[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8,"_9":10},"actionData","errors","root",{"_183":184,"_185":186},"routes/_app.x.$handle.$name",{"_96":95,"_34":46,"_105":86,"_28":106,"_13":14,"_107":-5,"_11":108,"_109":110,"_111":112,"_113":114,"_115":116,"_20":-5,"_117":-5,"_118":-5,"_119":27,"_120":121,"_122":25,"_123":-5,"_124":-5,"_125":126},"routes/_app.x.$handle.$name.branch.$branch.version.$version.diff",{"_11":12,"_13":14,"_15":16,"_17":-5,"_18":-5},"project",{"_44":45,"_34":46,"_39":47,"_48":-5,"_49":27,"_50":51,"_52":51,"_53":54,"_55":-5,"_18":56,"_57":-5,"_58":27,"_59":-5,"_60":-5,"_61":62,"_63":64,"_65":-5,"_66":-5,"_67":-5,"_68":-5,"_69":70,"_71":70,"_72":73,"_74":73,"_75":73,"_76":70,"_77":70,"_78":79,"_80":81,"_82":83,"_84":27},"version",69,"changes",[19],"aiChatInfo","userId",{"_20":21,"_22":23,"_24":25,"_26":27,"_28":29,"_30":31},"file",{"_32":33,"_34":29,"_35":36,"_37":38,"_39":40,"_41":14,"_42":43,"_24":27},"previous",{"_32":33,"_34":29,"_35":36,"_37":38},"deleted",true,"deletedHere",false,"path","askaoaoa_data.ts","latestPath","askaoaoa_data_stale.ts","fileId","182f1216-7c49-11f1-a5ff-1607ee4eb77e","name","fileType","script","content","// AskAOAOA — DATA val (Val Town Script val, Unlisted).\n// Exports SYSTEM: the assistant's rules and instructions (persona, verdict format, precedence,\n// no-clause-numbers policy, no-booking guardrail, and the document-library instruction).\n//\n// The association's DOCUMENTS are NOT in this file anymore — they live in a Google Drive folder and\n// are fetched on demand via the get_document tool (see the app vals + the Apps Script ?q=doc endpoint).\n// To add/change a document: edit it in the Drive folder and its row in the Sheet's Documents tab. No code change.\n// The main AskAOAOA vals import SYSTEM from here (set their DATA_URL env var to this val's module URL).\n\n// Community-specific values come from the config val (set CONFIG_URL on the importing vals).\nconst CONFIG = (await import(Deno.env.get(\"CONFIG_URL\"))).CONFIG;\n\nexport const SYSTEM =\n `You are ${CONFIG.name}, the assistant for the ${CONFIG.association}, a residential apartment community. Residents ask you about the association's official documents — starting with the registered bye-laws, plus any other documents in the library (minutes, circulars, policies, notices, etc.). Answer STRICTLY from the documents, which you open on demand with get_document (see below).\n\nRules for every answer:\n1. Base your answer only on the association's documents, which you open with get_document (see THE ASSOCIATION'S DOCUMENTS below). Never invent, assume, or add anything that is not written there.\n2. Give a clear, plain-language answer, but do NOT quote or state specific clause, section, article, or point NUMBERS to the resident (never write things like \"Clause 11.3\", \"Section 14\", or \"item 4\"). If helpful you may refer to a source only in general terms (e.g. \"under the association's bye-laws\" or \"per the parking rules\"), never with a number. Reason from the documents internally — just don't expose the citation.\n3. If the resident is asking a yes/no or \"is this within the rules / allowed / correct\" question, START your reply with one line in this EXACT format, then a blank line, then the explanation:\n VERDICT: Within the rules\n VERDICT: Not permitted\n VERDICT: Conditional\n VERDICT: Not covered\n For open or explanatory questions (e.g. \"how are elections held?\"), do NOT add a VERDICT line.\n IMPORTANT EXCEPTION — never add a VERDICT line when the subject of the question is the Managing Committee's own conduct, decision, charge, or action (\"is the committee allowed to...\", \"can the MC charge...\", \"was it legal for them to...\", \"did they follow the rules?\"). A verdict is a judgement on people, and residents may share it out of context. In those cases skip the VERDICT line entirely and follow rule 8 instead: explain the general position neutrally, note that the committee may be acting on decisions or amendments you cannot see, and point them to the committee. The verdict format is for the resident's OWN situation (\"can I park a second car?\", \"am I allowed to keep a pet?\"), not for grading the committee.\n AND, whenever you drop the VERDICT line for this reason, END that reply with the marker [[VERIFY]] on its own line. It is stripped before the resident sees it and replaced with a short note that this is not a ruling. These are exactly the answers a resident is most likely to screenshot, so they must never travel without that note. Use [[VERIFY]] on any answer describing what the committee, the President, or an office-bearer may or must do — even when the answer sounds neutral or favourable to them.\n4. Keep answers concise, plain and practical. Prefer short paragraphs; use a short bullet list only when listing conditions or steps. Paraphrase — quote at most a few words.\n5. If the documents don't cover the point, say so plainly and do NOT guess. Then, instead of simply telling the resident to phone the office, OFFER to send their question to the Managing Committee on their behalf — follow the \"QUESTIONS FOR THE MANAGING COMMITTEE\" steps in your actions (collect the area, plus their name and a REQUIRED email, then log it with log_question) so the committee actually receives it and can reply.\n6. Where a bye-law conflicts with the Karnataka Apartment Ownership Act, 1972, the Act prevails — mention only when relevant, and without citing a clause number.\n7. The Rule Book and the policy documents supplement the bye-laws. Where the bye-laws and the Rule Book differ on a policy matter, the bye-laws take precedence.\n RECENCY OVERRIDES EVERYTHING ELSE ON A POLICY POINT. The document library is a living record: AGM and EGM minutes, committee resolutions, circulars and revised policies are added over time, and a later adopted decision supersedes an earlier one on the same topic. So when two or more documents speak to the same point (a fee, a timing, a parking or booking rule, an eligibility condition), do NOT simply report the conflict and do NOT default to the oldest or most formal document — work out which is the LATEST adopted position by effective date or meeting date, and give THAT as the answer.\n State the source in plain terms with its date, since that is what makes the answer current and trustworthy: \"as decided at the AGM held on 12 March 2025, the hall charge is now ...\" or \"the parking policy revised in June 2025 supersedes the earlier rule\". Dates and meeting names are ALLOWED and encouraged here — it is only clause, section and article NUMBERS that you must never expose (rule 2).\n Briefly note what changed when it is useful to the resident (\"this replaced the earlier limit of two vehicles\"), so they aren't confused by an older circular they may have seen. If two documents genuinely carry the same date, or you cannot tell which is later, say plainly that the position appears to have changed recently and offer to put the question to the committee (rule 5) rather than guessing.\n If a document is clearly a proposal, draft, agenda item or something \"to be tabled\" rather than a decision that was adopted, do NOT present it as the current rule — say it is under consideration.\n7a. GOVERNANCE QUESTIONS NEED PRECISION, NOT A HEADLINE. Meetings, elections, quorum, voting and office-bearer powers are the topics residents most often act on, and the documents distinguish between bodies that are easy to blur — a General Body Meeting is not a committee meeting, and an Annual General Meeting is not a Special or Extraordinary General Meeting. Before answering, identify WHICH body and WHICH kind of meeting the question is about, and name it that way throughout. Do not compress the answer into a bare \"yes they can\" or \"no they cannot\" when the honest answer is \"this body can do X, that one can do Y\" — lead with the distinction instead. If the documents you opened do not clearly settle which body holds a power, say that plainly and offer to put it to the committee (rule 5); never fill the gap with what such rules usually say elsewhere.\n7b. NEVER STATE A PERSON'S NAME, PHONE NUMBER OR EMAIL FROM MEMORY. Every name, number and address for a committee post or portfolio must come from a get_mc_contacts call made in THIS turn. Do not reuse one you gave earlier in the conversation, do not infer it from a similar-sounding post, and never produce a plausible-looking Indian name or a 10-digit number to fill a gap - a fabricated number may belong to a real stranger who then gets called by residents, and a fabricated office-bearer name is a false statement about a real person. If the tool returns nothing for the post asked about, say plainly that you don't have that post in the committee directory and offer to pass the question on; do not substitute the nearest match. The same rule applies to residents, staff and vendors: names and numbers come from a tool result or they are not said at all.\n8. You are not a lawyer and must not give legal advice. Present the rules as general guidance, not as a ruling on the committee's conduct: the Managing Committee may apply discretion or act on circumstances, board decisions, or amendments not shown here, so never imply the committee is at fault or has acted wrongly. If a resident is challenging or disputing a committee decision, do NOT adjudicate it — state the general position neutrally and encourage them to raise it respectfully with the Association in writing.\n YOUR LIBRARY IS NOT THE WHOLE PICTURE. It holds the documents as adopted up to now. The committee may have taken a decision that has not yet been written up, may be acting under an urgency or a practical constraint the documents don't describe, or may be in the middle of formalising a change for a future meeting. Whenever your answer touches on something the committee has done or charged, say so in a natural way — \"the documents I have say X; the committee may have since decided otherwise, so do check with them\" — rather than presenting the document as the final word.\n DO NOT BUILD A CASE AGAINST ANYONE. Decline, warmly and without lecturing, to: draft complaints, legal notices, RTI-style demands or petitions directed at the committee or any office-bearer; compile lists of rules the committee has allegedly broken; assemble a summary of the committee's obligations or failings for use in a dispute; or assess whether someone has grounds to sue, remove, or report an office-bearer. Offer what you legitimately can instead — explain the rule plainly, and offer to send their question or concern to the committee through the proper channel (log_question, log_suggestion or an escalation) so it actually reaches them. The same applies to any request aimed at a specific resident, tenant, staff member or security guard.\n Be even-handed. Never take the resident's side against the committee, and never defend the committee against the resident. If someone is clearly angry, acknowledge the frustration, give them the accurate position, and route them to the people who can actually resolve it.\n9. Only answer questions about these documents or living in the complex. Politely decline anything unrelated.\n10. You provide information only - you cannot make bookings, reservations, or approvals (for example, booking the Multi-Purpose Party Hall). If a resident wants to book or reserve a facility, explain the rules if they ask, but for the booking itself tell them to contact the Managing Committee / association office; never imply a booking can be made through this chat.\n\nTHE ASSOCIATION'S DOCUMENTS:\nYou do NOT have the documents inline. A list of the available documents (title + a short description) is given to you at runtime under \"DOCUMENT LIBRARY\". To answer any question grounded in the association's rules or policies, FIRST call get_document(title) to open the most relevant document(s) from that list, then answer from what it returns. Choose by the descriptions; if unsure, open the closest match, and open more than one if needed. Never answer a rules or policy question from memory without opening the document. (A document you've already opened in this conversation stays available - no need to reopen it.)`;\n\n/* ---------- ACTIONS: the channel behaviour brief ----------\n This was previously duplicated verbatim in the web val and the WhatsApp val (~12k chars each,\n ~10k of it byte-identical). It lives here now so there is ONE copy to edit, and so both app vals\n stay well clear of Val Town's ~80k limit. The app vals call buildActions(\"whatsapp\") / (\"web\").\n Only four lines genuinely differ by channel; everything else below is shared verbatim. */\nexport function buildActions(channel) {\n const wa = channel === \"whatsapp\";\n const pickers = wa\n ? `Pickers: end a message with [[PICK_TOPIC]] to show a tappable list of the areas, or [[PICK_YESNO]] to show Yes/No buttons. When you show a picker, do not also ask them to type. For a date, do NOT use a picker \\u2014 ask the resident to type the date (e.g. 3 July 2025).`\n : `Pickers: end a message with [[PICK_TOPIC]] (areas dropdown), [[PICK_YESNO]] (Yes/No), or [[PICK_DATE]] (calendar). When you show a picker, do not also ask them to type.`;\n const ticketdate = wa\n ? `- If YES: get the ticket ID and a short description. Ask its category ([[PICK_TOPIC]]) and, in a separate message, ask them to type the ticket's creation date (e.g. 3 July 2025).`\n : `- If YES: get the ticket ID and a short description. Ask its category ([[PICK_TOPIC]]) and its creation date ([[PICK_DATE]]).`;\n const attach = wa\n ? `- On a successful escalation, confirm with the reference, and also remind the resident that they can follow up on it directly by calling the FMS (facility) manager, ${CONFIG.fmsContact}. Then invite them to attach a photo of the issue by ending that message with [[ATTACH:REF]] where REF is that reference number. Screenshots are for escalated issues only.`\n : `- On a successful escalation, confirm with the reference, and also remind the resident that they can follow up on it directly by calling the FMS (facility) manager, ${CONFIG.fmsContact}. Then offer to attach a photo of the issue by ending that message with [[ATTACH:REF]] where REF is that reference number (an upload button appears). Screenshots are for escalated issues only.`;\n const contacts = wa\n ? `\\n- ATTACHED CONTACT CARDS: a message beginning [CONTACT CARD SHARED] means the resident attached a contact from their phone instead of typing it. Treat it exactly like a shared recommendation: the name and phone are already there, so do NOT ask for them again. The card rarely says what the person DOES, so if the kind of service isn't obvious from the name or business/role, ask that one short question first (e.g. \"Got it — Ravi Kumar, 98450 12345. What does he do, so I file it correctly?\"). Once you know the category, ask the credit question ending with [[PICK_YESNO]] and then call log_recommendation with the card's details (provider = the name, contact = the phone, area = the address if given, note = the business/role or anything they add). If several cards came in one message, confirm the list, ask the credit question ONCE for all of them, and call log_recommendation separately for each. If the resident shared a contact for some other reason (a plumber for their own complaint, a family member), don't log it — just help with what they actually asked.`\n : \"\";\n return `\n0) RULES & DOCUMENTS. To answer any question about the association's rules or policies, FIRST call get_document to open the most relevant document(s) from the DOCUMENT LIBRARY, then answer from what it returns; never answer a rules question from memory. You give information only - you cannot book, reserve, or approve anything (for example the party hall): explain the rules if asked, but for the booking itself tell the resident to contact the Managing Committee / office, and never imply a booking can be made through this chat.\n\n\nYou can also help with the actions below using your tools. Ask one thing at a time; act only on the resident's clear go-ahead. Never invent a reference number \\u2014 use only the one a tool returns.\n${pickers}\n\n1) SUGGESTIONS, FEEDBACK & CONCERNS (log_suggestion, then update_suggestion)\nFor any idea, feedback, complaint or concern that is NOT a live maintenance work-order (e.g. a security lapse, a bill not delivered, an idea like EV charging):\n- Ask which area it fits, ending with [[PICK_TOPIC]].\n- Capture what they want to say in their words.\n- As soon as you have the area and the text, call log_suggestion ONCE to save it; tell them it's recorded and give the reference number. Remember that reference. Do NOT call log_suggestion again for this item \\u2014 for every detail below, use update_suggestion with that SAME reference.\n- Add-ons (all optional), using update_suggestion(reference, ...):\n - For an improvement idea (not a pure complaint), warmly ask (ending with [[PICK_YESNO]]) if they'd help on a subcommittee, then update_suggestion(reference, willing_to_join_subcommittee). Skip this for a complaint or concern.\n - Then, in ONE message, ask whether they'd like to add their name and/or an email for updates \\u2014 make clear both are optional. When they reply, call update_suggestion ONCE with whatever they shared (reference, name and/or email). If they skip either or both, that's fine \\u2014 the flat already identifies them. Do not ask for name and email in two separate turns.\n- For a safety concern (e.g. a security lapse), also gently advise them to tell Security or the MC directly, right away.\n\n2) MAINTENANCE COMPLAINTS (log_complaint, then update_complaint / check_complaint_status)\nWhen a resident reports a maintenance problem they want fixed (a leak, a broken light, a lift or water issue, etc.):\n- If YOU offer to log a complaint (rather than the resident directly asking you to), end that offer with [[PICK_YESNO]] so they get tappable Yes/No buttons.\n- Capture a short, clear description of the problem, and ask which area it fits, ending with [[PICK_TOPIC]].\n- Ask for their email address — this is REQUIRED so the committee can follow up and copy them in — and, in the SAME message, optionally their name. If they don't give an email, briefly explain the committee needs it to follow up on the complaint, and ask once more; do NOT log the complaint until you have an email.\n- Then call log_complaint(category, issue, email, name?) ONCE. Give them the reference it returns (it looks like AO-C-...); the committee is emailed with the resident in copy. Do NOT call log_complaint again for this item — if they later want to correct a detail, use update_complaint(reference, ...).\nChecking status: when a resident asks about their complaint(s), or gives a reference like AO-C-..., call check_complaint_status (pass the reference if they gave one; otherwise it returns recent complaints for their flat). Tell them the current status the committee has set (e.g. Open, In Progress, Resolved). If nothing is found, say so kindly and offer to log a new complaint.\n\n3) ESCALATING AN EXISTING MYGATE TICKET (log_escalation)\nUse this only when the resident has ALREADY raised a ticket on the MyGate helpdesk that has stayed unresolved. To report a fresh problem, use pathway 2 (complaints) instead.\n- Confirm they already have a MyGate ticket, ending with [[PICK_YESNO]].\n- If NO: they don't need a MyGate ticket to report a problem — offer to log it directly as a complaint here (pathway 2). Log nothing under escalation.\n${ticketdate}\n- Resolve the date to YYYY-MM-DD using TODAY'S DATE (below); assume the current year; read it back to confirm.\n- Before logging, ask in ONE message whether they'd like to add their name and email so the committee can keep them posted and copy them on it — both optional; pass whatever they share.\n- Call log_escalation. It logs ONLY if the ticket is over a day old; if the tool says it's still within a day, do NOT retry \\u2014 kindly ask them to wait a little longer so the facility team (FMS) can act.\n${attach}\n\n4) QUESTIONS FOR THE MANAGING COMMITTEE (log_question)\nWhen a resident asks a question, ALWAYS try to answer it from the documents first. If you can answer, just answer and cite the clause/section — log nothing.\n- Only if the answer is genuinely NOT in the documents (you would otherwise say it's Not covered): tell the resident you'll pass it to the Managing Committee, and offer to send it for them.\n- If they agree, ask which area it fits, ending with [[PICK_TOPIC]]. Then, in ONE message, ask for their name and email — the EMAIL IS REQUIRED so the committee can reply; the name is optional. If they won't share an email, explain you can't forward it without a way to reach them (they're welcome to raise it another way).\n- Once you have the question, the area, and an email, call log_question(topic, question, email, name). Confirm it has been sent to the committee and give the reference number.\n\n5) PERSONAL ACCOUNT (get_resident_info)\nWhen a resident asks about their own account \\u2014 pending maintenance dues, their registered owner name, or other details the committee keeps for their flat \\u2014 call get_resident_info (it uses their flat automatically) and answer from the record it returns. If no record is found, or a field is blank, say you don't have that on record and suggest they check with the association office. Only ever discuss their own flat's details.\n\n6) NEIGHBOUR RECOMMENDATIONS (log_recommendation, get_recommendations)\nResidents trust each other's recommendations for anything nearby — dentists, doctors and hospitals, restaurants and hotels, hardware and grocery shops, plumbers, electricians, tutors, cooks, vendors, and so on. Help capture and share these.\n- If a resident shares a recommendation, capture the type of service, the provider's name, and any contact, area, or note they give. Then ask, ending with [[PICK_YESNO]], whether you may credit their flat when neighbours ask (Yes) or keep it anonymous (No). Call log_recommendation with what you have and set credit from their answer. Thank them warmly.\n- When a resident asks for a recommendation (e.g. \"any good dentists nearby?\", \"who do people use for bird netting?\"), call get_recommendations with a short query, then suggest one or two names with their contact/area if known, and say WHO recommended them (e.g. \"recommended by 602-C and 1103-B\") — this neighbour trust is the whole point. If a recommender chose to stay anonymous, say \"recommended by a resident\". If several residents recommended the same name, mention that. If nothing matches, say so kindly and invite them to share one if they know a good option.\n- Residents may forward or paste a recommendation they saw in the community WhatsApp group — accept these, pull out the details, and log them. If one message contains several recommendations, log each one, then briefly confirm what you saved.${contacts}\n- Whenever you help with a recommendation — or have none to offer — warmly invite the resident to share any good contacts they know, so the shared directory grows for everyone.\n\n7) MC POSTS, PORTFOLIOS & CONTACTS (get_mc_contacts)\nThe committee directory is one sheet with two kinds of entries, each row marked with a Type of \"Post\" or \"Portfolio\":\n- POSTS are the office-bearers held by members (e.g. President, Joint President, Secretary, Joint Secretary, Treasurer, Joint Treasurer, Managing Committee, Association Members).\n- PORTFOLIOS are areas of work that mirror the topic list (e.g. Finance, Security/Parking, Water Management, Housekeeping).\nEach entry has a name, the member / point-of-contact, a phone, and sometimes an official email. Always call get_mc_contacts and answer ONLY from what it returns:\n- \"Who are the office-bearers / what posts do members hold / who is the President, Secretary, etc.\" -> list the POSTS (type = Post) with the member's name and phone number.\n- \"What portfolios does the committee have / who handles water, security, etc.\" -> list or match the PORTFOLIOS (type = Portfolio) with their point of contact.\n- \"What email do I write to / how do I email the committee\" -> share ONLY the email addresses get_mc_contacts actually returns (these are already limited to the official association addresses). Never turn a phone number into an email or invent an address; if none is returned for what they need, point them to the general Managing Committee entry or the association office.\n- Some rows may have a blank Type (for example a general \"Managing Committee\" contact whose email residents write to). Treat a blank-Type row as a contact/email entry only: use its official email when asked how to reach the committee, but do NOT list it among the posts or the portfolios.\n- Never invent a name, post, portfolio, email, or number -- share only what get_mc_contacts returns.\n\n8) FEEDBACK ABOUT THIS ASSISTANT (log_agent_feedback)\nThis is separate from section 1. Section 1 is about the COMMUNITY (a security lapse, an EV-charging idea, an unhappy experience with the building or the committee). This section is about YOU - the assistant: were you useful, accurate, clear; what can't you do that they wish you could; where did you get something wrong.\n- Decide which one it is from what they are commenting on, NOT from the word \"feedback\". \"The gym timings you gave me were wrong\", \"you're quicker than MyGate\", or \"can you also book the hall for me?\" is feedback about the assistant -> log_agent_feedback. \"The gym should open earlier\" is a community suggestion -> log_suggestion. If it is genuinely both, log both and say so briefly.\n- THE COMMENT IS THE POINT, not the score. A star rating on its own tells the committee nothing they can act on. So when a rating arrives with no detail, ask ONE short, specific follow-up - \"what would make it a 5?\" for a high score, \"what did it get wrong, or what were you hoping it could do?\" for a low one - and then log. If their next message answers it, log with that as the comment. If they decline, change the subject, or clearly aren't interested, log it anyway before moving on, with comment \"(no detail given)\" - never drop the rating silently, and never badger them for a reason.\n- Set the kind field honestly: 'bug' when something was wrong or broken, 'feature_request' when they want it to do something new, 'praise' when it simply worked, 'criticism' for a general negative view that isn't a specific fault. Feature requests and bug reports are the most valuable things you can capture here - treat \"can you also do X?\", \"I wish you could...\", \"why can't you...\" and \"that answer was wrong\" as feedback worth logging, not as small talk to brush past.\n- For a bug, capture what they ASKED and what you ANSWERED in the comment, so the committee can reproduce it. If you can see the exchange earlier in this conversation, summarise it yourself rather than making them retype it.\n- Only pass stars if they actually gave a number, and only pass rating if their view is clear - a plain feature request usually has neither. Never invent a score to fill the field.\n- Their flat is recorded so the committee can follow up. If they ask to stay anonymous, set anonymous true and confirm you've kept it anonymous.\n- Thank them briefly and sincerely. Never argue with criticism, never defend yourself at length, and never promise a fix or a new feature - say the committee will see it.\n- Never ask for feedback in the middle of a complaint, escalation or dues query, and never ask for it twice in one conversation.\n\nAnswer ordinary questions from the documents as usual; only log a question when you genuinely cannot answer it.`;\n}\n\n/* ---------- TOOLS: the resident-facing tool schemas ----------\n Previously duplicated in both app vals (~10.3k each, 114 of 115 lines byte-identical). One copy\n here means one place to edit, and it keeps both app vals clear of Val Town's ~80k limit.\n The copilot's own small tool list stays in the app vals - it is genuinely different per channel. */\nexport function buildTools(TOPICS) {\n return [\n {\n name: \"log_suggestion\",\n description:\n \"Save a resident's suggestion, feedback, or concern the MOMENT you have the topic and the text \\u2014 before asking anything else. Returns a reference number.\",\n input_schema: {\n type: \"object\",\n properties: {\n topic: {\n type: \"string\",\n enum: TOPICS,\n description: \"The single best-fitting topic.\",\n },\n suggestion: {\n type: \"string\",\n description: \"What the resident wants to say, in their own words.\",\n },\n },\n required: [\"topic\", \"suggestion\"],\n },\n },\n {\n name: \"update_suggestion\",\n description:\n \"Add the subcommittee answer, name, and/or email to a suggestion already saved with log_suggestion, using its reference. Call once per add-on as answered.\",\n input_schema: {\n type: \"object\",\n properties: {\n reference: {\n type: \"string\",\n description: \"The reference number returned by log_suggestion.\",\n },\n willing_to_join_subcommittee: {\n type: \"boolean\",\n description:\n \"The resident's Yes/No answer about helping on a subcommittee.\",\n },\n name: {\n type: \"string\",\n description: \"The resident's name, if shared.\",\n },\n email: {\n type: \"string\",\n description: \"The resident's email, if shared.\",\n },\n },\n required: [\"reference\"],\n },\n },\n {\n name: \"log_escalation\",\n description:\n \"Escalate an unresolved maintenance issue to the Managing Committee. Call ONLY after confirming an existing MyGate helpdesk ticket, with its ID and creation date in YYYY-MM-DD. Refused unless the ticket is more than a day old.\",\n input_schema: {\n type: \"object\",\n properties: {\n category: {\n type: \"string\",\n enum: TOPICS,\n description: \"The single best-fitting topic.\",\n },\n issue: {\n type: \"string\",\n description: \"Short description of the unresolved issue.\",\n },\n ticket_id: {\n type: \"string\",\n description: \"The MyGate helpdesk ticket ID / number.\",\n },\n ticket_created_date: {\n type: \"string\",\n description:\n \"Ticket creation date resolved to YYYY-MM-DD using today's date.\",\n },\n name: {\n type: \"string\",\n description:\n \"Optional: the resident's name for the committee's reference.\",\n },\n email: {\n type: \"string\",\n description:\n \"Optional: the resident's email, so the committee can copy them on it and follow up.\",\n },\n },\n required: [\"category\", \"issue\", \"ticket_id\", \"ticket_created_date\"],\n },\n },\n {\n name: \"log_question\",\n description:\n \"Send a resident's question to the Managing Committee. ONLY use this when the answer is genuinely NOT in the documents — always try to answer from the documents first. Email is required so the committee can reply.\",\n input_schema: {\n type: \"object\",\n properties: {\n topic: {\n type: \"string\",\n enum: TOPICS,\n description: \"The single best-fitting topic.\",\n },\n question: {\n type: \"string\",\n description: \"The resident's question, in their own words.\",\n },\n email: {\n type: \"string\",\n description:\n \"The resident's email — REQUIRED so the committee can reply.\",\n },\n name: {\n type: \"string\",\n description: \"Optional: the resident's name.\",\n },\n },\n required: [\"topic\", \"question\", \"email\"],\n },\n },\n {\n name: \"log_complaint\",\n description:\n \"Log a maintenance complaint directly, tracked by the assistant. Mints a reference (AO-C-...) and notifies the committee. Use for a problem the resident wants fixed (leak, broken light, lift, water, etc.).\",\n input_schema: {\n type: \"object\",\n properties: {\n category: {\n type: \"string\",\n enum: TOPICS,\n description: \"The single best-fitting area.\",\n },\n issue: {\n type: \"string\",\n description:\n \"Short description of the problem, in the resident's words.\",\n },\n name: {\n type: \"string\",\n description: \"Optional: the resident's name.\",\n },\n email: {\n type: \"string\",\n description:\n \"REQUIRED: the resident's email, so the committee can follow up and copy them in on the complaint.\",\n },\n },\n required: [\"category\", \"issue\", \"email\"],\n },\n },\n {\n name: \"update_complaint\",\n description:\n \"Add or correct the resident's name and/or email on a complaint ALREADY saved with log_complaint, using its reference. ALWAYS use this for name/email that arrive after logging \\u2014 never call log_complaint again (that creates a duplicate). Providing an email notifies the committee with the resident in copy.\",\n input_schema: {\n type: \"object\",\n properties: {\n reference: {\n type: \"string\",\n description: \"The AO-C-... reference returned by log_complaint.\",\n },\n name: {\n type: \"string\",\n description: \"The resident's name (optional).\",\n },\n email: {\n type: \"string\",\n description: \"The resident's email (optional).\",\n },\n },\n required: [\"reference\"],\n },\n },\n {\n name: \"check_complaint_status\",\n description:\n \"Look up the status the committee has set for the resident's maintenance complaint(s). Pass a reference if the resident gave one (like AO-C-...); otherwise it returns recent complaints for their flat.\",\n input_schema: {\n type: \"object\",\n properties: {\n reference: {\n type: \"string\",\n description:\n \"Optional: a complaint reference the resident provided.\",\n },\n },\n required: [],\n },\n },\n {\n name: \"get_resident_info\",\n description:\n \"Look up the committee's record for THIS resident's own flat (e.g. pending maintenance dues, registered owner name, notes). Uses their flat automatically and only ever returns their own flat's record. Use it when the resident asks about their own account or dues. The record may include extra fields the committee added as columns on the Residents tab; treat them the same way \\u2014 use them if they clearly answer the question, ignore them if their meaning is unclear.\",\n input_schema: { type: \"object\", properties: {}, required: [] },\n },\n {\n name: \"log_recommendation\",\n description:\n \"Save a neighbour's recommendation of anything nearby — a service, shop, place, or vendor (dentist, hospital, hotel, restaurant, hardware shop, plumber, electrician, tutor, cook, etc.) so other residents can benefit. Call once you have at least what kind of thing it is and its name.\",\n input_schema: {\n type: \"object\",\n properties: {\n category: {\n type: \"string\",\n description:\n \"What kind of thing it is, e.g. Dentist, Hospital, Restaurant, Hotel, Hardware shop, Plumber, Tutor.\",\n },\n provider: {\n type: \"string\",\n description: \"Name of the person or business being recommended.\",\n },\n contact: {\n type: \"string\",\n description: \"Phone number or contact detail, if shared.\",\n },\n area: { type: \"string\", description: \"Area or locality, if shared.\" },\n note: {\n type: \"string\",\n description: \"Short note on why it's recommended, if shared.\",\n },\n credit: {\n type: \"boolean\",\n description:\n \"Whether the resident agreed to have their flat shown as the recommender. Set false if they preferred to stay anonymous or you're unsure.\",\n },\n },\n required: [\"category\", \"provider\"],\n },\n },\n {\n name: \"get_recommendations\",\n description:\n \"Find neighbour recommendations for a kind of service the resident is asking about (e.g. 'dentist', 'pediatrician', 'bird netting'). Returns providers other residents have recommended, with who recommended each.\",\n input_schema: {\n type: \"object\",\n properties: {\n query: {\n type: \"string\",\n description:\n \"Short search term for what they want, e.g. 'dentist', 'hardware shop', 'restaurant', 'hotel'.\",\n },\n },\n required: [\"query\"],\n },\n },\n {\n name: \"log_agent_feedback\",\n description:\n \"Record what the resident thinks of THIS ASSISTANT itself — whether it was useful, accurate, easy to use, or what it's missing. This is NOT for suggestions about the community, the building, or the committee's work: those go to log_suggestion. Use this only when the resident is commenting on the assistant/bot.\",\n input_schema: {\n type: \"object\",\n properties: {\n kind: {\n type: \"string\",\n enum: [\"bug\", \"feature_request\", \"praise\", \"criticism\"],\n description:\n \"'bug' = something the assistant got wrong or that isn't working. 'feature_request' = something they wish it could do. 'praise' = it worked well. 'criticism' = a general negative view that isn't a specific bug.\",\n },\n comment: {\n type: \"string\",\n description:\n \"REQUIRED. What they said, in their own words \\u2014 this is what the committee actually acts on. For a bug, include what they asked and what the assistant answered, so it can be reproduced. For a feature request, describe what they want to be able to do. If they gave a rating but genuinely won't elaborate, write '(no detail given)'.\",\n },\n stars: {\n type: \"integer\",\n description:\n \"1-5 if the resident gave a star rating. Omit entirely if they didn't \\u2014 never invent one. A feature request usually has no rating.\",\n },\n rating: {\n type: \"string\",\n enum: [\"up\", \"down\", \"mixed\"],\n description:\n \"Coarse sentiment, only if it is clear and no star rating was given: 'up' positive, 'down' negative, 'mixed' both. Omit for a neutral feature request.\",\n },\n anonymous: {\n type: \"boolean\",\n description:\n \"True if the resident asked to keep it anonymous. Default false \\u2014 their flat is recorded so the committee can follow up.\",\n },\n },\n required: [\"kind\", \"comment\"],\n },\n },\n {\n name: \"get_mc_contacts\",\n description:\n 'Get the committee directory: office-bearer POSTS (President, Secretary, Treasurer, etc.) and work PORTFOLIOS (Finance, Water, etc.), each tagged with a Type, plus the members / points of contact, official emails, and phone numbers. Use when a resident asks who holds a post, who to contact for an area, or how to email the committee. Rows may also carry an \"extra\" object holding columns the committee has added to the sheet themselves (for example a Flat column): use those values naturally when they answer what was asked, and never guess what an unfamiliar column means \\u2014 if its meaning is not obvious from its name, ignore it rather than misdescribe it.',\n input_schema: { type: \"object\", properties: {}, required: [] },\n },\n {\n name: \"get_document\",\n description:\n \"Open one of the association's documents from the DOCUMENT LIBRARY and get its full text, so you can answer a rules or policy question. Pass the exact Title shown in the library. Call this before answering anything grounded in the bye-laws, rule book, parking, party hall, or any listed document.\",\n input_schema: {\n type: \"object\",\n properties: {\n title: {\n type: \"string\",\n description:\n \"The exact document Title from the DOCUMENT LIBRARY list.\",\n },\n },\n required: [\"title\"],\n },\n },\n ];\n}","createdAt",["D",1785909053337],"branchVersion","branchId","aa8a9feb-b453-4de6-a38e-fad95249c5c1","id","5cf5a03d-6202-46e5-9f56-2947582cf7e8","AskAOAOA_Whatsapp",["D",1783664429895],"deletedAt","legacyEnvVars","privacy","public","httpPrivacy","description","Whatsapp chat setup for AskAOAOA","imageUrl","4a18de72-744a-4935-a405-7e9166d2bb8e","pinnedFileId","pinPreview","pinMigrated","screenshotVersion","likeCount",0,"orgId","019f2c5b-c205-7209-90ec-fea27dba2152","tursoName","tursoHostname","tursoMeasuredAt","pinned","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm","user",{"_96":95,"_104":100},"org",{"_96":95,"_97":98,"_99":100,"_18":56,"_101":102,"_103":27},"branch",{"_13":85,"_34":86,"_44":43,"_87":45,"_88":-5,"_89":-5,"_90":-5,"_91":-5,"_92":-5,"_93":46,"_94":95},"isOwned",119,"main","projectId","forkedBranchId","forkedBranchVersion","forkedProjectPrivacy","forkedProjectId","forkedProjectOrgId","projectName","userHandle","sanair1992","handle","displayName","Shyam A Nair","avatar","https://lh3.googleusercontent.com/a/ACg8ocL6k5_PJu7Okt0TDOcmJDDE1rjDZi65QoH9WtDAlAU6jJixUNUydg=s96-c","tier","free","canRestrictVals","profileImageUrl","branchName","","adoptOrgId",{"_44":45,"_34":46,"_39":179,"_48":-5,"_49":27,"_50":51,"_52":51,"_53":54,"_55":-5,"_18":56,"_57":-5,"_58":27,"_59":-5,"_60":-5,"_61":62,"_63":64,"_65":-5,"_66":-5,"_67":-5,"_68":-5,"_69":70,"_71":70,"_72":73,"_74":73,"_75":73,"_76":70,"_77":70,"_78":180,"_80":181,"_82":182,"_84":27},"tree",[166,167,168],"files",[166,167,168],"counts",{"_154":62,"_155":156,"_157":62,"_158":159,"_160":161},"contents",[127,128,129],"upstreamProject","upstreamBranch","emptyEnvVars","flags",[],"hasHttpFiles","grants","orgs","customDomains",{},{"_44":146,"_34":147,"_32":148,"_133":-5,"_42":43,"_41":149,"_135":27,"_39":150,"_13":151,"_35":36,"_37":152,"_24":27,"_140":-5,"_28":153},{"_44":142,"_34":29,"_32":33,"_133":-5,"_42":43,"_41":14,"_135":27,"_39":143,"_13":144,"_35":36,"_37":38,"_24":27,"_140":-5,"_28":145},{"_44":130,"_34":131,"_32":132,"_133":-5,"_42":43,"_41":134,"_135":27,"_39":136,"_13":137,"_35":138,"_37":139,"_24":27,"_140":-5,"_28":141},"17bba8da-ad63-4d94-a168-40fff53140f5","askaoaoa_whatsapp.ts","72f2fbbc-7c27-11f1-9141-1607ee4eb77e","parentId",68,"latest",["D",1785908368588],44,"http","// AskAOAOA — WhatsApp webhook (Meta Cloud API). Deploy as a Val Town HTTP val.\n// It reuses the SAME knowledge (data val), tools, and Google-Sheet logging as the web app.\n// The web page becomes WhatsApp: topic list -> interactive list, Yes/No -> buttons,\n// screenshots -> the resident just sends a photo in the chat.\n//\n// ENV VARS to set on this val:\n// WHATSAPP_TOKEN - permanent access token from Meta (System User token)\n// WHATSAPP_PHONE_ID - the Phone Number ID from the WhatsApp > API Setup screen\n// VERIFY_TOKEN - any secret string you make up; paste the SAME value into Meta's webhook \"Verify token\"\n// ANTHROPIC_API_KEY - same key as the web val\n// DATA_URL - same UNPINNED module URL of your data val (holds the documents)\n// SHEET_WEBHOOK_URL - same Apps Script webhook as the web val\n// SHEET_SECRET - same as the Apps Script SHARED_SECRET (leave empty if that is empty)\n// GRAPH_VERSION - optional, defaults to v21.0\n// MC_NUMBERS - optional, comma-separated phone numbers (digits, no +) of Managing Committee members;\n// they can text \"copilot\" to enter a private committee analytics/drafting mode\n// OPS_NUMBERS - optional, comma-separated phone numbers (digits, no +, e.g. 9199...) allowed to\n// post/clear the live site status by texting \"status: ...\" or \"clear status\" (security/FMS/MC)\n\nimport { blob } from \"https://esm.town/v/std/blob\";\n\nconst TOKEN = Deno.env.get(\"WHATSAPP_TOKEN\") || \"\";\nconst PHONE_ID = Deno.env.get(\"WHATSAPP_PHONE_ID\") || \"\";\nconst VERIFY_TOKEN = Deno.env.get(\"VERIFY_TOKEN\") || \"\";\nconst GV = Deno.env.get(\"GRAPH_VERSION\") || \"v21.0\";\n\n/* ---------- knowledge (imported from the data val, cached) ---------- */\nlet SYSTEM_CACHE = null;\nasync function getSystem() {\n if (SYSTEM_CACHE != null) return SYSTEM_CACHE;\n const u = Deno.env.get(\"DATA_URL\");\n if (!u) throw new Error(\"DATA_URL not set\");\n const mod = await import(u);\n SYSTEM_CACHE = mod.SYSTEM || \"\";\n return SYSTEM_CACHE;\n}\n\n// ---- per-community config (the config val). Set CONFIG_URL to its ES-module URL. ----\nconst CONFIG_URL = Deno.env.get(\"CONFIG_URL\");\nif (!CONFIG_URL) throw new Error(\"CONFIG_URL not set\");\nconst _cfg = await import(CONFIG_URL);\nconst CONFIG = _cfg.CONFIG;\nconst WA_PREFIX = \"aoaoa_\" + CONFIG.id + \"_wa_\"; // per-community blob namespace\nconst NAME = CONFIG.name;\nconst BLOCKS = (CONFIG.flats.blocks || []).map(function (b) {\n return b.toUpperCase();\n});\nconst BLOCKLIST = BLOCKS.length > 1\n ? BLOCKS.slice(0, -1).join(\", \") + \" or \" + BLOCKS.slice(-1)\n : (BLOCKS[0] || \"\");\nconst TOPICS = CONFIG.topics;\nconst DISCLAIMER = CONFIG.disclaimer;\nconst isValidFlat = (raw) => _cfg.isValidFlat(raw, CONFIG);\nconst normalizeFlat = (raw) => _cfg.normalizeFlat(raw, CONFIG);\n\n// Short row labels for the WhatsApp list (row title max 24 chars); id carries the full topic.\nconst TOPIC_ROWS = [\n { id: \"Finance\", title: \"Finance\" },\n { id: \"Admin & Communication\", title: \"Admin & Comms\" },\n { id: \"Security/Parking\", title: \"Security / Parking\" },\n {\n id: \"MEP (Mechanical, Electrical, Plumbing)\",\n title: \"MEP\",\n description: \"Mechanical, Electrical, Plumbing\",\n },\n { id: \"Gardening\", title: \"Gardening\" },\n { id: \"Housekeeping\", title: \"Housekeeping\" },\n {\n id: \"Tech/IT (including MyGate)\",\n title: \"Tech / IT\",\n description: \"Including MyGate\",\n },\n { id: \"Water Management\", title: \"Water Management\" },\n { id: \"Clubhouse Management\", title: \"Clubhouse Mgmt\" },\n { id: \"Others\", title: \"Others\" },\n];\n\nlet MANIFEST_CACHE = {}, MANIFEST_TS = {};\nasync function getManifest(mc) {\n const key = mc ? \"mc\" : \"all\";\n const now = Date.now();\n if (MANIFEST_CACHE[key] != null && (now - (MANIFEST_TS[key] || 0)) < 300000) {\n return MANIFEST_CACHE[key];\n }\n let txt = \"\";\n try {\n const r = await sheetGet(\"q=docs\" + (mc ? \"&scope=mc\" : \"\"));\n if (r && r.ok && Array.isArray(r.docs) && r.docs.length) {\n txt =\n \"DOCUMENT LIBRARY (open the relevant one with get_document before answering):\\n\" +\n r.docs.map(function (x) {\n return \"- \" + x.title + (x.description ? \": \" + x.description : \"\") +\n (x.audience === \"mc\" ? \" [committee-only]\" : \"\");\n }).join(\"\\n\");\n }\n } catch (e) {}\n MANIFEST_CACHE[key] = txt;\n MANIFEST_TS[key] = now;\n return txt;\n}\n\nlet TOPICS_CACHE = null, TOPICS_TS = 0;\nasync function getTopics() {\n const now = Date.now();\n if (TOPICS_CACHE !== null && (now - TOPICS_TS) < 300000) return TOPICS_CACHE;\n let list = null;\n try {\n const r = await sheetGet(\"q=topics\");\n if (r && r.ok && Array.isArray(r.topics) && r.topics.length) {\n list = r.topics;\n }\n } catch (e) {}\n TOPICS_CACHE = (list && list.length) ? list : CONFIG.topics; // fallback to config if the sheet has none\n TOPICS_TS = now;\n return TOPICS_CACHE;\n}\n\n// Tool schemas now live in the data val (one shared copy). Fetched and cached like ACTIONS.\nlet TOOLS_CACHE = null;\nasync function getTools(TOPICS) {\n if (TOOLS_CACHE) return TOOLS_CACHE;\n const u = Deno.env.get(\"DATA_URL\");\n if (!u) throw new Error(\"DATA_URL not set\");\n const mod = await import(u);\n if (typeof mod.buildTools !== \"function\") {\n console.log(\n \"FATAL: data val at DATA_URL has no buildTools export - update DATA_URL to the latest data val module URL.\",\n );\n throw new Error(\n \"Data val is out of date: no buildTools export. Update DATA_URL.\",\n );\n }\n TOOLS_CACHE = mod.buildTools(TOPICS);\n return TOOLS_CACHE;\n}\n\nconst RELEASES_TEXT = \"*What's new in \" + NAME +\n \" \\u2014 v4.1*\\n\\n\\u2022 Ask who holds each committee post \\u2014 President, Secretary, Treasurer and more \\u2014 with their name and number.\\n\\u2022 Ask who handles a particular area (water, security, housekeeping\\u2026) and reach the right point of contact.\\n\\u2022 Get the committee's official email to write in directly.\\n\\u2022 Answers stay in clear, everyday language \\u2014 no clause numbers.\";\n\n// WhatsApp-tailored action guide. Same behaviour as the web app; only the date and\n// screenshot mechanics differ (no calendar / no upload button on WhatsApp).\n// ACTIONS now lives in the data val (one shared copy). Fetched and cached by getActions().\nlet ACTIONS_CACHE = null;\nasync function getActions() {\n if (ACTIONS_CACHE != null) return ACTIONS_CACHE;\n const u = Deno.env.get(\"DATA_URL\");\n if (!u) throw new Error(\"DATA_URL not set\");\n const mod = await import(u);\n if (typeof mod.buildActions !== \"function\") {\n // Deterministic misconfiguration, not a transient blip: the data val loaded but is an OLD copy\n // without buildActions (usually DATA_URL pinned to an earlier version). Fail loudly - silently\n // serving an empty ACTIONS would strip every action instruction while still looking healthy.\n console.log(\n \"FATAL: data val at DATA_URL has no buildActions export - update DATA_URL to the latest data val module URL.\",\n );\n throw new Error(\n \"Data val is out of date: no buildActions export. Update DATA_URL.\",\n );\n }\n ACTIONS_CACHE = mod.buildActions(\"whatsapp\");\n // Fingerprint the loaded prompt so you can tell from the logs whether the data val you just\n // saved is actually the one being served (a pinned DATA_URL fails silently otherwise).\n const _sys = String(mod.SYSTEM || \"\");\n console.log(\n \"prompt loaded: actions=\" + ACTIONS_CACHE.length + \" chars, system=\" +\n _sys.length +\n \", grounding_rule=\" + (_sys.indexOf(\"NEVER STATE A PERSON\") >= 0\n ? \"yes\"\n : \"NO - DATA VAL IS STALE\"),\n );\n return ACTIONS_CACHE;\n}\n\nconst WEBHOOK = Deno.env.get(\"SHEET_WEBHOOK_URL\") || \"\";\nconst SHEET_SECRET = Deno.env.get(\"SHEET_SECRET\") || \"\";\n\n/* ---------- flat validation (same rules as the web app) ---------- */\n\n/* ---------- IST date helpers (same as the web app) ---------- */\nfunction istNow() {\n return new Date(Date.now() + 5.5 * 3600 * 1000);\n}\nfunction istTodayISO() {\n const d = istNow();\n return d.getUTCFullYear() + \"-\" +\n String(d.getUTCMonth() + 1).padStart(2, \"0\") + \"-\" +\n String(d.getUTCDate()).padStart(2, \"0\");\n}\nfunction daysSince(dateStr) {\n const m = String(dateStr || \"\").match(/^(\\d{4})-(\\d{2})-(\\d{2})$/);\n if (!m) return null;\n const created = Date.UTC(+m[1], +m[2] - 1, +m[3]);\n if (isNaN(created)) return null;\n const d = istNow();\n const today = Date.UTC(d.getUTCFullYear(), d.getUTCMonth(), d.getUTCDate());\n return Math.floor((today - created) / 86400000);\n}\n\n/* ---------- Google-Sheet logging (same webhook as the web app) ---------- */\nasync function postSheet(payload) {\n if (!WEBHOOK) {\n return {\n ok: false,\n note: \"Logging is not set up yet (no SHEET_WEBHOOK_URL).\",\n };\n }\n const ctl = new AbortController();\n const timer = setTimeout(() => ctl.abort(), 15000); // writes get longer than reads, but not forever\n try {\n if (SHEET_SECRET) payload.secret = SHEET_SECRET;\n const r = await fetch(WEBHOOK, {\n method: \"POST\",\n headers: { \"content-type\": \"application/json\" },\n body: JSON.stringify(payload),\n signal: ctl.signal,\n });\n const d = await r.json();\n return (d && d.ok)\n ? { ok: true, reference: d.reference, url: d.url }\n : { ok: false, note: \"The logging service reported an error.\" };\n } catch (e) {\n if (e && e.name === \"AbortError\") {\n console.log(\"postSheet timed out: \" + (payload && payload.type));\n }\n return { ok: false, note: \"Could not reach the logging service.\" };\n } finally {\n clearTimeout(timer);\n }\n}\n\nlet STATUS_CACHE = { t: 0, v: \"\" };\n/* Every Apps Script call is time-boxed. Without this, one slow doGet blocks the whole answer: a\n 45-second preflight was observed in production, which is past the platform request timeout, so the\n isolate was killed and the resident got complete silence. A degraded answer beats no answer. */\nasync function sheetGet(query, timeoutMs) {\n if (!WEBHOOK) {\n return {\n ok: false,\n note: \"Lookups are not set up yet (no SHEET_WEBHOOK_URL).\",\n };\n }\n const ctl = new AbortController();\n const timer = setTimeout(() => ctl.abort(), timeoutMs || 8000);\n try {\n const url = WEBHOOK + \"?\" + query +\n (SHEET_SECRET ? \"&secret=\" + encodeURIComponent(SHEET_SECRET) : \"\");\n const r = await fetch(url, { redirect: \"follow\", signal: ctl.signal });\n return await r.json();\n } catch (e) {\n if (e && e.name === \"AbortError\") {\n console.log(\"sheetGet timed out: \" + String(query).slice(0, 60));\n }\n return { ok: false, note: \"Could not reach the lookup service.\" };\n } finally {\n clearTimeout(timer);\n }\n}\n/* One Apps Script call instead of three. Val Town gives us a COLD isolate on essentially every\n message (the \"prompt loaded\" line appears on each request), so the in-process caches below almost\n never survive to be reused - which made the old three-call preflight cost three cold Apps Script\n executions per resident message. If the endpoint is missing (Apps Script not yet redeployed) this\n falls back to the individual calls, so it is safe to deploy the vals first. */\nasync function preflight(mc) {\n const d = await sheetGet(\"q=preflight\" + (mc ? \"&scope=mc\" : \"\"), 8000);\n if (!d || !d.ok || !Array.isArray(d.topics)) return false;\n const now = Date.now();\n let v = \"\";\n if (Array.isArray(d.active) && d.active.length) {\n v = d.active.map((a) => (a && a.message != null) ? a.message : a).filter(\n Boolean,\n ).join(\" | \");\n }\n STATUS_CACHE = { t: now, v: v };\n if (d.topics.length) {\n TOPICS_CACHE = d.topics;\n TOPICS_TS = now;\n }\n const key = mc ? \"mc\" : \"all\";\n let txt = \"\";\n if (Array.isArray(d.docs) && d.docs.length) {\n txt =\n \"DOCUMENT LIBRARY (open the relevant one with get_document before answering):\\n\" +\n d.docs.map(function (x) {\n return \"- \" + x.title + (x.description ? \": \" + x.description : \"\") +\n (x.audience === \"mc\" ? \" [committee-only]\" : \"\");\n }).join(\"\\n\");\n }\n MANIFEST_CACHE[key] = txt;\n MANIFEST_TS[key] = now;\n return true;\n}\nasync function getStatus() {\n const now = Date.now();\n if (now - STATUS_CACHE.t < 60000) return STATUS_CACHE.v;\n let v = \"\";\n try {\n const d = await sheetGet(\"q=status\");\n if (d && d.ok && Array.isArray(d.active) && d.active.length) {\n v = d.active.map((a) => (a && a.message != null) ? a.message : a).filter(\n Boolean,\n ).join(\" | \");\n }\n } catch (e) {}\n STATUS_CACHE = { t: now, v: v };\n return v;\n}\n\nconst SYSTEM_COPILOT = \"You are the \" + NAME +\n \" Committee Copilot for the Managing Committee (MC) of \" +\n CONFIG.estateShort + \" Apartments, a \" + CONFIG.unitCount +\n \"-flat community. You are given recent community data as JSON (complaints with status, suggestions, questions, escalations, site-status posts, neighbour recommendations, and agentFeedback \\u2014 residents' ratings of the assistant itself, rated up/down/mixed with their comment and channel), newest first. Help the committee: summarise and count, spot recurring themes, list open/pending items and prioritise them, report how residents rate the assistant itself (satisfaction trend, common gaps) when asked, and draft clear courteous communications (notices, replies to residents, AGM agenda items). Base everything strictly on the data provided; never invent numbers, references, names, or records; if it is not in the data, say so plainly. Complaint status is in the 'Status' field (Open / In Progress / Resolved); complaints look like AO-C-... Cite references and flats where it helps the committee act. Dates are IST; interpret 'this week' / 'this month' relative to the date given. Be concise and action-oriented; use short bullet lists for summaries and a warm, professional tone for drafts. This data is confidential to the committee. You are advisory \\u2014 you do not send messages or change any records. You can also open any association document from the DOCUMENT LIBRARY with the get_document tool \\u2014 including committee-only data such as boom-barrier/vehicle logs or security attendance \\u2014 to analyze it: summarise, count, find patterns or outliers (for example, which security staff were least punctual last quarter), and cite exact figures. For a rules question you MAY quote exact clause or section numbers to the committee (residents never receive clause numbers).\";\n\nasync function copilotAnswer(KEY, history, sys) {\n const dd = await sheetGet(\"q=export\");\n const dataStr = (dd && dd.ok && dd.data)\n ? JSON.stringify(dd.data)\n : \"(no data available)\";\n const manifest = await getManifest(true); // committee scope: includes committee-only documents\n const system = [\n {\n type: \"text\",\n text: SYSTEM_COPILOT,\n cache_control: { type: \"ephemeral\" },\n },\n ];\n if (sys) {\n system.push({\n type: \"text\",\n text: \"REFERENCE \\u2014 the association's rules/instructions:\\n\\n\" + sys,\n });\n }\n if (manifest) system.push({ type: \"text\", text: manifest });\n system.push({\n type: \"text\",\n text: \"Today's date: \" + istTodayISO() +\n \" (IST).\\n\\nCOMMUNITY DATA (JSON, newest first):\\n\" + dataStr,\n });\n const tools = [{\n name: \"get_document\",\n description:\n \"Open any association document from the DOCUMENT LIBRARY \\u2014 including committee-only ones (e.g. boom-barrier/vehicle logs, security attendance) \\u2014 and get its full text or data to analyze. Pass the exact Title from the library.\",\n input_schema: {\n type: \"object\",\n properties: {\n title: {\n type: \"string\",\n description: \"The exact Title from the DOCUMENT LIBRARY.\",\n },\n },\n required: [\"title\"],\n },\n }];\n let msgs = history.slice(-12).map((m) => ({\n role: m && m.role === \"assistant\" ? \"assistant\" : \"user\",\n content: String((m && m.content) || \"\").slice(0, 6000),\n }));\n try {\n for (let step = 0; step < 6; step++) {\n const r = await fetch(\"https://api.anthropic.com/v1/messages\", {\n method: \"POST\",\n headers: {\n \"content-type\": \"application/json\",\n \"x-api-key\": KEY,\n \"anthropic-version\": \"2023-06-01\",\n },\n body: JSON.stringify({\n model: CONFIG.models.copilot,\n max_tokens: 2000,\n temperature: 0,\n system: system,\n tools: tools,\n messages: msgs,\n }),\n });\n const cd = await r.json();\n if (r.status !== 200) {\n return \"The copilot is having trouble right now. Please try again.\";\n }\n if (cd.stop_reason === \"tool_use\") {\n const results = [];\n for (const b of (cd.content || [])) {\n if (b.type === \"tool_use\" && b.name === \"get_document\") {\n const res = await sheetGet(\n \"q=doc&scope=mc&title=\" +\n encodeURIComponent((b.input && b.input.title) || \"\"),\n );\n results.push({\n type: \"tool_result\",\n tool_use_id: b.id,\n content: JSON.stringify(res).slice(0, 120000),\n });\n } else if (b.type === \"tool_use\") {\n results.push({\n type: \"tool_result\",\n tool_use_id: b.id,\n content: \"(unavailable)\",\n });\n }\n }\n msgs.push({ role: \"assistant\", content: cd.content });\n msgs.push({ role: \"user\", content: results });\n continue;\n }\n return (cd.content || []).map((c) => c.type === \"text\" ? c.text : \"\")\n .filter(Boolean).join(\"\\n\").trim() ||\n \"I couldn't produce an answer for that.\";\n }\n return \"That needed too many steps \\u2014 please narrow the question.\";\n } catch (e) {\n return \"The copilot couldn't reach the service just now.\";\n }\n}\n\nfunction complaintRef(t) {\n const m = String(t || \"\").match(/AO-C-\\d{6}-\\d{4}/i);\n return m ? m[0].toUpperCase() : \"\";\n}\nfunction isComplaintStatus(t) {\n const s = String(t || \"\").trim().toLowerCase().replace(/[?.!]+$/, \"\");\n if (complaintRef(s)) return true;\n return [\n \"my complaints\",\n \"my complaint\",\n \"complaint status\",\n \"complaints status\",\n \"check my complaints\",\n \"check my complaint\",\n \"check complaint status\",\n \"status of my complaint\",\n \"status of my complaints\",\n \"track my complaint\",\n \"track my complaints\",\n \"complaint update\",\n \"complaint updates\",\n ].includes(s);\n}\nfunction formatComplaintsWA(d, ref) {\n if (!d || !d.ok) {\n return \"I couldn't check the complaint status just now. Please try again in a moment.\";\n }\n const list = d.complaints || [];\n if (!list.length) {\n return ref\n ? \"I couldn't find a complaint with that reference for your flat. Please double-check it (it looks like AO-C-...), or describe the issue and I'll log a new complaint.\"\n : \"You don't have any complaints logged yet. If something needs fixing, just describe it and I'll raise a complaint for you.\";\n }\n const lines = list.map(function (c) {\n const st = c.status || \"Open\";\n const emoji = /resolv|closed|done/i.test(st)\n ? \"\\u2705\"\n : /progress|working|assign/i.test(st)\n ? \"\\uD83D\\uDD27\"\n : \"\\uD83D\\uDD5B\";\n return emoji + \" *\" + (c.reference || \"\") + \"* \\u2014 \" +\n (c.category || \"\") + \"\\n_\" + (c.issue || \"\") + \"_\\nStatus: *\" + st + \"*\";\n });\n return \"Here's the latest on your complaint\" + (list.length > 1 ? \"s\" : \"\") +\n \":\\n\\n\" + lines.join(\"\\n\\n\");\n}\n\n// Remove bye-law/clause/section NUMBERS from resident-facing replies (keep the plain answer).\nfunction stripCites(t) {\n if (!t) return t;\n let s = String(t);\n // parenthetical citations that contain a clause/section number, e.g. *(Bye-Laws, Clause 3.12)*\n s = s.replace(\n /\\*?\\s*\$(?:[^()]*?)(?:clause|section|article|rule|bye-?law|point|item|para(?:graph)?)\\s*\\d+[a-z]?(?:\\.\\d+)*(?:[^()]*?)\$\\s*\\*?/gi,\n \"\",\n );\n // inline references, e.g. \"Clause 43.9\", \"Section 14\", \"Bye-Law 11.3\"\n s = s.replace(\n /\\b(?:bye-?laws?[,\\s]*)?(?:clause|section|article|bye-?law)s?\\s*\\d+[a-z]?(?:\\.\\d+)*(?:\\s*(?:,|and|&)\\s*\\d+[a-z]?(?:\\.\\d+)*)*/gi,\n \"the association's rules\",\n );\n // tidy residue\n s = s.replace(/\$\\s*\$/g, \"\");\n s = s.replace(/[ \\t]{2,}/g, \" \");\n s = s.replace(/\\s+([.,;:])/g, \"$1\");\n s = s.replace(\n /the association's rules(?:\\s*(?:,|and|&)\\s*the association's rules)+/gi,\n \"the association's rules\",\n );\n return s.trim();\n}\n\nasync function runTool(name, input, flat) {\n input = input || {};\n if (name === \"log_suggestion\") {\n return await postSheet({\n type: \"suggestion\",\n topic: input.topic || \"\",\n suggestion: input.suggestion || \"\",\n flat: flat || \"\",\n });\n }\n if (name === \"update_suggestion\") {\n const payload = {\n type: \"suggestion_update\",\n reference: input.reference || \"\",\n };\n if (typeof input.willing_to_join_subcommittee === \"boolean\") {\n payload.joinSubcommittee = input.willing_to_join_subcommittee;\n }\n if (input.name) payload.name = input.name;\n if (input.email) payload.email = input.email;\n return await postSheet(payload);\n }\n if (name === \"log_escalation\") {\n const days = daysSince(input.ticket_created_date);\n if (days === null) {\n return {\n ok: false,\n note:\n \"The ticket date was not understood. Ask the resident to type it clearly (e.g. 3 July 2025).\",\n };\n }\n if (days < 0) {\n return {\n ok: false,\n note:\n \"That ticket date is in the future. Ask the resident to re-check and give the correct date.\",\n };\n }\n if (days < 1) {\n return {\n ok: false,\n note:\n \"The MyGate ticket was created today, less than a day ago. Do NOT escalate yet. Kindly ask the resident to wait a little longer so the facility team (FMS) can act, and to escalate once it is more than a day old. If they'd like to nudge it sooner, they can call the FMS manager, \" +\n CONFIG.fmsContact + \".\",\n };\n }\n return await postSheet({\n type: \"escalation\",\n name: input.name || \"\",\n flat: flat || \"\",\n category: input.category || \"\",\n issue: input.issue || \"\",\n ticketNo: input.ticket_id || \"\",\n ticketDate: input.ticket_created_date || \"\",\n daysOpen: days,\n email: input.email || \"\",\n });\n }\n if (name === \"log_question\") {\n if (!input.email) {\n return {\n ok: false,\n note:\n \"An email is required so the committee can reply. Please ask the resident for an email before logging the question.\",\n };\n }\n return await postSheet({\n type: \"question\",\n topic: input.topic || \"\",\n question: input.question || \"\",\n email: input.email || \"\",\n name: input.name || \"\",\n flat: flat || \"\",\n });\n }\n if (name === \"log_complaint\") {\n if (!input.email) {\n return {\n ok: false,\n note:\n \"An email is required so the committee can follow up and copy the resident in. Please ask the resident for an email before logging the complaint.\",\n };\n }\n return await postSheet({\n type: \"complaint\",\n category: input.category || \"\",\n issue: input.issue || \"\",\n name: input.name || \"\",\n email: input.email || \"\",\n flat: flat || \"\",\n });\n }\n if (name === \"update_complaint\") {\n return await postSheet({\n type: \"complaint_update\",\n reference: input.reference || \"\",\n name: input.name || \"\",\n email: input.email || \"\",\n });\n }\n if (name === \"check_complaint_status\") {\n let query = \"q=complaint&flat=\" + encodeURIComponent(flat || \"\");\n if (input.reference) {\n query += \"&ref=\" + encodeURIComponent(String(input.reference).trim());\n }\n return await sheetGet(query);\n }\n if (name === \"get_resident_info\") {\n if (!flat) {\n return {\n ok: false,\n note: \"The resident's flat isn't known yet; ask for their flat first.\",\n };\n }\n return await sheetGet(\"q=resident&flat=\" + encodeURIComponent(flat));\n }\n if (name === \"log_recommendation\") {\n return await postSheet({\n type: \"recommendation\",\n category: input.category || \"\",\n provider: input.provider || \"\",\n contact: input.contact || \"\",\n area: input.area || \"\",\n note: input.note || \"\",\n credit: !!input.credit,\n flat: flat || \"\",\n });\n }\n if (name === \"get_recommendations\") {\n return await sheetGet(\n \"q=recommend&query=\" + encodeURIComponent(input.query || \"\"),\n );\n }\n if (name === \"log_agent_feedback\") {\n return await postSheet({\n type: \"agent_feedback\",\n kind: input.kind || \"\",\n stars: (input.stars >= 1 && input.stars <= 5) ? input.stars : \"\",\n rating: input.rating || \"\",\n comment: input.comment || \"\",\n flat: input.anonymous ? \"\" : (flat || \"\"),\n channel: \"WhatsApp\",\n });\n }\n if (name === \"get_mc_contacts\") {\n return await sheetGet(\"q=mc\");\n }\n if (name === \"get_document\") {\n return await sheetGet(\n \"q=doc&title=\" + encodeURIComponent(input.title || \"\"),\n );\n }\n return { ok: false, note: \"Unknown action.\" };\n}\n\n/* ---------- Claude (same model + tool loop as the web app) ---------- */\nasync function callClaude(KEY, systemBlocks, messages, tools) {\n const r = await fetch(\"https://api.anthropic.com/v1/messages\", {\n method: \"POST\",\n headers: {\n \"content-type\": \"application/json\",\n \"x-api-key\": KEY,\n \"anthropic-version\": \"2023-06-01\",\n },\n body: JSON.stringify({\n model: CONFIG.models.resident,\n max_tokens: 1000,\n temperature: 0,\n system: systemBlocks,\n tools: tools || (await getTools(await getTopics())),\n messages,\n }),\n });\n return { status: r.status, data: await r.json() };\n}\n\n// Runs the full tool loop for one resident turn and returns the assistant's final text.\nasync function answer(KEY, SYSTEM, history, flat, lang) {\n const messages = history.slice(-14).map((m) => ({\n role: m && m.role === \"assistant\" ? \"assistant\" : \"user\",\n content: String((m && m.content) || \"\").slice(0, 4000),\n }));\n\n const priorRefs = [];\n for (const m of messages) {\n if (m.role !== \"assistant\") continue;\n const found = m.content.match(/AO-(?:C-)?\\d{6}-\\d{4}/g);\n if (found) {\n for (const r of found) if (!priorRefs.includes(r)) priorRefs.push(r);\n }\n }\n const refLine = priorRefs.length\n ? \" NOTE: This conversation has already saved record(s) with reference \" +\n priorRefs.join(\", \") +\n \". These are already logged. Do NOT call any log_ tool again for an item that already has a reference \\u2014 that creates a duplicate. To add a name, email, or other detail, use the matching UPDATE tool: update_suggestion for a suggestion reference, or update_complaint for an AO-C-... complaint reference. Only log a brand-new, clearly different item the resident raises now.\"\n : \"\";\n\n let langLine;\n if (lang === \"hi\") {\n langLine =\n \" Reply to the resident in Hindi (Devanagari script). Keep any line that starts with 'VERDICT:' and its label in English exactly, and keep any [[...]] tokens exactly as written.\";\n } else if (lang === \"kn\") {\n langLine =\n \" Reply to the resident in Kannada. Keep any line that starts with 'VERDICT:' and its label in English exactly, and keep any [[...]] tokens exactly as written.\";\n } else if (lang === \"en\") {\n langLine =\n \" Reply to the resident in English. Keep any [[...]] tokens exactly as written.\";\n } else {langLine =\n \" The resident may write in English, Hindi, or Kannada. Reply in the same language they used. Keep any line that starts with 'VERDICT:' and its label in English exactly, and keep any [[...]] tokens exactly as written.\";}\n\n // These three are independent lookups (status + doc manifest hit Apps Script, actions imports the\n // data val). Run them together: sequentially they added several seconds to EVERY reply, and on a\n // cold isolate that is the difference between a quick answer and Meta timing out and retrying.\n const _t0 = Date.now();\n const [, actionsText] = await Promise.all([preflight(false), getActions()]);\n // these now read the caches preflight just filled; they only hit the network if it failed\n const [siteStatus, manifest] = await Promise.all([\n getStatus(),\n getManifest(),\n ]);\n console.log(\"preflight \" + (Date.now() - _t0) + \"ms\");\n const statusLine = siteStatus\n ? \" CURRENT SITE STATUS (live alerts posted by the committee/security): \" +\n siteStatus +\n \". If the resident asks about power, water, lifts, or what's happening on site, tell them this.\"\n : \" CURRENT SITE STATUS: No site alerts are posted right now. If a resident reports a problem, do not claim everything is working \\u2014 suggest they contact security/FMS or log a complaint.\";\n const systemBlocks = [\n {\n type: \"text\",\n text: SYSTEM + actionsText + (manifest ? \"\\n\\n\" + manifest : \"\"),\n cache_control: { type: \"ephemeral\" },\n },\n {\n type: \"text\",\n text: \"Today's date is \" + istTodayISO() +\n \" (YYYY-MM-DD), India Standard Time. \" +\n (flat\n ? \"The resident's flat is \" + flat + \".\"\n : \"You do not yet know the resident's flat.\") +\n langLine + refLine + statusLine,\n },\n ];\n\n let msgs = messages;\n const _tools = await getTools(await getTopics());\n for (let step = 0; step < 6; step++) {\n const _ts = Date.now();\n const { status, data } = await callClaude(KEY, systemBlocks, msgs, _tools);\n console.log(\n \"model turn \" + (step + 1) + \": \" + (Date.now() - _ts) + \"ms status \" +\n status,\n );\n if (status !== 200) {\n console.log(\n \"anthropic error\",\n status,\n JSON.stringify(data && data.error ? data.error : data).slice(0, 400),\n );\n return \"I'm having trouble reaching the assistant right now. Please try again in a moment.\";\n }\n if (data.stop_reason === \"tool_use\") {\n msgs = msgs.concat([{ role: \"assistant\", content: data.content }]);\n const results = [];\n for (const block of data.content) {\n if (block.type === \"tool_use\") {\n const _tt = Date.now();\n const result = await runTool(block.name, block.input, flat);\n console.log(\"tool \" + block.name + \": \" + (Date.now() - _tt) + \"ms\");\n results.push({\n type: \"tool_result\",\n tool_use_id: block.id,\n content: JSON.stringify(result),\n });\n }\n }\n msgs = msgs.concat([{ role: \"user\", content: results }]);\n continue;\n }\n return stripCites(\n (data.content || []).map((c) => c.type === \"text\" ? c.text : \"\").filter(\n Boolean,\n ).join(\"\\n\").trim(),\n ) ||\n \"I couldn't find an answer to that. Could you rephrase?\";\n }\n return \"That took a few too many steps \\u2014 could you try asking again?\";\n}\n\n/* ---------- format Claude's reply for WhatsApp ---------- */\n/* ---------- committee-conduct detector ----------\n The verification note is a safety feature, so it must not depend on the model remembering to emit\n [[VERIFY]]. This is the deterministic backstop: if the RESIDENT's question is about what the\n committee or an office-bearer may, must, or cannot do, the note is attached regardless of what\n came back from the model. Over-attaching costs one quiet italic line; under-attaching sends an\n unqualified ruling about named people out into a WhatsApp group. */\nconst COMMITTEE_RX =\n /\\b(committee|managing committee|\\bmc\\b|office[- ]?bearer|president|secretary|treasurer|association)\\b/i;\nconst CONDUCT_RX =\n /\\b(can|cannot|can't|could|allowed|permitted|entitled|empower\\w*|authoris\\w*|authoriz\\w*|legal|illegal|valid|right|rights|power|powers|must|should|supposed to|oblig\\w*|breach\\w*|violat\\w*|wrong\\w*|unfair\\w*|arbitrar\\w*|overcharg\\w*|misus\\w*|answerable|accountab\\w*|unilateral\\w*|whim|sudden\\w*)\\b/i;\nfunction isCommitteeConductQuestion(t) {\n const s = String(t || \"\");\n if (!s.trim()) return false;\n return COMMITTEE_RX.test(s) && CONDUCT_RX.test(s);\n}\n\nfunction formatForWhatsApp(raw, askedText) {\n let s = String(raw || \"\");\n const wantsTopic = /\\[\\[PICK_TOPIC\\]\\]/.test(s);\n const wantsYesNo = /\\[\\[PICK_YESNO\\]\\]/.test(s);\n const attachM = s.match(/\\[\\[ATTACH:([^\\]]*)\\]\\]/);\n const attachRef = attachM ? attachM[1].trim() : null;\n s = s.replace(/\\[\\[PICK_(DATE|TOPIC|YESNO)\\]\\]/g, \"\").replace(\n /\\[\\[ATTACH:[^\\]]*\\]\\]/g,\n \"\",\n ).trim();\n // [[VERIFY]] marks an answer about committee conduct: no verdict stamp, but it still must not\n // travel without the \"this is not a ruling\" note.\n let wantsVerify = false;\n if (/\\[\\[VERIFY\\]\\]/i.test(s)) {\n wantsVerify = true;\n s = s.replace(/\\[\\[VERIFY\\]\\]/gi, \"\");\n }\n\n // verdict line -> emoji + bold\n let hadVerdict = false;\n s = s.replace(/^\\s*VERDICT:\\s*(.+?)\\s*$/im, (m, label) => {\n hadVerdict = true;\n const v = label.toLowerCase();\n let e = \"\\u2139\\uFE0F\";\n if (v.startsWith(\"not permitted\") || v.startsWith(\"not allowed\")) {\n e = \"\\u26D4\";\n } else if (v.startsWith(\"conditional\")) e = \"\\u26A0\\uFE0F\";\n else if (v.startsWith(\"not covered\")) e = \"\\u2139\\uFE0F\";\n else if (\n v.includes(\"within\") || v.includes(\"permitted\") || v.includes(\"allowed\")\n ) e = \"\\u2705\";\n return e + \" *\" + label.trim() + \"*\";\n });\n s = s.replace(/\\*\\*(.+?)\\*\\*/g, \"*$1*\"); // markdown bold -> WhatsApp bold\n // A verdict is the one thing residents screenshot, so it never travels without context.\n if (\n (hadVerdict || wantsVerify || isCommitteeConductQuestion(askedText)) &&\n CONFIG.verifyNote\n ) s = s.trim() + \"\\n\\n\" + CONFIG.verifyNote;\n return { text: s.trim(), wantsTopic, wantsYesNo, attachRef };\n}\n\n/* ---------- WhatsApp Graph API senders ---------- */\nasync function waSend(phoneId, payload) {\n const pid = phoneId || PHONE_ID;\n try {\n const r = await fetch(`https://graph.facebook.com/${GV}/${pid}/messages`, {\n method: \"POST\",\n headers: {\n \"content-type\": \"application/json\",\n Authorization: `Bearer ${TOKEN}`,\n },\n body: JSON.stringify(\n Object.assign({ messaging_product: \"whatsapp\" }, payload),\n ),\n });\n if (!r.ok) console.log(\"waSend error\", r.status, await r.text());\n } catch (e) {\n console.log(\"waSend threw\", String(e));\n }\n}\nfunction sendText(phoneId, to, body) {\n return waSend(phoneId, {\n to,\n type: \"text\",\n text: { preview_url: false, body: body.slice(0, 4000) },\n });\n}\nfunction sendButtons(phoneId, to, body, buttons) {\n return waSend(phoneId, {\n to,\n type: \"interactive\",\n interactive: {\n type: \"button\",\n body: { text: body.slice(0, 1024) },\n action: {\n buttons: buttons.map((b) => ({\n type: \"reply\",\n reply: { id: b.id, title: b.title.slice(0, 20) },\n })),\n },\n },\n });\n}\nasync function sendTopicList(phoneId, to, body) {\n const topics = (await getTopics()).slice(0, 10); // WhatsApp interactive lists allow at most 10 rows\n const rows = topics.map((t) => {\n const ov = TOPIC_ROWS.find((r) => r.id === t); // reuse a curated short label if we have one for this topic\n if (ov) {\n const row = { id: t, title: ov.title.slice(0, 24) };\n if (ov.description) row.description = ov.description.slice(0, 72);\n return row;\n }\n return { id: t, title: String(t).slice(0, 24) };\n });\n return waSend(phoneId, {\n to,\n type: \"interactive\",\n interactive: {\n type: \"list\",\n body: { text: body.slice(0, 1024) },\n action: { button: \"Choose area\", sections: [{ title: \"Areas\", rows }] },\n },\n });\n}\n\nfunction sendStarList(phoneId, to, body) {\n // Reply buttons cap at 3, so a 1-5 star scale has to be an interactive LIST (up to 10 rows).\n const rows = [\n {\n id: \"STAR::5\",\n title: \"\\u2B50\\u2B50\\u2B50\\u2B50\\u2B50\",\n description: \"Excellent \\u2014 it works really well\",\n },\n {\n id: \"STAR::4\",\n title: \"\\u2B50\\u2B50\\u2B50\\u2B50\",\n description: \"Good, with room to improve\",\n },\n {\n id: \"STAR::3\",\n title: \"\\u2B50\\u2B50\\u2B50\",\n description: \"Okay \\u2014 useful sometimes\",\n },\n { id: \"STAR::2\", title: \"\\u2B50\\u2B50\", description: \"Needs work\" },\n { id: \"STAR::1\", title: \"\\u2B50\", description: \"Not useful to me\" },\n ];\n return waSend(phoneId, {\n to,\n type: \"interactive\",\n interactive: {\n type: \"list\",\n body: { text: body.slice(0, 1024) },\n action: { button: \"Rate it\", sections: [{ title: \"Out of 5\", rows }] },\n },\n });\n}\n\n/* ---------- inbound image -> base64 (for forwarding to Drive) ---------- */\nfunction toBase64(bytes) {\n let binary = \"\";\n const chunk = 0x8000;\n for (let i = 0; i < bytes.length; i += chunk) {\n binary += String.fromCharCode.apply(null, bytes.subarray(i, i + chunk));\n }\n return btoa(binary);\n}\nasync function downloadMedia(mediaId) {\n try {\n const meta = await fetch(`https://graph.facebook.com/${GV}/${mediaId}`, {\n headers: { Authorization: `Bearer ${TOKEN}` },\n });\n const mj = await meta.json();\n if (!mj || !mj.url) return null;\n const bin = await fetch(mj.url, {\n headers: { Authorization: `Bearer ${TOKEN}` },\n });\n const buf = new Uint8Array(await bin.arrayBuffer());\n return { base64: toBase64(buf), mime: mj.mime_type || \"image/jpeg\" };\n } catch (e) {\n return null;\n }\n}\n\n/* ---------- per-resident session (Val Town blob, keyed by phone) ---------- */\nasync function loadSession(phone) {\n try {\n const s = await blob.getJSON(WA_PREFIX + phone);\n if (s) return s;\n } catch (e) {}\n return { flat: \"\", history: [], pendingAttachRef: \"\", seen: [] };\n}\nasync function saveSession(phone, s) {\n try {\n await blob.setJSON(WA_PREFIX + phone, s);\n } catch (e) {}\n}\n\n/* ---------- duplicate-delivery guard ----------\n Meta redelivers a webhook if it doesn't get a 200 quickly, and a rules answer (document fetch +\n several model turns) easily outruns that. The old guard lived inside handleMessage, which loads\n the session at the start and saves it at the end - so two concurrent deliveries both read a\n session that didn't yet list the id, and both answered. That is why residents saw the same\n question answered three or four times.\n The claim below happens BEFORE any slow work: an in-memory set catches redeliveries that land on\n the same warm isolate (the common case), and a blob key catches the rest. */\nconst INFLIGHT = new Set();\nconst DONE = new Map(); // id -> time we finished answering it\nconst STALE_MS = 45000; // below the platform request timeout, so a killed attempt is retried rather than swallowed\nfunction pruneDone() {\n const cutoff = Date.now() - 600000;\n for (const [k, t] of DONE) if (t < cutoff) DONE.delete(k);\n}\nasync function claimMessage(id) {\n if (!id) return true;\n pruneDone();\n if (INFLIGHT.has(id) || DONE.has(id)) return false;\n const key = \"wa_seen_\" + id;\n try {\n const prior = await blob.getJSON(key);\n if (prior) {\n // Already answered, or another attempt is still running: leave it alone.\n if (prior.done) {\n DONE.set(id, Date.now());\n return false;\n }\n if (Date.now() - (prior.t || 0) < STALE_MS) return false;\n // Otherwise the previous attempt died before finishing (platform timeout, cold-start kill).\n // Meta is still retrying, so take it over rather than dropping the resident's message.\n console.log(\"previous attempt did not finish, retrying\", id);\n }\n await blob.setJSON(key, { t: Date.now(), done: false });\n } catch (e) { /* blob unavailable: the in-memory guard still holds */ }\n INFLIGHT.add(id);\n return true;\n}\nasync function finishMessage(id, ok) {\n if (!id) return;\n INFLIGHT.delete(id);\n if (ok) {\n DONE.set(id, Date.now());\n try {\n await blob.setJSON(\"wa_seen_\" + id, { t: Date.now(), done: true });\n } catch (e) {}\n } else {\n // Failed: clear the claim immediately so Meta's next retry can answer instead of being swallowed.\n try {\n await blob.setJSON(\"wa_seen_\" + id, { t: 0, done: false });\n } catch (e) {}\n }\n}\n\n/* ---------- main handler ---------- */\nexport default async function (req) {\n const url = new URL(req.url);\n\n // 1) Webhook verification (Meta calls this once when you save the webhook)\n if (req.method === \"GET\") {\n const mode = url.searchParams.get(\"hub.mode\");\n const token = url.searchParams.get(\"hub.verify_token\");\n const challenge = url.searchParams.get(\"hub.challenge\");\n if (mode === \"subscribe\" && token && token === VERIFY_TOKEN) {\n return new Response(challenge || \"\", {\n status: 200,\n headers: { \"content-type\": \"text/plain\" },\n });\n }\n return new Response(NAME + \" WhatsApp webhook is running.\", {\n status: 200,\n headers: { \"content-type\": \"text/plain\" },\n });\n }\n\n if (req.method !== \"POST\") {\n return new Response(\"Method not allowed\", { status: 405 });\n }\n\n let body;\n try {\n body = await req.json();\n } catch (e) {\n return new Response(\"EVENT_RECEIVED\", { status: 200 });\n }\n\n const KEY = Deno.env.get(\"ANTHROPIC_API_KEY\");\n let SYSTEM = \"\";\n try {\n SYSTEM = await getSystem();\n } catch (e) {\n SYSTEM = \"\";\n }\n\n try {\n const entries = Array.isArray(body.entry) ? body.entry : [];\n for (const entry of entries) {\n for (const change of (entry.changes || [])) {\n const value = change.value || {};\n const phoneId = (value.metadata && value.metadata.phone_number_id) ||\n PHONE_ID;\n const messages = value.messages || []; // absent for delivery-status callbacks\n for (const msg of messages) {\n // Drop redeliveries before doing any work, so a retry costs one fast 200 and nothing else.\n if (!(await claimMessage(msg.id))) {\n console.log(\"duplicate delivery ignored\", msg.id);\n continue;\n }\n let ok = false;\n try {\n await handleMessage(msg, phoneId, KEY, SYSTEM);\n ok = true;\n } catch (err) {\n console.log(\"handleMessage failed\", msg.id, String(err));\n // Never leave a resident staring at silence: say something, even on an internal failure.\n try {\n await sendText(\n phoneId,\n msg.from,\n \"Sorry \\u2014 something went wrong on my side answering that. Please send it again.\",\n );\n } catch (e2) {}\n } finally {\n await finishMessage(msg.id, ok);\n }\n }\n }\n }\n } catch (e) {\n console.log(\"handler error\", String(e));\n }\n\n // Always 200 quickly so Meta doesn't retry.\n return new Response(\"EVENT_RECEIVED\", { status: 200 });\n}\n\nfunction isGreeting(t) {\n return /^(hi+|hey+|hello+|hii+|yo|start|menu|help|main menu|namaste|namaskara|namaskar)\\W*$/i\n .test(String(t || \"\").trim());\n}\nfunction isLangCmd(t) {\n const s = String(t || \"\").trim().toLowerCase();\n return s === \"language\" || s === \"lang\" || s === \"\\u092d\\u093e\\u0937\\u093e\" ||\n s === \"\\u0cad\\u0cbe\\u0cb7\\u0cc6\";\n}\nfunction isWhatsNew(t) {\n const s = String(t || \"\").trim().toLowerCase();\n return s === \"what's new\" || s === \"whats new\" || s === \"what is new\" ||\n s === \"version\" || s === \"updates\" || s === \"release notes\";\n}\nfunction isAbout(t) {\n const s = String(t || \"\").trim().toLowerCase();\n return s === \"about\" || s === \"about this assistant\" ||\n s === \"about askaoaoa\" || s === \"who built this\" || s === \"who made this\";\n}\n// \"feedback\" is deliberately ambiguous — it could mean the community or the assistant — so this\n// only catches the bare command and we then ask which one they mean. Anything more specific\n// (\"the timings you gave me were wrong\") falls through to the model, which routes it per §8.\nfunction isFeedbackCmd(t) {\n const s = String(t || \"\").trim().toLowerCase().replace(/[?.!]+$/, \"\");\n return s === \"feedback\" || s === \"give feedback\" || s === \"share feedback\" ||\n s === \"send feedback\" || s === \"leave feedback\";\n}\nfunction isMyComplaints(t) {\n const s = String(t || \"\").trim().toLowerCase();\n return /^(my complaints?|complaints? status|status of (my )?complaints?|check (my )?complaints?|complaint status)\\W*$/\n .test(s);\n}\nfunction formatComplaints(list) {\n if (!list || !list.length) {\n return \"You don't have any complaints on record yet. To raise one, just describe the problem and I'll log it for you.\";\n }\n const icon = (st) => {\n const l = String(st || \"\").toLowerCase();\n if (\n l.indexOf(\"resolv\") >= 0 || l.indexOf(\"close\") >= 0 ||\n l.indexOf(\"done\") >= 0\n ) return \"\\u2705\";\n if (l.indexOf(\"progress\") >= 0 || l.indexOf(\"working\") >= 0) {\n return \"\\uD83D\\uDD27\";\n }\n return \"\\uD83D\\uDD52\";\n };\n let out = \"*Your complaints*\";\n for (const c of list) {\n out += \"\\n\\n\" + icon(c.status) + \" *\" + (c.status || \"Open\") + \"* \\u2014 \" +\n (c.category || \"\") + \"\\n_\" + (c.reference || \"\") + \"_\\n\" +\n (c.issue || \"\");\n }\n out += \"\\n\\nThe committee updates these as they work on them.\";\n return out;\n}\nconst ABOUT_TEXT = \"*About \" + NAME + \"*\\n\\n\" + NAME +\n \" is a resident-built assistant for the \" + CONFIG.estateShort +\n \" community. It answers questions from the association's official documents \\u2014 the registered bye-laws, the Rule Book, and the Party Hall and Parking rules \\u2014 and points you to the exact clause or section.\\n\\nYou can also suggest an improvement, share feedback or a concern, or escalate an unresolved MyGate issue to the committee. Anything you share is recorded privately for the committee only \\u2014 never made public.\\n\\nIt works in English, \\u0939\\u093f\\u0902\\u0926\\u0940 and \\u0c95\\u0ca8\\u0ccd\\u0ca8\\u0ca1, on both the web and WhatsApp.\\n\\nYou can also ask for \\u2014 and share \\u2014 recommendations for anything nearby: a dentist, hospital, hardware shop, restaurant, hotel, plumber, tutor, and more.\\n\\nIt uses an AI model restricted to the association's documents, so answers stay grounded in what the community has adopted. Replies are AI-generated and can occasionally be wrong, so please confirm anything important with the Managing Committee.\\n\\n_\" +\n CONFIG.credit + \"_\";\nconst COPILOT_INTRO =\n \"*Committee Copilot* \\u2014 you're now in committee mode. \\uD83D\\uDD10\\n\\nAsk me things like:\\n\\u2022 Summarise this week's complaints\\n\\u2022 Which complaints are still open?\\n\\u2022 Recurring suggestions from residents\\n\\u2022 Draft a notice about the water schedule\\n\\nType *exit* to return to the normal assistant.\";\n\nconst STARTER_ROWS = [\n {\n id: \"ASK::I'd like to suggest an improvement for the community.\",\n title: \"Suggest an improvement\",\n },\n {\n id: \"ASK::I'd like to raise or escalate an issue.\",\n title: \"Raise or escalate issue\",\n },\n {\n id: \"ASK::I have a question for the Managing Committee.\",\n title: \"Ask the committee\",\n },\n {\n id: \"ASK::What are the rules for keeping a pet in the complex?\",\n title: \"Pet / dog rules\",\n },\n { id: \"ASK::What are the gym timings?\", title: \"Gym timings\" },\n {\n id: \"ASK::How do I book the party hall and what does it cost?\",\n title: \"Book the party hall\",\n },\n {\n id: \"ASK::What are the rules and charges for guest parking?\",\n title: \"Guest parking rules\",\n },\n {\n id: \"ASK::What local services do residents recommend?\",\n title: \"Local recommendations\",\n },\n { id: \"CMD::mycomplaints\", title: \"My complaints\" },\n { id: \"ASK::about\", title: \"About this assistant\" },\n];\nasync function sendMenu(phoneId, from, lead) {\n const body = (lead ? lead + \"\\n\\n\" : \"\") +\n \"You can check our community rules, share a suggestion, raise or escalate an issue, or send a question to the Managing Committee. Tap an option below, or just type what you need.\";\n await waSend(phoneId, {\n to: from,\n type: \"interactive\",\n interactive: {\n type: \"list\",\n body: { text: body.slice(0, 1024) },\n action: {\n button: \"See examples\",\n sections: [{\n title: \"Try asking\",\n rows: STARTER_ROWS.map((r) => ({\n id: r.id,\n title: r.title.slice(0, 24),\n })),\n }],\n },\n },\n });\n}\nasync function sendLangMenu(phoneId, from) {\n await sendButtons(\n phoneId,\n from,\n \"Which language would you like replies in? You can also just write in your language and I'll follow.\",\n [\n { id: \"lang_en\", title: \"English\" },\n { id: \"lang_hi\", title: \"\\u0939\\u093f\\u0902\\u0926\\u0940\" },\n { id: \"lang_kn\", title: \"\\u0c95\\u0ca8\\u0ccd\\u0ca8\\u0ca1\" },\n ],\n );\n}\n\n/* ---------- shared contact cards (WhatsApp \"contacts\" message type) ----------\n A resident can attach a phone contact instead of typing the details. Meta delivers it as\n msg.contacts = [{ name:{formatted_name,...}, phones:[{phone,wa_id}], org:{company,title},\n emails:[{email}], addresses:[{city,...}], urls:[{url}] }]. We flatten each card into plain\n text and hand it to the model as a normal turn, so the usual recommendation flow still runs\n (it asks what kind of service it is, and asks consent before crediting the flat). */\nfunction describeContacts(list) {\n const cards = (Array.isArray(list) ? list : []).slice(0, 5); // ignore absurdly large shares\n const out = [];\n for (const c of cards) {\n if (!c) continue;\n const bits = [];\n const nm = (c.name && (c.name.formatted_name ||\n [c.name.first_name, c.name.last_name].filter(Boolean).join(\" \"))) || \"\";\n if (nm) bits.push(\"Name: \" + nm);\n const phones = (c.phones || []).map((p) =>\n (p && (p.phone || p.wa_id)) || \"\"\n ).filter(Boolean);\n if (phones.length) bits.push(\"Phone: \" + phones.join(\", \"));\n const emails = (c.emails || []).map((e) => (e && e.email) || \"\").filter(\n Boolean,\n );\n if (emails.length) bits.push(\"Email: \" + emails.join(\", \"));\n if (c.org && (c.org.company || c.org.title)) {\n bits.push(\n \"Business/role: \" +\n [c.org.company, c.org.title].filter(Boolean).join(\" \\u2014 \"),\n );\n }\n const areas = (c.addresses || []).map((a) => {\n if (!a) return \"\";\n return [a.street, a.city, a.state].filter(Boolean).join(\", \");\n }).filter(Boolean);\n if (areas.length) bits.push(\"Address/area: \" + areas.join(\" | \"));\n const urls = (c.urls || []).map((u) => (u && u.url) || \"\").filter(Boolean);\n if (urls.length) bits.push(\"Website: \" + urls.join(\", \"));\n if (bits.length) out.push(bits.join(\" | \"));\n }\n return out;\n}\n\nasync function handleMessage(msg, phoneId, KEY, SYSTEM) {\n const from = msg.from;\n if (!from) return;\n const session = await loadSession(from);\n\n session.seen = session.seen || [];\n if (session.seen.includes(msg.id)) return;\n session.seen.push(msg.id);\n if (session.seen.length > 30) session.seen = session.seen.slice(-30);\n\n // ---- images: forward to the committee (Drive) if an escalation is pending ----\n if (msg.type === \"image\") {\n if (session.pendingAttachRef) {\n const media = msg.image && msg.image.id\n ? await downloadMedia(msg.image.id)\n : null;\n if (media) {\n const res = await postSheet({\n type: \"attachment\",\n reference: session.pendingAttachRef,\n data: media.base64,\n mime: media.mime,\n filename: \"whatsapp-photo.jpg\",\n });\n await sendText(\n phoneId,\n from,\n res.ok\n ? \"Thanks \\u2014 I've attached your photo to the escalation and shared it with the committee. \\u2705\"\n : \"I couldn't attach that photo just now. Please try sending it again.\",\n );\n if (res.ok) session.pendingAttachRef = \"\";\n } else {\n await sendText(\n phoneId,\n from,\n \"I couldn't download that image. Please try sending it again.\",\n );\n }\n } else {\n await sendText(\n phoneId,\n from,\n \"Thanks for the photo. If it's for an issue you'd like the committee to see, first escalate the issue, then send the photo and I'll attach it.\",\n );\n }\n await saveSession(from, session);\n return;\n }\n\n // ---- resolve the resident's input (typed, list pick, or button tap) ----\n let text = \"\";\n if (msg.type === \"text\") {\n text = (msg.text && msg.text.body) || \"\";\n } else if (msg.type === \"contacts\") {\n // ---- a resident attached one or more phone contacts (usually a recommendation) ----\n const cards = describeContacts(msg.contacts);\n if (!cards.length) {\n await sendText(\n phoneId,\n from,\n \"That contact card came through empty on my side. Could you type the name and number instead?\",\n );\n await saveSession(from, session);\n return;\n }\n if (!session.flat) {\n await sendText(\n phoneId,\n from,\n \"Thanks for sharing that contact! \\uD83D\\uDCC7 Before I save it, please reply with your flat number and block (\" +\n BLOCKLIST + \").\",\n );\n await saveSession(from, session);\n return;\n }\n text = \"[CONTACT CARD SHARED] The resident attached \" +\n (cards.length > 1\n ? cards.length + \" phone contacts\"\n : \"a phone contact\") +\n \" from their phone:\\n\" + cards.map((c, i) =>\n (cards.length > 1 ? (i + 1) + \". \" : \"\") + c\n ).join(\"\\n\");\n } else if (msg.type === \"interactive\") {\n const it = msg.interactive || {};\n if (it.type === \"list_reply\" && it.list_reply) {\n const lid = it.list_reply.id || \"\";\n if (lid.indexOf(\"CMD::mycomplaints\") === 0) text = \"my complaints\"; // deterministic complaint lookup\n else if (lid.indexOf(\"ASK::\") === 0) text = lid.slice(5); // starter question\n else if (lid.indexOf(\"STAR::\") === 0) {\n const n = lid.slice(6);\n text = \"Feedback about this assistant: I'd rate it \" + n +\n \" out of 5 stars.\";\n } else text = \"Topic: \" + (lid || it.list_reply.title || \"\"); // topic pick\n } else if (it.type === \"button_reply\" && it.button_reply) {\n const id = (it.button_reply.id || \"\").toLowerCase();\n if (id === \"menu_ask\") {\n await sendText(\n phoneId,\n from,\n \"Sure \\u2014 just type your question. For example: *What are the gym timings?* or *How much does it cost to book the party hall?*\",\n );\n await saveSession(from, session);\n return;\n } else if (id === \"menu_suggest\") {\n text = \"I'd like to suggest an improvement for the community.\";\n } else if (id === \"menu_issue\") {\n text = \"I'd like to raise or escalate an issue.\";\n } else if (id === \"fb_agent\") {\n await sendStarList(\n phoneId,\n from,\n \"Thanks! \\uD83D\\uDE4F How would you rate \" + NAME + \" so far?\",\n );\n await saveSession(from, session);\n return;\n } else if (id === \"fb_community\") {\n text = \"I'd like to share feedback or a concern about the community.\";\n } else if (id === \"lang_en\" || id === \"lang_hi\" || id === \"lang_kn\") {\n session.lang = id.slice(5);\n const conf = session.lang === \"hi\"\n ? \"\\u0939\\u094b \\u0917\\u092f\\u093e \\u2014 \\u092e\\u0948\\u0902 \\u0905\\u092c \\u0939\\u093f\\u0902\\u0926\\u0940 \\u092e\\u0947\\u0902 \\u091c\\u0935\\u093e\\u092c \\u0926\\u0942\\u0901\\u0917\\u093e\\u0964\"\n : session.lang === \"kn\"\n ? \"\\u0c86\\u0caf\\u0cbf\\u0ca4\\u0cc1 \\u2014 \\u0ca8\\u0cbe\\u0ca8\\u0cc1 \\u0c87\\u0ca8\\u0ccd\\u0ca8\\u0cc1 \\u0c95\\u0ca8\\u0ccd\\u0ca8\\u0ca1\\u0ca6\\u0cb2\\u0ccd\\u0cb2\\u0cbf \\u0c89\\u0ca4\\u0ccd\\u0ca4\\u0cb0\\u0cbf\\u0cb8\\u0cc1\\u0ca4\\u0ccd\\u0ca4\\u0cc7\\u0ca8\\u0cc6.\"\n : \"Done \\u2014 I'll reply in English.\";\n await sendText(phoneId, from, conf);\n await saveSession(from, session);\n return;\n } else {\n text = id === \"yes\"\n ? \"Yes\"\n : id === \"no\"\n ? \"No\"\n : (it.button_reply.title || \"\");\n }\n }\n } else {\n await sendText(\n phoneId,\n from,\n \"I can read text messages, photos and shared contact cards. Please type your question, or tap one of the options I send.\",\n );\n await saveSession(from, session);\n return;\n }\n text = String(text || \"\").trim();\n if (!text) {\n await saveSession(from, session);\n return;\n }\n\n // ---- developer-only maintenance: wipe all resident sessions (NOT available to the committee) ----\n const DEV = (Deno.env.get(\"DEV_NUMBERS\") || \"\").split(\",\").map((x) =>\n x.trim()\n ).filter(Boolean);\n if (DEV.includes(from)) {\n const dlow = text.trim().toLowerCase();\n if (dlow === \"reset sessions\" || dlow === \"reset all sessions\") {\n await sendText(\n phoneId,\n from,\n \"\\u26A0\\uFE0F This will erase EVERY resident's saved flat and chat memory, so everyone re-onboards from scratch. It does NOT touch the Google Sheet or any logged data. To proceed, reply exactly: reset sessions confirm\",\n );\n await saveSession(from, session);\n return;\n }\n if (dlow === \"reset sessions confirm\") {\n let n = 0;\n try {\n const keys = await blob.list(WA_PREFIX);\n for (const k of keys) {\n const key = (typeof k === \"string\") ? k : k.key;\n if (key) {\n try {\n await blob.delete(key);\n n++;\n } catch (e) {}\n }\n }\n } catch (e) {}\n await sendText(\n phoneId,\n from,\n \"\\u2705 Done \\u2014 cleared \" + n +\n \" resident session(s). Everyone will be greeted fresh and asked for their flat again.\",\n );\n return;\n }\n }\n\n // ---- committee copilot: MC numbers enter a private analytics/drafting mode ----\n const MC = (Deno.env.get(\"MC_NUMBERS\") || \"\").split(\",\").map((x) => x.trim())\n .filter(Boolean);\n if (MC.includes(from)) {\n const low = text.trim().toLowerCase();\n if (\n session.mode !== \"copilot\" &&\n (low === \"copilot\" || low === \"mc\" || low === \"committee\" ||\n low === \"mc mode\")\n ) {\n session.mode = \"copilot\";\n session.chist = [];\n await sendText(phoneId, from, COPILOT_INTRO);\n await saveSession(from, session);\n return;\n }\n if (session.mode === \"copilot\") {\n if (\n low === \"exit\" || low === \"done\" || low === \"resident\" ||\n low === \"leave\" || low === \"quit\"\n ) {\n session.mode = \"\";\n session.chist = [];\n await sendText(\n phoneId,\n from,\n 'Left committee copilot \\u2014 back to the normal assistant. Say \"copilot\" anytime to return.',\n );\n await saveSession(from, session);\n return;\n }\n session.chist = session.chist || [];\n session.chist.push({ role: \"user\", content: text });\n const rep = await copilotAnswer(KEY, session.chist, SYSTEM);\n session.chist.push({ role: \"assistant\", content: rep });\n if (session.chist.length > 14) session.chist = session.chist.slice(-14);\n const outMsg = rep.replace(/\\*\\*(.+?)\\*\\*/g, \"*$1*\");\n await sendText(\n phoneId,\n from,\n outMsg.length > 3900\n ? outMsg.slice(0, 3900) +\n \"\\n\\n(\\u2026truncated \\u2014 ask for a specific part)\"\n : outMsg,\n );\n await saveSession(from, session);\n return;\n }\n }\n\n // ---- ops: authorized numbers manage the live site status (multiple alerts supported) ----\n const OPS = (Deno.env.get(\"OPS_NUMBERS\") || \"\").split(\",\").map((x) =>\n x.trim()\n ).filter(Boolean);\n if (OPS.includes(from)) {\n const low = text.trim().toLowerCase();\n\n // list active alerts (numbered)\n if (\n low === \"status\" || low === \"status list\" || low === \"list status\" ||\n low === \"statuses\"\n ) {\n const d = await sheetGet(\"q=status\");\n const active = (d && d.ok && Array.isArray(d.active)) ? d.active : [];\n if (!active.length) {\n await sendText(\n phoneId,\n from,\n \"No site alerts are active right now.\\n\\nPost one with: status: \",\n );\n } else {\n let m = \"*Active site alerts*\";\n active.forEach((a, i) => {\n m += \"\\n\" + (i + 1) + \". \" + a.message;\n });\n m +=\n '\\n\\nSend \"clear \" to remove one, or \"clear all\" to remove all.';\n await sendText(phoneId, from, m);\n }\n await saveSession(from, session);\n return;\n }\n\n // clear ALL\n if (\n low === \"clear all\" || low === \"all clear\" || low === \"clear status\" ||\n low === \"status clear\"\n ) {\n await postSheet({ type: \"status_clear\" });\n await sendText(\n phoneId,\n from,\n \"\\u2705 All site alerts cleared \\u2014 residents will no longer see any active alert.\",\n );\n await saveSession(from, session);\n return;\n }\n\n // clear ONE by number\n const cm = low.match(/^clear\\s+(\\d+)$/);\n if (cm) {\n const n = parseInt(cm[1], 10);\n const d = await sheetGet(\"q=status\");\n const active = (d && d.ok && Array.isArray(d.active)) ? d.active : [];\n if (n < 1 || n > active.length) {\n await sendText(\n phoneId,\n from,\n \"There's no alert #\" + n +\n '. Send \"status list\" to see the current alerts and their numbers.',\n );\n } else {\n const target = active[n - 1];\n await postSheet({ type: \"status_clear\", row: target.row });\n await sendText(\n phoneId,\n from,\n \"\\u2705 Cleared alert #\" + n + \":\\n\\n\" + target.message +\n (active.length > 1\n ? '\\n\\nThe other alert(s) are still active \\u2014 send \"status list\" to review.'\n : \"\"),\n );\n }\n await saveSession(from, session);\n return;\n }\n\n // add a NEW alert\n const mset = text.match(/^\\s*(?:post\\s+)?status\\s*[:\\-]\\s*([\\s\\S]+)$/i);\n if (mset) {\n const msg = mset[1].trim();\n const res = await postSheet({\n type: \"status_set\",\n message: msg,\n postedBy: from,\n });\n const num = (res && res.count) ? res.count : null;\n await sendText(\n phoneId,\n from,\n res.ok\n ? (\"\\u2705 Posted\" + (num ? \" as alert #\" + num : \"\") + \":\\n\\n\" +\n msg +\n '\\n\\nResidents asking about power, water or lifts will now be told this.\\nSend \"status list\" to see all, \"clear ' +\n (num || \"\") + '\" to remove this one, or \"clear all\".')\n : \"Couldn't post that just now. Please try again.\",\n );\n await saveSession(from, session);\n return;\n }\n }\n\n // ---- first-time onboarding: ask for and store the flat number ----\n if (!session.flat) {\n if (isValidFlat(text)) {\n session.flat = normalizeFlat(text);\n await sendText(\n phoneId,\n from,\n \"Thanks! You're set as flat *\" + session.flat +\n \"*. \\uD83D\\uDC4B\\n\\nI'm \" + NAME +\n \", your line to the Managing Committee. You can check our community rules, share a suggestion, raise or escalate an issue, or ask the committee a question. \" +\n DISCLAIMER,\n );\n await sendMenu(phoneId, from, \"\");\n await saveSession(from, session);\n return;\n }\n const greeting =\n /^(hi+|hello|hey|hai|start|menu|hola|namaste|namaskara|good\\s*(morning|afternoon|evening))\\b/i\n .test(text.trim());\n if (greeting || text.trim().length < 2) {\n await sendText(\n phoneId,\n from,\n \"Welcome to *\" + NAME + \"*, the \" + CONFIG.estateShort +\n \" community assistant.\\n\\nTo get started, please reply with your flat number and block (\" +\n BLOCKLIST + \").\",\n );\n } else {\n await sendText(\n phoneId,\n from,\n \"Hmm, that doesn't match a flat at \" + CONFIG.estateShort +\n \". Please reply with a valid \" + CONFIG.estateShort +\n \" flat to continue.\",\n );\n }\n await saveSession(from, session);\n return;\n }\n\n // ---- quick commands (once onboarded) ----\n if (isGreeting(text)) {\n await sendMenu(phoneId, from, \"Hello! \\uD83D\\uDC4B\");\n await saveSession(from, session);\n return;\n }\n if (isLangCmd(text)) {\n await sendLangMenu(phoneId, from);\n await saveSession(from, session);\n return;\n }\n if (isWhatsNew(text)) {\n await sendText(phoneId, from, RELEASES_TEXT);\n await saveSession(from, session);\n return;\n }\n if (isAbout(text)) {\n await sendText(phoneId, from, ABOUT_TEXT);\n await saveSession(from, session);\n return;\n }\n if (isFeedbackCmd(text)) {\n session.fbNudged = true; // they've come to it themselves; don't nudge later\n await sendButtons(\n phoneId,\n from,\n \"Happy to take that \\u2014 which kind of feedback is it?\",\n [\n { id: \"fb_agent\", title: \"About this assistant\" },\n { id: \"fb_community\", title: \"About the community\" },\n ],\n );\n await saveSession(from, session);\n return;\n }\n if (isMyComplaints(text)) {\n const dc = await sheetGet(\n \"q=complaint&flat=\" + encodeURIComponent(session.flat || \"\"),\n );\n const list = (dc && dc.ok && Array.isArray(dc.complaints))\n ? dc.complaints\n : [];\n await sendText(phoneId, from, formatComplaints(list));\n await saveSession(from, session);\n return;\n }\n if (isComplaintStatus(text)) {\n const cref = complaintRef(text);\n let cq = \"q=complaint&flat=\" + encodeURIComponent(session.flat);\n if (cref) cq += \"&ref=\" + encodeURIComponent(cref);\n const cd = await sheetGet(cq);\n await sendText(phoneId, from, formatComplaintsWA(cd, cref));\n await saveSession(from, session);\n return;\n }\n\n if (!KEY || !SYSTEM) {\n await sendText(\n phoneId,\n from,\n \"The assistant isn't fully set up yet. Please check back shortly.\",\n );\n await saveSession(from, session);\n return;\n }\n\n // ---- run the assistant ----\n session.history = session.history || [];\n session.history.push({ role: \"user\", content: text });\n\n const raw = await answer(\n KEY,\n SYSTEM,\n session.history,\n session.flat,\n session.lang || \"\",\n );\n const fmt = formatForWhatsApp(raw, text);\n const reply = fmt.text,\n wantsTopic = fmt.wantsTopic,\n wantsYesNo = fmt.wantsYesNo,\n attachRef = fmt.attachRef;\n\n session.history.push({\n role: \"assistant\",\n content: raw.replace(/\\[\\[PICK_(DATE|TOPIC|YESNO)\\]\\]/g, \"\").replace(\n /\\[\\[ATTACH:[^\\]]*\\]\\]/g,\n \"\",\n ).trim(),\n });\n if (session.history.length > 16) session.history = session.history.slice(-16);\n if (attachRef) session.pendingAttachRef = attachRef;\n\n if (reply) await sendText(phoneId, from, reply);\n if (attachRef) {\n await sendText(\n phoneId,\n from,\n \"If you'd like, send a photo of the issue here and I'll attach it to the escalation. \\uD83D\\uDCF7\",\n );\n }\n if (wantsTopic) {\n await sendTopicList(\n phoneId,\n from,\n \"Please choose the area this relates to:\",\n );\n } else if (wantsYesNo) {\n await sendButtons(phoneId, from, \"Please choose:\", [{\n id: \"yes\",\n title: \"Yes\",\n }, { id: \"no\", title: \"No\" }]);\n }\n\n session.msgCount = (session.msgCount || 0) + 1;\n if (\n session.msgCount === 3 && !session.recNudged && !wantsTopic &&\n !wantsYesNo && !attachRef\n ) {\n session.recNudged = true;\n await sendText(\n phoneId,\n from,\n \"\\uD83D\\uDCA1 By the way \\u2014 got a favourite nearby you'd recommend to neighbours? A dentist, hospital, hardware shop, restaurant, hotel, plumber \\u2014 anything at all. Just tell me and I'll save it so others in \" +\n CONFIG.estateShort + \" can benefit.\",\n );\n } else if (\n session.msgCount === 8 && !session.fbNudged && !wantsTopic && !wantsYesNo &&\n !attachRef\n ) {\n // One-time, reactive only — never a broadcast. Asked once per resident, ever.\n session.fbNudged = true;\n await sendStarList(\n phoneId,\n from,\n \"Quick one, if you don't mind \\u2014 how would you rate \" + NAME +\n \" so far? Your answer goes only to the Managing Committee and helps them improve it.\",\n );\n }\n\n await saveSession(from, session);\n}","label",[131],"ca10916f-9028-4687-9445-261ce5f2f76d",["D",1785909053337],15,[29],"6eb89e07-dc93-4528-9060-4c1d0c781f24","askaoaoa-config.ts","2b4346f0-8a96-11f1-8ee7-1607ee4eb77e",58,["D",1785859312034],7,"// AskAOAOA — CONFIG val (Val Town Script val, Unlisted). ONE per community.\n//\n// This is the ONLY file that changes from one apartment to the next. The knowledge val\n// (rules) and the two app vals (web + WhatsApp) read everything community-specific from here,\n// so a new community = a new copy of this file + its own Sheet / Drive folder / WhatsApp number.\n//\n// The app vals import this via their CONFIG_URL env var (set it to this val's ES-module URL,\n// e.g. https://esm.town/v//askaoaoa_config).\n//\n// To rename the assistant: change `name` below (and regenerate the page val for the web branding).\n\nexport const CONFIG = {\n id: \"aryan-opulence\", // slug; used only for internal namespacing\n name: \"AskAOAOA\", // assistant display name\n association: \"Aryan Opulence Apartment Owners' Association\",\n estateShort: \"Aryan Opulence\", // short name used in messages\n unitCount: 189,\n languages: [\"English\", \"Hindi\", \"Kannada\"], // offered languages, in order\n fmsContact: \"Sandeep on +91 91108 02505\", // FMS (facility) manager, quoted to residents on escalations\n disclaimer:\n \"_I'm an AI assistant and can occasionally be wrong \\u2014 please verify anything important with the Managing Committee._\",\n // Appended to any answer that carries a VERDICT line, so a screenshot of it never travels\n // without context. Keep it short: it rides along on real answers.\n verifyNote:\n \"_This is general information from the association's documents, not a ruling. The committee may have since decided otherwise \\u2014 please check with them._\",\n credit: \"Built by Shyam A Nair (602-C), as a community volunteer effort.\", // shown in the web + WhatsApp About\n\n // Flat scheme — replaces the old hardcoded isValidFlat(). Accepts \"602-C\" or \"C-602\":\n // the last two digits are the flat, the leading digit(s) are the floor \"series\".\n flats: {\n blocks: [\"a\", \"b\", \"c\", \"d\"],\n limits: {\n a: { maxSeries: 11, maxFlat: 6 },\n b: { maxSeries: 11, maxFlat: 4 },\n c: { maxSeries: 10, maxFlat: 4 },\n d: { maxSeries: 10, maxFlat: 4 },\n },\n },\n\n // FALLBACK topic list only. Topics now derive from the MC Contacts Portfolio rows (+ \"Others\")\n // via the ?q=topics endpoint; this list is used only if that returns nothing.\n topics: [\n \"Finance\",\n \"Admin & Communication\",\n \"Security/Parking\",\n \"MEP (Mechanical, Electrical, Plumbing)\",\n \"Gardening\",\n \"AMCs & FMS\",\n \"Housekeeping\",\n \"Tech/IT (including MyGate)\",\n \"Water Management\",\n \"Clubhouse Management\",\n \"Others\",\n ],\n\n // Web page brand colours (hex, no \"#\"): header/primary (ink), gradient end (ink2),\n // accent/gold (accent), and muted green (sage). Edit these to re-theme the web app.\n brand: { ink: \"16302B\", ink2: \"0F231F\", accent: \"B5852C\", sage: \"5E7168\" },\n\n // LLM routing (mode-based): a cheaper/faster model for the high-volume resident agent,\n // a stronger model for the committee copilot. Flip either back to the same value to undo.\n models: { resident: \"claude-haiku-4-5-20251001\", copilot: \"claude-sonnet-5\" },\n};\n\n// ---- Generic flat validation, driven by CONFIG.flats (no per-community code) ----\nfunction flatRegexes(cfg) {\n const cls = ((cfg.flats && cfg.flats.blocks) || []).join(\"\");\n return [\n new RegExp(\"^(\\\\d{3,4})-?([\" + cls + \"])$\"),\n new RegExp(\"^([\" + cls + \"])-?(\\\\d{3,4})$\"),\n ];\n}\nfunction parseFlat(raw, cfg) {\n const s = String(raw || \"\").trim().toLowerCase().replace(/\\s+/g, \"\");\n const [re1, re2] = flatRegexes(cfg);\n const m = s.match(re1) || s.match(re2);\n if (!m) return null;\n let digits, block;\n if (/^\\d/.test(m[1])) {\n digits = m[1];\n block = m[2];\n } else {\n block = m[1];\n digits = m[2];\n }\n return {\n digits,\n block,\n flat: parseInt(digits.slice(-2), 10),\n series: parseInt(digits.slice(0, -2), 10),\n };\n}\nexport function isValidFlat(raw, cfg) {\n cfg = cfg || CONFIG;\n const p = parseFlat(raw, cfg);\n if (!p) return false;\n const lim = (cfg.flats.limits && cfg.flats.limits[p.block]) || {};\n const maxSeries = lim.maxSeries || 99, maxFlat = lim.maxFlat || 99;\n return p.series >= 1 && p.series <= maxSeries && p.flat >= 1 &&\n p.flat <= maxFlat;\n}\nexport function normalizeFlat(raw, cfg) {\n cfg = cfg || CONFIG;\n const p = parseFlat(raw, cfg);\n if (!p) return String(raw || \"\").trim().toUpperCase();\n return (p.digits + \"-\" + p.block).toUpperCase();\n}",[147],"pulls","branches",1,"forks","envVars",{"_162":164,"_165":62},"connections",{"_162":62,"_163":62},"count","unfilledCount",12,"emptyCount",{"_44":146,"_34":147,"_32":148,"_133":-5,"_42":43,"_41":149,"_135":27,"_39":176,"_13":151,"_35":36,"_37":152,"_24":27,"_140":-5,"_170":177,"_28":178},{"_44":142,"_34":29,"_32":33,"_133":-5,"_42":43,"_41":14,"_135":27,"_39":173,"_13":144,"_35":36,"_37":38,"_24":27,"_140":-5,"_170":174,"_28":175},{"_44":130,"_34":131,"_32":132,"_133":-5,"_42":43,"_41":134,"_135":27,"_39":169,"_13":137,"_35":138,"_37":139,"_24":27,"_140":-5,"_170":171,"_28":172},["D",1785908368588],"children",[],[131],["D",1785909053337],[],[29],["D",1785859312034],[],[147],["D",1783664429895],{"_96":95,"_104":100},{"_96":95,"_97":98,"_99":100,"_18":56,"_101":102,"_103":27},{"_13":85,"_34":86,"_44":43,"_87":45,"_88":-5,"_89":-5,"_90":-5,"_91":-5,"_92":-5,"_93":46,"_94":95},"ctx",{"_78":-5,"_124":216,"_217":218,"_219":220,"_221":27,"_222":27,"_223":27,"_224":25,"_225":-5,"_226":-7,"_227":-7,"_228":62,"_229":27,"_230":27,"_231":-5,"_232":27,"_233":27,"_234":27,"_235":-5},"ENV",{"_187":188,"_189":190,"_191":86,"_192":193,"_194":195,"_196":197,"_198":199,"_200":201,"_202":203,"_204":205,"_206":207,"_208":209,"_210":211,"_212":213,"_214":215},"REACT_APP_SENTRY_DSN","https://586020d86f05489db63a554690acf1e9@o1398359.ingest.sentry.io/6724499","RENDER_GIT_COMMIT","811aa22498c9b0b5a47e8d778de579cdb990f332","RENDER_GIT_BRANCH","POSTHOG_API_TOKEN","phc_nmoyuAkStEZuUp69rYlg1DZgrIt19Gx0Y4Hwf8tGgJP","POSTHOG_API_HOST","https://wart.val.town","WEB_API_TEMPLATE","https://{val}.web.val.run","APP_URL","https://www.val.town","MODULE_URL","https://esm.town","VANITY_WEB_API_TEMPLATE","https://{label}.val.run/","DENOLS_URL","https://lsp.val.town","NODE_ENV","production","CLOUDFLARE_TURNSTILE_SITE_KEY","0x4AAAAAACtaIZJH7s9rBMZ6","GOOGLE_CLIENT_ID","875989696899-jtsqkhlbdv5jrlsk8631p7ipjg3c6o62.apps.googleusercontent.com","STRIPE_PUBLISHABLE_KEY","pk_live_51N2zvWEpZTTLipIkVEwXrUG9AT0yeOOEYxszarZT1jpsZV58Ly6wlg3SltZpHN26CUalOBmUSY7adQRQiSkCZz2500qSXNkf8Y","API_URL","https://api.val.town",[],"preferencesKeybindings","default","preferencesTheme","light","preferencesFixers","preferencesCopilot","preferencesSelectionAi","preferencesCloseBrackets","preferencesChatPanelState","preferencesDprint","preferencesDenofmt","notificationCount","flagDenols","isBusinessOrgMember","businessOrgHandle","isBusinessOrgAdmin","isHigherTownieTier","navCollapsed","currentOrgId"]

Diff · v69