audience: { type: \"string\", description: \"Target audience (optional)\" },\n objective_id: { type: \"string\", description: \"Optional OKR objective id this campaign serves (the FINAL goal). Resolve it first via list_objectives, then pass its id so a completed campaign run advances that objective.\" },\n mode: { type: \"string\", enum: [\"sequential\", \"parallel\"], description: \"How the member agents cooperate. 'sequential' (default): each agent hands its output to the next. 'parallel': agents work independently and the director merges results.\" },\n agents: {\n type: \"array\",\n description: \"Optional custom team of member agents the director orchestrates. Omit to use the default research -> strategy -> compose pipeline.\",\n items: {\n type: \"object\",\n properties: {\n name: { type: \"string\", description: \"Member agent name, e.g. 'Research Agent'\" },\n role: { type: \"string\", description: \"Short role label, e.g. research | strategy | compose\" },\n system_prompt: { type: \"string\", description: \"Instructions defining what this member agent does\" },\n },\n required: [\"name\"],\n },\n },\n },\n required: [\"name\", \"channel\", \"goal\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"run_campaign\",\n description:\n \"Execute the active campaign — the member agents actually DO the work (research/strategy/compose/etc.) and the result is queued to Approvals. \" +\n \"Call this when the user asks you to run / do / execute / start the project's work (e.g. 'run a funnel analysis').\",\n parameters: {\n type: \"object\",\n properties: {\n campaign_id: {\n type: \"string\",\n description: \"The campaign id to execute. Optional — defaults to the campaign-scoped chat's campaign.\",\n },\n },\n required: [],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"create_drive_folder\",\n description: \"Create a folder in Google Drive. Runs immediately (internal, no approval).\",\n parameters: {\n type: \"object\",\n properties: { name: { type: \"string\", description: \"Folder name\" } },\n required: [\"name\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"create_google_doc\",\n description: \"Create a Google Doc, optionally with initial content. Runs immediately (internal, no approval).\",\n parameters: {\n type: \"object\",\n properties: {\n title: { type: \"string\", description: \"Document title\" },\n content: { type: \"string\", description: \"Optional initial body text\" },\n },\n required: [\"title\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"create_google_sheet\",\n description: \"Create a Google Sheet. Runs immediately (internal, no approval).\",\n parameters: {\n type: \"object\",\n properties: { title: { type: \"string\", description: \"Spreadsheet title\" } },\n required: [\"title\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"prepare_outlook_email\",\n description:\n \"Draft an email and QUEUE it for human approval (Outlook/Microsoft 365). NOT sent until the user approves it in Approvals.\",\n parameters: {\n type: \"object\",\n properties: {\n to: { type: \"string\", description: \"Recipient email address\" },\n subject: { type: \"string\", description: \"Email subject\" },\n body: { type: \"string\", description: \"Email body (plain text)\" },\n cc: { type: \"string\", description: \"Optional CC address\" },\n task_id: { type: \"string\", description: \"Optional task id to mark done once sent\" },\n },\n required: [\"to\", \"subject\", \"body\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"get_file_content\",\n description:\n \"Read the full content of a single file or message from a connected integration by its id. \" +\n \"Use for drive/onedrive files or outlook/gmail messages — pass the integration name and the item's id \" +\n \"(get the id first via get_integration_data, e.g. listing drive 'files' or gmail 'inbox').\",\n parameters: {\n type: \"object\",\n properties: {\n integration: {\n type: \"string\",\n description: \"Integration name: drive, onedrive, outlook, gmail\",\n },\n id: {\n type: \"string\",\n description: \"The file id (drive/onedrive) or message id (outlook/gmail) to read\",\n },\n },\n required: [\"integration\", \"id\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"apply_skill\",\n description:\n \"Load a marketing/sales skill playbook into context. Call this with the skill name (from the \" +\n \"MARKETING SKILLS INDEX) when a request matches one, then follow the returned playbook.\",\n parameters: {\n type: \"object\",\n properties: {\n skill_name: { type: \"string\", description: \"The skill name to apply (from the skills index)\" },\n },\n required: [\"skill_name\"],\n },\n },\n },\n {\n type: \"function\",\n function: {\n name: \"get_skill_reference\",\n description:\n \"List a skill's reference documents (omit reference) or fetch one reference doc's content \" +\n \"(pass reference). Use after apply_skill when the playbook points to a named reference.\",\n parameters: {\n type: \"object\",\n properties: {\n skill_name: { type: \"string\", description: \"The skill name\" },\n reference: { type: \"string\", description: \"Optional reference doc name to fetch; omit to list available references\" },\n },\n required: [\"skill_name\"],\n },\n },\n },\n ];\n}\n\n// ─── Tool executor ────────────────────────────────────────────────────────────\n\n// Run a WRITE/mutate endpoint on an integration immediately, POSTing through the same proxy\n// the approval queue uses. Returns parsed JSON (or {error}).\nasync function runIntegrationWrite(integration: string, endpoint: string, payload: any): Promise {\n const creds = await getCredentials(integration);\n if (!creds) return { error: `Integration '${integration}' is not connected. Add it in Settings first.` };\n const params = new URLSearchParams({ endpoint });\n const req = new Request(\n `https://aihive.internal/api/integrations/${integration}/data?endpoint=${encodeURIComponent(endpoint)}`,\n { method: \"POST\", headers: { \"Content-Type\": \"application/json\" }, body: JSON.stringify(payload ?? {}) },\n );\n try {\n const res = await handleIntegrationProxy(req, integration, params, creds);\n return await res.json().catch(() => ({}));\n } catch (err: any) {\n return { error: err?.message ?? \"Integration call failed\" };\n }\n}\n\nexport async function executeTool(\n toolName: string,\n toolArgs: Record,\n llmConfig: Awaited>,\n // When the chat is scoped to a campaign/project, this carries its campaign_id (+ objective_id)\n // so created tasks link back to the project. Optional — preserves all existing callers.\n scope?: CampaignScope\n): Promise {\n switch (toolName) {\n case \"get_integration_data\": {\n const { name, endpoint, params = {} } = toolArgs as {\n name: string;\n endpoint: string;\n params?: Record;\n };\n\n const creds = await getCredentials(name);\n if (!creds) {\n return JSON.stringify({\n error: `Integration '${name}' is not configured. Please add credentials first.`,\n });\n }\n\n // Build a synthetic URLSearchParams from args\n const searchParams = new URLSearchParams({ endpoint });\n for (const [k, v] of Object.entries(params)) {\n searchParams.set(k, String(v));\n }\n\n // Build a minimal Request so handlers can read body if needed\n const syntheticReq = new Request(\"https://aihive.internal/proxy\", {\n method: \"GET\",\n });\n\n try {\n const proxyRes = await handleIntegrationProxy(\n syntheticReq,\n name,\n searchParams,\n creds\n );\n const data = await proxyRes.json();\n return JSON.stringify(data);\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Proxy request failed\" });\n }\n }\n\n case \"create_agent\": {\n const { name, workspace, description, system_prompt, tools } = toolArgs as {\n name: string;\n workspace: \"marketing\" | \"sales\";\n description: string;\n system_prompt: string;\n tools: string[];\n };\n\n try {\n const agent = await createAgent({\n name,\n workspace,\n description,\n system_prompt,\n tools,\n });\n return JSON.stringify({ success: true, agent });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to create agent\" });\n }\n }\n\n case \"list_agents\": {\n const { workspace } = toolArgs as { workspace?: string };\n try {\n const agents = await listAgents(workspace);\n return JSON.stringify({ agents });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to list agents\" });\n }\n }\n\n case \"compose_content\": {\n const { type, topic, tone = \"professional\" } = toolArgs as {\n type: \"blog\" | \"linkedin\" | \"email\" | \"youtube_description\";\n topic: string;\n tone?: string;\n };\n\n const contentPrompts: Record = {\n blog: `Write a comprehensive, SEO-friendly blog post about: \"${topic}\". Tone: ${tone}. Include an introduction, 3-5 sections with headers, and a conclusion.`,\n linkedin: `Write an engaging LinkedIn post about: \"${topic}\". Tone: ${tone}. Keep it under 300 words, include relevant hashtags, and a call-to-action.`,\n email: `Write a professional email about: \"${topic}\". Tone: ${tone}. Include subject line, greeting, body paragraphs, and a clear call-to-action.`,\n youtube_description: `Write an optimized YouTube video description for a video about: \"${topic}\". Tone: ${tone}. Include a hook, timestamps placeholder, links section, and relevant tags.`,\n };\n\n const prompt = contentPrompts[type] ?? `Write content about: \"${topic}\". Tone: ${tone}.`;\n\n // W1 — prepend unified company context so generated content stays on-brand.\n const companyCtx = await getCompanyContextString(1500).catch(() => \"\");\n\n try {\n const contentMessages: ChatMessage[] = [\n { role: \"user\", content: prompt },\n ];\n\n const result = await callLLM(llmConfig, contentMessages, {\n systemPrompt:\n (companyCtx ? companyCtx + \"\\n\\n---\\n\\n\" : \"\") +\n \"You are an expert content writer for marketing and sales teams. Produce high-quality, engaging content tailored to the specified format and tone.\",\n maxTokens: 2048,\n });\n\n // Persist the composed output as a REVIEW-ONLY deliverable so it doesn't evaporate.\n // integration:\"content\" is internal — approveAction simply ACCEPTS it (no external send).\n // Guarded so a queue failure never breaks the compose tool result.\n let action_id: number | undefined;\n try {\n const title = `Content: ${type} — ${String(topic).slice(0, 80)}`;\n const action = await enqueueAction({\n integration: \"content\",\n endpoint: \"review\",\n payload: { type, topic, tone, content: result.content },\n title,\n preview: String(result.content ?? \"\").slice(0, 600),\n source_type: \"chat\",\n });\n action_id = action?.id;\n } catch { /* non-critical — still return the composed content */ }\n\n return JSON.stringify({\n type,\n topic,\n tone,\n content: result.content,\n action_id,\n queued_for_review: action_id != null,\n note: action_id != null\n ? \"Composed and saved to Approvals as a review-only deliverable.\"\n : undefined,\n });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to compose content\" });\n }\n }\n\n case \"prepare_email\": {\n const { to, subject, body, cc, task_id } = toolArgs as any;\n if (!to || !subject) return JSON.stringify({ error: \"to + subject required\" });\n const action = await enqueueAction({\n integration: \"gmail\", endpoint: \"send\",\n payload: { to, subject, body: body ?? \"\", cc },\n title: `Email: ${subject}`,\n preview: `To: ${to}\\nSubject: ${subject}\\n\\n${String(body ?? \"\").slice(0, 600)}`,\n task_id, source_type: \"chat\",\n });\n return JSON.stringify({ queued: true, status: \"prepared\", action_id: action?.id, note: \"Email drafted and QUEUED FOR APPROVAL — it will NOT send until the user approves it in Approvals.\" });\n }\n\n case \"prepare_linkedin_post\": {\n const { text, task_id } = toolArgs as any;\n if (!text) return JSON.stringify({ error: \"text required\" });\n const action = await enqueueAction({\n integration: \"linkedin\", endpoint: \"create-post\",\n payload: { text },\n title: \"LinkedIn post\",\n preview: String(text).slice(0, 600),\n task_id, source_type: \"chat\", workspace: \"sales\",\n });\n return JSON.stringify({ queued: true, status: \"prepared\", action_id: action?.id, note: \"LinkedIn post drafted and QUEUED FOR APPROVAL.\" });\n }\n\n case \"prepare_wordpress_post\": {\n const { title, content, status, task_id } = toolArgs as any;\n if (!title || !content) return JSON.stringify({ error: \"title + content required\" });\n const action = await enqueueAction({\n integration: \"wordpress\", endpoint: \"publish-post\",\n payload: { title, content, status: status === \"draft\" ? \"draft\" : \"publish\" },\n title: `WordPress: ${title}`,\n preview: `${title}\\n\\n${String(content).slice(0, 600)}`,\n task_id, source_type: \"chat\",\n });\n return JSON.stringify({ queued: true, status: \"prepared\", action_id: action?.id, note: \"WordPress post drafted and QUEUED FOR APPROVAL.\" });\n }\n\n case \"create_task\": {\n const a = toolArgs as any;\n if (!a.title) return JSON.stringify({ error: \"title required\" });\n try {\n // Active-project scope: when this chat is directing a campaign, link the task to that\n // campaign (and its parent objective) unless the model explicitly supplied its own.\n const campaign_id = a.campaign_id ?? scope?.campaign_id;\n const objective_id = a.objective_id ?? scope?.objective_id;\n const task = await createTask({\n title: a.title, description: a.description ?? null,\n priority: a.priority, workspace: a.workspace ?? \"marketing\",\n due_date: a.due_date, objective_id,\n key_result_id: a.key_result_id, experiment_id: a.experiment_id,\n campaign_id,\n } as any);\n return JSON.stringify({ success: true, task });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to create task\" });\n }\n }\n\n case \"list_tasks\": {\n const a = toolArgs as { workspace?: string; status?: string; campaign_id?: string };\n try {\n const filters: any = {};\n if (a.workspace) filters.workspace = a.workspace;\n if (a.status) filters.status = a.status;\n if (a.campaign_id ?? scope?.campaign_id) filters.campaign_id = a.campaign_id ?? scope?.campaign_id;\n const tasks = await listTasks(filters);\n return JSON.stringify({ tasks: tasks.map((t: any) => ({ id: t.id, title: t.title, status: t.status, priority: t.priority, due_date: t.due_date, campaign_id: t.campaign_id })) });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to list tasks\" });\n }\n }\n\n case \"update_task\": {\n const a = toolArgs as any;\n if (!a.task_id) return JSON.stringify({ error: \"task_id required (use list_tasks to find it)\" });\n try {\n const patch: any = {};\n for (const k of [\"title\", \"description\", \"status\", \"priority\", \"due_date\", \"assignee\", \"objective_id\", \"key_result_id\", \"campaign_id\"]) {\n if (a[k] !== undefined) patch[k] = a[k];\n }\n const task = await updateTask(String(a.task_id), patch);\n return JSON.stringify(task ? { success: true, task } : { error: `Task ${a.task_id} not found.` });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to update task\" });\n }\n }\n\n case \"delete_task\": {\n const a = toolArgs as { task_id?: string };\n if (!a.task_id) return JSON.stringify({ error: \"task_id required (use list_tasks to find it)\" });\n try {\n const ok = await deleteTask(String(a.task_id));\n return JSON.stringify(ok ? { success: true, deleted_task: a.task_id, note: \"Task permanently deleted.\" } : { error: `Task ${a.task_id} not found or already deleted.` });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to delete task\" });\n }\n }\n\n case \"schedule_calendar_event\": {\n const a = toolArgs as any;\n if (!a.title || !a.event_date) return JSON.stringify({ error: \"title + event_date (YYYY-MM-DD) required\" });\n try {\n const entry = await createCalendarEntry({\n title: a.title, event_date: a.event_date, event_time: a.event_time ?? null,\n workspace: a.workspace ?? \"marketing\", event_type: \"custom\",\n } as any);\n return JSON.stringify({ success: true, entry });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to schedule event\" });\n }\n }\n\n case \"list_objectives\": {\n const { workspace } = toolArgs as { workspace?: string };\n try {\n const objectives = await listObjectives(workspace ? { workspace } : {});\n return JSON.stringify({ objectives: objectives.map((o: any) => ({ id: o.id, title: o.title, status: o.status, progress: o.progress })) });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to list objectives\" });\n }\n }\n\n case \"create_campaign\": {\n const a = toolArgs as any;\n if (!a.name) return JSON.stringify({ error: \"name required\" });\n const workspace = a.workspace === \"sales\" ? \"sales\" : \"marketing\";\n try {\n // A campaign is a REAL multi-agent system: a DIRECTOR orchestrates ordered member agents\n // that cooperate. Use the model-supplied team if given, else createCampaign falls back to its\n // default research -> strategy -> compose pipeline.\n const agents = Array.isArray(a.agents) && a.agents.length ? a.agents : undefined;\n // Link the campaign to an OKR objective when the model resolved one via list_objectives.\n // The tool delivers objective_id as a string; createCampaign expects a number — coerce + guard.\n const objNum = a.objective_id != null ? Number(a.objective_id) : NaN;\n const objective_id = Number.isFinite(objNum) ? objNum : undefined;\n const campaign = await createCampaign({\n name: a.name,\n workspace,\n goal: a.goal ?? a.description ?? undefined,\n channel: a.channel ?? undefined,\n audience: a.audience ?? undefined,\n mode: a.mode,\n objective_id,\n agents,\n });\n const members = (campaign?.agents || []) as any[];\n const memberNames = members\n .map((m) => m?.agent?.name || m?.role || \"agent\")\n .filter(Boolean);\n const mode = campaign?.mode || \"sequential\";\n const summary =\n `Created campaign \"${campaign?.name ?? a.name}\" with a director orchestrating ` +\n `${memberNames.length} member agent${memberNames.length === 1 ? \"\" : \"s\"} (${mode}): ` +\n `${memberNames.join(\" -> \")}.`;\n return JSON.stringify({ success: true, campaign, message: summary });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to create campaign\" });\n }\n }\n\n case \"run_campaign\": {\n // EXECUTE the active campaign — the member agents actually do the work and the deliverable\n // lands in Approvals. CRITICAL: we do NOT run the ~77s orchestration inline here — that would\n // block the request / time out the SSE stream. Instead we resolve the campaign id and return a\n // short \"kicking off\" message. The BROWSER fires the real run: the streamed {tool} frame carries\n // name:\"run_campaign\" + the resolved campaign_id (in args/result), and the world entry POSTs to\n // /api/campaigns/:id/runs so the user watches the team work in the World.\n const a = toolArgs as { campaign_id?: string };\n const id = a.campaign_id ?? scope?.campaign_id;\n if (!id) {\n return JSON.stringify({\n error:\n \"No campaign to run — open a project (campaign-scoped chat) or pass campaign_id. Use list_campaigns to find one.\",\n });\n }\n return JSON.stringify({\n kicking_off: true,\n campaign_id: String(id),\n message:\n \"Kicking off the team now — the agents are executing the campaign; watch them work in the World and the deliverable will land in Approvals.\",\n });\n }\n\n case \"list_campaigns\": {\n const { workspace } = toolArgs as { workspace?: string };\n try {\n const campaigns = await listCampaigns(workspace);\n return JSON.stringify({ campaigns: campaigns.map((c: any) => ({ id: c.id, name: c.name, status: c.status, mode: c.mode, workspace: c.workspace })) });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to list campaigns\" });\n }\n }\n\n case \"delete_campaign\": {\n const a = toolArgs as { campaign_id?: string; name?: string };\n try {\n let id = a.campaign_id;\n if (!id && a.name) {\n const all = await listCampaigns();\n const q = String(a.name).toLowerCase();\n const match = all.find((c: any) => String(c.name).toLowerCase() === q)\n || all.find((c: any) => String(c.name).toLowerCase().includes(q));\n id = match?.id;\n }\n if (!id) return JSON.stringify({ error: `No campaign found matching ${a.campaign_id ?? a.name ?? \"(none)\"}. Use list_campaigns to see ids.` });\n const members = await getCampaignAgents(id).catch(() => []);\n const ok = await deleteCampaign(id);\n return JSON.stringify(ok\n ? { success: true, deleted_campaign: id, deleted_member_agents: members.length, note: \"Campaign and its member agents were permanently deleted.\" }\n : { error: `Campaign ${id} not found or already deleted.` });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to delete campaign\" });\n }\n }\n\n case \"delete_agent\": {\n const a = toolArgs as { agent_id?: string; name?: string };\n try {\n let id = a.agent_id;\n let agent = id ? await getAgent(id) : null;\n if (!agent && a.name) {\n const all = await listAgents();\n const q = String(a.name).toLowerCase();\n agent = all.find((g: any) => String(g.name).toLowerCase() === q)\n || all.find((g: any) => String(g.name).toLowerCase().includes(q)) || null;\n id = agent?.id;\n }\n if (!id) return JSON.stringify({ error: `No agent found matching ${a.agent_id ?? a.name ?? \"(none)\"}. Use list_agents to see ids.` });\n const ok = await deleteAgent(id);\n return JSON.stringify(ok\n ? { success: true, deleted_agent: id, name: agent?.name ?? null, note: \"Agent permanently deleted.\" }\n : { error: `Agent ${id} not found or already deleted.` });\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Failed to delete agent\" });\n }\n }\n\n case \"create_drive_folder\": {\n const a = toolArgs as any;\n if (!a.name) return JSON.stringify({ error: \"name required\" });\n const data = await runIntegrationWrite(\"drive\", \"createFolder\", { name: a.name });\n return JSON.stringify(data?.error ? data : { success: true, file: data });\n }\n\n case \"create_google_doc\": {\n const a = toolArgs as any;\n if (!a.title) return JSON.stringify({ error: \"title required\" });\n const data = await runIntegrationWrite(\"docs\", \"create\", { title: a.title, content: a.content });\n return JSON.stringify(data?.error ? data : { success: true, doc: data });\n }\n\n case \"create_google_sheet\": {\n const a = toolArgs as any;\n if (!a.title) return JSON.stringify({ error: \"title required\" });\n const data = await runIntegrationWrite(\"sheets\", \"create\", { title: a.title });\n return JSON.stringify(data?.error ? data : { success: true, sheet: data });\n }\n\n case \"prepare_outlook_email\": {\n const a = toolArgs as any;\n if (!a.to || !a.subject) return JSON.stringify({ error: \"to + subject required\" });\n const action = await enqueueAction({\n integration: \"outlook\", endpoint: \"send\",\n payload: {\n message: {\n subject: a.subject,\n body: { contentType: \"Text\", content: a.body ?? \"\" },\n toRecipients: [{ emailAddress: { address: a.to } }],\n ...(a.cc ? { ccRecipients: [{ emailAddress: { address: a.cc } }] } : {}),\n },\n saveToSentItems: true,\n },\n title: `Outlook email: ${a.subject}`,\n preview: `To: ${a.to}\\nSubject: ${a.subject}\\n\\n${String(a.body ?? \"\").slice(0, 600)}`,\n task_id: a.task_id, source_type: \"chat\",\n });\n return JSON.stringify({ queued: true, status: \"prepared\", action_id: action?.id, note: \"Outlook email drafted and QUEUED FOR APPROVAL.\" });\n }\n\n case \"get_file_content\": {\n // Read a single file/message from an integration via the SAME read path as get_integration_data.\n // drive/onedrive use endpoint 'read'; outlook/gmail use endpoint 'message'. The id is passed as ?id=.\n const { integration, id } = toolArgs as { integration: string; id: string };\n if (!integration || !id) return JSON.stringify({ error: \"integration + id required\" });\n\n const creds = await getCredentials(integration);\n if (!creds) {\n return JSON.stringify({\n error: `Integration '${integration}' is not configured. Please add credentials first.`,\n });\n }\n\n const endpoint =\n integration === \"outlook\" || integration === \"gmail\" ? \"message\" : \"read\";\n\n const searchParams = new URLSearchParams({ endpoint });\n searchParams.set(\"id\", String(id));\n\n const syntheticReq = new Request(\"https://aihive.internal/proxy\", {\n method: \"GET\",\n });\n\n try {\n const proxyRes = await handleIntegrationProxy(\n syntheticReq,\n integration,\n searchParams,\n creds\n );\n const data = await proxyRes.json();\n return JSON.stringify(data);\n } catch (err: any) {\n return JSON.stringify({ error: err?.message ?? \"Proxy request failed\" });\n }\n }\n\n case \"apply_skill\": {\n const ctx = await applySkillContext(toolArgs.skill_name);\n return ctx || JSON.stringify({ error: \"No such skill\" });\n }\n\n case \"get_skill_reference\": {\n if (!toolArgs.reference) {\n return JSON.stringify({ references: getSkillReferenceList(toolArgs.skill_name) });\n }\n const r = getSkillReference(toolArgs.skill_name, toolArgs.reference);\n return r || JSON.stringify({ error: \"No such reference\", available: getSkillReferenceList(toolArgs.skill_name) });\n }\n\n default:\n return JSON.stringify({ error: `Unknown tool: ${toolName}` });\n }\n}\n\n// ─── Main chat handler ────────────────────────────────────────────────────────\n\nexport async function handleChat(req: Request): Promise {\n if (req.method === \"OPTIONS\") {\n return new Response(null, { status: 204, headers: CORS });\n }\n\n if (req.method !== \"POST\") {\n return json({ error: \"Method not allowed\" }, 405);\n }\n\n let body: any;\n try {\n body = await req.json();\n } catch {\n return json({ error: \"Invalid JSON body\" }, 400);\n }\n\n const {\n message,\n agent_id,\n conversation_id: incomingConversationId,\n workspace,\n record_run,\n campaign: campaignId,\n } = body as {\n message?: string;\n agent_id?: string;\n conversation_id?: string;\n workspace?: string;\n record_run?: boolean;\n campaign?: string;\n };\n\n if (!message || typeof message !== \"string\" || message.trim() === \"\") {\n return json({ error: \"message field is required and must be a non-empty string\" }, 400);\n }\n\n // Part 14.1/14.2 — own the agent-run record + capture a structured trace.\n // wantRun is only true when an agent is attached AND the caller asked to record the run.\n const wantRun = !!(agent_id && record_run);\n let runId: string | null = null;\n if (wantRun) {\n runId = await startAgentRun(agent_id as string, \"manual\", message.trim()).catch(() => null);\n }\n // Trace steps accumulated as the tool loop runs (round/tool/message shapes).\n const trace: any[] = [];\n\n // Load LLM config\n const llmConfig = await getLLMConfig();\n\n // Load agent if specified\n let agent = null;\n if (agent_id) {\n agent = await getAgent(agent_id);\n }\n\n // Determine conversation ID\n const conversationId = incomingConversationId ?? crypto.randomUUID();\n\n // Load existing conversation history or start fresh\n let conversationHistory: ChatMessage[] = [];\n if (incomingConversationId) {\n const existing = await loadConversation(incomingConversationId);\n if (existing) {\n conversationHistory = existing.messages;\n }\n }\n\n // Add the new user message\n conversationHistory.push({ role: \"user\", content: message.trim() });\n\n // Build funnel-aware system prompt: inject live funnel state + top correlations\n // so the LLM always answers in causal/positional terms (TOFU/MOFU/BOFU).\n // Falls back silently if intelligence layer is unavailable.\n let funnelContext = \"\";\n try {\n const [summary, correlations] = await Promise.all([getFunnelSummary(), getCorrelations()]);\n const stageLines = summary.stages.map((s) =>\n ` ${s.stage} (${s.label}): health ${s.health_score}/100 · ${s.metrics.length} metrics · ${s.diagnosis}`\n ).join(\"\\n\");\n const bottleneck = summary.bottleneck ? `Bottleneck: ${summary.bottleneck.stage} — ${summary.bottleneck.reason}` : \"No bottleneck detected.\";\n const corrLines = correlations.slice(0, 5).map((c) =>\n ` - ${c.narrative} [${c.funnel_stage_from}→${c.funnel_stage_to}, ${c.strength}, lag ${c.lag_days}d, conf ${Math.round(c.confidence*100)}%]`\n ).join(\"\\n\");\n funnelContext = `\n\n## Current Funnel State (live, last 30d)\n${stageLines}\n${bottleneck}\nIntegrations connected: ${summary.total_integrations_connected} · Metrics observed: ${summary.total_metrics_observed}\n\n## Active Signal Correlations\n${corrLines || \" (no correlations yet — connect more integrations)\"}\n\n## Your Role\nYou are **Nexus** — a senior marketing & sales intelligence analyst embedded in the dashboard.\n- When a metric is mentioned, explain its funnel position (TOFU/MOFU/BOFU) and causal relationships\n- Proactively surface the highest-priority bottleneck in every response\n- Suggest specific actions tied to the integration data above\n- Use academic terminology: CAC, LTV, MQL, SQL, attribution, conversion velocity, message-market fit\n- End responses with a concrete next action when applicable\n`.trim();\n } catch { /* intelligence layer unavailable — proceed without context */ }\n\n // I3: agent role prompts. When agent.role is one of listener/topic/creator/analyst, prepend\n // the specialized prefix so the model behaves consistently with the role badge shown in the UI.\n const AGENT_ROLE_PROMPTS: Record = {\n listener: `You are a Listener Agent. Your job:\n1. Monitor every incoming signal from Gmail, Outlook, and LinkedIn for buyer intent.\n2. Flag mentions of: pain points, competitor references, pricing questions, budget signals.\n3. Surface weekly patterns (\"3 leads this week mentioned X\").\n4. Err on the side of flagging — never miss a commercially-relevant signal.`,\n topic: `You are a Topic Agent. Your job:\n1. Take buyer pain points + intent signals (from the Listener Agent or Signals workspace).\n2. Generate content ideas: blog titles, LinkedIn angles, email subject lines.\n3. Score each idea by search-volume potential (Ahrefs), funnel-stage fit, estimated ICE impact.\n4. Output a prioritized content backlog formatted as: Title | Funnel Stage | AARRR Lever | Est. ICE Score.\n5. Link every output to an active OKR Key Result when possible.`,\n creator: `You are a Creator Agent. Your job:\n1. Take approved topic ideas and draft complete marketing assets.\n2. Match brand voice: direct, data-backed, B2B SaaS tone.\n3. Blog posts: H1/H2 structure, 800–1200 words, CTA at end.\n4. LinkedIn posts: hook in first line, insight in body, question as CTA, max 250 words.\n5. Email sequences: subject + 3-email sequence (awareness → consideration → CTA).\n6. Always cite one data point from the connected integrations as social proof.`,\n analyst: `You are an Analyst Agent. Your job:\n1. Analyze the funnel summary across all 14 integrations.\n2. Identify the single biggest bottleneck in the current TOFU/MOFU/BOFU funnel.\n3. Propose the top 3 ICE-scored experiments to address it.\n4. Track experiment outcomes and update affected OKR Key Results.\n5. Produce a weekly growth digest: wins, losses, signals, recommended experiments.`,\n };\n const rolePrompt = agent?.role && AGENT_ROLE_PROMPTS[agent.role] ? AGENT_ROLE_PROMPTS[agent.role] : \"\";\n\n // Part 15.4 / W1 — CONTEXT STACK. Each source is fetched in its own try/catch (default \"\")\n // so a failure in any one never breaks chat. Non-empty blocks are appended additively to the\n // system prompt below, separated by blank lines — existing prompt content is left untouched.\n let companyContext = \"\";\n try { companyContext = await getCompanyContextString(2000); } catch { companyContext = \"\"; }\n let memoryContext = \"\";\n try { memoryContext = await getMemoryContextString(2000); } catch { memoryContext = \"\"; }\n let knowledgeContext = \"\";\n try { knowledgeContext = await loadEnabledKnowledgeContext({ target: \"chat\", maxChars: 4000 }); } catch { knowledgeContext = \"\"; }\n let skillsIndex = \"\";\n try { skillsIndex = await getSkillsIndex({ maxChars: 6000 }); } catch { skillsIndex = \"\"; }\n\n // GOAL AWARENESS — the FINAL GOAL is the OKR. Inject a concise summary of the ACTIVE objective(s)\n // + their key results (title · progress · target) so every answer is steered toward the objective.\n // Short (a few hundred chars) and fully guarded — never breaks chat if OKRs are unavailable.\n let goalContext = \"\";\n try {\n const objs = await listObjectives(workspace ? { workspace } : {});\n const active = objs.filter((o: any) => o.status === \"active\").slice(0, 3);\n if (active.length) {\n const lines = active.map((o: any) => {\n const krs = (o.key_results || []).slice(0, 3).map((k: any) =>\n ` • ${k.title} — ${k.progress}% (now ${k.current_value}${k.unit ? \" \" + k.unit : \"\"} → target ${k.target_value}${k.unit ? \" \" + k.unit : \"\"})`\n ).join(\"\\n\");\n return ` - ${o.title} [${o.progress}% complete]${krs ? \"\\n\" + krs : \"\"}`;\n }).join(\"\\n\");\n goalContext = (\n \"## FINAL GOAL — Active OKRs (the priority)\\n\" +\n \"Everything you do should advance these objectives + key results. Tie tasks, campaigns and recommendations back to them.\\n\" +\n lines\n ).slice(0, 1200);\n }\n } catch { goalContext = \"\"; }\n\n // ACTIVE PROJECT — when scoped to a campaign, load it + its objective + member agents and prime\n // the Director with it. scope carries campaign_id/objective_id so created tasks link to the project.\n const { context: campaignContext, scope: campaignScope } = await buildCampaignContext(campaignId);\n\n const baseSystemPrompt: string =\n agent?.system_prompt?.trim()\n ? agent.system_prompt.trim()\n : DEFAULT_SYSTEM_PROMPT;\n const systemPrompt = [\n rolePrompt, baseSystemPrompt, EXECUTION_GUIDE, campaignContext, goalContext, funnelContext,\n companyContext, memoryContext, knowledgeContext, skillsIndex,\n ]\n .filter((s) => s && s.length > 0)\n .join(\"\\n\\n\");\n\n // Determine tools — always give all tools; agents narrow by integration name via get_integration_data\n // Agent.tools stores integration names (e.g. [\"ga4\",\"gsc\"]), not tool function names,\n // so filtering by function name would always produce an empty list. Give all tools instead.\n const tools = buildTools();\n\n // Read temperature + frequency_penalty from app_config so Settings sliders are real, not cosmetic.\n let temperature: number | undefined = undefined;\n let frequencyPenalty: number | undefined = undefined;\n try {\n const cfgRows = await sqlite.execute(\"SELECT key, value FROM app_config WHERE key IN ('temperature','frequency_penalty') AND tenant_id = ?\", [currentTenantId()]);\n for (const row of cfgRows.rows) {\n const [k, v] = row as any[];\n const num = parseFloat(String(v).replace(/^\"|\"$/g, \"\"));\n if (!isNaN(num)) {\n if (k === \"temperature\") temperature = num;\n if (k === \"frequency_penalty\") frequencyPenalty = num;\n }\n }\n } catch { /* fall back to provider defaults */ }\n\n // Part 14.2 — wrap the LLM/tool work so we can finalize the agent-run record + trace on both\n // success and failure. Behavior is otherwise unchanged from the v551 baseline.\n try {\n // First LLM call\n let llmResult = await callLLM(llmConfig, conversationHistory, {\n systemPrompt,\n tools,\n maxTokens: 4096,\n temperature,\n frequencyPenalty,\n });\n\n const allToolCalls: any[] = [];\n\n // Handle tool use — bounded multi-round loop (was a single round, which truncated\n // any plan needing chained tool calls). Cap rounds to avoid runaway cost/latency.\n let toolRounds = 0;\n const MAX_TOOL_ROUNDS = 5;\n while (llmResult.tool_calls && llmResult.tool_calls.length > 0 && toolRounds < MAX_TOOL_ROUNDS) {\n toolRounds++;\n // Trace: mark the start of this tool round.\n trace.push({ t: \"round\", n: toolRounds, ts: new Date().toISOString() });\n // Record the assistant message with tool calls.\n // reasoning_content must be echoed back for DeepSeek thinking-mode multi-turn.\n const assistantMsg: any = {\n role: \"assistant\",\n content: llmResult.content || \"\",\n tool_calls: llmResult.tool_calls,\n };\n if ((llmResult as any).reasoning_content) {\n assistantMsg.reasoning_content = (llmResult as any).reasoning_content;\n }\n conversationHistory.push(assistantMsg);\n\n // Execute each tool call and collect results\n const toolResults: Array<{ tool_call_id: string; role: \"tool\"; content: string }> = [];\n\n for (const toolCall of llmResult.tool_calls) {\n allToolCalls.push(toolCall);\n\n let args: Record = {};\n try {\n args =\n typeof toolCall.function.arguments === \"string\"\n ? JSON.parse(toolCall.function.arguments)\n : toolCall.function.arguments ?? {};\n } catch {\n args = {};\n }\n\n // Measure wall-clock elapsed around the existing executeTool call (no double-call).\n const _start = performance.now();\n const toolResult = await executeTool(toolCall.function.name, args, llmConfig, campaignScope);\n const _ms = Math.round(performance.now() - _start);\n\n // Trace: record this tool call with short arg/result previews.\n trace.push({\n t: \"tool\",\n name: toolCall.function.name,\n args_preview: JSON.stringify(args).slice(0, 300),\n result_preview: String(toolResult).slice(0, 300),\n ok: true,\n ms: _ms,\n ts: new Date().toISOString(),\n });\n\n toolResults.push({\n tool_call_id: toolCall.id,\n role: \"tool\",\n content: toolResult,\n });\n }\n\n // Add tool results to history\n for (const tr of toolResults) {\n conversationHistory.push(tr as any);\n }\n\n // Second LLM call with tool results\n llmResult = await callLLM(llmConfig, conversationHistory, {\n systemPrompt,\n tools,\n maxTokens: 4096,\n temperature,\n frequencyPenalty,\n });\n }\n\n // Add final assistant response to history\n conversationHistory.push({\n role: \"assistant\",\n content: llmResult.content,\n });\n\n // Trace: record the final assistant message.\n trace.push({\n t: \"message\",\n role: \"assistant\",\n preview: String(llmResult.content).slice(0, 300),\n ts: new Date().toISOString(),\n });\n\n // Persist conversation\n await saveConversation(\n conversationId,\n conversationHistory,\n agent_id,\n workspace\n );\n\n // W1 — fire-and-forget: distill durable memory from this conversation. Never blocks the response.\n captureMemoryFromConversation(conversationHistory as any).catch(() => {});\n\n // Part 14.1/14.2 — finalize the agent-run record + persist the trace on success.\n if (wantRun && runId) {\n await finishAgentRun(runId, \"completed\", llmResult.content).catch(() => {});\n await sqlite.execute(\"UPDATE agent_runs SET trace=? WHERE id=? AND tenant_id = ?\", [JSON.stringify(trace), runId, currentTenantId()]).catch(() => {});\n }\n\n return json({\n response: llmResult.content,\n conversation_id: conversationId,\n tool_calls: allToolCalls.length > 0 ? allToolCalls : undefined,\n agent_id: agent_id ?? undefined,\n trace,\n });\n } catch (err: any) {\n // Part 14.1/14.2 — mark the run failed and persist whatever partial trace we captured.\n if (wantRun && runId) {\n await finishAgentRun(runId, \"failed\", undefined, String(err)).catch(() => {});\n await sqlite.execute(\"UPDATE agent_runs SET trace=? WHERE id=? AND tenant_id = ?\", [JSON.stringify(trace), runId, currentTenantId()]).catch(() => {});\n }\n // Preserve prior behavior: surface the error as a 500 JSON response.\n return json({ error: err?.message ?? \"Chat failed\" }, 500);\n }\n}\n\n// ─── Streaming chat handler (SSE) ────────────────────────────────────────────\n\nexport async function handleChatStream(req: Request): Promise {\n if (req.method === \"OPTIONS\") {\n return new Response(null, { status: 204, headers: CORS });\n }\n if (req.method !== \"POST\") {\n return new Response(JSON.stringify({ error: \"Method not allowed\" }), { status: 405, headers: { \"Content-Type\": \"application/json\", ...CORS } });\n }\n\n let body: any;\n try { body = await req.json(); } catch {\n return new Response(JSON.stringify({ error: \"Invalid JSON body\" }), { status: 400, headers: { \"Content-Type\": \"application/json\", ...CORS } });\n }\n\n const { message, agent_id, conversation_id: incomingConversationId, workspace, campaign: campaignId } = body as {\n message?: string; agent_id?: string; conversation_id?: string; workspace?: string; campaign?: string;\n };\n\n if (!message || typeof message !== \"string\" || message.trim() === \"\") {\n return new Response(JSON.stringify({ error: \"message field is required\" }), { status: 400, headers: { \"Content-Type\": \"application/json\", ...CORS } });\n }\n\n const llmConfig = await getLLMConfig();\n let agent = null;\n if (agent_id) { agent = await getAgent(agent_id); }\n\n const conversationId = incomingConversationId ?? crypto.randomUUID();\n let conversationHistory: ChatMessage[] = [];\n if (incomingConversationId) {\n const existing = await loadConversation(incomingConversationId);\n if (existing) conversationHistory = existing.messages;\n }\n conversationHistory.push({ role: \"user\", content: message.trim() });\n\n const baseSystemPrompt: string = agent?.system_prompt?.trim() ? agent.system_prompt.trim() :\n \"You are an AI assistant for this marketing and sales platform.\";\n\n // W1 — inject unified company context (cheap single app_config read; safe on failure).\n const companyCtx = await getCompanyContextString(2000).catch(() => \"\");\n // ACTIVE PROJECT — the /world Director chat sends body.campaign (campaign id). Load that project\n // + its parent objective + member agents so the Director answers WITH project context instead of\n // generically. scope carries campaign_id/objective_id so created tasks link back to the project.\n const { context: campaignContext, scope: campaignScope } = await buildCampaignContext(campaignId);\n // Ground the streaming chat in its REAL tool capabilities (same guide as handleChat) so it\n // actually runs create_agent/create_campaign/create_task/prepare_* instead of narrating.\n const streamSystemPrompt = [baseSystemPrompt, EXECUTION_GUIDE, campaignContext, companyCtx]\n .filter((s) => s && s.trim()).join(\"\\n\\n\");\n\n // The streaming chat gets the SAME toolset as handleChat so it can DO work mid-stream.\n const tools = buildTools();\n\n const encoder = new TextEncoder();\n const stream = new ReadableStream({\n async start(controller) {\n const send = (data: object) => {\n controller.enqueue(encoder.encode(`data: ${JSON.stringify(data)}\\n\\n`));\n };\n\n let fullResponse = \"\";\n try {\n // ── TOOL LOOP (mirrors handleChat's executeTool loop, but over SSE) ──────────\n // Phase 1 (tool rounds): call the model NON-streaming so we can detect tool_calls.\n // For each requested tool we emit a {tool} event, execute it (create_agent /\n // create_campaign / create_task / prepare_* …), append the result, and loop —\n // exactly like handleChat. Any assistant prose accompanying a tool round is emitted\n // as a {delta} so no text is lost.\n // Phase 2 (final answer): once the model stops requesting tools, we STREAM the\n // closing assistant turn token-by-token via callLLMStream — preserving the existing\n // text streaming behaviour for the user-visible answer.\n const MAX_TOOL_ROUNDS = 5;\n let toolRounds = 0;\n\n while (toolRounds < MAX_TOOL_ROUNDS) {\n const llmResult = await callLLM(llmConfig, conversationHistory, {\n systemPrompt: streamSystemPrompt, tools, maxTokens: 4096,\n });\n\n // No tools requested → the model is ready to answer. Break to the streaming pass,\n // which re-prompts once and streams the final text token-by-token.\n if (!llmResult.tool_calls || llmResult.tool_calls.length === 0) break;\n\n // Tool round: surface any assistant prose, then run + report each tool call.\n toolRounds++;\n if (llmResult.content) {\n fullResponse += llmResult.content;\n send({ delta: llmResult.content, done: false });\n }\n const assistantMsg: any = {\n role: \"assistant\",\n content: llmResult.content || \"\",\n tool_calls: llmResult.tool_calls,\n };\n if ((llmResult as any).reasoning_content) {\n assistantMsg.reasoning_content = (llmResult as any).reasoning_content;\n }\n conversationHistory.push(assistantMsg);\n\n for (const toolCall of llmResult.tool_calls) {\n let args: Record = {};\n try {\n args = typeof toolCall.function.arguments === \"string\"\n ? JSON.parse(toolCall.function.arguments)\n : toolCall.function.arguments ?? {};\n } catch { args = {}; }\n\n // run_campaign defaults to the campaign-scoped chat's campaign — make sure the streamed\n // frame ALWAYS carries that resolved campaign_id (in args) so the world entry can fire the\n // real run even when the model didn't pass campaign_id explicitly.\n if (toolCall.function.name === \"run_campaign\" && args.campaign_id == null && campaignScope?.campaign_id) {\n args = { ...args, campaign_id: campaignScope.campaign_id };\n }\n\n // Tell the client a tool is running (name + arg preview) before we execute it.\n send({ tool: { name: toolCall.function.name, args, status: \"running\" } });\n\n const toolResult = await executeTool(toolCall.function.name, args, llmConfig, campaignScope);\n\n // Report completion + a short result preview. For run_campaign also lift the resolved\n // campaign_id onto the {tool} frame itself so the world entry can act on it (fire the run).\n const doneFrame: any = { name: toolCall.function.name, status: \"done\", result_preview: String(toolResult).slice(0, 300) };\n if (toolCall.function.name === \"run_campaign\") {\n let cid: string | undefined;\n try { cid = JSON.parse(toolResult)?.campaign_id; } catch { /* keep undefined */ }\n cid = cid ?? args.campaign_id ?? campaignScope?.campaign_id;\n if (cid != null) { doneFrame.campaign_id = String(cid); doneFrame.args = { campaign_id: String(cid) }; }\n }\n send({ tool: doneFrame });\n\n conversationHistory.push({\n tool_call_id: toolCall.id,\n role: \"tool\",\n content: toolResult,\n } as any);\n }\n // loop again — the model now sees the tool results\n }\n\n // Phase 2 — STREAM the final assistant turn token-by-token (existing text streaming).\n let streamed = \"\";\n const gen = callLLMStream(llmConfig, conversationHistory, {\n systemPrompt: streamSystemPrompt, maxTokens: 4096,\n });\n for await (const token of gen) {\n if (token) { streamed += token; fullResponse += token; send({ delta: token, done: false }); }\n }\n conversationHistory.push({ role: \"assistant\", content: streamed });\n\n await saveConversation(conversationId, conversationHistory, agent_id, workspace);\n captureMemoryFromConversation(conversationHistory as any).catch(() => {});\n send({ delta: \"\", done: true, response: fullResponse, conversation_id: conversationId });\n } catch (err: any) {\n send({ delta: \"\", done: true, error: err?.message ?? \"Stream error\", conversation_id: conversationId });\n } finally {\n controller.close();\n }\n },\n });\n\n return new Response(stream, {\n headers: {\n \"Content-Type\": \"text/event-stream\",\n \"Cache-Control\": \"no-cache\",\n \"Connection\": \"keep-alive\",\n ...CORS,\n },\n });\n}\n\n// ─── Conversation history CRUD ────────────────────────────────────────────────\n\nexport async function listConversations(): Promise {\n try {\n const result = await sqlite.execute(\n `SELECT id, workspace, agent_id, created_at, updated_at, messages\n FROM conversations WHERE tenant_id = ? ORDER BY updated_at DESC LIMIT 50`,\n [currentTenantId()]\n );\n return result.rows.map((row) => {\n const obj: Record = {};\n for (let i = 0; i < result.columns.length; i++) obj[result.columns[i]] = row[i];\n // Extract first user message as title\n let firstMsg = \"\";\n try {\n const msgs: ChatMessage[] = JSON.parse(String(obj.messages || \"[]\"));\n const first = msgs.find((m) => m.role === \"user\");\n firstMsg = first ? String(first.content).slice(0, 60) : \"\";\n } catch {}\n return {\n id: obj.id,\n workspace: obj.workspace,\n agent_id: obj.agent_id,\n created_at: obj.created_at,\n updated_at: obj.updated_at,\n title: firstMsg || \"(no messages)\",\n };\n });\n } catch { return []; }\n}\n\nexport async function getConversationById(id: string): Promise {\n try {\n const result = await sqlite.execute(\n \"SELECT id, workspace, agent_id, created_at, updated_at, messages FROM conversations WHERE id = ? AND tenant_id = ?\",\n [id, currentTenantId()]\n );\n if (!result.rows?.length) return null;\n const obj: Record = {};\n for (let i = 0; i < result.columns.length; i++) obj[result.columns[i]] = result.rows[0][i];\n let messages: ChatMessage[] = [];\n try { messages = JSON.parse(String(obj.messages || \"[]\")); } catch {}\n return { id: obj.id, workspace: obj.workspace, agent_id: obj.agent_id,\n created_at: obj.created_at, updated_at: obj.updated_at, messages };\n } catch { return null; }\n}\n\nexport async function deleteConversation(id: string): Promise {\n try {\n await sqlite.execute(\"DELETE FROM conversations WHERE id = ? AND tenant_id = ?\", [id, currentTenantId()]);\n return true;\n } catch { return false; }\n}\n","createdAt",["D",1780483959462],"branchVersion","branchId","ecb58646-56c9-11f1-abd8-ee650bb23af1","id","ecb4ccf6-56c9-11f1-abd8-ee650bb23af1","aihivecommand",["D",1779556068488],"deletedAt","legacyEnvVars","privacy","public","httpPrivacy","description","AIHive Command Center","imageUrl","e8c17657-6210-4c45-bd2b-dbcf2e020293","pinnedFileId","32167c2c-56ca-11f1-955a-ee650bb23af1","pinPreview",true,"pinMigrated","screenshotVersion","likeCount",0,"orgId","019e4de5-f6b1-7488-a017-a3a028bc2ba8","tursoName","ecb4ccf656c911f1abd8ee650bb23af1","tursoHostname","ecb4ccf656c911f1abd8ee650bb23af1-val-town.aws-us-east-2.turso.io","tursoMeasuredAt",["D",1789503703835],"pinned","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm","user",{"_101":100,"_108":104},"org",{"_101":100,"_102":100,"_103":104,"_18":56,"_105":106,"_107":25},"branch",{"_13":90,"_32":91,"_44":43,"_92":45,"_93":-5,"_94":-5,"_95":-5,"_96":-5,"_97":-5,"_98":46,"_99":100},"isOwned",1111,"main","projectId","forkedBranchId","forkedBranchVersion","forkedProjectPrivacy","forkedProjectId","forkedProjectOrgId","projectName","userHandle","vlduy123","handle","displayName","avatar","https://avatars.githubusercontent.com/u/208025931?v=4","tier","free","canRestrictVals","profileImageUrl","branchName","","adoptOrgId",{"_44":45,"_32":46,"_39":1898,"_48":-5,"_49":25,"_50":51,"_52":51,"_53":54,"_55":-5,"_18":56,"_57":58,"_59":60,"_61":-5,"_62":-5,"_63":64,"_65":66,"_67":68,"_69":70,"_71":1899,"_73":-5,"_74":75,"_76":75,"_77":78,"_79":78,"_80":78,"_81":75,"_82":75,"_83":1900,"_85":1901,"_87":1902,"_89":25},"tree",[555,641,659,688,689,690,691,692,693,694,695,696,697,698,699,700,701,702,703,704,705,706,707,708,709,710,711,712,713,714,715,716,717,718,719,720,721,722,723,724,725,726,727,728,729,730,731,732,733],"files",[555,556,557,558,559,560,561,562,563,564,565,566,567,568,569,570,571,572,573,574,575,576,577,578,579,580,581,582,583,584,585,586,587,588,589,590,591,592,593,594,595,596,597,598,599,600,601,602,603,604,605,606,607,608,609,610,611,612,613,614,615,616,617,618,619,620,621,622,623,624,625,626,627,628,629,630,631,632,633,634,635,636,637,638,639,640,641,642,643,644,645,646,647,648,649,650,651,652,653,654,655,656,657,658,659,660,661,662,663,664,665,666,667,668,669,670,671,672,673,674,675,676,677,678,679,680,681,682,683,684,685,686,687,688,689,690,691,692,693,694,695,696,697,698,699,700,701,702,703,704,705,706,707,708,709,710,711,712,713,714,715,716,717,718,719,720,721,722,723,724,725,726,727,728,729,730,731,732,733],"counts",{"_545":64,"_546":303,"_547":64,"_548":549,"_550":551},"contents",[131,132,133,134,135,136,137,138,139,140,141,142,143,144,145,146,147,148,149,150,151,152,153,154,155,156,157,158,159,160,161,162,163,164,165,166,167,168,169,170,171,172,173,174,175,176,177,178,179],"upstreamProject","upstreamBranch","emptyEnvVars","flags",[],"hasHttpFiles","grants","orgs","customDomains",{},{"_44":540,"_32":541,"_30":542,"_183":-5,"_42":43,"_41":196,"_185":60,"_39":543,"_13":64,"_34":532,"_36":-5,"_24":25,"_189":-5,"_27":544},{"_44":534,"_32":535,"_30":536,"_183":-5,"_42":43,"_41":537,"_185":60,"_39":538,"_13":64,"_34":532,"_36":-5,"_24":25,"_189":-5,"_27":539},{"_44":527,"_32":528,"_30":529,"_183":-5,"_42":43,"_41":530,"_185":60,"_39":531,"_13":64,"_34":532,"_36":-5,"_24":25,"_189":-5,"_27":533},{"_44":520,"_32":521,"_30":522,"_183":-5,"_42":43,"_41":523,"_185":60,"_39":524,"_13":236,"_34":35,"_36":525,"_24":25,"_189":-5,"_27":526},{"_44":513,"_32":514,"_30":515,"_183":-5,"_42":43,"_41":516,"_185":25,"_39":517,"_13":220,"_34":35,"_36":518,"_24":25,"_189":-5,"_27":519},{"_44":506,"_32":507,"_30":508,"_183":-5,"_42":43,"_41":509,"_185":60,"_39":510,"_13":196,"_34":35,"_36":511,"_24":25,"_189":-5,"_27":512},{"_44":498,"_32":499,"_30":500,"_183":-5,"_42":43,"_41":501,"_185":25,"_39":502,"_13":503,"_34":35,"_36":504,"_24":25,"_189":-5,"_27":505},{"_44":490,"_32":491,"_30":492,"_183":-5,"_42":43,"_41":493,"_185":25,"_39":494,"_13":495,"_34":35,"_36":496,"_24":25,"_189":-5,"_27":497},{"_44":483,"_32":484,"_30":485,"_183":-5,"_42":43,"_41":486,"_185":60,"_39":487,"_13":303,"_34":35,"_36":488,"_24":25,"_189":-5,"_27":489},{"_44":475,"_32":476,"_30":477,"_183":-5,"_42":43,"_41":478,"_185":60,"_39":479,"_13":480,"_34":35,"_36":481,"_24":25,"_189":-5,"_27":482},{"_44":467,"_32":468,"_30":469,"_183":-5,"_42":43,"_41":470,"_185":60,"_39":471,"_13":472,"_34":35,"_36":473,"_24":25,"_189":-5,"_27":474},{"_44":460,"_32":461,"_30":462,"_183":-5,"_42":43,"_41":463,"_185":60,"_39":464,"_13":258,"_34":35,"_36":465,"_24":25,"_189":-5,"_27":466},{"_44":452,"_32":453,"_30":454,"_183":-5,"_42":43,"_41":455,"_185":60,"_39":456,"_13":457,"_34":35,"_36":458,"_24":25,"_189":-5,"_27":459},{"_44":445,"_32":446,"_30":447,"_183":-5,"_42":43,"_41":448,"_185":60,"_39":449,"_13":220,"_34":35,"_36":450,"_24":25,"_189":-5,"_27":451},{"_44":438,"_32":439,"_30":440,"_183":-5,"_42":43,"_41":441,"_185":60,"_39":442,"_13":64,"_34":35,"_36":443,"_24":25,"_189":-5,"_27":444},{"_44":431,"_32":432,"_30":433,"_183":-5,"_42":43,"_41":434,"_185":60,"_39":435,"_13":326,"_34":35,"_36":436,"_24":25,"_189":-5,"_27":437},{"_44":424,"_32":425,"_30":426,"_183":-5,"_42":43,"_41":427,"_185":60,"_39":428,"_13":220,"_34":35,"_36":429,"_24":25,"_189":-5,"_27":430},{"_44":416,"_32":417,"_30":418,"_183":-5,"_42":43,"_41":419,"_185":25,"_39":420,"_13":421,"_34":35,"_36":422,"_24":25,"_189":-5,"_27":423},{"_44":408,"_32":409,"_30":410,"_183":-5,"_42":43,"_41":411,"_185":25,"_39":412,"_13":413,"_34":35,"_36":414,"_24":25,"_189":-5,"_27":415},{"_44":401,"_32":402,"_30":403,"_183":-5,"_42":43,"_41":404,"_185":60,"_39":405,"_13":334,"_34":35,"_36":406,"_24":25,"_189":-5,"_27":407},{"_44":394,"_32":395,"_30":396,"_183":-5,"_42":43,"_41":397,"_185":25,"_39":398,"_13":326,"_34":35,"_36":399,"_24":25,"_189":-5,"_27":400},{"_44":387,"_32":388,"_30":389,"_183":-5,"_42":43,"_41":390,"_185":60,"_39":391,"_13":236,"_34":35,"_36":392,"_24":25,"_189":-5,"_27":393},{"_44":380,"_32":381,"_30":382,"_183":-5,"_42":43,"_41":383,"_185":60,"_39":384,"_13":64,"_34":35,"_36":385,"_24":25,"_189":-5,"_27":386},{"_44":373,"_32":374,"_30":375,"_183":-5,"_42":43,"_41":376,"_185":60,"_39":377,"_13":64,"_34":35,"_36":378,"_24":25,"_189":-5,"_27":379},{"_44":366,"_32":367,"_30":368,"_183":-5,"_42":43,"_41":369,"_185":60,"_39":370,"_13":64,"_34":35,"_36":371,"_24":25,"_189":-5,"_27":372},{"_44":359,"_32":360,"_30":361,"_183":-5,"_42":43,"_41":362,"_185":25,"_39":363,"_13":64,"_34":35,"_36":364,"_24":25,"_189":-5,"_27":365},{"_44":352,"_32":353,"_30":354,"_183":-5,"_42":43,"_41":355,"_185":60,"_39":356,"_13":326,"_34":35,"_36":357,"_24":25,"_189":-5,"_27":358},{"_44":344,"_32":345,"_30":346,"_183":-5,"_42":43,"_41":347,"_185":60,"_39":348,"_13":349,"_34":35,"_36":350,"_24":25,"_189":-5,"_27":351},{"_44":337,"_32":338,"_30":339,"_183":-5,"_42":43,"_41":340,"_185":60,"_39":341,"_13":196,"_34":35,"_36":342,"_24":25,"_189":-5,"_27":343},{"_44":329,"_32":330,"_30":331,"_183":-5,"_42":43,"_41":332,"_185":60,"_39":333,"_13":334,"_34":35,"_36":335,"_24":25,"_189":-5,"_27":336},{"_44":321,"_32":322,"_30":323,"_183":-5,"_42":43,"_41":324,"_185":60,"_39":325,"_13":326,"_34":35,"_36":327,"_24":25,"_189":-5,"_27":328},{"_44":313,"_32":314,"_30":315,"_183":-5,"_42":43,"_41":316,"_185":25,"_39":317,"_13":318,"_34":35,"_36":319,"_24":25,"_189":-5,"_27":320},{"_44":306,"_32":307,"_30":308,"_183":-5,"_42":43,"_41":309,"_185":60,"_39":310,"_13":196,"_34":35,"_36":311,"_24":25,"_189":-5,"_27":312},{"_44":298,"_32":299,"_30":300,"_183":-5,"_42":43,"_41":301,"_185":25,"_39":302,"_13":303,"_34":35,"_36":304,"_24":25,"_189":-5,"_27":305},{"_44":291,"_32":292,"_30":293,"_183":-5,"_42":43,"_41":294,"_185":60,"_39":295,"_13":236,"_34":35,"_36":296,"_24":25,"_189":-5,"_27":297},{"_44":284,"_32":285,"_30":286,"_183":-5,"_42":43,"_41":287,"_185":60,"_39":288,"_13":220,"_34":35,"_36":289,"_24":25,"_189":-5,"_27":290},{"_44":277,"_32":278,"_30":279,"_183":-5,"_42":43,"_41":280,"_185":60,"_39":281,"_13":196,"_34":35,"_36":282,"_24":25,"_189":-5,"_27":283},{"_44":269,"_32":270,"_30":271,"_183":-5,"_42":43,"_41":272,"_185":60,"_39":273,"_13":274,"_34":35,"_36":275,"_24":25,"_189":-5,"_27":276},{"_44":261,"_32":262,"_30":263,"_183":-5,"_42":43,"_41":264,"_185":60,"_39":265,"_13":266,"_34":35,"_36":267,"_24":25,"_189":-5,"_27":268},{"_44":253,"_32":254,"_30":255,"_183":-5,"_42":43,"_41":256,"_185":60,"_39":257,"_13":258,"_34":35,"_36":259,"_24":25,"_189":-5,"_27":260},{"_44":246,"_32":247,"_30":248,"_183":-5,"_42":43,"_41":249,"_185":60,"_39":250,"_13":64,"_34":35,"_36":251,"_24":25,"_189":-5,"_27":252},{"_44":239,"_32":240,"_30":241,"_183":-5,"_42":43,"_41":242,"_185":60,"_39":243,"_13":64,"_34":35,"_36":244,"_24":25,"_189":-5,"_27":245},{"_44":231,"_32":232,"_30":233,"_183":-5,"_42":43,"_41":234,"_185":60,"_39":235,"_13":236,"_34":35,"_36":237,"_24":25,"_189":-5,"_27":238},{"_44":223,"_32":224,"_30":225,"_183":-5,"_42":43,"_41":226,"_185":60,"_39":227,"_13":228,"_34":35,"_36":229,"_24":25,"_189":-5,"_27":230},{"_44":215,"_32":216,"_30":217,"_183":-5,"_42":43,"_41":218,"_185":60,"_39":219,"_13":220,"_34":35,"_36":221,"_24":25,"_189":-5,"_27":222},{"_44":207,"_32":208,"_30":58,"_183":-5,"_42":43,"_41":209,"_185":25,"_39":210,"_13":211,"_34":212,"_36":213,"_24":25,"_189":-5,"_27":214},{"_44":199,"_32":200,"_30":201,"_183":-5,"_42":43,"_41":202,"_185":25,"_39":203,"_13":204,"_34":35,"_36":205,"_24":25,"_189":-5,"_27":206},{"_44":191,"_32":192,"_30":193,"_183":-5,"_42":43,"_41":194,"_185":25,"_39":195,"_13":196,"_34":35,"_36":197,"_24":25,"_189":-5,"_27":198},{"_44":180,"_32":181,"_30":182,"_183":-5,"_42":43,"_41":184,"_185":60,"_39":186,"_13":187,"_34":35,"_36":188,"_24":25,"_189":-5,"_27":190},"026e63e8-5d7a-11f1-a7f5-1607ee4eb77e","types.ts","151549be-56ca-11f1-bc41-ee650bb23af1","parentId",699,"latest",["D",1780291403301],16,"export interface Integration {\n id: string;\n name: string; // 'ahrefs'|'ga4'|'gsc'|'linkedin'|'outlook'|'wordpress'|'youtube'|'producthunt'\n api_key?: string;\n api_secret?: string;\n access_token?: string;\n refresh_token?: string;\n extra_config?: string; // JSON string\n connected_at?: string;\n updated_at?: string;\n}\n\nexport interface LLMConfig {\n id: string;\n // QA fix: gemini was already routed in callLLM but missing from the union, causing TS noise.\n provider: 'anthropic' | 'openai' | 'gemini' | 'custom';\n api_key: string;\n endpoint?: string;\n model: string;\n updated_at?: string;\n // Per-provider key presence flags (computed from the 4 api_key_ columns) —\n // used by Settings UI to render \"(saved)\" pills without leaking the actual key.\n has_key?: { anthropic: boolean; openai: boolean; gemini: boolean; custom: boolean };\n}\n\nexport interface Agent {\n id: string;\n name: string;\n workspace: 'marketing' | 'sales';\n description: string;\n system_prompt: string;\n tools: string; // JSON array string\n model?: string;\n enabled: number; // 0 or 1\n created_at: string;\n updated_at: string;\n created_by: 'user' | 'ai';\n // B2: budget cap (USD/month) + running spend, joined from agent_budgets\n monthly_cap?: number;\n current_spend?: number;\n}\n\nexport interface Conversation {\n id: string;\n agent_id?: string;\n workspace: string;\n messages: string; // JSON array string\n created_at: string;\n updated_at: string;\n}\n\nexport interface Task {\n id: string;\n agent_id?: string;\n workspace: string;\n title: string;\n status: 'pending' | 'running' | 'done' | 'failed';\n input?: string;\n output?: string;\n created_at: string;\n completed_at?: string;\n}\n\nexport interface ChatMessage {\n role: 'user' | 'assistant' | 'system' | 'tool';\n content: string;\n timestamp?: string;\n // Tool-use fields (OpenAI-style) carried at runtime by chat.ts / llm.ts.\n tool_calls?: any[];\n tool_call_id?: string;\n reasoning_content?: string;\n // Part 14.2: structured execution trace (round/tool/action/message steps) shown in the UI.\n trace?: Array<{ t: string; role?: string; name?: string; preview?: string; args_preview?: string; result_preview?: string; ok?: boolean; ms?: number; n?: number; ts?: string }>;\n}\n","label",[181],"55fca990-5f3a-11f1-b12a-1607ee4eb77e","schema.ts","4ad25812-5f2b-11f1-9413-1607ee4eb77e",1070,["D",1780483958033],2,"import { sqlite, DEFAULT_AGENT_SEEDS } from \"./db.ts\";\n\n// The schema setup body lives here (moved verbatim out of db.ts) to keep db.ts under\n// val.town's 80KB/file cap. db.ts lazily `import(\"./schema.ts\")`s this and calls runSchema().\nexport async function runSchema(): Promise {\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS integrations (\n id TEXT PRIMARY KEY,\n name TEXT NOT NULL,\n api_key TEXT,\n api_secret TEXT,\n access_token TEXT,\n refresh_token TEXT,\n extra_config TEXT,\n connected_at TEXT,\n updated_at TEXT,\n tenant_id TEXT,\n UNIQUE(tenant_id, name)\n )\n `);\n\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS llm_config (\n id TEXT PRIMARY KEY DEFAULT 'default',\n provider TEXT NOT NULL DEFAULT 'anthropic',\n api_key TEXT NOT NULL DEFAULT '',\n endpoint TEXT,\n model TEXT NOT NULL DEFAULT 'claude-sonnet-4-6',\n updated_at TEXT\n )\n `);\n\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS agents (\n id TEXT PRIMARY KEY,\n name TEXT NOT NULL,\n workspace TEXT NOT NULL,\n description TEXT,\n system_prompt TEXT,\n tools TEXT DEFAULT '[]',\n model TEXT,\n enabled INTEGER DEFAULT 1,\n created_at TEXT NOT NULL,\n updated_at TEXT NOT NULL,\n created_by TEXT DEFAULT 'user'\n )\n `);\n\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS conversations (\n id TEXT PRIMARY KEY,\n agent_id TEXT,\n workspace TEXT NOT NULL,\n messages TEXT DEFAULT '[]',\n created_at TEXT NOT NULL,\n updated_at TEXT NOT NULL\n )\n `);\n\n // F4 migration: the original tasks table (TEXT id, agent_id, input/output) is replaced by\n // the framework-layer tasks schema (INTEGER id, ICE fields, OKR/experiment links). If a\n // legacy table exists, rename it so the new CREATE further down can build the right shape.\n try {\n const cols = await sqlite.execute(\"PRAGMA table_info(tasks)\");\n const have = new Set((cols.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n // Old schema is identifiable by `agent_id` + `input` + no `priority`. New schema has `priority` + `ice_score`.\n const isLegacy = have.has(\"agent_id\") && have.has(\"input\") && !have.has(\"ice_score\");\n if (isLegacy) {\n await sqlite.execute(\"ALTER TABLE tasks RENAME TO tasks_legacy\");\n }\n } catch { /* table doesn't exist yet — fine, new CREATE will run */ }\n\n // Co-working: shared notes for team collaboration on strategies.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS notes (\n id TEXT PRIMARY KEY,\n author TEXT,\n content TEXT NOT NULL,\n workspace TEXT,\n pinned INTEGER DEFAULT 0,\n created_at TEXT NOT NULL,\n updated_at TEXT\n )\n `);\n\n // Intelligence layer: per-metric 30d baselines used for health scoring.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS metric_baselines (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n integration TEXT NOT NULL,\n metric_key TEXT NOT NULL,\n baseline_value REAL,\n baseline_period TEXT DEFAULT '30d',\n last_updated TEXT DEFAULT (datetime('now')),\n UNIQUE(integration, metric_key)\n )\n `);\n\n // Intelligence cache — funnel summary, correlations, recommendations, LLM-generated feed.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS intelligence_cache (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n cache_key TEXT UNIQUE NOT NULL,\n payload TEXT NOT NULL,\n generated_at TEXT DEFAULT (datetime('now')),\n expires_at TEXT NOT NULL\n )\n `);\n\n // App-wide key/value config: ROI inputs (avg_deal_size, est_marketing_spend),\n // workspace preferences (ultra_density, translucency, hardware_accel, etc.).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS app_config (\n key TEXT PRIMARY KEY,\n value TEXT,\n updated_at TEXT DEFAULT (datetime('now'))\n )\n `);\n\n // Metric history — rolling timeseries for delta_7d / delta_30d computations.\n // Pruned to last 60 days on insert to keep size bounded.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS metric_history (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n integration TEXT NOT NULL,\n metric_key TEXT NOT NULL,\n value REAL NOT NULL,\n recorded_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_metric_history_key ON metric_history(integration, metric_key, recorded_at DESC)`);\n\n // A5: rolling baseline columns — added on existing metric_baselines tables via\n // PRAGMA-guarded ALTER so we never fail on re-run.\n try {\n const cols = await sqlite.execute(\"PRAGMA table_info(metric_baselines)\");\n const have = new Set((cols.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n if (!have.has(\"rolling_avg\")) {\n await sqlite.execute(\"ALTER TABLE metric_baselines ADD COLUMN rolling_avg REAL\");\n }\n if (!have.has(\"sample_count\")) {\n await sqlite.execute(\"ALTER TABLE metric_baselines ADD COLUMN sample_count INTEGER DEFAULT 0\");\n }\n } catch { /* non-critical: warm-start baselines still work */ }\n\n // QA fix: per-provider key columns on llm_config so switching providers doesn't\n // overwrite the previous provider's key. PRAGMA-guarded ALTER (re-run safe).\n try {\n const cols = await sqlite.execute(\"PRAGMA table_info(llm_config)\");\n const have = new Set((cols.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n for (const col of [\"api_key_anthropic\", \"api_key_openai\", \"api_key_gemini\", \"api_key_custom\"]) {\n if (!have.has(col)) await sqlite.execute(`ALTER TABLE llm_config ADD COLUMN ${col} TEXT DEFAULT ''`);\n }\n if (!have.has(\"endpoint_custom\")) {\n await sqlite.execute(\"ALTER TABLE llm_config ADD COLUMN endpoint_custom TEXT DEFAULT ''\");\n }\n } catch { /* non-critical: legacy single-key still works */ }\n\n // Agent budget tracking — real monthly LLM-cost cap per agent.\n // current_spend incremented on each LLM call, reset at period_start rollover.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS agent_budgets (\n agent_id TEXT PRIMARY KEY,\n monthly_cap REAL DEFAULT 250,\n current_spend REAL DEFAULT 0,\n period_start TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now'))\n )\n `);\n\n // C3: agent execution log. One row per run (manual or scheduled). Latest 50 power the\n // AgentCard timeline. status: queued → running → completed | failed | cancelled.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS agent_runs (\n id TEXT PRIMARY KEY,\n agent_id TEXT NOT NULL,\n status TEXT NOT NULL,\n trigger TEXT DEFAULT 'manual',\n started_at TEXT DEFAULT (datetime('now')),\n finished_at TEXT,\n duration_ms INTEGER,\n input TEXT,\n output TEXT,\n error TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_agent_runs_agent ON agent_runs(agent_id, started_at DESC)`);\n\n // D1: customer cohorts. One row per cohort (e.g. signup month). Powers cohort-survival LTV\n // beyond the simple avg_deal × retention_months heuristic. retention_curve is a JSON array\n // [m1, m2, m3 ...] of retained customer counts. arpu = average revenue per user (monthly).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS customer_cohorts (\n id TEXT PRIMARY KEY,\n cohort_label TEXT NOT NULL,\n cohort_size INTEGER NOT NULL,\n arpu REAL NOT NULL,\n retention_curve TEXT NOT NULL,\n created_at TEXT DEFAULT (datetime('now'))\n )\n `);\n\n // D3: lead state changes — every MQL→SQL transition gets a row so we can compute\n // engagement velocity (avg days lead spent in each stage).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS lead_state_changes (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n lead_email TEXT NOT NULL,\n from_state TEXT,\n to_state TEXT NOT NULL,\n changed_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_lead_state_email ON lead_state_changes(lead_email, changed_at)`);\n\n // D5: utm_conversions — channel attribution log. Populated when a conversion fires\n // (manually pushed via POST /api/attribution or scraped from GA4 in a later phase).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS utm_conversions (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n utm_source TEXT,\n utm_medium TEXT,\n utm_campaign TEXT,\n conversion_value REAL DEFAULT 0,\n converted_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_utm_source ON utm_conversions(utm_source, converted_at DESC)`);\n\n // ─── FRAMEWORK LAYER (F1-F4): NSM, OKRs, Experiments, Tasks ─────────────────\n\n // F1: North Star Metric (singleton — only one active row; previous deleted on set)\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS north_star (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n metric_name TEXT NOT NULL,\n metric_source TEXT NOT NULL,\n metric_key TEXT,\n current_value REAL,\n target_value REAL NOT NULL,\n target_date TEXT NOT NULL,\n unit TEXT DEFAULT 'count',\n description TEXT,\n set_at TEXT DEFAULT (datetime('now')),\n set_by TEXT DEFAULT 'admin'\n )\n `);\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS nsm_history (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n nsm_id INTEGER REFERENCES north_star(id) ON DELETE CASCADE,\n value REAL NOT NULL,\n recorded_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_nsm_history_id ON nsm_history(nsm_id, recorded_at DESC)`);\n\n // F2: OKRs (Objectives + Key Results)\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS objectives (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n title TEXT NOT NULL,\n description TEXT,\n quarter TEXT NOT NULL,\n workspace TEXT DEFAULT 'marketing',\n owner TEXT,\n status TEXT DEFAULT 'active',\n progress REAL DEFAULT 0,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS key_results (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n objective_id INTEGER NOT NULL REFERENCES objectives(id) ON DELETE CASCADE,\n title TEXT NOT NULL,\n metric_source TEXT,\n metric_key TEXT,\n start_value REAL DEFAULT 0,\n current_value REAL DEFAULT 0,\n target_value REAL NOT NULL,\n unit TEXT DEFAULT 'count',\n confidence REAL DEFAULT 0.5,\n progress REAL DEFAULT 0,\n updated_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_kr_objective ON key_results(objective_id)`);\n\n // F3: ICE Experiments (ice_score + uplift_pct computed in app code — sqlite GENERATED\n // column support is uneven; safer to project them client-side from impact/confidence/ease).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS experiments (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n title TEXT NOT NULL,\n hypothesis TEXT,\n objective_id INTEGER REFERENCES objectives(id) ON DELETE SET NULL,\n funnel_stage TEXT,\n aarrr_lever TEXT,\n status TEXT DEFAULT 'idea',\n impact REAL DEFAULT 5,\n confidence REAL DEFAULT 5,\n ease REAL DEFAULT 5,\n owner TEXT,\n start_date TEXT,\n end_date TEXT,\n channel TEXT,\n success_metric TEXT,\n baseline_value REAL,\n result_value REAL,\n outcome TEXT,\n learnings TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_exp_status ON experiments(status, updated_at DESC)`);\n\n // F4: Tasks (linked to OKRs / KRs / experiments — at least one for governance).\n // NOTE: the gap-closure prompt also referred to a \"tasks\" table — the older one (string id,\n // status enum 'pending|running|done|failed') is replaced by this richer schema. The two\n // were never wired to a shared UI so the swap is safe.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS tasks (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n title TEXT NOT NULL,\n description TEXT,\n status TEXT DEFAULT 'todo',\n priority TEXT DEFAULT 'medium',\n workspace TEXT DEFAULT 'marketing',\n assignee TEXT,\n due_date TEXT,\n objective_id INTEGER REFERENCES objectives(id) ON DELETE SET NULL,\n key_result_id INTEGER REFERENCES key_results(id) ON DELETE SET NULL,\n experiment_id INTEGER REFERENCES experiments(id) ON DELETE SET NULL,\n ice_score REAL,\n impact REAL,\n confidence REAL,\n ease REAL,\n tags TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now')),\n completed_at TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_tasks_status ON tasks(status, due_date)`);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_tasks_links ON tasks(experiment_id, key_result_id, objective_id)`);\n\n // OKR→Campaign→Task link: a task can belong to a campaign (the higher-level container).\n // Completing the task rolls progress up through the campaign to its objective. PRAGMA-guarded\n // ALTER so existing tasks tables gain the column without failing on re-run.\n try {\n const tc = await sqlite.execute(\"PRAGMA table_info(tasks)\");\n const tHave = new Set((tc.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n if (!tHave.has(\"campaign_id\")) await sqlite.execute(\"ALTER TABLE tasks ADD COLUMN campaign_id TEXT\");\n } catch { /* non-critical: tasks without a campaign link still work */ }\n\n // ─── INTELLIGENCE LAYER EXTENSIONS (I1, I4): Signals + Digests ──────────────\n\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS intent_signals (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n lead_email TEXT NOT NULL,\n lead_name TEXT,\n company TEXT,\n signal_type TEXT NOT NULL,\n signal_source TEXT NOT NULL,\n signal_strength INTEGER DEFAULT 1,\n raw_data TEXT,\n detected_at TEXT DEFAULT (datetime('now')),\n actioned INTEGER DEFAULT 0,\n actioned_at TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_signals_email ON intent_signals(lead_email, detected_at DESC)`);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_signals_actioned ON intent_signals(actioned, detected_at DESC)`);\n\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS growth_digests (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n week_of TEXT NOT NULL,\n payload TEXT NOT NULL,\n nsm_value_at_generation REAL,\n generated_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_digest_week ON growth_digests(week_of DESC)`);\n\n // ─── UNIFIED CALENDAR (CAL1): standalone custom calendar entries ────────────\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS calendar_entries (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n title TEXT NOT NULL,\n description TEXT,\n event_date TEXT NOT NULL,\n event_time TEXT,\n end_date TEXT,\n end_time TEXT,\n event_type TEXT DEFAULT 'custom',\n workspace TEXT DEFAULT 'marketing',\n color TEXT DEFAULT 'blue',\n all_day INTEGER DEFAULT 0,\n recurrence TEXT,\n created_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_calendar_date ON calendar_entries(event_date)`);\n\n // Knowledge files — growth playbooks/frameworks/formulas injected into AI features (see api/knowledge.ts).\n // This table was missing from initDB, so the knowledge UI + /api/knowledge endpoints failed (\"no such table\").\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS knowledge_files (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n name TEXT NOT NULL,\n kind TEXT DEFAULT 'playbook',\n content TEXT NOT NULL,\n summary TEXT,\n targets TEXT DEFAULT '[]',\n source TEXT DEFAULT 'user',\n tokens_estimate INTEGER DEFAULT 0,\n enabled INTEGER DEFAULT 1,\n uploaded_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_knowledge_enabled ON knowledge_files(enabled, uploaded_at DESC)`);\n\n // Action queue — human-in-the-loop execution. The AI/agents PREPARE external actions\n // (send email, publish post, post to LinkedIn, create calendar event) here; nothing is\n // sent until a human approves, at which point the integration call runs and the result\n // is recorded. status: prepared → approved → executing → done | failed | rejected.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS action_queue (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n source_type TEXT DEFAULT 'chat',\n source_id TEXT,\n integration TEXT NOT NULL,\n endpoint TEXT NOT NULL,\n payload TEXT NOT NULL,\n preview TEXT,\n title TEXT,\n workspace TEXT DEFAULT 'marketing',\n status TEXT DEFAULT 'prepared',\n result TEXT,\n task_id TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n approved_at TEXT,\n executed_at TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_action_queue_status ON action_queue(status, created_at DESC)`);\n\n // CRM: sales pipeline deals. Hybrid model — rows are auto-seeded from qualified leads\n // (getLeadScores) via POST /api/deals/sync AND can be created/edited/dragged by the user.\n // stage ∈ lead | qualified | proposal | negotiation | won | lost. lead_email links a deal back\n // to its source lead so the detail view can show that contact's intent_signals as activity.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS deals (\n id TEXT PRIMARY KEY,\n title TEXT,\n company TEXT,\n contact_name TEXT,\n contact_email TEXT,\n stage TEXT NOT NULL DEFAULT 'lead',\n amount REAL DEFAULT 0,\n currency TEXT DEFAULT 'USD',\n probability INTEGER DEFAULT 0,\n owner TEXT,\n close_date TEXT,\n source TEXT DEFAULT 'manual',\n lead_email TEXT,\n notes TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now')),\n last_activity_at TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_deals_stage ON deals(stage, updated_at DESC)`);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_deals_lead ON deals(lead_email)`);\n\n // ─── RELATIONAL CRM FOUNDATION (WAVE 1): Accounts + Contacts ────────────────\n // First-class company (accounts) and person (contacts) records. Deals link to them via\n // account_id + primary_contact_id, plus a many-to-many deal_contacts join (role per contact).\n // Backfilled NON-DESTRUCTIVELY from the legacy deals.company/contact_name/contact_email text\n // columns by the mig_crm_relational_v1 migration below. All tenant-scoped (tenant_id on every\n // row; every query filters/sets it). The legacy deal text columns are KEPT for back-compat.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS accounts (\n id TEXT PRIMARY KEY,\n name TEXT,\n domain TEXT,\n industry TEXT,\n size TEXT,\n website TEXT,\n owner TEXT,\n tags TEXT DEFAULT '[]',\n custom_fields TEXT DEFAULT '{}',\n notes TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_accounts_name ON accounts(tenant_id, name)`);\n\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS contacts (\n id TEXT PRIMARY KEY,\n account_id TEXT,\n first_name TEXT,\n last_name TEXT,\n email TEXT,\n phone TEXT,\n title TEXT,\n owner TEXT,\n tags TEXT DEFAULT '[]',\n custom_fields TEXT DEFAULT '{}',\n notes TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_contacts_email ON contacts(tenant_id, email)`);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_contacts_account ON contacts(tenant_id, account_id)`);\n\n // Many-to-many: a deal can involve several contacts, each with a buying-committee role\n // (decision_maker | champion | influencer | user | contact; the migration uses \"primary\"\n // for the deal's primary_contact_id). Composite PK prevents duplicate links.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS deal_contacts (\n deal_id TEXT NOT NULL,\n contact_id TEXT NOT NULL,\n role TEXT DEFAULT 'contact',\n tenant_id TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n PRIMARY KEY (deal_id, contact_id)\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_deal_contacts_contact ON deal_contacts(tenant_id, contact_id)`);\n\n // deals gains account_id + primary_contact_id (guarded ALTER, re-run safe). The legacy\n // company/contact_name/contact_email columns are intentionally KEPT for back-compat.\n try {\n const dc = await sqlite.execute(\"PRAGMA table_info(deals)\");\n const dHave = new Set((dc.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n if (!dHave.has(\"account_id\")) await sqlite.execute(\"ALTER TABLE deals ADD COLUMN account_id TEXT\");\n if (!dHave.has(\"primary_contact_id\")) await sqlite.execute(\"ALTER TABLE deals ADD COLUMN primary_contact_id TEXT\");\n } catch { /* non-critical: relational links added next warm start */ }\n\n // ─── ENGAGEMENT + PIPELINE DEPTH (WAVE 2): Activities + Line Items + Stages ──\n // Unified activity timeline (logged calls/meetings/notes/emails/tasks) attachable to a\n // contact, account, and/or deal. The getTimeline() feed MERGES these rows with intent_signals\n // (by contact email) and linked tasks. All tenant-scoped (tenant_id on every row).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS activities (\n id TEXT PRIMARY KEY,\n type TEXT,\n subject TEXT,\n body TEXT,\n contact_id TEXT,\n account_id TEXT,\n deal_id TEXT,\n owner TEXT,\n due_date TEXT,\n completed INTEGER DEFAULT 0,\n occurred_at TEXT,\n created_at TEXT,\n updated_at TEXT,\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_activities_deal ON activities(tenant_id, deal_id)`);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_activities_contact ON activities(tenant_id, contact_id)`);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_activities_account ON activities(tenant_id, account_id)`);\n\n // Deal LINE ITEMS — products/services on a deal. The deal amount rolls up as SUM(amount) when\n // line items exist (else the manual deals.amount is kept). amount is persisted = quantity * unit_price.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS deal_products (\n id TEXT PRIMARY KEY,\n deal_id TEXT,\n name TEXT,\n quantity REAL DEFAULT 1,\n unit_price REAL DEFAULT 0,\n amount REAL DEFAULT 0,\n tenant_id TEXT,\n created_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_deal_products_deal ON deal_products(tenant_id, deal_id)`);\n\n // Configurable PIPELINE STAGES (label + win-probability per stage). Seeded once below with the\n // 6 existing stages so deal stage probabilities/labels become editable per tenant.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS pipeline_stages (\n id TEXT PRIMARY KEY,\n stage_key TEXT,\n label TEXT,\n probability INTEGER,\n sort_order INTEGER,\n is_won INTEGER DEFAULT 0,\n is_lost INTEGER DEFAULT 0,\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_pipeline_stages_tenant ON pipeline_stages(tenant_id, sort_order)`);\n\n // deals gains next_step + win/loss REASONS + recurring/MRR (guarded ALTER, re-run safe).\n try {\n const d2 = await sqlite.execute(\"PRAGMA table_info(deals)\");\n const d2Have = new Set((d2.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n for (const [col, def] of [\n [\"next_step\", \"TEXT\"], [\"win_reason\", \"TEXT\"], [\"lost_reason\", \"TEXT\"],\n [\"recurring_type\", \"TEXT\"], [\"mrr\", \"REAL DEFAULT 0\"],\n ] as Array<[string, string]>) {\n if (!d2Have.has(col)) await sqlite.execute(`ALTER TABLE deals ADD COLUMN ${col} ${def}`);\n }\n } catch { /* non-critical: deal next-step/reasons/MRR added next warm start */ }\n\n // tasks gains CRM links (deal_id / contact_id / account_id) so an existing Task can be a CRM\n // follow-up attached to a deal/contact/account — preserves ALL current task behavior.\n try {\n const t2 = await sqlite.execute(\"PRAGMA table_info(tasks)\");\n const t2Have = new Set((t2.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n for (const col of [\"deal_id\", \"contact_id\", \"account_id\"]) {\n if (!t2Have.has(col)) await sqlite.execute(`ALTER TABLE tasks ADD COLUMN ${col} TEXT`);\n }\n } catch { /* non-critical: CRM task links added next warm start */ }\n\n // ONE-TIME seed of the 6 default pipeline stages, gated by app_config \"mig_pipeline_stages_v1\".\n // Seeds for every tenant that already has pipeline_stages-eligible data (any present tenant_id\n // across tenants) OR at least the default tenant. Idempotent: INSERT OR IGNORE on a deterministic\n // id (tenant + stage_key) + an existence check, so a re-run never duplicates rows.\n try {\n const flag = await sqlite.execute(\"SELECT value FROM app_config WHERE key = 'mig_pipeline_stages_v1'\");\n if (flag.rows.length === 0) {\n const STAGE_SEEDS: Array<[string, string, number, number, number, number]> = [\n // stage_key, label, probability, sort_order, is_won, is_lost\n [\"lead\", \"Lead\", 10, 0, 0, 0],\n [\"qualified\", \"Qualified\", 30, 1, 0, 0],\n [\"proposal\", \"Proposal\", 50, 2, 0, 0],\n [\"negotiation\", \"Negotiation\", 70, 3, 0, 0],\n [\"won\", \"Won\", 100, 4, 1, 0],\n [\"lost\", \"Lost\", 0, 5, 0, 1],\n ];\n // Collect the set of tenants to seed: every distinct tenant on the tenants table, always\n // including t_default. Falls back to just t_default if the tenants table is empty/missing.\n const tenantIds = new Set([\"t_default\"]);\n try {\n const tns = await sqlite.execute(\"SELECT id FROM tenants\");\n for (const r of (tns.rows || [])) {\n const tid = String(Array.isArray(r) ? r[0] : (r as any).id ?? \"\").trim();\n if (tid) tenantIds.add(tid);\n }\n } catch { /* tenants table may not exist on a very fresh DB — default tenant still seeded */ }\n for (const tid of tenantIds) {\n for (const [key, label, prob, sort, isWon, isLost] of STAGE_SEEDS) {\n await sqlite.execute(\n `INSERT OR IGNORE INTO pipeline_stages (id, stage_key, label, probability, sort_order, is_won, is_lost, tenant_id)\n VALUES (?, ?, ?, ?, ?, ?, ?, ?)`,\n [\"ps_\" + tid + \"_\" + key, key, label, prob, sort, isWon, isLost, tid]\n );\n }\n }\n await sqlite.execute(\n \"INSERT OR REPLACE INTO app_config (key, value, tenant_id, updated_at) VALUES ('mig_pipeline_stages_v1', 'done', 't_default', datetime('now'))\"\n );\n }\n } catch { /* non-critical: stage seed retried next warm start (flag only set on success) */ }\n\n // ─── SALES ANALYTICS + LEAD OPS + PRODUCTIVITY (WAVE 3) ─────────────────────\n // Eight new tenant-scoped tables overlaid on the existing deals/contacts/intent_signals/users\n // foundation. All carry tenant_id on every row (every query filters/sets it) and are also added\n // to TENANT_SCOPED_TABLES below so the mig_tenant_v1 backfill covers them on legacy DBs.\n\n // A) SALES ANALYTICS ────────────────────────────────────────────────────────\n // deal_stage_history: one row per deal stage change (deals.ts logs it). Powers stage-conversion,\n // deal-velocity (time in stage), and the funnel/leaderboard analytics.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS deal_stage_history (\n id TEXT PRIMARY KEY,\n deal_id TEXT,\n from_stage TEXT,\n to_stage TEXT,\n owner TEXT,\n changed_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_deal_stage_history_deal ON deal_stage_history(tenant_id, deal_id)`);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_deal_stage_history_to ON deal_stage_history(tenant_id, to_stage)`);\n\n // sales_quotas: per-owner revenue target for a period. period_type ∈ month|quarter|year; period is\n // the label (e.g. 2026-Q2 or 2026-06). Powers quota-attainment (won vs target).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS sales_quotas (\n id TEXT PRIMARY KEY,\n owner TEXT,\n period TEXT,\n period_type TEXT,\n target_amount REAL DEFAULT 0,\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_sales_quotas_owner ON sales_quotas(tenant_id, owner, period)`);\n\n // pipeline_snapshots: one row PER STAGE per snapshot_date (point-in-time pipeline state). Powers\n // pipeline trend over time. weighted_amount = total_amount × stage probability at capture time.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS pipeline_snapshots (\n id TEXT PRIMARY KEY,\n snapshot_date TEXT,\n stage TEXT,\n deal_count INTEGER DEFAULT 0,\n total_amount REAL DEFAULT 0,\n weighted_amount REAL DEFAULT 0,\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_pipeline_snapshots_date ON pipeline_snapshots(tenant_id, snapshot_date)`);\n\n // B) LEAD OPS ─────────────────────────────────────────────────────────────────\n // lead_status: grade/owner/SLA overlay keyed by lead_email (one row per lead per tenant). grade ∈\n // A|B|C|D; status ∈ new|working|qualified|disqualified. sla_due_at = first-response deadline,\n // first_response_at = when the SLA clock was stopped. Indexed (tenant_id, lead_email) for upsert.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS lead_status (\n id TEXT PRIMARY KEY,\n lead_email TEXT,\n grade TEXT,\n owner TEXT,\n status TEXT DEFAULT 'new',\n first_response_at TEXT,\n sla_due_at TEXT,\n assigned_at TEXT,\n updated_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n // UNIQUE so the one-row-per-(tenant,lead) overlay can't be duplicated by a concurrent grade/assign race.\n await sqlite.execute(`CREATE UNIQUE INDEX IF NOT EXISTS idx_lead_status_email ON lead_status(tenant_id, lead_email)`);\n\n // C) PRODUCTIVITY ─────────────────────────────────────────────────────────────\n // email_templates: reusable subject/body with {{merge_field}} placeholders; rendered then handed\n // to the existing Gmail/Outlook draft path.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS email_templates (\n id TEXT PRIMARY KEY,\n name TEXT,\n subject TEXT,\n body TEXT,\n category TEXT,\n owner TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_email_templates_tenant ON email_templates(tenant_id, category)`);\n\n // sequences: a named multi-step cadence (active=1 by default).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS sequences (\n id TEXT PRIMARY KEY,\n name TEXT,\n description TEXT,\n active INTEGER DEFAULT 1,\n created_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_sequences_tenant ON sequences(tenant_id, active)`);\n\n // sequence_steps: ordered steps of a sequence. channel ∈ email|task|call. delay_days = wait after\n // the prior step. template_id optionally references an email_templates row; else inline subject/body.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS sequence_steps (\n id TEXT PRIMARY KEY,\n sequence_id TEXT,\n step_order INTEGER DEFAULT 0,\n channel TEXT DEFAULT 'email',\n delay_days INTEGER DEFAULT 0,\n template_id TEXT,\n subject TEXT,\n body TEXT,\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_sequence_steps_seq ON sequence_steps(tenant_id, sequence_id, step_order)`);\n\n // sequence_enrollments: a contact (and optionally deal) enrolled in a sequence. current_step is the\n // 0-based pointer; status ∈ active|paused|completed; next_action_at drives the cadence runner.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS sequence_enrollments (\n id TEXT PRIMARY KEY,\n sequence_id TEXT,\n contact_id TEXT,\n deal_id TEXT,\n current_step INTEGER DEFAULT 0,\n status TEXT DEFAULT 'active',\n enrolled_at TEXT DEFAULT (datetime('now')),\n next_action_at TEXT,\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_seq_enroll_seq ON sequence_enrollments(tenant_id, sequence_id, status)`);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_seq_enroll_contact ON sequence_enrollments(tenant_id, contact_id)`);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_seq_enroll_next ON sequence_enrollments(tenant_id, next_action_at)`);\n\n // ─── MARKETING WORKSPACE (WAVE 1) ─────────────────────────────────────────\n // Persona/ICP/segment library for the Marketing Audiences view. kind ∈ icp|persona|segment;\n // attributes is a JSON blob; source ∈ enrichment|manual (importPersonasFromEnrichment writes\n // 'enrichment' rows non-destructively). All tenant-scoped (tenant_id on every row).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS marketing_personas (\n id TEXT PRIMARY KEY,\n name TEXT,\n kind TEXT,\n description TEXT,\n attributes TEXT,\n source TEXT DEFAULT 'manual',\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_marketing_personas_kind ON marketing_personas(tenant_id, kind)`);\n\n // utm_links: saved campaign-tagged URLs for the UTM builder. full_url is the composed\n // base_url + utm params. All tenant-scoped (tenant_id on every row).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS utm_links (\n id TEXT PRIMARY KEY,\n name TEXT,\n base_url TEXT,\n utm_source TEXT,\n utm_medium TEXT,\n utm_campaign TEXT,\n utm_term TEXT,\n utm_content TEXT,\n full_url TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_utm_links_tenant ON utm_links(tenant_id)`);\n\n // brand_kit: one row per tenant (upsert) — voice/tone/values/dos/donts/boilerplate plus a JSON\n // palette and logo_url for the Brand Kit editor. tenant_id on every row.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS brand_kit (\n id TEXT PRIMARY KEY,\n voice TEXT,\n tone TEXT,\n brand_values TEXT,\n dos TEXT,\n donts TEXT,\n boilerplate TEXT,\n palette TEXT,\n logo_url TEXT,\n updated_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_brand_kit_tenant ON brand_kit(tenant_id)`);\n\n // ─── MARKETING CAPTURE + AUDIENCES (WAVE 2) ─────────────────────────────────\n // Six new tenant-scoped tables giving Marketing a place to CONVERT traffic (forms +\n // landing pages) and people to TARGET (segments + lists). All carry tenant_id on every\n // row (every query filters/sets it) EXCEPT the two PUBLIC read paths, which resolve the\n // owning tenant explicitly by id/slug. Added to TENANT_SCOPED_TABLES below so the\n // mig_tenant_v1 backfill covers them on legacy DBs. NEVER name a column with a SQL\n // reserved word (e.g. values/order/group) — a prior bug came from exactly that.\n\n // marketing_forms: a lead-capture form definition. fields is JSON [{name,label,type,required}].\n // asset_url = gated lead-magnet download; redirect_url = post-submit redirect. submit_count is\n // bumped on each public submission (by the form-owning tenant).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS marketing_forms (\n id TEXT PRIMARY KEY,\n name TEXT,\n fields TEXT,\n success_message TEXT,\n asset_url TEXT,\n redirect_url TEXT,\n submit_count INTEGER DEFAULT 0,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_marketing_forms_tenant ON marketing_forms(tenant_id)`);\n\n // form_submissions: one row per public submit. data is JSON of the raw posted fields.\n // tenant_id is the OWNING tenant of form_id (resolved by id at submit time — never the\n // anonymous caller's session). Indexed (tenant_id, form_id) for the submissions table.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS form_submissions (\n id TEXT PRIMARY KEY,\n form_id TEXT,\n data TEXT,\n lead_email TEXT,\n lead_name TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_form_submissions_form ON form_submissions(tenant_id, form_id)`);\n\n // landing_pages: a public landing page. slug is GLOBALLY unique (used for the public\n // /api/marketing/p/:slug render across tenants). blocks is JSON [{type,...}]. form_id\n // optionally embeds a marketing_forms form. published=1 gates the public render; view_count\n // is bumped on each public view (by the page-owning tenant).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS landing_pages (\n id TEXT PRIMARY KEY,\n slug TEXT,\n title TEXT,\n subtitle TEXT,\n blocks TEXT,\n form_id TEXT,\n theme TEXT,\n published INTEGER DEFAULT 0,\n view_count INTEGER DEFAULT 0,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_landing_pages_tenant ON landing_pages(tenant_id)`);\n // slug is GLOBALLY unique (public render resolves a page by slug across all tenants).\n await sqlite.execute(`CREATE UNIQUE INDEX IF NOT EXISTS idx_landing_pages_slug ON landing_pages(slug)`);\n\n // segments: a saved audience definition. kind ∈ dynamic|static. rules is JSON\n // [{field,op,value}] evaluated over contacts + lead_status + intent_signals within the tenant.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS segments (\n id TEXT PRIMARY KEY,\n name TEXT,\n description TEXT,\n kind TEXT,\n rules TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_segments_tenant ON segments(tenant_id)`);\n\n // marketing_lists: a named contact list. kind ∈ static|dynamic. For dynamic lists, segment_id\n // references the segment whose rules materialize the membership.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS marketing_lists (\n id TEXT PRIMARY KEY,\n name TEXT,\n kind TEXT,\n segment_id TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_marketing_lists_tenant ON marketing_lists(tenant_id)`);\n\n // list_members: membership join (a contact in a list). lead_email is the stable key when a\n // contact_id is unavailable. Indexed (tenant_id, list_id) for the members view.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS list_members (\n id TEXT PRIMARY KEY,\n list_id TEXT,\n contact_id TEXT,\n lead_email TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_list_members_list ON list_members(tenant_id, list_id)`);\n\n // ─── CONTENT + CAMPAIGN PRODUCTION ENGINE (WAVE 3) ──────────────────────────\n // Four new tenant-scoped tables powering the editorial calendar, SEO briefs, asset\n // library, and campaign budgeting/performance. All carry tenant_id on every row (every\n // query filters/sets it) and are added to TENANT_SCOPED_TABLES below so the mig_tenant_v1\n // backfill covers them on legacy DBs. NEVER name a column with a SQL reserved word.\n\n // content_items: a piece of marketing content moving through the pipeline. type ∈\n // blog|social|video|email|landing|ad; status ∈ idea|brief|draft|review|scheduled|published.\n // campaign_id links to campaigns; brief_id links to content_briefs; asset_ids is JSON of\n // marketing_assets ids. scheduled_at drives the editorial calendar; body holds the draft.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS content_items (\n id TEXT PRIMARY KEY,\n title TEXT,\n type TEXT,\n status TEXT DEFAULT 'idea',\n channel TEXT,\n campaign_id TEXT,\n brief_id TEXT,\n body TEXT,\n asset_ids TEXT DEFAULT '[]',\n owner TEXT,\n scheduled_at TEXT,\n published_at TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_content_items_status ON content_items(tenant_id, status)`);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_content_items_scheduled ON content_items(tenant_id, scheduled_at)`);\n\n // content_briefs: an SEO brief backing a content_item. outline/competitors/internal_links are\n // JSON or free text; word_count_target is the writer's target. content_id links back to the item.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS content_briefs (\n id TEXT PRIMARY KEY,\n content_id TEXT,\n target_keyword TEXT,\n search_intent TEXT,\n outline TEXT,\n competitors TEXT,\n internal_links TEXT,\n word_count_target INTEGER,\n notes TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_content_briefs_tenant ON content_briefs(tenant_id)`);\n\n // marketing_assets: the media/asset registry. asset_type ∈ image|video|doc|pdf|other;\n // source ∈ upload|wordpress|drive|link; tags is JSON or CSV; bytes is the file size.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS marketing_assets (\n id TEXT PRIMARY KEY,\n name TEXT,\n url TEXT,\n asset_type TEXT,\n tags TEXT,\n source TEXT DEFAULT 'upload',\n bytes INTEGER,\n created_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_marketing_assets_tenant ON marketing_assets(tenant_id)`);\n\n // campaign_budgets: planned vs actual spend per channel per period for a campaign. period is a\n // label (e.g. 2026-Q2 or 2026-06). actual_amount accrues as spend is recorded; powers pacing.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS campaign_budgets (\n id TEXT PRIMARY KEY,\n campaign_id TEXT,\n channel TEXT,\n period TEXT,\n planned_amount REAL DEFAULT 0,\n actual_amount REAL DEFAULT 0,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_campaign_budgets_campaign ON campaign_budgets(tenant_id, campaign_id)`);\n\n // ─── ACQUISITION CHANNELS + AUTOMATION DEPTH (WAVE 4) ───────────────────────\n // Six new tenant-scoped tables: paid media (ad_campaigns), triggered automation\n // (journeys + journey_runs), account-based marketing (abm_targets), email broadcasts\n // (broadcasts), and partner/referral channels (partners). All carry tenant_id on every\n // row (every query filters/sets it) and are added to TENANT_SCOPED_TABLES below so the\n // mig_tenant_v1 backfill covers them on legacy DBs. NEVER name a column with a SQL\n // reserved word — all column names here are non-reserved (e.g. no values/order/group).\n\n // ad_campaigns: paid-media spend rows. platform ∈ google_ads|meta|linkedin|other. external_id is\n // the platform's own campaign id; campaign_id links to an internal campaigns row. period is a label\n // (e.g. 2026-Q2 or 2026-06). Real data only — manual entry or (future) connected creds. Indexed\n // (tenant_id, platform) for the per-platform table.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS ad_campaigns (\n id TEXT PRIMARY KEY,\n platform TEXT,\n external_id TEXT,\n name TEXT,\n status TEXT,\n spend REAL DEFAULT 0,\n impressions INTEGER DEFAULT 0,\n clicks INTEGER DEFAULT 0,\n conversions INTEGER DEFAULT 0,\n campaign_id TEXT,\n period TEXT,\n synced_at TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_ad_campaigns_platform ON ad_campaigns(tenant_id, platform)`);\n\n // journeys: triggered-automation rules. trigger_type ∈ form_submit|intent_signal|deal_stage|\n // list_added|manual; trigger_config is JSON matched against the runtime ctx. action_type ∈\n // enroll_sequence|add_to_list|create_task|notify|grade_lead; action_config is JSON for the action.\n // active=1 gates execution; run_count bumps on each match. Indexed (tenant_id, trigger_type, active)\n // for the runtime lookup in runJourneyTriggers.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS journeys (\n id TEXT PRIMARY KEY,\n name TEXT,\n description TEXT,\n trigger_type TEXT,\n trigger_config TEXT,\n action_type TEXT,\n action_config TEXT,\n active INTEGER DEFAULT 1,\n run_count INTEGER DEFAULT 0,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_journeys_trigger ON journeys(tenant_id, trigger_type, active)`);\n\n // journey_runs: execution log. One row per journey evaluation. trigger_type mirrors the journey's;\n // entity_ref is the triggering entity (e.g. lead_email or deal_id). status ∈ matched|acted|skipped|\n // error; detail holds a short outcome message. Indexed (tenant_id, journey_id) for the run log.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS journey_runs (\n id TEXT PRIMARY KEY,\n journey_id TEXT,\n trigger_type TEXT,\n entity_ref TEXT,\n status TEXT,\n detail TEXT,\n ran_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_journey_runs_journey ON journey_runs(tenant_id, journey_id)`);\n\n // abm_targets: account-based marketing target accounts. account_id links to a CRM accounts row.\n // tier ∈ 1|2|3 (stored as text); engagement_score computed from real signals/activities/deals.\n // status ∈ target|engaged|opportunity|won|dropped. Indexed (tenant_id, account_id) for upsert/board.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS abm_targets (\n id TEXT PRIMARY KEY,\n account_id TEXT,\n tier TEXT,\n engagement_score REAL DEFAULT 0,\n status TEXT DEFAULT 'target',\n owner TEXT,\n notes TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_abm_targets_account ON abm_targets(tenant_id, account_id)`);\n\n // broadcasts: one-off/scheduled email sends. audience_type ∈ list|segment|all_contacts; audience_id\n // references the chosen list/segment (NULL for all_contacts). channel ∈ gmail|outlook. status ∈\n // draft|scheduled|sending|sent. sent_count accrues as the send resolves. Indexed (tenant_id, status).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS broadcasts (\n id TEXT PRIMARY KEY,\n name TEXT,\n subject TEXT,\n body TEXT,\n audience_type TEXT,\n audience_id TEXT,\n channel TEXT,\n status TEXT DEFAULT 'draft',\n scheduled_at TEXT,\n sent_count INTEGER DEFAULT 0,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_broadcasts_status ON broadcasts(tenant_id, status)`);\n\n // partners: referral/affiliate/reseller/agency channel partners. type ∈ referral|affiliate|reseller|\n // agency. referral_code is the code carried on a referred deal/contact source. status ∈ active|paused.\n // deals_referred + revenue accrue from referral attribution. Indexed (tenant_id).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS partners (\n id TEXT PRIMARY KEY,\n name TEXT,\n type TEXT,\n contact_email TEXT,\n referral_code TEXT,\n status TEXT DEFAULT 'active',\n deals_referred INTEGER DEFAULT 0,\n revenue REAL DEFAULT 0,\n created_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_partners_tenant ON partners(tenant_id)`);\n\n // ── Wave 5: analytics / optimization / market-intel layer ──\n // competitors — tracked rival domains. metrics is a JSON snapshot (domain_rating/backlinks/\n // ref_domains/organic_keywords) refreshed from the Ahrefs integration. tracked_keywords is a\n // JSON array. Tenant-scoped, indexed (tenant_id).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS competitors (\n id TEXT PRIMARY KEY,\n name TEXT,\n domain TEXT,\n notes TEXT,\n tracked_keywords TEXT,\n metrics TEXT,\n last_checked TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_competitors_tenant ON competitors(tenant_id)`);\n\n // cro_tests — conversion-rate-optimization tests linking an experiment + a landing_page.\n // status ∈ draft|running|complete. winner names the winning arm once resolved. Results read\n // from the linked experiment (stat-significance reused from api/experiments.ts). Indexed\n // (tenant_id, status).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS cro_tests (\n id TEXT PRIMARY KEY,\n name TEXT,\n hypothesis TEXT,\n experiment_id INTEGER,\n landing_page_id TEXT,\n status TEXT DEFAULT 'draft',\n winner TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_cro_tests_tenant ON cro_tests(tenant_id, status)`);\n\n // market_digests — caches the AI market-intel digest. summary is the synthesized text; signals\n // is a JSON list of the real sources used (ProductHunt/YouTube/intent_signals/competitor metrics).\n // Indexed (tenant_id).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS market_digests (\n id TEXT PRIMARY KEY,\n period TEXT,\n summary TEXT,\n signals TEXT,\n generated_at TEXT DEFAULT (datetime('now')),\n tenant_id TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_market_digests_tenant ON market_digests(tenant_id)`);\n\n // I3: agents.role column for Listener/Topic/Creator/Analyst/General roles (ALTER guarded)\n try {\n const ac = await sqlite.execute(\"PRAGMA table_info(agents)\");\n const aHave = new Set((ac.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n if (!aHave.has(\"role\")) await sqlite.execute(\"ALTER TABLE agents ADD COLUMN role TEXT DEFAULT 'general'\");\n } catch { /* non-critical */ }\n\n // Part 15.3: persistent business memory (evolving learned facts). Additive on v551.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS memory (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n type TEXT DEFAULT 'fact',\n content TEXT NOT NULL,\n source TEXT DEFAULT 'manual',\n confidence REAL DEFAULT 0.7,\n pinned INTEGER DEFAULT 0,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now'))\n )\n `);\n\n // Part 1: skills (marketing playbooks) STATE table — enabled/pinned + custom skills.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS skills (\n name TEXT PRIMARY KEY,\n enabled INTEGER DEFAULT 1,\n pinned INTEGER DEFAULT 0,\n source TEXT DEFAULT 'builtin',\n category TEXT,\n description TEXT,\n body TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now'))\n )\n `);\n\n // Part 3 fix 7: experiment significance + sample_size (PRAGMA-guarded ALTER).\n try {\n const ec = await sqlite.execute(\"PRAGMA table_info(experiments)\");\n const eHave = new Set((ec.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n if (!eHave.has(\"sample_size\")) await sqlite.execute(\"ALTER TABLE experiments ADD COLUMN sample_size INTEGER\");\n if (!eHave.has(\"significance\")) await sqlite.execute(\"ALTER TABLE experiments ADD COLUMN significance REAL\");\n } catch { /* non-critical */ }\n\n // Part 14.2: agent_runs.trace JSON column (structured TraceStep[]).\n try {\n const rc = await sqlite.execute(\"PRAGMA table_info(agent_runs)\");\n const rHave = new Set((rc.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n if (!rHave.has(\"trace\")) await sqlite.execute(\"ALTER TABLE agent_runs ADD COLUMN trace TEXT\");\n } catch { /* non-critical */ }\n\n // Multi-agent CAMPAIGNS — the higher-level container the user wants: a campaign has many\n // inputs + MANY member agents that cooperate. A campaign RUN is the Director orchestrating\n // its member agents (sequential handoff or parallel) and synthesizing a final output.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS campaigns (\n id TEXT PRIMARY KEY,\n name TEXT NOT NULL,\n workspace TEXT NOT NULL DEFAULT 'marketing',\n description TEXT,\n goal TEXT,\n channel TEXT,\n audience TEXT,\n mode TEXT DEFAULT 'sequential',\n status TEXT DEFAULT 'active',\n objective_id INTEGER,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now')),\n created_by TEXT DEFAULT 'user'\n )\n `);\n // Ordered membership: which agents are in the campaign + their pipeline order + role.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS campaign_agents (\n campaign_id TEXT NOT NULL,\n agent_id TEXT NOT NULL,\n sequence INTEGER DEFAULT 0,\n role TEXT,\n PRIMARY KEY (campaign_id, agent_id)\n )\n `);\n // One coordinated pipeline execution (the Director run).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS campaign_runs (\n id TEXT PRIMARY KEY,\n campaign_id TEXT NOT NULL,\n status TEXT NOT NULL DEFAULT 'running',\n input TEXT,\n final_output TEXT,\n trace TEXT,\n started_at TEXT DEFAULT (datetime('now')),\n finished_at TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_campaign_runs ON campaign_runs(campaign_id, started_at DESC)`);\n // Campaign TRIGGERS + LOOP (run continuously, not just on/off). trigger_type: manual | interval\n // | schedule | event. trigger_config: JSON (e.g. {minutes:60} or {event:\"new_lead\"}). loop=1 ⇒\n // reschedule after each run. next_run_at drives the cron-based runner. (Guarded ALTER, re-run safe.)\n try {\n const cc = await sqlite.execute(\"PRAGMA table_info(campaigns)\");\n const cHave = new Set((cc.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n for (const [col, def] of [[\"trigger_type\", \"TEXT DEFAULT 'manual'\"], [\"trigger_config\", \"TEXT\"], [\"next_run_at\", \"TEXT\"], [\"last_run_at\", \"TEXT\"], [\"loop\", \"INTEGER DEFAULT 0\"]] as Array<[string, string]>) {\n if (!cHave.has(col)) await sqlite.execute(`ALTER TABLE campaigns ADD COLUMN ${col} ${def}`);\n }\n } catch { /* non-critical */ }\n // agent_runs gains campaign context: which campaign run + the handed-off context from prior agents.\n try {\n const arc = await sqlite.execute(\"PRAGMA table_info(agent_runs)\");\n const arHave = new Set((arc.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n if (!arHave.has(\"campaign_run_id\")) await sqlite.execute(\"ALTER TABLE agent_runs ADD COLUMN campaign_run_id TEXT\");\n if (!arHave.has(\"received_input\")) await sqlite.execute(\"ALTER TABLE agent_runs ADD COLUMN received_input TEXT\");\n // task_id links an agent_run back to the task it executed (POST /api/tasks/:id/run).\n if (!arHave.has(\"task_id\")) await sqlite.execute(\"ALTER TABLE agent_runs ADD COLUMN task_id TEXT\");\n } catch { /* non-critical */ }\n\n // ─── AUTH / MULTI-TENANCY FOUNDATION ───────────────────────────────────────\n // New auth tables. These have NO tenant_id (they ARE the tenant/identity layer).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS tenants (\n id TEXT PRIMARY KEY,\n name TEXT NOT NULL,\n slug TEXT UNIQUE,\n plan TEXT DEFAULT 'trial',\n status TEXT DEFAULT 'active',\n created_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS users (\n id TEXT PRIMARY KEY,\n email TEXT UNIQUE NOT NULL,\n name TEXT,\n password_hash TEXT,\n avatar_url TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n last_login_at TEXT\n )\n `);\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS memberships (\n tenant_id TEXT NOT NULL,\n user_id TEXT NOT NULL,\n role TEXT DEFAULT 'owner',\n created_at TEXT DEFAULT (datetime('now')),\n PRIMARY KEY (tenant_id, user_id)\n )\n `);\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS sessions (\n id TEXT PRIMARY KEY,\n user_id TEXT NOT NULL,\n tenant_id TEXT NOT NULL,\n created_at TEXT DEFAULT (datetime('now')),\n expires_at TEXT,\n user_agent TEXT,\n revoked INTEGER DEFAULT 0\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_sessions_user ON sessions(user_id)`);\n\n // Phase D: read-only impersonation (\"view as tenant\"). A platform admin toggles this flag on\n // THEIR OWN session row; resolveSessionToken then redirects currentTenantId() to this target\n // tenant for reads only (all writes blocked by the main.ts guard). NULL = not impersonating.\n // PRAGMA-guarded ALTER (idempotent, re-run safe — same pattern as the other ALTERs above).\n try {\n const sc = await sqlite.execute(\"PRAGMA table_info(sessions)\");\n const sHave = new Set((sc.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n if (!sHave.has(\"impersonate_tenant_id\")) await sqlite.execute(\"ALTER TABLE sessions ADD COLUMN impersonate_tenant_id TEXT\");\n } catch { /* non-critical: impersonation disabled until column exists; added next warm start */ }\n\n // ─── PLATFORM SUPERADMIN LAYER (above tenants) ─────────────────────────────\n // The single, audited exception to per-tenant isolation: a platform owner who can read\n // ACROSS all tenants. users gains is_platform_admin (the master flag) + status (active/…).\n // PRAGMA-guarded ALTERs (idempotent, re-run safe — same pattern as the other ALTERs above).\n try {\n const uc = await sqlite.execute(\"PRAGMA table_info(users)\");\n const uHave = new Set((uc.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n if (!uHave.has(\"is_platform_admin\")) await sqlite.execute(\"ALTER TABLE users ADD COLUMN is_platform_admin INTEGER DEFAULT 0\");\n if (!uHave.has(\"status\")) await sqlite.execute(\"ALTER TABLE users ADD COLUMN status TEXT DEFAULT 'active'\");\n } catch { /* non-critical: master flag set by migration below on next warm start */ }\n\n // platform_audit — append-only log of every master-only action (suspend/delete/impersonate in\n // LATER phases). Created now so the audit surface exists from the start. No tenant_id: it spans\n // all tenants by design (this is the cross-tenant exception).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS platform_audit (\n id TEXT PRIMARY KEY,\n actor_user_id TEXT,\n actor_email TEXT,\n action TEXT,\n target_type TEXT,\n target_id TEXT,\n detail TEXT,\n created_at TEXT DEFAULT (datetime('now'))\n )\n `);\n\n // tenant_id column on EVERY existing business table. PRAGMA-guarded ALTER (idempotent, same\n // pattern used for other ALTERs above). Do NOT add to tenants/users/memberships/sessions.\n const TENANT_SCOPED_TABLES = [\n \"integrations\", \"llm_config\", \"agents\", \"conversations\", \"tasks\", \"tasks_legacy\",\n \"notes\", \"metric_baselines\", \"intelligence_cache\", \"app_config\", \"metric_history\",\n \"agent_budgets\", \"agent_runs\", \"customer_cohorts\", \"lead_state_changes\", \"utm_conversions\",\n \"north_star\", \"nsm_history\", \"objectives\", \"key_results\", \"experiments\", \"intent_signals\",\n \"growth_digests\", \"calendar_entries\", \"knowledge_files\", \"action_queue\", \"deals\", \"memory\",\n \"skills\", \"campaigns\", \"campaign_agents\", \"campaign_runs\",\n \"accounts\", \"contacts\", \"deal_contacts\",\n \"activities\", \"deal_products\", \"pipeline_stages\",\n \"deal_stage_history\", \"sales_quotas\", \"pipeline_snapshots\", \"lead_status\",\n \"email_templates\", \"sequences\", \"sequence_steps\", \"sequence_enrollments\",\n \"marketing_personas\", \"utm_links\", \"brand_kit\",\n \"marketing_forms\", \"form_submissions\", \"landing_pages\",\n \"segments\", \"marketing_lists\", \"list_members\",\n \"content_items\", \"content_briefs\", \"marketing_assets\", \"campaign_budgets\",\n \"ad_campaigns\", \"journeys\", \"journey_runs\", \"abm_targets\", \"broadcasts\", \"partners\",\n ];\n for (const tbl of TENANT_SCOPED_TABLES) {\n try {\n const cols = await sqlite.execute(`PRAGMA table_info(${tbl})`);\n const have = new Set((cols.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n // Only ALTER if the table actually exists (PRAGMA returns 0 cols for missing tables) and lacks tenant_id.\n if (have.size > 0 && !have.has(\"tenant_id\")) {\n await sqlite.execute(`ALTER TABLE ${tbl} ADD COLUMN tenant_id TEXT`);\n }\n } catch { /* non-critical: table may not exist yet (e.g. tasks_legacy) */ }\n }\n\n // ONE-TIME backfill + seed, gated by app_config \"mig_tenant_v1\" so it runs once. Backfill\n // every business table's tenant_id to \"t_default\" and insert the default tenant row. Owner\n // user creation is left to api/auth.ts bootstrapOwner (no password seeded here).\n try {\n const flag = await sqlite.execute(\"SELECT value FROM app_config WHERE key = 'mig_tenant_v1'\");\n if (flag.rows.length === 0) {\n for (const tbl of TENANT_SCOPED_TABLES) {\n try {\n const cols = await sqlite.execute(`PRAGMA table_info(${tbl})`);\n const have = new Set((cols.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n if (have.has(\"tenant_id\")) {\n await sqlite.execute(`UPDATE ${tbl} SET tenant_id = 't_default' WHERE tenant_id IS NULL`);\n }\n } catch { /* non-critical: skip missing tables */ }\n }\n await sqlite.execute(\n \"INSERT OR IGNORE INTO tenants (id, name, slug) VALUES ('t_default', 'Default Workspace', 'default')\"\n );\n // NOTE: app_config now has tenant_id; write the migration flag scoped to t_default so the\n // gate read above (no tenant filter) still finds it on subsequent boots.\n await sqlite.execute(\n \"INSERT OR REPLACE INTO app_config (key, value, tenant_id, updated_at) VALUES ('mig_tenant_v1', 'done', 't_default', datetime('now'))\"\n );\n }\n } catch { /* non-critical: migration retried next warm start */ }\n\n // ─── PLATFORM MASTER PROMOTION (mig_platform_admin_v1) ──────────────────────\n // ONE-TIME: promote the platform owner (PLATFORM_OWNER_EMAIL env, default the bootstrapped\n // master) to is_platform_admin=1. Gated by app_config \"mig_platform_admin_v1\" so it runs once.\n // api/auth.ts bootstrapOwner ALSO sets the flag at signup time; this covers the already-existing\n // master row. Idempotent: re-running the UPDATE is harmless even if the flag write ever fails.\n try {\n const flag = await sqlite.execute(\"SELECT value FROM app_config WHERE key = 'mig_platform_admin_v1'\");\n if (flag.rows.length === 0) {\n const masterEmail = (Deno.env.get(\"PLATFORM_OWNER_EMAIL\") || \"\").trim().toLowerCase();\n await sqlite.execute(\n \"UPDATE users SET is_platform_admin = 1 WHERE lower(email) = ?\",\n [masterEmail]\n );\n await sqlite.execute(\n \"INSERT OR REPLACE INTO app_config (key, value, tenant_id, updated_at) VALUES ('mig_platform_admin_v1', 'done', 't_default', datetime('now'))\"\n );\n }\n } catch { /* non-critical: promotion retried next warm start (flag only set on success) */ }\n\n // ─── INTEGRATIONS UNIQUENESS → PER-TENANT (mig_integrations_v2) ──────────────\n // The original integrations table has `name TEXT NOT NULL UNIQUE` (a GLOBAL unique\n // constraint). Multi-tenant needs each tenant to own its OWN 'google'/'outlook'/etc.\n // row, so uniqueness must be PER TENANT: UNIQUE(tenant_id, name). SQLite cannot ALTER a\n // UNIQUE constraint in place, so we do a ONE-TIME guarded table rebuild gated by an\n // app_config flag (\"mig_integrations_v2\"). Idempotent + safe on the live DB: we only\n // rebuild when the flag is absent AND the live table still carries the global UNIQUE(name)\n // index. By this point Stage 1's ALTER + backfill guarantee every existing integrations\n // row already has tenant_id='t_default', so the copy preserves correct ownership.\n try {\n const flag = await sqlite.execute(\"SELECT value FROM app_config WHERE key = 'mig_integrations_v2'\");\n if (flag.rows.length === 0) {\n // Confirm the table exists before rebuilding. A FRESH DB already gets the per-tenant\n // shape from the top-level CREATE (tenant_id present), so detect the OLD shape — the\n // global UNIQUE(name) index lacks the tenant_id column — and only rebuild then.\n const icols = await sqlite.execute(\"PRAGMA table_info(integrations)\");\n const iHave = new Set((icols.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n let needsRebuild = false;\n if (iHave.size > 0) {\n // index_list → for each unique index, index_info tells us its columns. The legacy global\n // constraint indexes only (name); the new one indexes (tenant_id, name).\n try {\n const idxList = await sqlite.execute(\"PRAGMA index_list(integrations)\");\n for (const row of (idxList.rows || [])) {\n const idxName = String(Array.isArray(row) ? row[1] : row.name);\n const idxUnique = Number(Array.isArray(row) ? row[2] : row.unique);\n if (!idxUnique) continue;\n const info = await sqlite.execute(`PRAGMA index_info(${idxName})`);\n const idxCols = (info.rows || []).map((r: any) => String(Array.isArray(r) ? r[2] : r.name));\n // Old-shape: a unique index over exactly [name] with no tenant_id ⇒ rebuild required.\n if (idxCols.includes(\"name\") && !idxCols.includes(\"tenant_id\")) needsRebuild = true;\n }\n } catch { needsRebuild = true; /* if we can't introspect, rebuild to be safe */ }\n }\n if (needsRebuild) {\n // Build v2 with the SAME columns (now including tenant_id) but UNIQUE(tenant_id, name).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS integrations_v2 (\n id TEXT PRIMARY KEY,\n name TEXT NOT NULL,\n api_key TEXT,\n api_secret TEXT,\n access_token TEXT,\n refresh_token TEXT,\n extra_config TEXT,\n connected_at TEXT,\n updated_at TEXT,\n tenant_id TEXT,\n UNIQUE(tenant_id, name)\n )\n `);\n // Copy every row (existing rows already carry tenant_id='t_default' from Stage 1).\n await sqlite.execute(`\n INSERT OR IGNORE INTO integrations_v2\n (id, name, api_key, api_secret, access_token, refresh_token, extra_config, connected_at, updated_at, tenant_id)\n SELECT id, name, api_key, api_secret, access_token, refresh_token, extra_config, connected_at, updated_at,\n COALESCE(tenant_id, 't_default')\n FROM integrations\n `);\n await sqlite.execute(\"DROP TABLE integrations\");\n await sqlite.execute(\"ALTER TABLE integrations_v2 RENAME TO integrations\");\n }\n await sqlite.execute(\n \"INSERT OR REPLACE INTO app_config (key, value, tenant_id, updated_at) VALUES ('mig_integrations_v2', 'done', 't_default', datetime('now'))\"\n );\n }\n } catch { /* non-critical: rebuild retried next warm start (flag only set on success) */ }\n\n // ─── RELATIONAL CRM BACKFILL (mig_crm_relational_v1) ────────────────────────\n // ONE-TIME, gated by app_config \"mig_crm_relational_v1\". For EVERY existing deal, derive the\n // relational records NON-DESTRUCTIVELY from its legacy text fields:\n // • upsert an account by (tenant_id, lower(company)) [skip if company empty]\n // • upsert a contact by (tenant_id, lower(contact_email)) [skip if email empty], name from\n // contact_name, linked to that account\n // • set deal.account_id + deal.primary_contact_id, and insert a deal_contacts(role='primary') row\n // Runs at boot under the default ctx, so it is a GLOBAL pass: it reads each deal's OWN tenant_id\n // and writes the new rows carrying THAT tenant_id (never the boot ctx's). Fully idempotent —\n // existence checks + INSERT OR IGNORE — and never deletes/edits existing deal data.\n try {\n const flag = await sqlite.execute(\"SELECT value FROM app_config WHERE key = 'mig_crm_relational_v1'\");\n if (flag.rows.length === 0) {\n // Pull every deal with its own tenant_id + legacy text fields. SELECT * keeps this robust\n // regardless of column order; we read fields by name from the row object.\n const dealsRes = await sqlite.execute(\n \"SELECT id, tenant_id, company, contact_name, contact_email FROM deals\"\n );\n const dealRows = (dealsRes.rows || []) as any[];\n for (const row of dealRows) {\n // libSQL returns array OR object rows depending on driver — read defensively by index/name.\n const dealId = String(Array.isArray(row) ? row[0] : row.id ?? \"\");\n const tenantId = String((Array.isArray(row) ? row[1] : row.tenant_id) ?? \"t_default\") || \"t_default\";\n const company = String((Array.isArray(row) ? row[2] : row.company) ?? \"\").trim();\n const contactName = String((Array.isArray(row) ? row[3] : row.contact_name) ?? \"\").trim();\n const contactEmail = String((Array.isArray(row) ? row[4] : row.contact_email) ?? \"\").trim();\n if (!dealId) continue;\n\n let accountId: string | null = null;\n let contactId: string | null = null;\n\n // ── ACCOUNT upsert by (tenant_id, lower(company)) ──\n if (company) {\n const existAcct = await sqlite.execute(\n \"SELECT id FROM accounts WHERE tenant_id = ? AND lower(name) = lower(?) LIMIT 1\",\n [tenantId, company]\n );\n if (existAcct.rows.length > 0) {\n const ar = existAcct.rows[0] as any;\n accountId = String(Array.isArray(ar) ? ar[0] : ar.id);\n } else {\n accountId = crypto.randomUUID();\n await sqlite.execute(\n `INSERT INTO accounts (id, name, tags, custom_fields, created_at, updated_at, tenant_id)\n VALUES (?, ?, '[]', '{}', datetime('now'), datetime('now'), ?)`,\n [accountId, company, tenantId]\n );\n }\n }\n\n // ── CONTACT upsert by (tenant_id, lower(contact_email)) ──\n if (contactEmail) {\n const existContact = await sqlite.execute(\n \"SELECT id, account_id FROM contacts WHERE tenant_id = ? AND lower(email) = lower(?) LIMIT 1\",\n [tenantId, contactEmail]\n );\n if (existContact.rows.length > 0) {\n const cr = existContact.rows[0] as any;\n contactId = String(Array.isArray(cr) ? cr[0] : cr.id);\n const existingAcctId = Array.isArray(cr) ? cr[1] : cr.account_id;\n // Backfill the contact's account link if it has none yet and we found one here.\n if (accountId && (existingAcctId === null || existingAcctId === undefined || existingAcctId === \"\")) {\n await sqlite.execute(\n \"UPDATE contacts SET account_id = ?, updated_at = datetime('now') WHERE id = ? AND tenant_id = ?\",\n [accountId, contactId, tenantId]\n );\n }\n } else {\n // Split contact_name into first / last (first token = first name, remainder = last).\n let firstName = \"\", lastName = \"\";\n if (contactName) {\n const parts = contactName.split(/\\s+/);\n firstName = parts.shift() || \"\";\n lastName = parts.join(\" \");\n }\n contactId = crypto.randomUUID();\n await sqlite.execute(\n `INSERT INTO contacts (id, account_id, first_name, last_name, email, tags, custom_fields, created_at, updated_at, tenant_id)\n VALUES (?, ?, ?, ?, ?, '[]', '{}', datetime('now'), datetime('now'), ?)`,\n [contactId, accountId, firstName, lastName, contactEmail, tenantId]\n );\n }\n }\n\n // ── Link the deal: set account_id / primary_contact_id only if still empty (idempotent) ──\n if (accountId || contactId) {\n await sqlite.execute(\n `UPDATE deals\n SET account_id = COALESCE(NULLIF(account_id, ''), ?),\n primary_contact_id = COALESCE(NULLIF(primary_contact_id, ''), ?)\n WHERE id = ? AND tenant_id = ?`,\n [accountId, contactId, dealId, tenantId]\n );\n }\n\n // ── deal_contacts(role='primary') join row (INSERT OR IGNORE on composite PK) ──\n if (contactId) {\n await sqlite.execute(\n `INSERT OR IGNORE INTO deal_contacts (deal_id, contact_id, role, tenant_id, created_at)\n VALUES (?, ?, 'primary', ?, datetime('now'))`,\n [dealId, contactId, tenantId]\n );\n }\n }\n await sqlite.execute(\n \"INSERT OR REPLACE INTO app_config (key, value, tenant_id, updated_at) VALUES ('mig_crm_relational_v1', 'done', 't_default', datetime('now'))\"\n );\n }\n } catch { /* non-critical: relational backfill retried next warm start (flag only set on success) */ }\n\n // Seed default LLM config if not exists\n const existing = await sqlite.execute(\"SELECT id FROM llm_config WHERE id = 'default'\");\n if (existing.rows.length === 0) {\n await sqlite.execute(\n \"INSERT INTO llm_config (id, provider, api_key, model, updated_at) VALUES ('default', 'anthropic', '', 'claude-sonnet-4-6', ?)\",\n [new Date().toISOString()]\n );\n }\n\n // Demo-agent seeding disabled — starting from a clean agents table.\n // Re-enable by uncommenting the call below.\n // await seedDefaultAgents();\n}\n\nasync function seedDefaultAgents() {\n const existing = await sqlite.execute(\"SELECT id FROM agents LIMIT 1\");\n if (existing.rows.length > 0) return;\n\n const now = new Date().toISOString();\n for (const agent of DEFAULT_AGENT_SEEDS) {\n await sqlite.execute(\n `INSERT INTO agents (id, name, workspace, description, system_prompt, tools, model, enabled, created_at, updated_at, created_by)\n VALUES (?, ?, ?, ?, ?, ?, ?, 1, ?, ?, ?)`,\n [crypto.randomUUID(), agent.name, agent.workspace, agent.description, agent.system_prompt, agent.tools, agent.model, now, now, \"user\"]\n );\n }\n}\n",[192],"0cd8e370-5f15-11f1-bf54-1607ee4eb77e","router.ts","30479b74-56ca-11f1-a296-ee650bb23af1",1010,["D",1780467943919],126,"import { sqlite, initDB, DEFAULT_AGENT_SEEDS, currentRole, currentTenantId, currentCtx } from \"./db.ts\";\nimport { registerUser, loginUser, getMe, upsertOAuthUser, bootstrapOwner } from \"./api/auth.ts\";\nimport { resolveSession, sessionSetCookie, sessionClearCookie } from \"./api/session.ts\";\nimport { newOAuthState, CLEAR_OAUTH_STATE, redirectWithCookie, oauthStateValid, identityLoginStart, identityLoginCallback } from \"./api/oauth_login.ts\";\nimport { listAgents, createAgent, updateAgent, deleteAgent } from \"./api/agents.ts\";\nimport { getLLMConfig, saveLLMConfig } from \"./api/llm.ts\";\nimport { handleChat, handleChatStream, listConversations, getConversationById, deleteConversation } from \"./api/chat.ts\";\nimport { handleIntegrationProxy, getCredentials, GOOGLE_SERVICES, MICROSOFT_SCOPES, getIntegrationSchema, listDescriptorNames } from \"./api/integrations.ts\";\nimport { serveAppModule } from \"./app_modules.ts\";\nimport { getFunnelSummary, getCorrelations, getRecommendations, getLeadScores, getAIOKpis, getIntelligenceFeed } from \"./api/intelligence.ts\";\nimport { listSignals, recordSignal, actionSignal, getHotLeads, getHotLeadsCount } from \"./api/signals.ts\";\nimport { listObjectives, createObjective, updateObjective, deleteObjective, createKeyResult, updateKeyResult, getOKRDashboard, syncObjectivesFromFunnel } from \"./api/okrs.ts\";\nimport { getNSM, setNSM, recordNSMValue, getNSMProgress } from \"./api/nsm.ts\";\nimport { listCohorts, createCohort, deleteCohort, getCohortSummary, getEngagementVelocity, getAttributionSummary, recordConversion, getRetention, getRevenueModel } from \"./api/aio.ts\";\nimport { getLatestDigest, generateDigest, getDigestHistory } from \"./api/digest.ts\";\nimport { listExperiments, createExperiment, updateExperiment, deleteExperiment, runExperiment, completeExperiment, getExperimentStats, suggestICE } from \"./api/experiments.ts\";\nimport { getUnifiedCalendar, getCalendarSummary, createCalendarEntry, updateCalendarEntry, deleteCalendarEntry } from \"./api/calendar_unified.ts\";\nimport { getGraphify } from \"./api/graphify.ts\";\nimport { listTasks, createTask, updateTask, deleteTask, listIceRankedTasks, generateTasksFromSource, runTask } from \"./api/tasks.ts\";\nimport { listDeals, getDeal, createDeal, updateDeal, deleteDeal, syncDealsFromLeads, listCandidates, importDeals } from \"./api/deals.ts\";\nimport { listKnowledgeFiles, getKnowledgeFile, createKnowledgeFile, updateKnowledgeFile, deleteKnowledgeFile, summarizeKnowledgeFile, buildKnowledgeGraph, AI_FEATURES } from \"./api/knowledge.ts\";\nimport { getCompanyProfile, setCompanyProfile, getBrandContext, setBrandContext } from \"./api/company.ts\";\nimport { enrichCompanyProfile } from \"./api/company-enrichment.ts\";\nimport { listMemory, addMemory, updateMemory, deleteMemory } from \"./api/memory.ts\";\nimport { listSkills, getSkill, setSkillState, createCustomSkill } from \"./api/skills.ts\";\nimport { getMethodConfig, saveMethodConfig, getMethodsMetadata } from \"./api/method_config.ts\";\nimport { listActions, enqueueAction, getAction, approveAction, rejectAction } from \"./api/actions.ts\";\nimport { processPendingRuns } from \"./cron.ts\";\nimport { getExecutedWorkCounts } from \"./api/propagate.ts\";\nimport { listCampaigns, getCampaign, getCampaignAgents, createCampaign, deleteCampaign, listCampaignRuns, orchestrateCampaignRun, updateCampaign, setCampaignTrigger } from \"./api/campaigns.ts\";\nimport { getWorldActivity, getWorldBoard, getWorldProject } from \"./api/world.ts\";\nimport { handlePlatformRoutes } from \"./api/platform.ts\";\nimport { isCrmPath, handleCrmRoutes } from \"./api/crm_routes.ts\";\nimport { isMarketingPath, handleMarketingRoutes } from \"./api/marketing_routes.ts\";\n\nconst corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"GET, POST, PUT, DELETE, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type, Authorization\",\n};\n\n// App version for the data-free health probe. Prefer the val.town deploy id (DENO_DEPLOYMENT_ID\n// is set by the runtime per deploy); fall back to a hardcoded release tag when running locally.\nconst APP_VERSION = (() => {\n try { return Deno.env.get(\"DENO_DEPLOYMENT_ID\") || \"p3-s3\"; } catch { return \"p3-s3\"; }\n})();\n\nfunction json(data: unknown, status = 200): Response {\n return new Response(JSON.stringify(data), {\n status,\n headers: {\n \"Content-Type\": \"application/json\",\n ...corsHeaders,\n },\n });\n}\n// JSON response that also carries a Set-Cookie header (session set/clear on auth endpoints).\nfunction jsonWithCookie(data: unknown, setCookie: string, status = 200): Response {\n const res = json(data, status);\n res.headers.append(\"Set-Cookie\", setCookie);\n return res;\n}\n// Best-effort client IP for per-IP rate limiting (val.town/Cloudflare forwarded headers).\nfunction clientIp(req: Request): string {\n return req.headers.get(\"x-forwarded-for\")?.split(\",\")[0].trim() || req.headers.get(\"cf-connecting-ip\") || \"\";\n}\n\n// Mask an integration row before returning it: NO endpoint may return secrets in cleartext\n// — only presence flags + a short masked hint.\nfunction maskIntegration(obj: any, googleConnected = false) {\n return {\n ...obj,\n connected: !!(obj.api_key || obj.access_token) ||\n (GOOGLE_SERVICES.includes(String(obj.name)) && googleConnected),\n has_api_key: !!obj.api_key,\n has_api_secret: !!obj.api_secret,\n has_access_token: !!obj.access_token,\n api_key: obj.api_key ? String(obj.api_key).slice(0, 4) + \"***\" : undefined,\n api_secret: obj.api_secret ? \"***\" : undefined,\n access_token: obj.access_token ? \"***\" : undefined,\n refresh_token: obj.refresh_token ? \"***\" : undefined,\n };\n}\n\n\nexport async function handleRequest(req: Request, url: URL): Promise {\n const { pathname } = url;\n const method = req.method;\n const tid = currentTenantId(); // tenant scope for every business query (stable within request)\n\n // CORS preflight\n if (method === \"OPTIONS\") {\n return new Response(null, { status: 204, headers: corsHeaders });\n }\n\n try {\n // GET /api/health — data-free liveness probe: { ok, version } (+ legacy status/timestamp).\n if (pathname === \"/api/health\" && method === \"GET\") {\n return json({ ok: true, version: APP_VERSION, status: \"ok\", timestamp: new Date().toISOString() });\n }\n\n // GET /api/health/auth - gated token verifier. Reaching this (past the main.ts gate) means the\n // APP_TOKEN is valid; the login screen uses it to confirm the password.\n if (pathname === \"/api/health/auth\" && method === \"GET\") {\n return json({ ok: true });\n }\n\n // --- Auth: email/password + session ----\n // Thin handlers — logic lives in api/auth.ts; the \"nexus_session\" cookie via api/session.ts.\n if (pathname === \"/api/auth/register\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n const r = await registerUser({ email: body.email, password: body.password, name: body.name, honeypot: body.honeypot ?? body.company_url, turnstileToken: body.turnstileToken, ip: clientIp(req) });\n if (!r.ok) return json({ ok: false, error: r.error }, 400);\n return jsonWithCookie({ ok: true }, sessionSetCookie(r.ctxToken!));\n }\n if (pathname === \"/api/auth/login\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n const r = await loginUser({ email: body.email, password: body.password, ip: clientIp(req) });\n if (!r.ok) return json({ ok: false, error: r.error }, 401);\n return jsonWithCookie({ ok: true }, sessionSetCookie(r.ctxToken!));\n }\n if (pathname === \"/api/auth/logout\" && method === \"POST\") {\n return jsonWithCookie({ ok: true }, sessionClearCookie());\n }\n if (pathname === \"/api/auth/me\" && method === \"GET\") {\n const ctx = await resolveSession(req);\n if (!ctx) return json({ ok: false, user: null }, 200);\n return json({ ok: true, ...(await getMe(ctx)) });\n }\n\n // GET /api/config - LLM runtime config + all app_config key/values\n if (pathname === \"/api/config\" && method === \"GET\") {\n const llmConfig = await getLLMConfig();\n const merged: Record = {\n provider: llmConfig.provider,\n model: llmConfig.model,\n has_key: (llmConfig as any).has_key,\n };\n try {\n const r = await sqlite.execute(\"SELECT key, value FROM app_config WHERE tenant_id = ?\", [tid]);\n for (const row of (r.rows || []) as any[]) {\n const k = String(row[0]);\n let v: any = row[1];\n if (typeof v === \"string\") { try { v = JSON.parse(v); } catch { /* keep raw string */ } }\n merged[k] = v;\n }\n } catch { /* app_config missing → just return LLM fields */ }\n return json(merged);\n }\n // POST /api/config - upsert arbitrary key/value into app_config (Settings toggles, ROI inputs…).\n if (pathname === \"/api/config\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n const updated: Record = {};\n try {\n for (const [k, v] of Object.entries(body)) {\n if (k === \"provider\" || k === \"model\" || k === \"has_key\") continue;\n const value = typeof v === \"string\" ? v : JSON.stringify(v);\n await sqlite.execute(\n \"INSERT INTO app_config (key, value, tenant_id) VALUES (?, ?, ?) ON CONFLICT(key) DO UPDATE SET value = excluded.value, tenant_id = excluded.tenant_id, updated_at = datetime('now')\",\n [k, value, tid],\n );\n updated[k] = v;\n }\n return json({ success: true, updated });\n } catch (e: any) {\n return json({ error: String(e?.message || e) }, 500);\n }\n }\n\n // --- Frontend JS module routes (table lives in app_modules.ts to keep router under 80KB) ----\n if (method === \"GET\") {\n const _moduleSrc = serveAppModule(pathname);\n if (_moduleSrc !== null) {\n return new Response(_moduleSrc, {\n headers: {\n \"Content-Type\": \"application/javascript\",\n \"Cache-Control\": \"no-cache, no-store, must-revalidate\",\n ...corsHeaders,\n },\n });\n }\n }\n\n // --- Intelligence routes ----\n if (pathname === \"/api/intelligence/funnel-summary\" && method === \"GET\") {\n const force = url.searchParams.get(\"force\") === \"1\";\n const days = parseInt(url.searchParams.get(\"days\") ?? \"30\", 10);\n return json(await getFunnelSummary(force, days));\n }\n if (pathname === \"/api/intelligence/correlations\" && method === \"GET\") {\n const force = url.searchParams.get(\"force\") === \"1\";\n return json(await getCorrelations(force));\n }\n if (pathname === \"/api/intelligence/recommendations\" && method === \"GET\") {\n return json(await getRecommendations());\n }\n if (pathname === \"/api/intelligence/leads\" && method === \"GET\") {\n return json(await getLeadScores());\n }\n if (pathname === \"/api/intelligence/kpis\" && method === \"GET\") {\n return json(await getAIOKpis());\n }\n if (pathname === \"/api/intelligence/feed\" && method === \"GET\") {\n return json(await getIntelligenceFeed());\n }\n // Execution-aware metric: how much real work was completed (tasks/experiments).\n if (pathname === \"/api/intelligence/executed-work\" && method === \"GET\") {\n const days = parseInt(url.searchParams.get(\"days\") ?? \"30\", 10);\n return json(await getExecutedWorkCounts(days));\n }\n\n // --- Signals routes ----\n if (pathname === \"/api/signals\" && method === \"GET\") {\n const actioned = url.searchParams.get(\"actioned\");\n const lead_email = url.searchParams.get(\"lead_email\") ?? undefined;\n return json(await listSignals({ lead_email, actioned: actioned === \"true\" ? true : actioned === \"false\" ? false : undefined }));\n }\n if (pathname === \"/api/signals\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n if (!body.lead_email || !body.signal_type) {\n return json({ error: \"lead_email and signal_type are required\" }, 400);\n }\n return json(await recordSignal(body), 201);\n }\n const signalActionMatch = pathname.match(/^\\/api\\/signals\\/([^/]+)\\/action$/);\n if (signalActionMatch && method === \"PUT\") {\n const result = await actionSignal(signalActionMatch[1]);\n return result ? json(result) : json({ error: \"Signal not found\" }, 404);\n }\n if (pathname === \"/api/signals/hot-leads\" && method === \"GET\") {\n return json(await getHotLeads());\n }\n if (pathname === \"/api/signals/hot-leads/count\" && method === \"GET\") {\n return json({ count: await getHotLeadsCount() });\n }\n\n // --- OKR / Objectives routes ----\n if (pathname === \"/api/okrs\" && method === \"GET\") {\n const workspace = url.searchParams.get(\"workspace\") ?? undefined;\n return json(await listObjectives({ workspace }));\n }\n if (pathname === \"/api/okrs\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await createObjective(body), 201);\n }\n if (pathname === \"/api/okrs/dashboard\" && method === \"GET\") {\n return json(await getOKRDashboard());\n }\n // Explicit funnel→OKR sync — refreshes stored KR/objective progress on demand.\n if (pathname === \"/api/okrs/sync\" && method === \"POST\") {\n const workspace = url.searchParams.get(\"workspace\") ?? undefined;\n return json(await syncObjectivesFromFunnel({ workspace }));\n }\n const okrIdMatch = pathname.match(/^\\/api\\/okrs\\/([^/]+)$/);\n if (okrIdMatch && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n return json(await updateObjective(okrIdMatch[1], body));\n }\n if (okrIdMatch && method === \"DELETE\") {\n const ok = await deleteObjective(okrIdMatch[1]);\n return ok ? json({ success: true }) : json({ error: \"Not found\" }, 404);\n }\n const okrKrMatch = pathname.match(/^\\/api\\/okrs\\/([^/]+)\\/key-results$/);\n if (okrKrMatch && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await createKeyResult(okrKrMatch[1], body), 201);\n }\n const krIdMatch = pathname.match(/^\\/api\\/okrs\\/key-results\\/([^/]+)$/);\n if (krIdMatch && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n return json(await updateKeyResult(krIdMatch[1], body));\n }\n\n // --- NSM routes ----\n if (pathname === \"/api/nsm\" && method === \"GET\") {\n return json(await getNSM());\n }\n if (pathname === \"/api/nsm\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await setNSM(body));\n }\n if (pathname === \"/api/nsm/record\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n await recordNSMValue(body.value);\n return json({ success: true });\n }\n if (pathname === \"/api/nsm/progress\" && method === \"GET\") {\n return json(await getNSMProgress());\n }\n\n // --- AIO / Cohort routes ----\n if (pathname === \"/api/aio/cohorts\" && method === \"GET\") {\n return json(await listCohorts());\n }\n if (pathname === \"/api/aio/cohorts\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await createCohort(body), 201);\n }\n const cohortIdMatch = pathname.match(/^\\/api\\/aio\\/cohorts\\/([^/]+)$/);\n if (cohortIdMatch && method === \"DELETE\") {\n return json({ success: await deleteCohort(cohortIdMatch[1]) });\n }\n if (pathname === \"/api/aio/cohort-summary\" && method === \"GET\") {\n return json(await getCohortSummary());\n }\n if (pathname === \"/api/aio/engagement-velocity\" && method === \"GET\") {\n return json(await getEngagementVelocity());\n }\n\n // --- Frontend-expected aliases (cohorts/attribution/key-results) ----\n if (pathname === \"/api/cohorts\" && method === \"GET\") {\n return json(await getCohortSummary());\n }\n if (pathname === \"/api/cohorts\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await createCohort(body), 201);\n }\n const cohortAliasMatch = pathname.match(/^\\/api\\/cohorts\\/([^/]+)$/);\n if (cohortAliasMatch && method === \"DELETE\") {\n return json({ success: await deleteCohort(cohortAliasMatch[1]) });\n }\n if (pathname === \"/api/attribution\" && method === \"GET\") {\n return json(await getAttributionSummary());\n }\n // retention (GRR/NRR + curve) + revenue model endpoints.\n if (pathname === \"/api/intelligence/retention\" && method === \"GET\") {\n return json(await getRetention());\n }\n if (pathname === \"/api/intelligence/revenue-model\" && method === \"GET\") {\n return json(await getRevenueModel());\n }\n if (pathname === \"/api/attribution\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n await recordConversion(body);\n return json({ success: true }, 201);\n }\n const krAliasMatch = pathname.match(/^\\/api\\/key-results\\/([^/]+)$/);\n if (krAliasMatch && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n return json(await updateKeyResult(krAliasMatch[1], body));\n }\n\n // --- Digest routes ----\n if (pathname === \"/api/digest\" && method === \"GET\") {\n return json(await getLatestDigest());\n }\n if (pathname === \"/api/digest/generate\" && method === \"POST\") {\n return json(await generateDigest());\n }\n if (pathname === \"/api/digest/history\" && method === \"GET\") {\n const limit = parseInt(url.searchParams.get(\"limit\") ?? \"10\", 10);\n return json(await getDigestHistory(limit));\n }\n\n // --- Experiments routes ----\n if (pathname === \"/api/experiments\" && method === \"GET\") {\n const status = url.searchParams.get(\"status\") ?? undefined;\n const funnel_stage = url.searchParams.get(\"stage\") ?? url.searchParams.get(\"funnel_stage\") ?? undefined;\n const aarrr_lever = url.searchParams.get(\"aarrr_lever\") ?? undefined;\n return json(await listExperiments({ status, funnel_stage, aarrr_lever }));\n }\n if (pathname === \"/api/experiments\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await createExperiment(body), 201);\n }\n if (pathname === \"/api/experiments/stats\" && method === \"GET\") {\n return json(await getExperimentStats());\n }\n // suggest-ice: alias of ice-suggest (same handler).\n if ((pathname === \"/api/experiments/ice-suggest\" || pathname === \"/api/experiments/suggest-ice\") && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await suggestICE(body));\n }\n // GET /api/experiments/:id/stats — sub-resource form; getExperimentStats() is tenant-aggregate (portfolio-wide).\n // Declared before the run|complete and bare :id matchers so \"stats\" isn't read as an id.\n const expStatsMatch = pathname.match(/^\\/api\\/experiments\\/([^/]+)\\/stats$/);\n if (expStatsMatch && method === \"GET\") {\n return json(await getExperimentStats());\n }\n const expActionMatch = pathname.match(/^\\/api\\/experiments\\/([^/]+)\\/(run|complete)$/);\n if (expActionMatch && method === \"POST\") {\n const id = expActionMatch[1];\n const body = await req.json().catch(() => ({}));\n const result = expActionMatch[2] === \"run\"\n ? await runExperiment(id, Number(body.baseline_value) || 0)\n : await completeExperiment(id, {\n result_value: Number(body.result_value) || 0,\n outcome: body.outcome ?? \"inconclusive\",\n learnings: body.learnings,\n control: body.control,\n variant: body.variant,\n sample_size: body.sample_size,\n });\n return result ? json(result) : json({ error: \"Experiment not found\" }, 404);\n }\n const expIdMatch = pathname.match(/^\\/api\\/experiments\\/([^/]+)$/);\n if (expIdMatch && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n const updated = await updateExperiment(expIdMatch[1], body);\n return updated ? json(updated) : json({ error: \"Experiment not found\" }, 404);\n }\n if (expIdMatch && method === \"DELETE\") {\n const ok = await deleteExperiment(expIdMatch[1]);\n return ok ? json({ success: true }) : json({ error: \"Not found\" }, 404);\n }\n\n // --- Calendar unified route ----\n if (pathname === \"/api/calendar/unified\" && method === \"GET\") {\n const today = new Date().toISOString().slice(0, 10);\n const from = url.searchParams.get(\"from\") ?? url.searchParams.get(\"start\") ?? today;\n const to = url.searchParams.get(\"to\") ?? url.searchParams.get(\"end\") ?? from;\n const workspace = url.searchParams.get(\"workspace\") ?? undefined;\n const sourcesParam = url.searchParams.getAll(\"source\");\n return json(await getUnifiedCalendar({ from, to, workspace, sources: sourcesParam.length ? sourcesParam as any : undefined }));\n }\n if (pathname === \"/api/calendar/summary\" && method === \"GET\") {\n const date = url.searchParams.get(\"date\") ?? new Date().toISOString().slice(0, 10);\n return json(await getCalendarSummary(date));\n }\n if (pathname === \"/api/calendar/entries\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await createCalendarEntry(body), 201);\n }\n const calEntryMatch = pathname.match(/^\\/api\\/calendar\\/entries\\/([^/]+)$/);\n if (calEntryMatch && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n const updated = await updateCalendarEntry(calEntryMatch[1], body);\n return updated ? json(updated) : json({ error: \"Not found\" }, 404);\n }\n if (calEntryMatch && method === \"DELETE\") {\n const ok = await deleteCalendarEntry(calEntryMatch[1]);\n return ok ? json({ success: true }) : json({ error: \"Not found\" }, 404);\n }\n\n // --- Knowledge files (AI knowledge injection + graph view) ----\n if (pathname === \"/api/knowledge\" && method === \"GET\") {\n return json(await listKnowledgeFiles());\n }\n if (pathname === \"/api/knowledge\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n const created = await createKnowledgeFile(body);\n return created ? json(created, 201) : json({ error: \"content required\" }, 400);\n }\n if (pathname === \"/api/knowledge/graph\" && method === \"GET\") {\n return json(await buildKnowledgeGraph());\n }\n if (pathname === \"/api/knowledge/features\" && method === \"GET\") {\n return json(AI_FEATURES);\n }\n const knSummarizeMatch = pathname.match(/^\\/api\\/knowledge\\/([^/]+)\\/summarize$/);\n if (knSummarizeMatch && method === \"POST\") {\n const summary = await summarizeKnowledgeFile(knSummarizeMatch[1]);\n const file = await getKnowledgeFile(knSummarizeMatch[1]);\n return json({ summary, tokens_estimate: file?.tokens_estimate ?? 0 });\n }\n const knIdMatch = pathname.match(/^\\/api\\/knowledge\\/([^/]+)$/);\n if (knIdMatch && method === \"GET\") {\n const f = await getKnowledgeFile(knIdMatch[1]);\n return f ? json(f) : json({ error: \"Not found\" }, 404);\n }\n if (knIdMatch && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n const updated = await updateKnowledgeFile(knIdMatch[1], body);\n return updated ? json(updated) : json({ error: \"Not found\" }, 404);\n }\n if (knIdMatch && method === \"DELETE\") {\n const ok = await deleteKnowledgeFile(knIdMatch[1]);\n return ok ? json({ success: true }) : json({ error: \"Not found\" }, 404);\n }\n\n // --- Company Profile + Brand Context (Part 15.2) — the business identity ---\n if (pathname === \"/api/company-profile\" && method === \"GET\") {\n return json(await getCompanyProfile());\n }\n if (pathname === \"/api/company-profile\" && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n return json(await setCompanyProfile(body));\n }\n if (pathname === \"/api/company-profile/enrich\" && method === \"POST\") {\n try {\n const enriched = await enrichCompanyProfile();\n return json({ enriched, message: \"AI enrichment complete — review and save.\" });\n } catch (e) {\n return json({ error: String(e?.message ?? e) }, 500);\n }\n }\n if (pathname === \"/api/brand-context\" && method === \"GET\") {\n return json({ text: await getBrandContext() });\n }\n if (pathname === \"/api/brand-context\" && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n await setBrandContext(String(body.text ?? \"\"));\n return json({ ok: true });\n }\n\n // --- Memory (Part 15.3) — persistent evolving business facts ----\n if (pathname === \"/api/memory\" && method === \"GET\") {\n return json(await listMemory());\n }\n if (pathname === \"/api/memory\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n const m = await addMemory(body);\n return m ? json(m, 201) : json({ error: \"content required\" }, 400);\n }\n const memIdMatch = pathname.match(/^\\/api\\/memory\\/([0-9]+)$/);\n if (memIdMatch && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n const m = await updateMemory(Number(memIdMatch[1]), body);\n return m ? json(m) : json({ error: \"Not found\" }, 404);\n }\n if (memIdMatch && method === \"DELETE\") {\n const ok = await deleteMemory(Number(memIdMatch[1]));\n return ok ? json({ success: true }) : json({ error: \"Not found\" }, 404);\n }\n\n // --- Skills (Playbooks) — marketing skill catalog + per-skill state ----\n if (pathname === \"/api/skills\" && method === \"GET\") {\n return json(await listSkills());\n }\n if (pathname === \"/api/skills\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n const created = await createCustomSkill(body);\n return created.ok ? json(created, 201) : json({ error: created.error || \"invalid skill\" }, 400);\n }\n const skillNameMatch = pathname.match(/^\\/api\\/skills\\/([^/]+)$/);\n if (skillNameMatch && method === \"GET\") {\n const s = await getSkill(decodeURIComponent(skillNameMatch[1]));\n return s ? json(s) : json({ error: \"Not found\" }, 404);\n }\n if (skillNameMatch && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n const ok = await setSkillState(decodeURIComponent(skillNameMatch[1]), body);\n return ok ? json({ ok: true }) : json({ error: \"Not found\" }, 404);\n }\n\n // --- Method config (prioritization methodology: ICE/RICE/WSJF) ----\n if (pathname === \"/api/method-config/methods\" && method === \"GET\") {\n return json(getMethodsMetadata());\n }\n if (pathname === \"/api/method-config\" && method === \"GET\") {\n return json(await getMethodConfig());\n }\n if (pathname === \"/api/method-config\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await saveMethodConfig(body));\n }\n\n // --- Action queue (human-in-the-loop execution) ----\n if (pathname === \"/api/actions\" && method === \"GET\") {\n const status = url.searchParams.get(\"status\") ?? undefined;\n return json(await listActions(status));\n }\n if (pathname === \"/api/actions\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n const created = await enqueueAction(body);\n return created ? json(created, 201) : json({ error: \"integration + endpoint required\" }, 400);\n }\n const actionApproveMatch = pathname.match(/^\\/api\\/actions\\/([^/]+)\\/approve$/);\n if (actionApproveMatch && method === \"POST\") {\n const a = await approveAction(actionApproveMatch[1]);\n return a ? json(a) : json({ error: \"Not found\" }, 404);\n }\n const actionRejectMatch = pathname.match(/^\\/api\\/actions\\/([^/]+)\\/reject$/);\n if (actionRejectMatch && method === \"POST\") {\n const a = await rejectAction(actionRejectMatch[1]);\n return a ? json(a) : json({ error: \"Not found\" }, 404);\n }\n const actionIdMatch = pathname.match(/^\\/api\\/actions\\/([^/]+)$/);\n if (actionIdMatch && method === \"GET\") {\n const a = await getAction(actionIdMatch[1]);\n return a ? json(a) : json({ error: \"Not found\" }, 404);\n }\n\n // Manual trigger for the execution engine (the interval val also calls processPendingRuns).\n if (pathname === \"/api/agents/run-pending\" && method === \"POST\") {\n const limit = parseInt(url.searchParams.get(\"limit\") ?? \"3\", 10);\n return json(await processPendingRuns(limit));\n }\n\n // GET /api/agents/activity?limit=N — cross-agent run feed; before the /api/agents/:id matchers.\n if (pathname === \"/api/agents/activity\" && method === \"GET\") {\n const limit = parseInt(url.searchParams.get(\"limit\") ?? \"20\", 10);\n const r = await sqlite.execute(\n `SELECT r.id, r.agent_id, a.name, a.workspace, r.status, r.started_at, r.finished_at, r.output\n FROM agent_runs r LEFT JOIN agents a ON a.id = r.agent_id AND a.tenant_id = ?\n WHERE r.tenant_id = ?\n ORDER BY r.started_at DESC LIMIT ?`,\n [tid, tid, limit]\n );\n return json({ activity: r.rows.map((row) => rowToObject(r.columns, row)) });\n }\n\n // --- Agent runs ----\n const agentRunsMatch = pathname.match(/^\\/api\\/agents\\/([^/]+)\\/runs$/);\n if (agentRunsMatch && method === \"GET\") {\n const agent_id = agentRunsMatch[1];\n const limit = parseInt(url.searchParams.get(\"limit\") ?? \"20\", 10);\n const result = await sqlite.execute(\n \"SELECT * FROM agent_runs WHERE agent_id = ? AND tenant_id = ? ORDER BY started_at DESC LIMIT ?\",\n [agent_id, tid, limit]\n );\n return json(result.rows.map((row) => rowToObject(result.columns, row)));\n }\n if (agentRunsMatch && method === \"POST\") {\n const agent_id = agentRunsMatch[1];\n const body = await req.json().catch(() => ({}));\n const id = crypto.randomUUID();\n const now = new Date().toISOString();\n await sqlite.execute(\n \"INSERT INTO agent_runs (id, agent_id, trigger, status, started_at, tenant_id) VALUES (?, ?, ?, ?, ?, ?)\",\n [id, agent_id, body.trigger ?? \"manual\", \"pending\", now, tid]\n );\n // App-level automation (no cron): run the queued job in-request. Fire-and-forget; status='running' claim guards dupes.\n processPendingRuns(1).catch(() => {});\n return json({ id, agent_id, trigger: body.trigger ?? \"manual\", status: \"pending\", started_at: now }, 201);\n }\n\n // Integrations routes\n // GET /api/integrations - list all integrations\n if (pathname === \"/api/integrations\" && method === \"GET\") {\n const result = await sqlite.execute(\"SELECT * FROM integrations WHERE tenant_id = ? ORDER BY name ASC\", [tid]);\n const rows = result.rows.map((row) => rowToObject(result.columns, row));\n // Google-family services inherit the central 'google' OAuth token, so they are \"connected\"\n // whenever google is — even without their own api_key/access_token row.\n const googleConnected = rows.some((o) => o.name === \"google\" && o.access_token);\n const integrations = rows.map((obj) => maskIntegration(obj, googleConnected));\n return json(integrations);\n }\n\n // POST /api/integrations/:name - upsert integration credentials\n const integrationUpsertMatch = pathname.match(/^\\/api\\/integrations\\/([^/]+)$/);\n if (integrationUpsertMatch && method === \"POST\") {\n const name = integrationUpsertMatch[1];\n const body = await req.json().catch(() => ({}));\n const now = new Date().toISOString();\n\n const existing = await sqlite.execute(\n \"SELECT id, api_key, api_secret, access_token, refresh_token, extra_config FROM integrations WHERE name = ? AND tenant_id = ?\",\n [name, tid]\n );\n\n // Keep the existing secret when the incoming value is blank/missing/masked (\"***\"),\n // so saving the masked Settings form does not wipe stored credentials.\n const keepSecret = (incoming: any, current: any) => {\n if (incoming == null) return current ?? null;\n const s = String(incoming);\n if (s === \"\" || s.includes(\"***\")) return current ?? null;\n return incoming;\n };\n\n if (existing.rows.length > 0) {\n const cur = existing.rows[0] as any[]; // [id, api_key, api_secret, access_token, refresh_token, extra_config]\n await sqlite.execute(\n `UPDATE integrations SET\n api_key = ?,\n api_secret = ?,\n access_token = ?,\n refresh_token = ?,\n extra_config = ?,\n updated_at = ?\n WHERE name = ? AND tenant_id = ?`,\n [\n keepSecret(body.api_key, cur[1]),\n keepSecret(body.api_secret, cur[2]),\n keepSecret(body.access_token, cur[3]),\n keepSecret(body.refresh_token, cur[4]),\n body.extra_config != null ? JSON.stringify(body.extra_config) : (cur[5] ?? null),\n now,\n name,\n tid,\n ]\n );\n } else {\n // Insert\n const id = crypto.randomUUID();\n await sqlite.execute(\n `INSERT INTO integrations (id, name, api_key, api_secret, access_token, refresh_token, extra_config, connected_at, updated_at, tenant_id)\n VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`,\n [\n id,\n name,\n body.api_key ?? null,\n body.api_secret ?? null,\n body.access_token ?? null,\n body.refresh_token ?? null,\n body.extra_config != null ? JSON.stringify(body.extra_config) : null,\n now,\n now,\n tid,\n ]\n );\n }\n\n const updated = await sqlite.execute(\n \"SELECT * FROM integrations WHERE name = ? AND tenant_id = ?\",\n [name, tid]\n );\n const raw = rowToObject(updated.columns, updated.rows[0]);\n // NEVER return the just-saved secrets in cleartext — mask like the list endpoint.\n const googleConnected = raw.name === \"google\"\n ? !!raw.access_token\n : !!(await getCredentials(\"google\"))?.access_token;\n return json({ success: true, integration: maskIntegration(raw, googleConnected) });\n }\n\n // DELETE /api/integrations/:name\n const integrationDeleteMatch = pathname.match(/^\\/api\\/integrations\\/([^/]+)$/);\n if (integrationDeleteMatch && method === \"DELETE\") {\n const name = integrationDeleteMatch[1];\n await sqlite.execute(\"DELETE FROM integrations WHERE name = ? AND tenant_id = ?\", [name, tid]);\n return json({ success: true, message: `Integration '${name}' deleted` });\n }\n\n // Part 13: GET /api/integrations/:name/schema — descriptor-generated action/trigger schema\n // (UI form fields + AI tool hints). null when the integration is still hand-written.\n const integrationSchemaMatch = pathname.match(/^\\/api\\/integrations\\/([^/]+)\\/schema$/);\n if (integrationSchemaMatch && method === \"GET\") {\n const schema = getIntegrationSchema(integrationSchemaMatch[1]);\n return schema ? json(schema) : json({ error: \"no descriptor for this integration yet\", descriptor_integrations: listDescriptorNames() }, 404);\n }\n\n // GET|POST /api/integrations/:name/data - proxy to integration API\n const integrationDataMatch = pathname.match(/^\\/api\\/integrations\\/([^/]+)\\/data$/);\n if (integrationDataMatch && (method === \"GET\" || method === \"POST\")) {\n const name = integrationDataMatch[1];\n const credentials = await getCredentials(name);\n if (!credentials) {\n return json({ error: `Integration '${name}' not found or not configured` }, 404);\n }\n return handleIntegrationProxy(req, name, url.searchParams, credentials);\n }\n\n // Agents routes\n // GET /api/agents - list agents (optional ?workspace= query param)\n if (pathname === \"/api/agents\" && method === \"GET\") {\n const workspace = url.searchParams.get(\"workspace\") ?? undefined;\n const agents = await listAgents(workspace);\n return json(agents);\n }\n\n // POST /api/agents - create agent\n if (pathname === \"/api/agents\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n const agent = await createAgent(body);\n return json(agent, 201);\n }\n\n // PUT /api/agents/:id - update agent\n const agentUpdateMatch = pathname.match(/^\\/api\\/agents\\/([^/]+)$/);\n if (agentUpdateMatch && method === \"PUT\") {\n const id = agentUpdateMatch[1];\n const body = await req.json().catch(() => ({}));\n const agent = await updateAgent(id, body);\n if (!agent) {\n return json({ error: `Agent '${id}' not found` }, 404);\n }\n return json(agent);\n }\n\n // DELETE /api/agents/:id - delete agent\n const agentDeleteMatch = pathname.match(/^\\/api\\/agents\\/([^/]+)$/);\n if (agentDeleteMatch && method === \"DELETE\") {\n const id = agentDeleteMatch[1];\n const deleted = await deleteAgent(id);\n if (!deleted) {\n return json({ error: `Agent '${id}' not found` }, 404);\n }\n return json({ success: true, message: `Agent '${id}' deleted` });\n }\n\n // LLM config routes\n // GET /api/llm/config - get LLM config (mask api_key)\n if (pathname === \"/api/llm/config\" && method === \"GET\") {\n const config = await getLLMConfig();\n if (!config) {\n return json({ error: \"LLM config not found\" }, 404);\n }\n // Mask api_key: show first 8 chars + ***\n const masked = {\n ...config,\n api_key: config.api_key\n ? config.api_key.slice(0, 8) + \"***\"\n : \"\",\n };\n return json(masked);\n }\n\n // POST /api/llm/config - save LLM config\n if (pathname === \"/api/llm/config\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n const config = await saveLLMConfig(body);\n // Mask api_key in response\n const masked = {\n ...config,\n api_key: config.api_key\n ? config.api_key.slice(0, 8) + \"***\"\n : \"\",\n };\n return json(masked);\n }\n\n // Chat route\n // POST /api/chat - AI chat handler\n if (pathname === \"/api/chat\" && method === \"POST\") {\n return handleChat(req);\n }\n\n // POST /api/chat/stream - SSE streaming chat\n if (pathname === \"/api/chat/stream\" && method === \"POST\") {\n return handleChatStream(req);\n }\n\n // --- Conversation history routes ----\n if (pathname === \"/api/conversations\" && method === \"GET\") {\n return json(await listConversations());\n }\n const convIdMatch = pathname.match(/^\\/api\\/conversations\\/([^/]+)$/);\n if (convIdMatch && method === \"GET\") {\n const conv = await getConversationById(convIdMatch[1]);\n return conv ? json(conv) : json({ error: \"Not found\" }, 404);\n }\n if (convIdMatch && method === \"DELETE\") {\n const ok = await deleteConversation(convIdMatch[1]);\n return ok ? json({ success: true }) : json({ error: \"Not found\" }, 404);\n }\n\n // Tasks routes\n // GET /api/tasks - list tasks (optional ?workspace= query param)\n if (pathname === \"/api/tasks\" && method === \"GET\") {\n const workspace = url.searchParams.get(\"workspace\") ?? undefined;\n const status = url.searchParams.get(\"status\") ?? undefined;\n return json(await listTasks({ workspace, status }));\n }\n\n // POST /api/tasks - create task\n if (pathname === \"/api/tasks\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await createTask(body), 201);\n }\n\n // GET /api/tasks/ice - ICE-ranked backlog (sorted by ICE score desc).\n // Declared before the /api/tasks/:id matchers so \"ice\" isn't read as an id.\n if (pathname === \"/api/tasks/ice\" && method === \"GET\") {\n const workspace = url.searchParams.get(\"workspace\") ?? undefined;\n const limit = parseInt(url.searchParams.get(\"limit\") ?? \"50\", 10);\n return json(await listIceRankedTasks({ workspace, limit }));\n }\n\n // POST /api/tasks/generate - LLM-plan tasks from an experiment/KR/objective/campaign.\n if (pathname === \"/api/tasks/generate\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n if (!body.source_type || !body.source_id) {\n return json({ error: \"source_type and source_id are required\" }, 400);\n }\n return json(await generateTasksFromSource(body), 201);\n }\n\n // POST /api/tasks/:id/run - execute the task with an agent (sub-resource; before the bare :id matchers).\n const taskRunMatch = pathname.match(/^\\/api\\/tasks\\/([^/]+)\\/run$/);\n if (taskRunMatch && method === \"POST\") {\n const result = await runTask(taskRunMatch[1]);\n if (!result.task) return json({ error: result.error || \"Not found\" }, 404);\n return json(result);\n }\n\n // PUT /api/tasks/:id - update task (e.g. toggle status)\n const taskUpdateMatch = pathname.match(/^\\/api\\/tasks\\/([^/]+)$/);\n if (taskUpdateMatch && method === \"PUT\") {\n const id = taskUpdateMatch[1];\n const body = await req.json().catch(() => ({}));\n const updated = await updateTask(id, body);\n if (!updated) return json({ error: \"Not found\" }, 404);\n return json(updated);\n }\n\n // DELETE /api/tasks/:id - delete task\n const taskDeleteMatch = pathname.match(/^\\/api\\/tasks\\/([^/]+)$/);\n if (taskDeleteMatch && method === \"DELETE\") {\n const ok = await deleteTask(taskDeleteMatch[1]);\n return ok ? json({ success: true }) : json({ error: \"Not found\" }, 404);\n }\n\n // --- Relational CRM (accounts + contacts + deal↔contact links) ----\n // Delegate to api/crm_routes.ts; tenant context is already set by main.ts. Kept thin to stay under the 80KB file limit.\n if (isCrmPath(pathname)) return handleCrmRoutes(req, url, method);\n if (isMarketingPath(pathname)) return handleMarketingRoutes(req, url, method);\n\n // --- CRM Deals (sales pipeline) ----\n if (pathname === \"/api/deals\" && method === \"GET\") {\n return json(await listDeals(url.searchParams.get(\"stage\") ?? undefined));\n }\n if (pathname === \"/api/deals\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await createDeal(body), 201);\n }\n // Seed deals from qualified leads (explicit; checked before the :id matcher).\n if (pathname === \"/api/deals/sync\" && method === \"POST\") {\n return json(await syncDealsFromLeads());\n }\n // Interactive import: candidate leads to review, then import the selected emails.\n if (pathname === \"/api/deals/candidates\" && method === \"GET\") {\n return json(await listCandidates());\n }\n if (pathname === \"/api/deals/import\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await importDeals(Array.isArray(body.emails) ? body.emails : []));\n }\n const dealIdMatch = pathname.match(/^\\/api\\/deals\\/([^/]+)$/);\n if (dealIdMatch && dealIdMatch[1] !== \"sync\" && method === \"GET\") {\n const d = await getDeal(dealIdMatch[1]);\n return d ? json(d) : json({ error: \"Not found\" }, 404);\n }\n if (dealIdMatch && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n const updated = await updateDeal(dealIdMatch[1], body);\n return updated ? json(updated) : json({ error: \"Not found\" }, 404);\n }\n if (dealIdMatch && method === \"DELETE\") {\n const ok = await deleteDeal(dealIdMatch[1]);\n return ok ? json({ success: true }) : json({ error: \"Not found\" }, 404);\n }\n\n // --- Data export (CSV) ----\n if (pathname === \"/api/export/signals\" && method === \"GET\") {\n const r = await sqlite.execute(\"SELECT id, lead_email, lead_name, signal_type, signal_source, signal_strength, actioned, detected_at FROM intent_signals WHERE tenant_id = ? ORDER BY detected_at DESC\", [tid]);\n const cols = r.columns;\n let csv = cols.join(\",\") + \"\\n\";\n for (const row of r.rows) {\n csv += (row as any[]).map((v) => `\"${String(v ?? \"\").replace(/\"/g, '\"\"')}\"`).join(\",\") + \"\\n\";\n }\n return new Response(csv, { headers: { \"Content-Type\": \"text/csv\", \"Content-Disposition\": 'attachment; filename=\"signals.csv\"', ...corsHeaders } });\n }\n if (pathname === \"/api/export/okrs\" && method === \"GET\") {\n const objR = await sqlite.execute(\"SELECT id, title, description, workspace, status, progress, quarter, created_at FROM objectives WHERE tenant_id = ? ORDER BY created_at DESC\", [tid]);\n const krR = await sqlite.execute(\"SELECT id, objective_id, title, target_value, current_value, unit, progress, updated_at FROM key_results WHERE tenant_id = ? ORDER BY updated_at DESC\", [tid]);\n let csv = \"type,\" + objR.columns.join(\",\") + \"\\n\";\n for (const row of objR.rows) {\n csv += \"objective,\" + (row as any[]).map((v) => `\"${String(v ?? \"\").replace(/\"/g, '\"\"')}\"`).join(\",\") + \"\\n\";\n }\n if (krR.rows.length > 0) {\n csv += \"\\ntype,\" + krR.columns.join(\",\") + \"\\n\";\n for (const row of krR.rows) {\n csv += \"key_result,\" + (row as any[]).map((v) => `\"${String(v ?? \"\").replace(/\"/g, '\"\"')}\"`).join(\",\") + \"\\n\";\n }\n }\n return new Response(csv, { headers: { \"Content-Type\": \"text/csv\", \"Content-Disposition\": 'attachment; filename=\"okrs.csv\"', ...corsHeaders } });\n }\n if (pathname === \"/api/export/tasks\" && method === \"GET\") {\n const r = await sqlite.execute(\"SELECT id, title, description, status, priority, workspace, due_date, created_at, updated_at FROM tasks WHERE tenant_id = ? ORDER BY created_at DESC\", [tid]);\n const cols = r.columns;\n let csv = cols.join(\",\") + \"\\n\";\n for (const row of r.rows) {\n csv += (row as any[]).map((v) => `\"${String(v ?? \"\").replace(/\"/g, '\"\"')}\"`).join(\",\") + \"\\n\";\n }\n return new Response(csv, { headers: { \"Content-Type\": \"text/csv\", \"Content-Disposition\": 'attachment; filename=\"tasks.csv\"', ...corsHeaders } });\n }\n\n // --- LinkedIn OAuth flow (legacy aliases) ----\n if (pathname === \"/api/oauth/linkedin/start\" && method === \"GET\") {\n return Response.redirect(`${url.origin}/api/auth/linkedin`);\n }\n if (pathname === \"/api/oauth/linkedin/callback\" && method === \"GET\") {\n return Response.redirect(`${url.origin}/api/auth/linkedin/callback?${url.search.slice(1)}`);\n }\n\n // --- Graphify - app map for MCP / AI-native navigation ----\n if (pathname === \"/api/graphify\" && method === \"GET\") {\n return json(await getGraphify());\n }\n\n // --- Calendar today shortcut (TodayPanel widget) ----\n if (pathname === \"/api/calendar/today\" && method === \"GET\") {\n const today = new Date().toISOString().slice(0, 10);\n return json(await getUnifiedCalendar({ from: today, to: today }));\n }\n\n // --- Digest latest alias ----\n if (pathname === \"/api/digest/latest\" && method === \"GET\") {\n return json(await getLatestDigest());\n }\n\n // --- PLATFORM SUPERADMIN (master-only; the single audited exception to tenant isolation) ----\n // GATE first, then delegate all /api/platform/* routes to handlePlatformRoutes (api/platform.ts).\n if (pathname.startsWith(\"/api/platform/\")) {\n if (!currentCtx().isPlatformAdmin) return json({ error: \"Forbidden: platform admin only\" }, 403);\n return handlePlatformRoutes(req, url, method);\n }\n\n // --- ADMIN LOCKDOWN: every destructive /api/admin/* route is owner-only (plus per-route confirm\n // tokens below). No-op under default SYSTEM_CTX (role \"owner\"); locks out trial members. ----\n if (pathname.startsWith(\"/api/admin/\") && currentRole() !== \"owner\") {\n return json({ error: \"Forbidden: owner role required\" }, 403);\n }\n\n // Owner bootstrap (owner-only): create/attach the platform owner to t_default for email+password login. Idempotent.\n if (pathname === \"/api/admin/bootstrap-owner\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n if (!body.email || !body.password) return json({ ok: false, error: \"email + password required\" }, 400);\n const r = await bootstrapOwner(String(body.email), String(body.password));\n return json({ ok: r.ok, created: r.created });\n }\n\n // Admin backup/export (owner-only): read-only, tenant-scoped (WHERE tenant_id=?) JSON snapshot of this tenant's rows.\n // Confirm is optional: pass { confirm: \"EXPORT\" } to be explicit, otherwise it just runs.\n if (pathname === \"/api/admin/export\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n if (body.confirm != null && body.confirm !== \"EXPORT\") {\n return json({ ok: false, error: \"Send confirm: EXPORT (or omit confirm)\" }, 400);\n }\n const EXPORT_TABLES = [\n \"tasks\", \"deals\", \"campaigns\", \"objectives\", \"key_results\", \"experiments\",\n \"agents\", \"action_queue\", \"calendar_entries\", \"knowledge_files\", \"notes\",\n ];\n const data: Record = {};\n const counts: Record = {};\n for (const table of EXPORT_TABLES) {\n try {\n const r = await sqlite.execute(`SELECT * FROM ${table} WHERE tenant_id = ?`, [tid]);\n const rows = (r.rows || []).map((row) => rowToObject(r.columns, row));\n data[table] = rows;\n counts[table] = rows.length;\n } catch {\n // Missing table (e.g. not yet created): record empty rather than failing the whole export.\n data[table] = [];\n counts[table] = 0;\n }\n }\n const snapshot = {\n ok: true,\n tenant_id: tid,\n version: APP_VERSION,\n exported_at: new Date().toISOString(),\n tables: EXPORT_TABLES,\n counts,\n data,\n };\n const filename = `aihive-backup-${tid}-${new Date().toISOString().slice(0, 10)}.json`;\n return new Response(JSON.stringify(snapshot, null, 2), {\n headers: {\n \"Content-Type\": \"application/json\",\n \"Content-Disposition\": `attachment; filename=\"${filename}\"`,\n \"Cache-Control\": \"no-store\",\n ...corsHeaders,\n },\n });\n }\n\n // --- Admin purge (clear intelligence cache) ----\n if (pathname === \"/api/admin/purge\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n if (body.confirm !== \"PURGE\") return json({ error: \"Send confirm: PURGE\" }, 400);\n await sqlite.execute(\"DELETE FROM intelligence_cache WHERE tenant_id = ?\", [tid]).catch(() => {});\n await sqlite.execute(\"UPDATE app_config SET value=NULL WHERE key='funnel_cache' AND tenant_id = ?\", [tid]).catch(() => {});\n return json({ success: true, message: \"Cache purged\" });\n }\n\n // Remove demo/seed data: starter agents matched exactly on (name, description, system_prompt) so edits are never touched.\n if (pathname === \"/api/admin/seed-status\" && method === \"GET\") {\n let agents = 0;\n for (const seed of DEFAULT_AGENT_SEEDS) {\n const r = await sqlite.execute(\n \"SELECT COUNT(*) FROM agents WHERE name = ? AND description = ? AND system_prompt = ? AND tenant_id = ?\",\n [seed.name, seed.description, seed.system_prompt, tid],\n ).catch(() => null);\n if (r?.rows?.length) agents += Number((r.rows[0] as any[])[0]) || 0;\n }\n return json({ ok: true, total: agents, counts: { agents } });\n }\n if (pathname === \"/api/admin/cleanup-mock\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n if (body.confirm !== \"CLEANUP\") return json({ ok: false, error: \"Send confirm: CLEANUP\" }, 400);\n let removed = 0;\n for (const seed of DEFAULT_AGENT_SEEDS) {\n const r = await sqlite.execute(\n \"DELETE FROM agents WHERE name = ? AND description = ? AND system_prompt = ? AND tenant_id = ?\",\n [seed.name, seed.description, seed.system_prompt, tid],\n ).catch(() => null);\n removed += Number((r as any)?.rowsAffected ?? 0) || 0;\n }\n return json({ ok: true, total: removed });\n }\n\n // Bulk reset: wipe ALL agents (= agents + campaigns, same table) plus related\n // run/budget/conversation rows that deleteAgent() does not cascade. Confirm-guarded.\n if (pathname === \"/api/admin/reset-agents\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n if (body.confirm !== \"CLEANUP\") return json({ ok: false, error: \"Send confirm: CLEANUP\" }, 400);\n const counts: Record = {};\n for (const table of [\"agents\", \"agent_runs\", \"agent_budgets\", \"conversations\"]) {\n const r = await sqlite.execute(`DELETE FROM ${table} WHERE tenant_id = ?`, [tid]).catch(() => null);\n counts[table] = Number((r as any)?.rowsAffected ?? 0) || 0;\n }\n return json({ ok: true, counts });\n }\n\n // Factory reset (this app + tenant only): wipe the data tables below, keeping integrations/llm_config/knowledge_files. Confirm-guarded.\n if (pathname === \"/api/admin/reset-data\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n if (body.confirm !== \"CLEANUP\") return json({ ok: false, error: \"Send confirm: CLEANUP\" }, 400);\n const WIPE_TABLES = [\n \"agents\", \"conversations\", \"agent_budgets\", \"agent_runs\",\n \"objectives\", \"key_results\", \"north_star\", \"nsm_history\",\n \"experiments\", \"tasks\", \"tasks_legacy\", \"notes\",\n \"intent_signals\", \"growth_digests\", \"calendar_entries\", \"action_queue\",\n \"customer_cohorts\", \"lead_state_changes\", \"utm_conversions\",\n \"metric_baselines\", \"metric_history\", \"intelligence_cache\", \"app_config\",\n ];\n const counts: Record = {};\n for (const table of WIPE_TABLES) {\n const r = await sqlite.execute(`DELETE FROM ${table} WHERE tenant_id = ?`, [tid]).catch(() => null);\n counts[table] = Number((r as any)?.rowsAffected ?? 0) || 0;\n }\n return json({ ok: true, kept: [\"integrations\", \"llm_config\", \"knowledge_files\"], counts });\n }\n\n // Precise working-data wipe (KEEPS knowledge_files/skills/memory/app_config/integrations/llm_config). Confirm-guarded.\n if (pathname === \"/api/admin/wipe-working-data\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n if (body.confirm !== \"WIPE\") return json({ ok: false, error: \"Send confirm: WIPE\" }, 400);\n const WIPE = [\n \"agents\", \"agent_runs\", \"agent_budgets\", \"conversations\", \"deals\",\n \"experiments\", \"tasks\", \"tasks_legacy\", \"objectives\", \"key_results\",\n \"north_star\", \"nsm_history\", \"intent_signals\", \"growth_digests\",\n \"calendar_entries\", \"action_queue\", \"customer_cohorts\", \"lead_state_changes\",\n \"utm_conversions\", \"metric_baselines\", \"metric_history\", \"intelligence_cache\", \"notes\",\n ];\n const counts: Record = {};\n for (const table of WIPE) {\n const r = await sqlite.execute(`DELETE FROM ${table} WHERE tenant_id = ?`, [tid]).catch(() => null);\n counts[table] = Number((r as any)?.rowsAffected ?? 0) || 0;\n }\n return json({\n ok: true,\n kept: [\"knowledge_files\", \"skills\", \"memory\", \"app_config\", \"integrations\", \"llm_config\"],\n counts,\n });\n }\n\n // --- Targeted wipe: tasks + run history + agents/budgets + campaign↔agent links ONLY ----\n // Keeps campaigns/objectives/everything else. Confirm-guarded.\n if (pathname === \"/api/admin/wipe-tasks-agents\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n if (body.confirm !== \"WIPE\") return json({ ok: false, error: \"Send confirm: WIPE\" }, 400);\n const WIPE = [\"tasks\", \"tasks_legacy\", \"agent_runs\", \"agent_budgets\", \"campaign_agents\", \"agents\", \"campaign_runs\"];\n const counts: Record = {};\n for (const table of WIPE) {\n const r = await sqlite.execute(`DELETE FROM ${table} WHERE tenant_id = ?`, [tid]).catch(() => null);\n counts[table] = Number((r as any)?.rowsAffected ?? 0) || 0;\n }\n // Also clear internal completion events (metric_history integration='internal') and task-done\n // calendar markers (event_type='completion'); keep real integration metric history intact.\n const mh = await sqlite.execute(\"DELETE FROM metric_history WHERE integration = 'internal' AND tenant_id = ?\", [tid]).catch(() => null);\n counts[\"metric_history(internal)\"] = Number((mh as any)?.rowsAffected ?? 0) || 0;\n const cm = await sqlite.execute(\"DELETE FROM calendar_entries WHERE event_type = 'completion' AND tenant_id = ?\", [tid]).catch(() => null);\n counts[\"calendar_entries(completion)\"] = Number((cm as any)?.rowsAffected ?? 0) || 0;\n return json({ ok: true, wiped: [...WIPE, \"metric_history(internal)\", \"calendar_entries(completion)\"], counts });\n }\n\n // --- Targeted wipe: action_queue (approvals / prepared+sent action history) ONLY ----\n if (pathname === \"/api/admin/wipe-actions\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n if (body.confirm !== \"WIPE\") return json({ ok: false, error: \"Send confirm: WIPE\" }, 400);\n const r = await sqlite.execute(\"DELETE FROM action_queue WHERE tenant_id = ?\", [tid]).catch(() => null);\n return json({ ok: true, wiped: [\"action_queue\"], counts: { action_queue: Number((r as any)?.rowsAffected ?? 0) || 0 } });\n }\n\n // --- Google OAuth redirect flow ----\n if (pathname === \"/api/auth/google\" && method === \"GET\") {\n const creds = await getCredentials(\"google\");\n // OAuth client is env-driven by design (GOOGLE_CLIENT_ID/SECRET); fall back to the DB row.\n const clientId = Deno.env.get(\"GOOGLE_CLIENT_ID\") || creds?.extra_config?.client_id || creds?.extra_config?.clientId;\n const redirectUri = creds?.extra_config?.redirect_uri || creds?.extra_config?.redirectUri || `${url.origin}/api/auth/google/callback`;\n if (!clientId) {\n return new Response(\"Google OAuth not configured. Add Client ID in Settings - Google.\", { status: 400 });\n }\n const scopes = [\n \"https://www.googleapis.com/auth/userinfo.email\",\n \"https://www.googleapis.com/auth/analytics.readonly\",\n \"https://www.googleapis.com/auth/webmasters.readonly\",\n \"https://www.googleapis.com/auth/gmail.modify\",\n \"https://www.googleapis.com/auth/drive.readonly\",\n \"https://www.googleapis.com/auth/spreadsheets.readonly\",\n \"https://www.googleapis.com/auth/documents.readonly\",\n \"https://www.googleapis.com/auth/calendar\",\n \"https://www.googleapis.com/auth/youtube.readonly\",\n \"https://www.googleapis.com/auth/yt-analytics.readonly\",\n ].join(\" \");\n const { state, setCookie } = newOAuthState();\n const params = new URLSearchParams({\n client_id: clientId,\n redirect_uri: redirectUri,\n response_type: \"code\",\n scope: scopes,\n access_type: \"offline\",\n prompt: \"consent\",\n state,\n });\n return redirectWithCookie(`https://accounts.google.com/o/oauth2/v2/auth?${params}`, setCookie);\n }\n\n if (pathname === \"/api/auth/google/callback\" && method === \"GET\") {\n const code = url.searchParams.get(\"code\");\n const error = url.searchParams.get(\"error\");\n if (error || !code) {\n return Response.redirect(`${url.origin}/?auth_error=${encodeURIComponent(error || \"no_code\")}`);\n }\n if (!oauthStateValid(req, url)) {\n return Response.redirect(`${url.origin}/?auth_error=state_mismatch`);\n }\n const creds = await getCredentials(\"google\");\n // Trim env/DB values to avoid whitespace producing a malformed (invalid_request) token request.\n const clientId = (Deno.env.get(\"GOOGLE_CLIENT_ID\") || creds?.extra_config?.client_id || creds?.extra_config?.clientId || \"\").trim();\n const clientSecret = (Deno.env.get(\"GOOGLE_CLIENT_SECRET\") || creds?.api_secret || \"\").trim();\n const redirectUri = (creds?.extra_config?.redirect_uri || creds?.extra_config?.redirectUri || `${url.origin}/api/auth/google/callback`).trim();\n\n // Fail fast with a clear cause if a required param is empty (Google would return invalid_request).\n if (!clientId || !clientSecret) {\n const missing = !clientId && !clientSecret ? \"client_id_and_secret\"\n : !clientId ? \"client_id\" : \"client_secret\";\n return Response.redirect(`${url.origin}/?auth_error=${encodeURIComponent(\"missing_\" + missing)}`);\n }\n\n const tokenRes = await fetch(\"https://oauth2.googleapis.com/token\", {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/x-www-form-urlencoded\" },\n body: new URLSearchParams({ code, client_id: clientId, client_secret: clientSecret, redirect_uri: redirectUri, grant_type: \"authorization_code\" }),\n });\n const tokens = await tokenRes.json() as any;\n if (!tokenRes.ok || !tokens.access_token) {\n const detail = [tokens.error, tokens.error_description].filter(Boolean).join(\": \") || \"token_exchange_failed\";\n return Response.redirect(`${url.origin}/?auth_error=${encodeURIComponent(detail)}`);\n }\n\n const now = new Date().toISOString();\n const newExtraConfig = { ...(creds?.extra_config || {}), token_expiry: Date.now() + (tokens.expires_in || 3600) * 1000 };\n const existing = await sqlite.execute(\"SELECT id FROM integrations WHERE name = ? AND tenant_id = ?\", [\"google\", tid]);\n if (existing.rows.length > 0) {\n await sqlite.execute(\n \"UPDATE integrations SET access_token=?, refresh_token=?, extra_config=?, updated_at=? WHERE name='google' AND tenant_id = ?\",\n [tokens.access_token, tokens.refresh_token || creds?.refresh_token || null, JSON.stringify(newExtraConfig), now, tid]\n );\n } else {\n await sqlite.execute(\n \"INSERT INTO integrations (id, name, access_token, refresh_token, extra_config, connected_at, updated_at, tenant_id) VALUES (?,?,?,?,?,?,?,?)\",\n [crypto.randomUUID(), \"google\", tokens.access_token, tokens.refresh_token || null, JSON.stringify(newExtraConfig), now, now, tid]\n );\n }\n return redirectWithCookie(`${url.origin}/?auth=google_ok`, CLEAR_OAUTH_STATE);\n }\n\n // --- LinkedIn OAuth flow ----\n if (pathname === \"/api/auth/linkedin\" && method === \"GET\") {\n const creds = await getCredentials(\"linkedin\");\n const clientId = creds?.extra_config?.clientId || creds?.extra_config?.client_id;\n const redirectUri = `${url.origin}/api/auth/linkedin/callback`;\n if (!clientId) {\n return new Response(\"LinkedIn OAuth not configured. Add Client ID in Settings - LinkedIn.\", { status: 400 });\n }\n const { state, setCookie } = newOAuthState();\n const params = new URLSearchParams({\n response_type: \"code\",\n client_id: clientId,\n redirect_uri: redirectUri,\n scope: \"openid profile email w_member_social r_basicprofile\",\n state,\n });\n return redirectWithCookie(`https://www.linkedin.com/oauth/v2/authorization?${params}`, setCookie);\n }\n\n if (pathname === \"/api/auth/linkedin/callback\" && method === \"GET\") {\n const code = url.searchParams.get(\"code\");\n const error = url.searchParams.get(\"error\");\n if (error || !code) {\n return Response.redirect(`${url.origin}/?auth_error=${encodeURIComponent(error || \"no_code\")}`);\n }\n if (!oauthStateValid(req, url)) {\n return Response.redirect(`${url.origin}/?auth_error=state_mismatch`);\n }\n const creds = await getCredentials(\"linkedin\");\n const clientId = creds?.extra_config?.clientId || creds?.extra_config?.client_id || \"\";\n const clientSecret = creds?.api_secret || \"\";\n const redirectUri = `${url.origin}/api/auth/linkedin/callback`;\n\n const tokenRes = await fetch(\"https://www.linkedin.com/oauth/v2/accessToken\", {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/x-www-form-urlencoded\" },\n body: new URLSearchParams({ grant_type: \"authorization_code\", code, redirect_uri: redirectUri, client_id: clientId, client_secret: clientSecret }),\n });\n const tokens = await tokenRes.json() as any;\n if (!tokenRes.ok || !tokens.access_token) {\n return Response.redirect(`${url.origin}/?auth_error=${encodeURIComponent(tokens.error || \"linkedin_token_failed\")}`);\n }\n\n // Fetch LinkedIn profile to get the author URN needed for posts endpoint.\n let authorId = creds?.extra_config?.author_id || \"\";\n try {\n const profileRes = await fetch(\"https://api.linkedin.com/v2/me\", {\n headers: { Authorization: `Bearer ${tokens.access_token}` },\n });\n if (profileRes.ok) {\n const profile = await profileRes.json() as any;\n if (profile.id) authorId = `urn:li:person:${profile.id}`;\n }\n } catch { /* non-fatal - author_id can be set manually later */ }\n\n const now = new Date().toISOString();\n const newExtraConfig = { ...(creds?.extra_config || {}), token_expiry: Date.now() + (tokens.expires_in || 5184000) * 1000, author_id: authorId };\n const existing = await sqlite.execute(\"SELECT id FROM integrations WHERE name = ? AND tenant_id = ?\", [\"linkedin\", tid]);\n if (existing.rows.length > 0) {\n await sqlite.execute(\n \"UPDATE integrations SET access_token=?, extra_config=?, updated_at=? WHERE name='linkedin' AND tenant_id = ?\",\n [tokens.access_token, JSON.stringify(newExtraConfig), now, tid]\n );\n } else {\n await sqlite.execute(\n \"INSERT INTO integrations (id, name, access_token, extra_config, connected_at, updated_at, tenant_id) VALUES (?,?,?,?,?,?,?)\",\n [crypto.randomUUID(), \"linkedin\", tokens.access_token, JSON.stringify(newExtraConfig), now, now, tid]\n );\n }\n return redirectWithCookie(`${url.origin}/?auth=linkedin_ok`, CLEAR_OAUTH_STATE);\n }\n\n // Microsoft (Outlook) OAuth: env-first client (MICROSOFT_CLIENT_ID/SECRET), `common` tenant, DB fallback. Mirrors Google.\n if (pathname === \"/api/auth/microsoft\" && method === \"GET\") {\n const mcreds = await getCredentials(\"outlook\");\n const clientId = (Deno.env.get(\"MICROSOFT_CLIENT_ID\") || mcreds?.extra_config?.client_id || mcreds?.extra_config?.clientId || \"\").trim();\n if (!clientId) {\n return new Response(\"Microsoft OAuth not configured. Set MICROSOFT_CLIENT_ID (env) or add the Client ID in Settings → Outlook.\", { status: 400 });\n }\n const redirectUri = `${url.origin}/api/auth/microsoft/callback`;\n const scopes = MICROSOFT_SCOPES;\n const { state, setCookie } = newOAuthState();\n const params = new URLSearchParams({\n client_id: clientId,\n response_type: \"code\",\n redirect_uri: redirectUri,\n response_mode: \"query\",\n scope: scopes,\n state,\n });\n return redirectWithCookie(`https://login.microsoftonline.com/common/oauth2/v2.0/authorize?${params}`, setCookie);\n }\n\n if (pathname === \"/api/auth/microsoft/callback\" && method === \"GET\") {\n const code = url.searchParams.get(\"code\");\n const error = url.searchParams.get(\"error\");\n if (error || !code) {\n return Response.redirect(`${url.origin}/?auth_error=${encodeURIComponent(error || \"no_code\")}`);\n }\n if (!oauthStateValid(req, url)) {\n return Response.redirect(`${url.origin}/?auth_error=state_mismatch`);\n }\n const creds = await getCredentials(\"outlook\");\n const clientId = (Deno.env.get(\"MICROSOFT_CLIENT_ID\") || creds?.extra_config?.client_id || creds?.extra_config?.clientId || \"\").trim();\n const clientSecret = (Deno.env.get(\"MICROSOFT_CLIENT_SECRET\") || creds?.api_secret || \"\").trim();\n const redirectUri = `${url.origin}/api/auth/microsoft/callback`;\n if (!clientId || !clientSecret) {\n const missing = !clientId && !clientSecret ? \"client_id_and_secret\" : !clientId ? \"client_id\" : \"client_secret\";\n return Response.redirect(`${url.origin}/?auth_error=${encodeURIComponent(\"missing_microsoft_\" + missing)}`);\n }\n const scopes = MICROSOFT_SCOPES;\n const tokenRes = await fetch(\"https://login.microsoftonline.com/common/oauth2/v2.0/token\", {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/x-www-form-urlencoded\" },\n body: new URLSearchParams({ client_id: clientId, client_secret: clientSecret, code, redirect_uri: redirectUri, grant_type: \"authorization_code\", scope: scopes }),\n });\n const tokens = await tokenRes.json() as any;\n if (!tokenRes.ok || !tokens.access_token) {\n const detail = [tokens.error, tokens.error_description].filter(Boolean).join(\": \") || \"token_exchange_failed\";\n return Response.redirect(`${url.origin}/?auth_error=${encodeURIComponent(detail)}`);\n }\n\n const now = new Date().toISOString();\n const newExtraConfig = { ...(creds?.extra_config || {}), token_expiry: Date.now() + (tokens.expires_in || 3600) * 1000 };\n const existing = await sqlite.execute(\"SELECT id FROM integrations WHERE name = ? AND tenant_id = ?\", [\"outlook\", tid]);\n if (existing.rows.length > 0) {\n await sqlite.execute(\n \"UPDATE integrations SET access_token=?, refresh_token=?, extra_config=?, updated_at=? WHERE name='outlook' AND tenant_id = ?\",\n [tokens.access_token, tokens.refresh_token || creds?.refresh_token || null, JSON.stringify(newExtraConfig), now, tid]\n );\n } else {\n await sqlite.execute(\n \"INSERT INTO integrations (id, name, access_token, refresh_token, extra_config, connected_at, updated_at, tenant_id) VALUES (?,?,?,?,?,?,?,?)\",\n [crypto.randomUUID(), \"outlook\", tokens.access_token, tokens.refresh_token || null, JSON.stringify(newExtraConfig), now, now, tid]\n );\n }\n return redirectWithCookie(`${url.origin}/?auth=outlook_ok`, CLEAR_OAUTH_STATE);\n }\n\n // Identity login OAuth (sign-in, NOT data scopes): scope \"openid email profile\"; callback mints an app session.\n const loginMatch = pathname.match(/^\\/api\\/login\\/(google|microsoft)(\\/callback)?$/);\n if (loginMatch && method === \"GET\") {\n const provider = loginMatch[1] as \"google\" | \"microsoft\";\n return loginMatch[2]\n ? await identityLoginCallback(provider, req, url)\n : identityLoginStart(provider, url);\n }\n\n // --- Notes routes (TeamHub collaborative notes) ----\n if (pathname === \"/api/notes\" && method === \"GET\") {\n const workspace = url.searchParams.get(\"workspace\") ?? \"marketing\";\n const result = await sqlite.execute(\n \"SELECT * FROM notes WHERE workspace=? AND tenant_id = ? ORDER BY pinned DESC, created_at DESC\",\n [workspace, tid]\n );\n return json(result.rows.map((r) => rowToObject(result.columns, r)));\n }\n if (pathname === \"/api/notes\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n const id = crypto.randomUUID();\n const now = new Date().toISOString();\n await sqlite.execute(\n \"INSERT INTO notes (id, content, pinned, workspace, created_at, updated_at, tenant_id) VALUES (?,?,?,?,?,?,?)\",\n [id, body.content ?? \"\", body.pinned ? 1 : 0, body.workspace ?? \"marketing\", now, now, tid]\n );\n return json({ id, content: body.content ?? \"\", pinned: !!body.pinned, workspace: body.workspace ?? \"marketing\", created_at: now, updated_at: now }, 201);\n }\n const noteMatch = pathname.match(/^\\/api\\/notes\\/([^/]+)$/);\n if (noteMatch && method === \"PUT\") {\n const id = noteMatch[1];\n const body = await req.json().catch(() => ({}));\n const now = new Date().toISOString();\n const setParts: string[] = [\"updated_at=?\"];\n const setArgs: any[] = [now];\n if (body.pinned !== undefined) { setParts.push(\"pinned=?\"); setArgs.push(body.pinned ? 1 : 0); }\n if (body.content !== undefined) { setParts.push(\"content=?\"); setArgs.push(body.content); }\n await sqlite.execute(\n `UPDATE notes SET ${setParts.join(\", \")} WHERE id=? AND tenant_id = ?`,\n [...setArgs, id, tid]\n );\n const r = await sqlite.execute(\"SELECT * FROM notes WHERE id=? AND tenant_id = ?\", [id, tid]);\n if (!r.rows.length) return json({ error: \"Not found\" }, 404);\n return json(rowToObject(r.columns, r.rows[0]));\n }\n if (noteMatch && method === \"DELETE\") {\n await sqlite.execute(\"DELETE FROM notes WHERE id=? AND tenant_id = ?\", [noteMatch[1], tid]);\n return json({ success: true });\n }\n\n // --- World dashboard rollups (logs / kanban / project panels) ----\n // /api/world/activity — cross-agent runs; ?campaign[_id]=ID scopes to that campaign, bare ?campaign = legacy filter.\n if (pathname === \"/api/world/activity\" && method === \"GET\") {\n const limit = parseInt(url.searchParams.get(\"limit\") ?? \"40\", 10);\n const status = url.searchParams.get(\"status\") ?? undefined;\n const agent_id = url.searchParams.get(\"agent_id\") ?? undefined;\n const campaignParam = url.searchParams.get(\"campaign_id\") ?? url.searchParams.get(\"campaign\");\n const campaign_id = campaignParam ? campaignParam : undefined;\n const campaign = url.searchParams.get(\"campaign\") != null && !campaign_id;\n return json(await getWorldActivity({ limit, status, agent_id, campaign, campaign_id }));\n }\n // /api/world/board — tasks grouped into the 4 kanban columns; ?campaign[_id]=ID scopes them.\n if (pathname === \"/api/world/board\" && method === \"GET\") {\n const workspace = url.searchParams.get(\"workspace\") ?? undefined;\n const campaign_id = url.searchParams.get(\"campaign_id\") ?? url.searchParams.get(\"campaign\") ?? undefined;\n return json(await getWorldBoard({ workspace, campaign_id: campaign_id || undefined }));\n }\n // /api/world/project — active campaign brief + real run-usage rollup; ?campaign[_id]=ID focuses it.\n if (pathname === \"/api/world/project\" && method === \"GET\") {\n const workspace = url.searchParams.get(\"workspace\") ?? undefined;\n const campaign_id = url.searchParams.get(\"campaign_id\") ?? url.searchParams.get(\"campaign\") ?? undefined;\n return json(await getWorldProject({ workspace, campaign_id: campaign_id || undefined }));\n }\n\n // --- Campaigns (Director: multi-agent orchestrated pipelines) ----\n if (pathname === \"/api/campaigns\" && method === \"GET\") {\n const workspace = url.searchParams.get(\"workspace\") ?? undefined;\n return json(await listCampaigns(workspace));\n }\n if (pathname === \"/api/campaigns\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await createCampaign(body), 201);\n }\n // Sub-resources (:id/runs, :id/agents, :id/trigger) before the bare :id matcher.\n const campaignRunsMatch = pathname.match(/^\\/api\\/campaigns\\/([^/]+)\\/runs$/);\n if (campaignRunsMatch && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n // Runs the Director pipeline synchronously and returns the run result.\n return json(await orchestrateCampaignRun(campaignRunsMatch[1], body.input));\n }\n if (campaignRunsMatch && method === \"GET\") {\n return json(await listCampaignRuns(campaignRunsMatch[1]));\n }\n const campaignAgentsMatch = pathname.match(/^\\/api\\/campaigns\\/([^/]+)\\/agents$/);\n if (campaignAgentsMatch && method === \"GET\") {\n return json(await getCampaignAgents(campaignAgentsMatch[1]));\n }\n // Clicking a campaign returns its tasks (OKR→Campaign→Task link).\n const campaignTasksMatch = pathname.match(/^\\/api\\/campaigns\\/([^/]+)\\/tasks$/);\n if (campaignTasksMatch && method === \"GET\") {\n return json(await listTasks({ campaign_id: campaignTasksMatch[1] }));\n }\n const campaignTriggerMatch = pathname.match(/^\\/api\\/campaigns\\/([^/]+)\\/trigger$/);\n if (campaignTriggerMatch && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await setCampaignTrigger(campaignTriggerMatch[1], body));\n }\n const campaignIdMatch = pathname.match(/^\\/api\\/campaigns\\/([^/]+)$/);\n if (campaignIdMatch && method === \"GET\") {\n const c = await getCampaign(campaignIdMatch[1]);\n return c ? json(c) : json({ error: \"Not found\" }, 404);\n }\n if (campaignIdMatch && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n return json(await updateCampaign(campaignIdMatch[1], body));\n }\n if (campaignIdMatch && method === \"DELETE\") {\n const ok = await deleteCampaign(campaignIdMatch[1]);\n return ok ? json({ success: true }) : json({ error: \"Not found\" }, 404);\n }\n\n // No route matched\n return json({ error: \"Not found\", path: pathname }, 404);\n } catch (err) {\n const message = err instanceof Error ? err.message : String(err);\n console.error(`[router] Error handling ${method} ${pathname}:`, message);\n return json({ error: \"Internal server error\", detail: message }, 500);\n }\n}\n\n/**\n * Converts a sqlite result row (array of values) to a plain object\n * using the columns array from the result set.\n */\nfunction rowToObject(columns: string[], row: unknown[]): Record {\n const obj: Record = {};\n for (let i = 0; i < columns.length; i++) {\n obj[columns[i]] = row[i];\n }\n return obj;\n}\n",[200],"731c8246-5f37-11f1-acf9-1607ee4eb77e","main.ts",1066,["D",1780482718390],23,"http","import { initDB, tenantStore, SYSTEM_CTX } from \"./db.ts\";\nimport { handleRequest } from \"./router.ts\";\nimport { getFrontendHTML, getWorldHTML } from \"./frontend.ts\";\nimport { seedKnowledge } from \"./api/knowledge_seed.ts\";\nimport { resolveSession } from \"./api/session.ts\";\n\n// Foundation tenant context: a session resolves to a TenantCtx; absent one we fall back to the\n// system/default SYSTEM_CTX (imported from db.ts, typed as TenantCtx) so all existing\n// (t_default-backfilled) data keeps working exactly as today.\n\nlet initialized = false;\n\n// Shared-secret gate. Set APP_TOKEN in the val's environment. Every /api/* route\n// requires `Authorization: Bearer ` (or `?token=` for download links),\n// EXCEPT the public paths below. Fail-closed: if APP_TOKEN is unset, data routes 503.\nconst APP_TOKEN = (Deno.env.get(\"APP_TOKEN\") || \"\").trim();\n\n// Auth ENABLED (Phase 3): /api/* requires a valid nexus_session cookie OR the APP_TOKEN bearer\n// (recovery hatch). Public paths (app modules, /api/auth/*, /api/login/*, /api/health) stay open so\n// the login screen always loads. Verified: owner session authorizes, anonymous → 401 → login screen.\nconst AUTH_DISABLED = false;\n\n// Public /api paths that must stay reachable without the app token:\n// - /api/app-* : the frontend's own JS modules (static code, no secrets; loaded by \n \n \n \n \n \n \n \n \n\n\n

\n \n\n`;\n}\n\n// Part 10: the Agent World shell — a \"the-delegation\" simulation DASHBOARD at /world. Same origin\n// (shares nexus_token + every /api/* route), so the agents + panels run over the REAL backend.\n// This shell is the SOLE creator of every fixed DOM id the world modules target. It is a CSS-grid\n// dashboard (light theme): a top bar, a LEFT logs panel, a CENTER stage (scene header + a BOUNDED\n// canvas + a kanban under it) and a RIGHT project panel. The canvas is NOT fullscreen — it lives\n// inside #world-canvas-wrap and the scene sizes to that wrapper. three.js (+ addons) load from a\n// CDN via the importmap; the 3D + panel logic is in /api/app-world (the entry) and its sub-modules.\nexport function getWorldHTML(): string {\n return `\n\n\n \n \n AIHive · Agent World\n \n \n \n

Diff · v1072