[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8,"_9":10},"actionData","errors","root",{"_325":326,"_327":328},"routes/_app.x.$handle.$name",{"_99":98,"_32":44,"_107":89,"_27":108,"_13":14,"_109":-5,"_11":110,"_111":112,"_113":114,"_115":116,"_117":118,"_20":-5,"_119":-5,"_120":-5,"_121":25,"_122":123,"_124":58,"_125":-5,"_126":-5,"_127":128},"routes/_app.x.$handle.$name.branch.$branch.version.$version.diff",{"_11":12,"_13":14,"_15":16,"_17":-5,"_18":-5},"project",{"_42":43,"_32":44,"_37":45,"_46":-5,"_47":25,"_48":49,"_50":49,"_51":52,"_53":-5,"_18":54,"_55":56,"_57":58,"_59":-5,"_60":-5,"_61":62,"_63":64,"_65":66,"_67":68,"_69":70,"_71":-5,"_72":73,"_74":73,"_75":76,"_77":76,"_78":76,"_79":73,"_80":73,"_81":82,"_83":84,"_85":86,"_87":25},"version",117,"changes",[19],"aiChatInfo","userId",{"_20":21,"_22":23,"_24":25,"_26":25,"_27":28,"_29":28},"file",{"_30":31,"_32":28,"_33":20,"_34":36,"_37":38,"_39":14,"_40":41,"_24":25},"previous",{"_30":31,"_32":28,"_33":20,"_34":35},"deleted",false,"deletedHere","path","updates.md","latestPath","fileId","ab11182a-aaf9-11f1-9fff-1607ee4eb77e","name","fileType","content","# mibbs-bot Updates\n\nTrack feature ideas and implementation status here. Mark items `[x]` when implemented.\n\n---\n\n## Implemented Features\n\n### [x] Email Forwarding\nForward any email to the bot's Val Town email address. Claude reads the email, takes action (add to calendar, create task, store info), posts a summary to the Telegram group, and replies to the sender confirming what was done. The forwarder can add instructions at the top of the email body (e.g. \"add this to calendar\" or \"just FYI\") — these are parsed out and followed.\n\n**Files:** `email.ts` (Val Town email type)\n\n---\n\n### [x] Timed Reminders + Task Owners\nTasks now track an **owner** (jeff/marisa/both) and an optional **reminder time**. A new 15-minute cron (`reminder-cron.ts`) fires a proactive Telegram ping — with a real @mention — when a task's reminder time arrives. Owner→Telegram-ID mapping is auto-learned from `message.from` on every inbound message (no config required). Timed tasks ping at the exact specified time; date-only tasks ping at 8am ET. The `update_task` tool lets you assign or reschedule after creation.\n\nThis delivers the owner-tracking half of the [Responsibility Model](#-responsibility-model) backlog item.\n\n---\n\n## Planned Features\n\n### [ ] Smart Responses (LLM Response Gating)\nToday the bot only replies in the group when @mentioned (unless `RESPOND_TO_ALL=true`, which is noisy). @mention is reliable but cumbersome. Add a cheap, fast per-message LLM \"gate\" that decides whether the bot *should* respond when it wasn't tagged — because the message is directed at it, is a reply to one of its messages, continues a conversation with it, or is something it can clearly help with. Posture is **conservative**: only chime in when fairly clearly warranted, to avoid butting into ordinary chatter. A third mode between \"@mention only\" and \"respond to all.\"\n\n**Approach — two-stage gate for non-mention group messages:**\n1. **Free deterministic short-circuit:** a direct reply to a bot message (`message.reply_to_message.from.is_bot`) always gets a response — zero LLM cost.\n2. **Cheap LLM classifier:** returns `{respond, confidence, reason}` via a forced tool call, gated by a confidence threshold (default 0.8 = conservative). Fail-safe: any error/parse failure ⇒ stay silent (matches today's default). Provider-flexible — reuses the existing `isGptModel()` branch so it works on a `gpt-*` or `claude-*` gate model.\n\n**Implementation steps:**\n1. `llm.ts` — add config exports:\n - `SMART_RESPONSES = (Deno.env.get(\"SMART_RESPONSES\") ?? \"true\") === \"true\"`\n - `GATE_MODEL = Deno.env.get(\"GATE_MODEL\") ?? \"claude-haiku-4-5\"` (cheapest; can point at any gpt-*/claude-* id)\n - `GATE_CONFIDENCE_THRESHOLD = Number(Deno.env.get(\"GATE_CONFIDENCE_THRESHOLD\") ?? \"0.8\")`\n2. `llm.ts` — add `classifyShouldRespond({ text, history, botUsername }) => { respond, confidence, reason }`:\n - Conservative system prompt describing capabilities (tasks/calendar/info/weather) + criteria; return `respond: true` only when clearly directed at the bot, clearly continuing a thread with it, or an unambiguous actionable request; quiet on chatter / messages aimed at the other person / ambiguity.\n - Structured output via a forced single tool `decide` (schema `{respond: boolean, confidence: number, reason: string}`):\n - Anthropic branch: `tool_choice: {type: \"tool\", name: \"decide\"}`, `max_tokens: 128`, parse `tool_use.input`.\n - GPT branch: mirror the `OPENAI_TOOLS_RESPONSES` shape, forced `tool_choice`, `max_output_tokens: 128`, `reasoning: {effort: \"low\"}`, parse function-call `arguments`.\n - Fail-safe default `{respond: false, confidence: 0}`; `console.log` the decision + reason.\n - Pass last ~8 messages; mark image-only messages as `[photo]` so the gate stays text-only/cheap.\n3. `main.ts` — replace the gate at lines 784-788: if group && not `RESPOND_TO_ALL` && not @mentioned → respond free if `replyToBot`; else if `SMART_RESPONSES` run `classifyShouldRespond` and only proceed when `respond && confidence >= GATE_CONFIDENCE_THRESHOLD`; else legacy `return new Response(\"OK\")`. Import the new symbols. Downstream (typing, photo download, `processMessage`) unchanged — the full model call only happens once a message clears the gate, same as @mentions today.\n4. New env vars (all optional): `SMART_RESPONSES` (default true), `GATE_MODEL` (default claude-haiku-4-5), `GATE_CONFIDENCE_THRESHOLD` (default 0.8). Add to `CLAUDE.md` env table.\n5. Changelog: minor release within the Blue era → **Cerulean (v2.2)**. Update `changelog.md` + `CHANGELOG_LATEST` in `main.ts`. No Telegram announcement (minor).\n\n**Cost estimate:** per check ≈ ~800 input + ~40 output tokens (model-independent — multiply by your gate model's rate). At the default Haiku 4.5 ($1 in / $5 out per MTok) ≈ **$0.001/check**:\n\n| Volume (non-mention group msgs) | Checks/mo | Cost/mo |\n|---|---|---|\n| Light (~20/day) | ~600 | ~$0.60 |\n| Moderate (~50/day) | ~1,500 | ~$1.50 |\n| Heavy (~150/day) | ~4,500 | ~$4.50 |\n\nDirect replies to the bot cost $0 (deterministic path). The gate is strictly *new* cost — it replaces silent drops with a ~$0.001 check; only passing messages hit the main model, same as @mentions today (~10–50× cheaper than a full response turn). Prompt caching won't help (stable prefix ~370 tokens is below the min cacheable prefix; volume is low). Note: `MODEL` is currently a GPT model whose current per-token rate is not quoted here — apply your OpenAI rate to the ~800-in/~40-out figures if keeping the gate on a `gpt-*` model.\n\n**Verification:**\n1. Deploy with `SMART_RESPONSES=true`; confirm the gate model's API key is set.\n2. In group (not @mentioned): obvious request (\"someone add milk to the list\") → responds; direct reply to a bot message → responds (logs show no gate call); ordinary chatter (\"running 5 min late\") → silent.\n3. Check Val Town logs for `gate: respond=… conf=… — …` to confirm/tune the threshold.\n4. Confirm @mention and `RESPOND_TO_ALL=true` still work unchanged.\n\n---\n\n### [ ] Unstructured Notes Storage\nStore freeform text under a topic (e.g. \"store these ideas for our 2027 ski trip\"), distinct from the key/value `fm_info` table.\n\n**Implementation steps:**\n1. Add `fm_notes` table: `id, topic, content, created_by, created_at`\n2. Add migration to `initDb()` for the new table\n3. Add `addNote(topic, content, createdBy)` and `searchNotes(query)` db functions\n4. Add two tools to `TOOLS` array in `main.ts`:\n - `add_note(topic, content)` — store freeform notes under a topic label\n - `search_notes(query)` — full-text search across topic and content\n5. Add cases for both in the `executeTool` switch\n\n---\n\n### [x] Weather Lookup\nNatural-language weather queries: \"what's the weather in Boston this weekend?\"\n\n**Implementation steps:**\n1. Add `get_weather(location, days_ahead?)` tool to `TOOLS` array\n2. Implement `getWeather(location, daysAhead)` using Open-Meteo (no API key required):\n - Geocode location name: `https://geocoding-api.open-meteo.com/v1/search`\n - Fetch forecast: `https://api.open-meteo.com/v1/forecast`\n3. Format response as a readable summary (high/low, condition, precipitation)\n4. Add case in `executeTool` switch\n5. No new env vars required\n\n---\n\n### [x] Edit Existing Calendar Events\nUpdate an upcoming event's title, start/end time, or description via natural language — \"move the dentist appointment to 3pm\" or \"rename Mara's recital to Spring Concert\".\n\n**Files:** `main.ts` — `findCalendarEventByQuery`, `updateCalendarEvent`, `update_calendar_event` tool\n\n---\n\n### [x] Image Parsing (Vision)\nSend a photo to the bot and Claude will describe it, extract text, read a receipt, parse a whiteboard, etc.\n\n**Implementation steps:**\n1. In the main handler (`main.ts` around line 468), detect `message.photo` in addition to `message.text`\n2. Add `downloadTelegramPhoto(fileId)` function: calls `getFile` API then fetches the image bytes\n3. Convert image to base64 and pass to Claude as an `image` content block in the user message\n4. Claude responds naturally — no explicit tool needed; can still call existing tools (e.g. `add_task` items parsed from a photo of a handwritten list)\n5. Handle caption text alongside the image if present (`message.caption`)\n\n---\n\n### [ ] LLM Evals\nLightweight regression testing for the LLM's tool-routing and response quality.\n\n**Approach: two layers**\n\n**Layer 1 — Trace logging (production data, ~30min)**\nAdd an `fm_evals` SQLite table that captures every real LLM interaction: input message, tools called, and final response. Zero test-writing upfront; surfaces failure patterns from real traffic. Add a `/evals` command or HTTP endpoint to browse the log.\n\n**Layer 2 — Golden test val (~1h)**\nCreate `evals.ts` as a script val with ~8–10 fixed test cases that verify tool routing:\n\n```typescript\nconst cases = [\n { input: \"Jeff: add milk to the grocery list\", expect_tools: [\"add_task\"] },\n { input: \"what's on the calendar this week?\", expect_tools: [\"get_calendar_events\"] },\n { input: \"remember the wifi password is Hunter2\", expect_tools: [\"store_info\"] },\n { input: \"mark task 5 as done\", expect_tools: [\"complete_task\"] },\n { input: \"hey how's everyone doing\", expect_tools: [] }, // should NOT call tools\n];\n// For each: call Anthropic with the same system prompt + test input,\n// assert tool_use blocks match expected tools, log PASS/FAIL to console + SQLite\n```\n\nRun manually or on a cron; results visible in Val Town logs.\n\n**Implementation steps:**\n1. Add `fm_evals` table: `id, input, tools_called, response, created_at, passed`\n2. In `processMessage`, after each completed agentic loop, insert a row into `fm_evals`\n3. Create `evals.ts` script val importing the same `TOOLS` and system prompt from `main.ts`\n4. Write 8–10 golden test cases covering each tool + the no-tool case\n5. Run each case, compare actual `tool_use` blocks to expected, write results to `fm_evals`\n6. Optionally add a `/evals` slash command in `main.ts` to show recent pass/fail counts\n\n---\n\n### [ ] Image → Calendar/Task Extraction\nExtend existing vision to automatically extract structured events and todos from photos — school flyers, birthday invitations, sports schedules, camp confirmation emails, etc.\n\nThe current image parsing describes what's in a photo; this goes further: Claude identifies dates, times, places, and required actions, then proposes adding them to the calendar or task list in a single step. The chat becomes the family inbox — you shouldn't need to decide whether something goes on the calendar or the task list.\n\n**Implementation steps:**\n1. Extend the image system prompt to explicitly ask Claude: \"If this image contains dates, events, deadlines, or action items, propose adding them using `add_calendar_event`, `add_task`, or `add_note`\"\n2. Add optional `confirm_before_adding` flag in the message handler so Claude can ask \"Found: Mara's fall concert Nov 14 at 7pm. Add it?\" before writing\n3. Handle caption text as additional context (\"this is Cass's soccer schedule\")\n4. Test against: school flyers, party invitations, prescription bottles, forms with due dates\n\n---\n\n### [ ] Open Loops (Unresolved Family Decisions)\nTrack things that need figuring out but have no date attached — distinct from tasks (\"do this\") and calendar events (\"be there at X\"). Examples: \"we need to figure out summer camp\", \"decide on Mara's birthday party venue\", \"sort out ski rentals for Vermont trip\".\n\nOpen loops persist until someone explicitly closes them. The bot surfaces them in the Sunday weekly briefing and when contextually relevant (e.g., upcoming trip → check open loops for that trip).\n\n**Implementation steps:**\n1. Add `fm_loops` table: `id, title, context, status (open/closed), created_by, created_at, closed_at`\n2. Add tools:\n - `add_open_loop(title, context?)` — create an unresolved item\n - `close_loop(id_or_title)` — mark resolved\n - `list_open_loops()` — show all unresolved items\n3. Claude should recognize loop-shaped language (\"we need to figure out\", \"still need to decide\", \"haven't sorted\") and proactively offer to add a loop\n4. Include open loops in the Sunday briefing output (see Proactive Briefings feature)\n\n---\n\n### [ ] Proactive Scheduled Briefings\nTwo recurring messages that turn the bot from reactive to proactive:\n\n**Morning briefing (daily, ~7am):** \"Today: Mara school 8:15; Cass daycare; Jeff has late meeting until 5:30; rain at 3 — bring raincoats. Mara needs her library book today.\"\n\n**Sunday week-ahead (~8am):** Reconciles both calendars + memory + open loops. Doesn't just list events — answers \"what does this week require us to do?\" Example output: \"Mara's birthday party Saturday → gift not bought yet; Jeff out Thursday evening → Marisa solo bedtime; ski weekend coming → Cass rentals unresolved.\"\n\n**Implementation steps:**\n1. Create `cron.ts` (or extend existing) with two interval jobs\n2. Morning briefing: fetch today's calendar events, check weather, query open tasks with today's due date, surface any prep notes from `fm_info`/`fm_notes`\n3. Sunday briefing: fetch full week's events, run Claude with the prompt \"what does this week require the family to prepare, decide, or do?\" — pulling from calendar + tasks + open loops + family memory\n4. Send to the family Telegram chat via `sendTelegramMessage`\n5. Make briefing opt-in via `/briefing on|off` slash command stored in `fm_info`\n\n---\n\n### [ ] Family Memory (Structured)\nA richer layer on top of `fm_info` for the facts family life constantly needs: kids' clothing/shoe sizes, school schedules, babysitter contact info, gift ideas, activities the kids love, dietary restrictions, packing preferences, places you've stayed, things Marisa or Jeff have mentioned wanting.\n\nThe goal: natural queries work years later. \"What size shoes does Mara wear?\" \"What was that restaurant in Rotterdam the kids loved?\" \"What are gift ideas for Cass?\"\n\n**Implementation steps:**\n1. Add `fm_memory` table: `id, person, category, key, value, created_at, updated_at`\n - `person`: \"mara\", \"cass\", \"jeff\", \"marisa\", \"family\"\n - `category`: \"sizes\", \"gifts\", \"activities\", \"places\", \"contacts\", \"preferences\"\n2. Add tools:\n - `remember(person, category, key, value)` — store a memory\n - `recall(person?, category?, query)` — retrieve matching memories\n3. Claude should recognize memory-worthy statements (\"Cass is obsessed with trains\", \"Mara said she wants a telescope\") and proactively store them\n4. Memories feed into trip planning, outing suggestions, and gift tracking\n5. Supersedes some ad-hoc `fm_info` usage — migrate key entries on first run\n\n---\n\n### [ ] Trip Planner\nMaintain a \"trip object\" in memory for each upcoming family trip. Tracks logistics as a checklist: lodging, tickets/passes, rentals, restaurant reservations, packing list. A few days before departure, auto-generates a packing list based on destination + weather forecast + planned activities + family ages.\n\n**Implementation steps:**\n1. Add `fm_trips` table: `id, name, destination, start_date, end_date, status, notes`\n2. Add `fm_trip_items` table: `id, trip_id, category (lodging/transport/activity/rental/other), description, status (open/done), owner`\n3. Add tools:\n - `create_trip(name, destination, start_date, end_date)` — create trip record\n - `add_trip_item(trip_name, category, description, owner?)` — add a checklist item\n - `get_trip_status(trip_name)` — show full trip with open/done items\n3. 3 days before `start_date`: auto-send trip status + generated packing list (via cron) using weather forecast for the destination\n4. Open loops referencing a trip (e.g., \"Cass rentals — Vermont\") should link to the trip object\n\n---\n\n### [ ] Web Search\nLet the bot look things up: local events, restaurant hours, camp registration dates, activity options, nearby attractions. Turns \"remember and coordinate\" into \"figure this out for us.\"\n\nThis is the unlock for: \"find a good Italian restaurant near us for Friday\", \"when does [summer camp] registration open?\", \"what's open on Columbus Day weekend?\", \"find something fun to do Saturday with a 5-year-old.\"\n\n**Implementation steps:**\n1. Add `web_search(query)` tool using a search API (Brave Search or Tavily — both have free tiers; store key in Val Town env)\n2. Claude decides when a query needs live information vs. stored memory\n3. Combine with weather and family memory for proactive suggestions: \"Saturday is sunny and free — here are 3 ideas based on what the kids like\"\n4. Add `BRAVE_SEARCH_API_KEY` or `TAVILY_API_KEY` to Val Town env vars\n\n---\n\n### [ ] Proactive Outing Suggestions\nOn Friday afternoons when the weekend has unscheduled time + good weather, the bot proactively sends outing ideas — drawing on family memory (what activities the kids like, places you've been) + weather + calendar gaps.\n\nAlso: flag weather-sensitive plans (\"rain forecast Saturday — your park plan might need a backup\"), first snow (\"sledding Sunday?\"), or local events (concert, festival) matching the family's interests.\n\n**Implementation steps:**\n1. Add a Friday ~3pm cron job that checks: Saturday/Sunday calendar for free blocks of 2+ hours; weather forecast for the weekend\n2. If a free block + decent weather exists, call Claude with: free times, weather, family memory (activities, kids' ages, recent outings), and ask for 2–3 suggestions\n3. Send to family chat: \"Saturday 10am–2pm looks free and sunny. A few ideas: [...]\"\n4. Requires Web Search tool for event discovery beyond stored memory\n5. Track \"recently done\" to avoid repeating suggestions (`fm_notes` topic: \"recent outings\")\n\n---\n\n### [ ] Responsibility Model\nTrack who owns what for a given event: preparation, transportation, and follow-through. When both parents are busy at the same time, the bot flags the conflict rather than leaving it implicit.\n\nExample: \"Mara's birthday party Saturday 2pm — Jeff driving, gift still needed, card still needed.\" Or: \"Jeff is out Thursday evening → Marisa is solo bedtime — confirm?\"\n\n**Implementation steps:**\n1. Add `owner` field to calendar events and tasks (values: \"jeff\", \"marisa\", \"both\", \"tbd\")\n2. Add tool: `assign_owner(event_or_task_id, owner)` — set responsibility\n3. In weekly briefing, for each event/task with `owner = tbd` or no owner, flag it explicitly\n4. When a new event is added that overlaps with an existing commitment for the same owner, surface the conflict: \"Jeff already has a meeting 5–6:30pm Thursday — who handles pickup?\"\n5. Eventually: \"I can do pickup Wednesday\" → Claude parses speaker identity (from Telegram user) and updates the owner field automatically\n","# mibbs-bot Updates\n\nTrack feature ideas and implementation status here. Mark items `[x]` when implemented.\n\n---\n\n## Implemented Features\n\n### [x] Email Forwarding\nForward any email to the bot's Val Town email address. Claude reads the email, takes action (add to calendar, create task, store info), posts a summary to the Telegram group, and replies to the sender confirming what was done. The forwarder can add instructions at the top of the email body (e.g. \"add this to calendar\" or \"just FYI\") — these are parsed out and followed.\n\n**Files:** `email.ts` (Val Town email type)\n\n---\n\n### [x] Timed Reminders + Task Owners\nTasks now track an **owner** (jeff/marisa/both) and an optional **reminder time**. A new 15-minute cron (`reminder-cron.ts`) fires a proactive Telegram ping — with a real @mention — when a task's reminder time arrives. Owner→Telegram-ID mapping is auto-learned from `message.from` on every inbound message (no config required). Timed tasks ping at the exact specified time; date-only tasks ping at 8am ET. The `update_task` tool lets you assign or reschedule after creation.\n\nThis delivers the owner-tracking half of the [Responsibility Model](#-responsibility-model) backlog item.\n\n---\n\n## Implemented Features\n\n### [x] Emoji Reactions\nWhen the bot completes a simple, unambiguous action (adding a task, setting a reminder, adding a calendar event, storing info), it now acknowledges with an emoji reaction on the triggering message instead of writing a full sentence. Questions, lookups, and anything that needs a conversational response still get text replies.\n\n**How it works:** A `react` tool in `tools.ts` calls `setMessageReaction` (Telegram Bot API 7.0+), attaching the emoji directly to the user's message. If Telegram rejects the reaction (group settings, old message), it falls back to sending the emoji as a plain message. The LLM decides which mode fits based on guidance in the system prompt.\n\n**Files:** `tools.ts` (react tool + ToolCtx type), `main.ts` (setMessageReaction helper, updated processMessage, updated system prompt)\n\n---\n\n## Planned Features\n\n### [ ] Smart Responses (LLM Response Gating)\nToday the bot only replies in the group when @mentioned (unless `RESPOND_TO_ALL=true`, which is noisy). @mention is reliable but cumbersome. Add a cheap, fast per-message LLM \"gate\" that decides whether the bot *should* respond when it wasn't tagged — because the message is directed at it, is a reply to one of its messages, continues a conversation with it, or is something it can clearly help with. Posture is **conservative**: only chime in when fairly clearly warranted, to avoid butting into ordinary chatter. A third mode between \"@mention only\" and \"respond to all.\"\n\n**Approach — two-stage gate for non-mention group messages:**\n1. **Free deterministic short-circuit:** a direct reply to a bot message (`message.reply_to_message.from.is_bot`) always gets a response — zero LLM cost.\n2. **Cheap LLM classifier:** returns `{respond, confidence, reason}` via a forced tool call, gated by a confidence threshold (default 0.8 = conservative). Fail-safe: any error/parse failure ⇒ stay silent (matches today's default). Provider-flexible — reuses the existing `isGptModel()` branch so it works on a `gpt-*` or `claude-*` gate model.\n\n**Implementation steps:**\n1. `llm.ts` — add config exports:\n - `SMART_RESPONSES = (Deno.env.get(\"SMART_RESPONSES\") ?? \"true\") === \"true\"`\n - `GATE_MODEL = Deno.env.get(\"GATE_MODEL\") ?? \"claude-haiku-4-5\"` (cheapest; can point at any gpt-*/claude-* id)\n - `GATE_CONFIDENCE_THRESHOLD = Number(Deno.env.get(\"GATE_CONFIDENCE_THRESHOLD\") ?? \"0.8\")`\n2. `llm.ts` — add `classifyShouldRespond({ text, history, botUsername }) => { respond, confidence, reason }`:\n - Conservative system prompt describing capabilities (tasks/calendar/info/weather) + criteria; return `respond: true` only when clearly directed at the bot, clearly continuing a thread with it, or an unambiguous actionable request; quiet on chatter / messages aimed at the other person / ambiguity.\n - Structured output via a forced single tool `decide` (schema `{respond: boolean, confidence: number, reason: string}`):\n - Anthropic branch: `tool_choice: {type: \"tool\", name: \"decide\"}`, `max_tokens: 128`, parse `tool_use.input`.\n - GPT branch: mirror the `OPENAI_TOOLS_RESPONSES` shape, forced `tool_choice`, `max_output_tokens: 128`, `reasoning: {effort: \"low\"}`, parse function-call `arguments`.\n - Fail-safe default `{respond: false, confidence: 0}`; `console.log` the decision + reason.\n - Pass last ~8 messages; mark image-only messages as `[photo]` so the gate stays text-only/cheap.\n3. `main.ts` — replace the gate at lines 784-788: if group && not `RESPOND_TO_ALL` && not @mentioned → respond free if `replyToBot`; else if `SMART_RESPONSES` run `classifyShouldRespond` and only proceed when `respond && confidence >= GATE_CONFIDENCE_THRESHOLD`; else legacy `return new Response(\"OK\")`. Import the new symbols. Downstream (typing, photo download, `processMessage`) unchanged — the full model call only happens once a message clears the gate, same as @mentions today.\n4. New env vars (all optional): `SMART_RESPONSES` (default true), `GATE_MODEL` (default claude-haiku-4-5), `GATE_CONFIDENCE_THRESHOLD` (default 0.8). Add to `CLAUDE.md` env table.\n5. Changelog: minor release within the Blue era → **Cerulean (v2.2)**. Update `changelog.md` + `CHANGELOG_LATEST` in `main.ts`. No Telegram announcement (minor).\n\n**Cost estimate:** per check ≈ ~800 input + ~40 output tokens (model-independent — multiply by your gate model's rate). At the default Haiku 4.5 ($1 in / $5 out per MTok) ≈ **$0.001/check**:\n\n| Volume (non-mention group msgs) | Checks/mo | Cost/mo |\n|---|---|---|\n| Light (~20/day) | ~600 | ~$0.60 |\n| Moderate (~50/day) | ~1,500 | ~$1.50 |\n| Heavy (~150/day) | ~4,500 | ~$4.50 |\n\nDirect replies to the bot cost $0 (deterministic path). The gate is strictly *new* cost — it replaces silent drops with a ~$0.001 check; only passing messages hit the main model, same as @mentions today (~10–50× cheaper than a full response turn). Prompt caching won't help (stable prefix ~370 tokens is below the min cacheable prefix; volume is low). Note: `MODEL` is currently a GPT model whose current per-token rate is not quoted here — apply your OpenAI rate to the ~800-in/~40-out figures if keeping the gate on a `gpt-*` model.\n\n**Verification:**\n1. Deploy with `SMART_RESPONSES=true`; confirm the gate model's API key is set.\n2. In group (not @mentioned): obvious request (\"someone add milk to the list\") → responds; direct reply to a bot message → responds (logs show no gate call); ordinary chatter (\"running 5 min late\") → silent.\n3. Check Val Town logs for `gate: respond=… conf=… — …` to confirm/tune the threshold.\n4. Confirm @mention and `RESPOND_TO_ALL=true` still work unchanged.\n\n---\n\n### [ ] Unstructured Notes Storage\nStore freeform text under a topic (e.g. \"store these ideas for our 2027 ski trip\"), distinct from the key/value `fm_info` table.\n\n**Implementation steps:**\n1. Add `fm_notes` table: `id, topic, content, created_by, created_at`\n2. Add migration to `initDb()` for the new table\n3. Add `addNote(topic, content, createdBy)` and `searchNotes(query)` db functions\n4. Add two tools to `TOOLS` array in `main.ts`:\n - `add_note(topic, content)` — store freeform notes under a topic label\n - `search_notes(query)` — full-text search across topic and content\n5. Add cases for both in the `executeTool` switch\n\n---\n\n### [x] Weather Lookup\nNatural-language weather queries: \"what's the weather in Boston this weekend?\"\n\n**Implementation steps:**\n1. Add `get_weather(location, days_ahead?)` tool to `TOOLS` array\n2. Implement `getWeather(location, daysAhead)` using Open-Meteo (no API key required):\n - Geocode location name: `https://geocoding-api.open-meteo.com/v1/search`\n - Fetch forecast: `https://api.open-meteo.com/v1/forecast`\n3. Format response as a readable summary (high/low, condition, precipitation)\n4. Add case in `executeTool` switch\n5. No new env vars required\n\n---\n\n### [x] Edit Existing Calendar Events\nUpdate an upcoming event's title, start/end time, or description via natural language — \"move the dentist appointment to 3pm\" or \"rename Mara's recital to Spring Concert\".\n\n**Files:** `main.ts` — `findCalendarEventByQuery`, `updateCalendarEvent`, `update_calendar_event` tool\n\n---\n\n### [x] Image Parsing (Vision)\nSend a photo to the bot and Claude will describe it, extract text, read a receipt, parse a whiteboard, etc.\n\n**Implementation steps:**\n1. In the main handler (`main.ts` around line 468), detect `message.photo` in addition to `message.text`\n2. Add `downloadTelegramPhoto(fileId)` function: calls `getFile` API then fetches the image bytes\n3. Convert image to base64 and pass to Claude as an `image` content block in the user message\n4. Claude responds naturally — no explicit tool needed; can still call existing tools (e.g. `add_task` items parsed from a photo of a handwritten list)\n5. Handle caption text alongside the image if present (`message.caption`)\n\n---\n\n### [ ] LLM Evals\nLightweight regression testing for the LLM's tool-routing and response quality.\n\n**Approach: two layers**\n\n**Layer 1 — Trace logging (production data, ~30min)**\nAdd an `fm_evals` SQLite table that captures every real LLM interaction: input message, tools called, and final response. Zero test-writing upfront; surfaces failure patterns from real traffic. Add a `/evals` command or HTTP endpoint to browse the log.\n\n**Layer 2 — Golden test val (~1h)**\nCreate `evals.ts` as a script val with ~8–10 fixed test cases that verify tool routing:\n\n```typescript\nconst cases = [\n { input: \"Jeff: add milk to the grocery list\", expect_tools: [\"add_task\"] },\n { input: \"what's on the calendar this week?\", expect_tools: [\"get_calendar_events\"] },\n { input: \"remember the wifi password is Hunter2\", expect_tools: [\"store_info\"] },\n { input: \"mark task 5 as done\", expect_tools: [\"complete_task\"] },\n { input: \"hey how's everyone doing\", expect_tools: [] }, // should NOT call tools\n];\n// For each: call Anthropic with the same system prompt + test input,\n// assert tool_use blocks match expected tools, log PASS/FAIL to console + SQLite\n```\n\nRun manually or on a cron; results visible in Val Town logs.\n\n**Implementation steps:**\n1. Add `fm_evals` table: `id, input, tools_called, response, created_at, passed`\n2. In `processMessage`, after each completed agentic loop, insert a row into `fm_evals`\n3. Create `evals.ts` script val importing the same `TOOLS` and system prompt from `main.ts`\n4. Write 8–10 golden test cases covering each tool + the no-tool case\n5. Run each case, compare actual `tool_use` blocks to expected, write results to `fm_evals`\n6. Optionally add a `/evals` slash command in `main.ts` to show recent pass/fail counts\n\n---\n\n### [ ] Image → Calendar/Task Extraction\nExtend existing vision to automatically extract structured events and todos from photos — school flyers, birthday invitations, sports schedules, camp confirmation emails, etc.\n\nThe current image parsing describes what's in a photo; this goes further: Claude identifies dates, times, places, and required actions, then proposes adding them to the calendar or task list in a single step. The chat becomes the family inbox — you shouldn't need to decide whether something goes on the calendar or the task list.\n\n**Implementation steps:**\n1. Extend the image system prompt to explicitly ask Claude: \"If this image contains dates, events, deadlines, or action items, propose adding them using `add_calendar_event`, `add_task`, or `add_note`\"\n2. Add optional `confirm_before_adding` flag in the message handler so Claude can ask \"Found: Mara's fall concert Nov 14 at 7pm. Add it?\" before writing\n3. Handle caption text as additional context (\"this is Cass's soccer schedule\")\n4. Test against: school flyers, party invitations, prescription bottles, forms with due dates\n\n---\n\n### [ ] Open Loops (Unresolved Family Decisions)\nTrack things that need figuring out but have no date attached — distinct from tasks (\"do this\") and calendar events (\"be there at X\"). Examples: \"we need to figure out summer camp\", \"decide on Mara's birthday party venue\", \"sort out ski rentals for Vermont trip\".\n\nOpen loops persist until someone explicitly closes them. The bot surfaces them in the Sunday weekly briefing and when contextually relevant (e.g., upcoming trip → check open loops for that trip).\n\n**Implementation steps:**\n1. Add `fm_loops` table: `id, title, context, status (open/closed), created_by, created_at, closed_at`\n2. Add tools:\n - `add_open_loop(title, context?)` — create an unresolved item\n - `close_loop(id_or_title)` — mark resolved\n - `list_open_loops()` — show all unresolved items\n3. Claude should recognize loop-shaped language (\"we need to figure out\", \"still need to decide\", \"haven't sorted\") and proactively offer to add a loop\n4. Include open loops in the Sunday briefing output (see Proactive Briefings feature)\n\n---\n\n### [ ] Proactive Scheduled Briefings\nTwo recurring messages that turn the bot from reactive to proactive:\n\n**Morning briefing (daily, ~7am):** \"Today: Mara school 8:15; Cass daycare; Jeff has late meeting until 5:30; rain at 3 — bring raincoats. Mara needs her library book today.\"\n\n**Sunday week-ahead (~8am):** Reconciles both calendars + memory + open loops. Doesn't just list events — answers \"what does this week require us to do?\" Example output: \"Mara's birthday party Saturday → gift not bought yet; Jeff out Thursday evening → Marisa solo bedtime; ski weekend coming → Cass rentals unresolved.\"\n\n**Implementation steps:**\n1. Create `cron.ts` (or extend existing) with two interval jobs\n2. Morning briefing: fetch today's calendar events, check weather, query open tasks with today's due date, surface any prep notes from `fm_info`/`fm_notes`\n3. Sunday briefing: fetch full week's events, run Claude with the prompt \"what does this week require the family to prepare, decide, or do?\" — pulling from calendar + tasks + open loops + family memory\n4. Send to the family Telegram chat via `sendTelegramMessage`\n5. Make briefing opt-in via `/briefing on|off` slash command stored in `fm_info`\n\n---\n\n### [ ] Family Memory (Structured)\nA richer layer on top of `fm_info` for the facts family life constantly needs: kids' clothing/shoe sizes, school schedules, babysitter contact info, gift ideas, activities the kids love, dietary restrictions, packing preferences, places you've stayed, things Marisa or Jeff have mentioned wanting.\n\nThe goal: natural queries work years later. \"What size shoes does Mara wear?\" \"What was that restaurant in Rotterdam the kids loved?\" \"What are gift ideas for Cass?\"\n\n**Implementation steps:**\n1. Add `fm_memory` table: `id, person, category, key, value, created_at, updated_at`\n - `person`: \"mara\", \"cass\", \"jeff\", \"marisa\", \"family\"\n - `category`: \"sizes\", \"gifts\", \"activities\", \"places\", \"contacts\", \"preferences\"\n2. Add tools:\n - `remember(person, category, key, value)` — store a memory\n - `recall(person?, category?, query)` — retrieve matching memories\n3. Claude should recognize memory-worthy statements (\"Cass is obsessed with trains\", \"Mara said she wants a telescope\") and proactively store them\n4. Memories feed into trip planning, outing suggestions, and gift tracking\n5. Supersedes some ad-hoc `fm_info` usage — migrate key entries on first run\n\n---\n\n### [ ] Trip Planner\nMaintain a \"trip object\" in memory for each upcoming family trip. Tracks logistics as a checklist: lodging, tickets/passes, rentals, restaurant reservations, packing list. A few days before departure, auto-generates a packing list based on destination + weather forecast + planned activities + family ages.\n\n**Implementation steps:**\n1. Add `fm_trips` table: `id, name, destination, start_date, end_date, status, notes`\n2. Add `fm_trip_items` table: `id, trip_id, category (lodging/transport/activity/rental/other), description, status (open/done), owner`\n3. Add tools:\n - `create_trip(name, destination, start_date, end_date)` — create trip record\n - `add_trip_item(trip_name, category, description, owner?)` — add a checklist item\n - `get_trip_status(trip_name)` — show full trip with open/done items\n3. 3 days before `start_date`: auto-send trip status + generated packing list (via cron) using weather forecast for the destination\n4. Open loops referencing a trip (e.g., \"Cass rentals — Vermont\") should link to the trip object\n\n---\n\n### [ ] Web Search\nLet the bot look things up: local events, restaurant hours, camp registration dates, activity options, nearby attractions. Turns \"remember and coordinate\" into \"figure this out for us.\"\n\nThis is the unlock for: \"find a good Italian restaurant near us for Friday\", \"when does [summer camp] registration open?\", \"what's open on Columbus Day weekend?\", \"find something fun to do Saturday with a 5-year-old.\"\n\n**Implementation steps:**\n1. Add `web_search(query)` tool using a search API (Brave Search or Tavily — both have free tiers; store key in Val Town env)\n2. Claude decides when a query needs live information vs. stored memory\n3. Combine with weather and family memory for proactive suggestions: \"Saturday is sunny and free — here are 3 ideas based on what the kids like\"\n4. Add `BRAVE_SEARCH_API_KEY` or `TAVILY_API_KEY` to Val Town env vars\n\n---\n\n### [ ] Proactive Outing Suggestions\nOn Friday afternoons when the weekend has unscheduled time + good weather, the bot proactively sends outing ideas — drawing on family memory (what activities the kids like, places you've been) + weather + calendar gaps.\n\nAlso: flag weather-sensitive plans (\"rain forecast Saturday — your park plan might need a backup\"), first snow (\"sledding Sunday?\"), or local events (concert, festival) matching the family's interests.\n\n**Implementation steps:**\n1. Add a Friday ~3pm cron job that checks: Saturday/Sunday calendar for free blocks of 2+ hours; weather forecast for the weekend\n2. If a free block + decent weather exists, call Claude with: free times, weather, family memory (activities, kids' ages, recent outings), and ask for 2–3 suggestions\n3. Send to family chat: \"Saturday 10am–2pm looks free and sunny. A few ideas: [...]\"\n4. Requires Web Search tool for event discovery beyond stored memory\n5. Track \"recently done\" to avoid repeating suggestions (`fm_notes` topic: \"recent outings\")\n\n---\n\n### [ ] Responsibility Model\nTrack who owns what for a given event: preparation, transportation, and follow-through. When both parents are busy at the same time, the bot flags the conflict rather than leaving it implicit.\n\nExample: \"Mara's birthday party Saturday 2pm — Jeff driving, gift still needed, card still needed.\" Or: \"Jeff is out Thursday evening → Marisa is solo bedtime — confirm?\"\n\n**Implementation steps:**\n1. Add `owner` field to calendar events and tasks (values: \"jeff\", \"marisa\", \"both\", \"tbd\")\n2. Add tool: `assign_owner(event_or_task_id, owner)` — set responsibility\n3. In weekly briefing, for each event/task with `owner = tbd` or no owner, flag it explicitly\n4. When a new event is added that overlaps with an existing commitment for the same owner, surface the conflict: \"Jeff already has a meeting 5–6:30pm Thursday — who handles pickup?\"\n5. Eventually: \"I can do pickup Wednesday\" → Claude parses speaker identity (from Telegram user) and updates the owner field automatically\n","createdAt",["D",1788988992452],"branchVersion","branchId","05facaa4-90f3-48b5-bf13-4bf8572bc454","id","ea90a44b-e5c3-4abb-8fb5-1fdf1a363f60","mibbs-bot",["D",1788733325122],"deletedAt","legacyEnvVars","privacy","public","httpPrivacy","description","mibbs bot code","imageUrl","e8e1fec0-46e7-11f0-a801-9b9c2b15cc6a","pinnedFileId","63f1ffa8-aa41-11f1-aa27-1607ee4eb77e","pinPreview",true,"pinMigrated","screenshotVersion","likeCount",0,"orgId","7a1a376d-d8d2-47ee-b61d-1016a5b7e33d","tursoName","ea90a44be5c34abb8fb51fdf1a363f60","tursoHostname","ea90a44be5c34abb8fb51fdf1a363f60-val-town.aws-us-east-2.turso.io","tursoMeasuredAt",["D",1789522252250],"pinned","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm","user",{"_99":98,"_106":102},"org",{"_99":98,"_100":-5,"_101":102,"_18":54,"_103":104,"_105":25},"branch",{"_13":88,"_32":89,"_42":41,"_90":43,"_91":-5,"_92":-5,"_93":-5,"_94":-5,"_95":-5,"_96":44,"_97":98},"isOwned",130,"main","projectId","forkedBranchId","forkedBranchVersion","forkedProjectPrivacy","forkedProjectId","forkedProjectOrgId","projectName","userHandle","jeffvincent","handle","displayName","avatar","https://img.clerk.com/eyJ0eXBlIjoicHJveHkiLCJzcmMiOiJodHRwczovL2ltYWdlcy5jbGVyay5kZXYvb2F1dGhfZ2l0aHViL2ltZ18yeU41V0tQaE1XdnpOME5pR05pMVFRNkh0bUIifQ","tier","free","canRestrictVals","profileImageUrl","branchName","","adoptOrgId",{"_42":43,"_32":44,"_37":320,"_46":-5,"_47":25,"_48":49,"_50":49,"_51":52,"_53":-5,"_18":54,"_55":56,"_57":58,"_59":-5,"_60":-5,"_61":62,"_63":64,"_65":66,"_67":68,"_69":321,"_71":-5,"_72":73,"_74":73,"_75":76,"_77":76,"_78":76,"_79":73,"_80":73,"_81":322,"_83":323,"_85":324,"_87":25},"tree",[263,264,265,266,267,268,269,270,271,272,273,274,275,276],"files",[263,264,265,266,267,268,269,270,271,272,273,274,275,276],"counts",{"_252":62,"_253":242,"_254":62,"_255":256,"_257":258},"contents",[129,130,131,132,133,134,135,136,137,138,139,140,141,142],"upstreamProject","upstreamBranch","emptyEnvVars","flags",[],"hasHttpFiles","grants","orgs","customDomains",{},{"_42":245,"_32":246,"_30":247,"_146":-5,"_40":41,"_39":248,"_148":25,"_37":249,"_13":180,"_33":20,"_34":250,"_24":25,"_152":-5,"_27":251},{"_42":237,"_32":238,"_30":239,"_146":-5,"_40":41,"_39":240,"_148":58,"_37":241,"_13":242,"_33":150,"_34":243,"_24":25,"_152":-5,"_27":244},{"_42":230,"_32":231,"_30":232,"_146":-5,"_40":41,"_39":233,"_148":25,"_37":234,"_13":163,"_33":20,"_34":235,"_24":25,"_152":-5,"_27":236},{"_42":222,"_32":223,"_30":224,"_146":-5,"_40":41,"_39":225,"_148":58,"_37":226,"_13":227,"_33":172,"_34":228,"_24":25,"_152":-5,"_27":229},{"_42":215,"_32":216,"_30":217,"_146":-5,"_40":41,"_39":218,"_148":58,"_37":219,"_13":62,"_33":150,"_34":220,"_24":25,"_152":-5,"_27":221},{"_42":205,"_32":206,"_30":207,"_146":-5,"_40":41,"_39":208,"_148":58,"_37":209,"_13":210,"_33":211,"_34":212,"_24":25,"_152":213,"_27":214},{"_42":198,"_32":199,"_30":200,"_146":-5,"_40":41,"_39":201,"_148":25,"_37":202,"_13":62,"_33":150,"_34":203,"_24":25,"_152":-5,"_27":204},{"_42":191,"_32":192,"_30":193,"_146":-5,"_40":41,"_39":194,"_148":58,"_37":195,"_13":171,"_33":150,"_34":196,"_24":25,"_152":-5,"_27":197},{"_42":183,"_32":184,"_30":56,"_146":-5,"_40":41,"_39":185,"_148":25,"_37":186,"_13":187,"_33":188,"_34":189,"_24":25,"_152":-5,"_27":190},{"_42":175,"_32":176,"_30":177,"_146":-5,"_40":41,"_39":178,"_148":58,"_37":179,"_13":180,"_33":172,"_34":181,"_24":25,"_152":-5,"_27":182},{"_42":166,"_32":167,"_30":168,"_146":-5,"_40":41,"_39":169,"_148":58,"_37":170,"_13":171,"_33":172,"_34":173,"_24":25,"_152":-5,"_27":174},{"_42":158,"_32":159,"_30":160,"_146":-5,"_40":41,"_39":161,"_148":25,"_37":162,"_13":163,"_33":150,"_34":164,"_24":25,"_152":-5,"_27":165},{"_42":154,"_32":28,"_30":31,"_146":-5,"_40":41,"_39":14,"_148":25,"_37":155,"_13":156,"_33":20,"_34":36,"_24":25,"_152":-5,"_27":157},{"_42":143,"_32":144,"_30":145,"_146":-5,"_40":41,"_39":147,"_148":58,"_37":149,"_13":62,"_33":150,"_34":151,"_24":25,"_152":-5,"_27":153},"b4e21e3a-a304-42a3-8586-0e49f3d181fe","weather.ts","25aaa176-ac69-11f1-92fd-1607ee4eb77e","parentId",101,"latest",["D",1788970302763],"script","// Shared weather helpers for mibbs-bot using Open-Meteo (no API key required).\n\nexport const WMO_CODES: Record = {\n 0: \"Clear\", 1: \"Mainly clear\", 2: \"Partly cloudy\", 3: \"Overcast\",\n 45: \"Foggy\", 48: \"Foggy\", 51: \"Light drizzle\", 53: \"Drizzle\", 55: \"Heavy drizzle\",\n 61: \"Light rain\", 63: \"Rain\", 65: \"Heavy rain\", 71: \"Light snow\", 73: \"Snow\", 75: \"Heavy snow\",\n 77: \"Snow grains\", 80: \"Showers\", 81: \"Showers\", 82: \"Heavy showers\",\n 85: \"Snow showers\", 86: \"Heavy snow showers\", 95: \"Thunderstorm\", 96: \"Thunderstorm\", 99: \"Thunderstorm\",\n};\n\nexport function wmoDesc(code: number): string {\n return WMO_CODES[code] ?? WMO_CODES[Math.floor(code / 10) * 10] ?? \"Unknown\";\n}\n\nexport interface DailyForecast {\n date: string;\n hi: number;\n lo: number;\n precip: number;\n weathercode: number;\n}\n\n// Low-level: returns raw daily data so crons can format their own way.\nexport async function fetchForecast(\n location: string,\n days: number,\n): Promise<{ location: string; daily: DailyForecast[] } | null> {\n try {\n const geoResp = await fetch(\n `https://geocoding-api.open-meteo.com/v1/search?name=${encodeURIComponent(location)}&count=1&language=en&format=json`,\n );\n const geoData = await geoResp.json();\n if (!geoData.results?.length) return null;\n const { latitude, longitude, name, admin1, country } = geoData.results[0];\n const locationLabel = [name, admin1, country].filter(Boolean).join(\", \");\n const clampedDays = Math.min(Math.max(Math.round(days), 1), 7);\n const forecastResp = await fetch(\n `https://api.open-meteo.com/v1/forecast?latitude=${latitude}&longitude=${longitude}&daily=temperature_2m_max,temperature_2m_min,precipitation_sum,weathercode&temperature_unit=fahrenheit&precipitation_unit=inch&forecast_days=${clampedDays}&timezone=auto`,\n );\n const forecast = await forecastResp.json();\n if (!forecast.daily) return null;\n const { time, temperature_2m_max, temperature_2m_min, precipitation_sum, weathercode } = forecast.daily;\n const daily: DailyForecast[] = (time as string[]).map((date: string, i: number) => ({\n date,\n hi: Math.round(temperature_2m_max[i]),\n lo: Math.round(temperature_2m_min[i]),\n precip: precipitation_sum[i] as number,\n weathercode: weathercode[i],\n }));\n return { location: locationLabel, daily };\n } catch {\n return null;\n }\n}\n\n// High-level: formatted string returned as a tool result or cron message.\nexport async function getWeather(location: string, daysAhead = 3): Promise {\n const result = await fetchForecast(location, daysAhead);\n if (!result) return `Couldn't find location or fetch forecast for: ${location}`;\n const lines = result.daily.map((d) => {\n const label = new Date(d.date + \"T12:00:00\").toLocaleDateString(\"en-US\", {\n weekday: \"short\", month: \"short\", day: \"numeric\",\n });\n const precipStr = d.precip > 0.05 ? ` · ${d.precip.toFixed(2)}\"` : \"\";\n return `${label}: ${wmoDesc(d.weathercode)}, ${d.hi}°/${d.lo}°F${precipStr}`;\n });\n return `📍 *${result.location}*\\n${lines.join(\"\\n\")}`;\n}\n","label",[144],"d6347d6c-8702-439d-b67c-fc03cd2ff7bd",["D",1788988992452],9,[28],"999b25e3-ab7a-4e09-bcd7-4213e63ddb16","tools.ts","3c581426-ac69-11f1-820a-1607ee4eb77e",114,["D",1788988887845],3,"// Single source of truth for the mibbs-bot family-manager toolset.\n// TOOLS: full Anthropic-shaped tool definitions (11 tools).\n// executeTool: the implementation switch; caller passes ctx as context.\n\nimport Anthropic from \"npm:@anthropic-ai/sdk\";\nimport { addTask, listTasks, completeTask, updateTask, storeInfo, getInfo } from \"./db.ts\";\nimport { getCalendarEvents, createCalendarEvent, updateCalendarEvent } from \"./google.ts\";\nimport { getWeather } from \"./weather.ts\";\n\n// Execution context passed to executeTool.\n// The `react` callback is only provided by main.ts (Telegram surface); the\n// email surface omits it and the react tool safely no-ops in that case.\nexport type ToolCtx = {\n userName: string;\n /** Send an emoji reaction on the triggering message. Returns true if the\n * reaction was applied; false if the API rejected it (caller should fall back\n * to sending the emoji as a message). */\n react?: (emoji: string) => Promise;\n /** Set to true by the react case after a reaction is delivered. Read by\n * processMessage to decide whether to suppress the text reply. */\n reacted?: boolean;\n};\n\nexport const TOOLS: Anthropic.Tool[] = [\n {\n name: \"add_task\",\n description: \"Add a new task to the household task list. Tasks can have an owner (who should do it) and an optional due date/time. A timed task (full ISO 8601 UTC datetime) sends an automatic reminder ping to the group at that exact time. A date-only task sends a reminder at 8am ET on that day.\",\n input_schema: {\n type: \"object\",\n properties: {\n text: { type: \"string\", description: \"Task description\" },\n due_date: { type: \"string\", description: \"Optional due date. Use YYYY-MM-DD for date-only (pings at 8am ET) or full ISO 8601 UTC datetime for an exact time (e.g. 2026-09-08T18:00:00Z). For relative times like 'in 2 hours', calculate from the current UTC time.\" },\n owner: { type: \"string\", description: \"Optional: who should do it — 'jeff', 'marisa', or 'both'. Sets responsibility and determines who gets @mentioned in the reminder.\" },\n },\n required: [\"text\"],\n },\n },\n {\n name: \"update_task\",\n description: \"Update a task's owner or due date/time. When due_date changes, the reminder resets and will fire again at the new time. Use this to assign tasks ('assign task 3 to Marisa') or reschedule reminders ('move task 5 to 3pm today').\",\n input_schema: {\n type: \"object\",\n properties: {\n task_id: { type: \"number\", description: \"Task ID to update\" },\n owner: { type: \"string\", description: \"New owner: 'jeff', 'marisa', or 'both'\" },\n due_date: { type: \"string\", description: \"New due date. Use YYYY-MM-DD for date-only or full ISO 8601 UTC datetime for an exact time.\" },\n },\n required: [\"task_id\"],\n },\n },\n {\n name: \"list_tasks\",\n description: \"Get all incomplete household tasks\",\n input_schema: { type: \"object\", properties: {}, required: [] },\n },\n {\n name: \"complete_task\",\n description: \"Mark a task as completed by its ID\",\n input_schema: {\n type: \"object\",\n properties: { task_id: { type: \"number\", description: \"Task ID to complete\" } },\n required: [\"task_id\"],\n },\n },\n {\n name: \"get_calendar_events\",\n description: \"Get upcoming calendar events\",\n input_schema: {\n type: \"object\",\n properties: { days_ahead: { type: \"number\", description: \"How many days ahead to look (default 7)\" } },\n required: [],\n },\n },\n {\n name: \"add_calendar_event\",\n description: \"Add an event to the family calendar\",\n input_schema: {\n type: \"object\",\n properties: {\n summary: { type: \"string\", description: \"Event title\" },\n start_dt: { type: \"string\", description: \"Start datetime in ISO 8601 Eastern Time WITHOUT UTC offset — e.g. '2026-09-09T15:00:00' for 3pm ET. Do not use Z or +00:00.\" },\n end_dt: { type: \"string\", description: \"End datetime in ISO 8601 Eastern Time WITHOUT UTC offset — e.g. '2026-09-09T16:00:00' for 4pm ET. Do not use Z or +00:00.\" },\n description: { type: \"string\", description: \"Optional description\" },\n },\n required: [\"summary\", \"start_dt\", \"end_dt\"],\n },\n },\n {\n name: \"update_calendar_event\",\n description: \"Update an existing calendar event — change its title, start/end time, or description. Searches for the event by name among the next 60 days.\",\n input_schema: {\n type: \"object\",\n properties: {\n event_query: { type: \"string\", description: \"Part of the event title to search for (case-insensitive)\" },\n new_summary: { type: \"string\", description: \"New title for the event\" },\n new_start_dt: { type: \"string\", description: \"New start datetime in ISO 8601 Eastern Time WITHOUT UTC offset — e.g. '2026-09-09T15:00:00' for 3pm ET.\" },\n new_end_dt: { type: \"string\", description: \"New end datetime in ISO 8601 Eastern Time WITHOUT UTC offset — e.g. '2026-09-09T16:00:00' for 4pm ET.\" },\n new_description: { type: \"string\", description: \"New description\" },\n },\n required: [\"event_query\"],\n },\n },\n {\n name: \"store_info\",\n description: \"Store important household info (doctor names, account details, important dates, etc.)\",\n input_schema: {\n type: \"object\",\n properties: {\n key: { type: \"string\", description: \"Label for the info\" },\n value: { type: \"string\", description: \"The info to store\" },\n },\n required: [\"key\", \"value\"],\n },\n },\n {\n name: \"retrieve_info\",\n description: \"Search stored household information\",\n input_schema: {\n type: \"object\",\n properties: { query: { type: \"string\", description: \"Search term\" } },\n required: [],\n },\n },\n {\n name: \"get_weather\",\n description: \"Get a weather forecast for any location. Use for questions like 'what's the weather in Boston?' or 'will it rain this weekend in Chicago?'\",\n input_schema: {\n type: \"object\",\n properties: {\n location: { type: \"string\", description: \"City name, e.g. 'Boston' or 'Boston, MA'\" },\n days_ahead: { type: \"number\", description: \"Number of days to forecast (1-7, default 3)\" },\n },\n required: [\"location\"],\n },\n },\n {\n name: \"react\",\n description: \"Acknowledge the user's message with an emoji reaction instead of a text reply. Call this AFTER completing a simple, unambiguous action (setting a reminder, adding a task, adding a calendar event, storing info) when no information needs to be sent back. After calling react, produce NO text reply. Do NOT call this when the user asked a question, needs a list of items, asked for info (weather, calendar, tasks), or the outcome needs explaining.\",\n input_schema: {\n type: \"object\",\n properties: {\n emoji: {\n type: \"string\",\n description: \"A single emoji to react with — 👍 for general completions, 🎉 for added events, ❤ for stored info, ✅ is not supported so use 👍 instead.\",\n enum: [\"👍\", \"👎\", \"❤\", \"🔥\", \"🎉\", \"🙏\", \"👀\", \"😍\", \"🤔\", \"💯\", \"😁\", \"🤩\"],\n },\n },\n required: [\"emoji\"],\n },\n },\n];\n\nexport async function executeTool(name: string, input: any, ctx: ToolCtx): Promise {\n const { userName } = ctx;\n try {\n switch (name) {\n case \"add_task\": {\n await addTask(input.text, userName, input.due_date, input.owner);\n const ownerNote = input.owner ? ` (owner: ${input.owner})` : \"\";\n const timeNote = input.due_date ? ` — reminder set for ${input.due_date}` : \"\";\n return `Task added: ${input.text}${ownerNote}${timeNote}`;\n }\n case \"update_task\": {\n const ok = await updateTask(input.task_id, input.owner, input.due_date);\n return ok ? `Task ${input.task_id} updated.` : `No incomplete task with ID ${input.task_id}.`;\n }\n case \"list_tasks\": {\n const tasks = await listTasks();\n if (!tasks.length) return \"No pending tasks.\";\n return tasks.map((t) => {\n const ownerStr = t.owner ? ` — ${t.owner}` : \"\";\n const dueStr = t.due_date ? ` (due: ${t.due_date})` : \"\";\n return `[${t.id}] ${t.text}${ownerStr}${dueStr}`;\n }).join(\"\\n\");\n }\n case \"complete_task\": {\n const ok = await completeTask(input.task_id);\n return ok ? \"Task marked complete ✅\" : `No incomplete task with ID ${input.task_id}.`;\n }\n case \"get_calendar_events\":\n return await getCalendarEvents(input.days_ahead || 7);\n case \"add_calendar_event\":\n return await createCalendarEvent(input.summary, input.start_dt, input.end_dt, input.description || \"\");\n case \"update_calendar_event\":\n return await updateCalendarEvent(input.event_query, {\n summary: input.new_summary,\n start_dt: input.new_start_dt,\n end_dt: input.new_end_dt,\n description: input.new_description,\n });\n case \"store_info\":\n await storeInfo(input.key, input.value, userName);\n return `Saved: ${input.key} = ${input.value}`;\n case \"retrieve_info\": {\n const items = await getInfo(input.query);\n if (!items.length) return \"Nothing found.\";\n return items.map((i) => `• ${i.key}: ${i.value}`).join(\"\\n\");\n }\n case \"get_weather\":\n return await getWeather(input.location, input.days_ahead ?? 3);\n case \"react\": {\n if (ctx.react) {\n const ok = await ctx.react(input.emoji);\n ctx.reacted = true;\n return ok\n ? `Reacted with ${input.emoji}`\n : `Sent ${input.emoji} as message (reaction API unavailable)`;\n }\n // No Telegram context (e.g. email surface) — no-op silently\n ctx.reacted = true;\n return `Acknowledged ${input.emoji}`;\n }\n default:\n return `Unknown tool: ${name}`;\n }\n } catch (err) {\n return `Error in ${name}: ${String(err)}`;\n }\n}\n",[159],"61874cac-1eb1-4838-8231-1f8842119ae7","reminder-cron.ts","589751f0-ab93-11f1-bb1d-1607ee4eb77e",108,["D",1788970674252],2,"interval","// Family Manager Bot — Reminder Cron\n// Cron: */15 * * * *\n// Runs every 15 minutes. Finds tasks whose remind_at has passed and haven't been fired yet,\n// sends a proactive Telegram ping (with @mention of the owner), then marks them as reminded.\n\nimport { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { rowsToObjects, initDb } from \"./db.ts\";\n\nconst TELEGRAM_TOKEN = Deno.env.get(\"TELEGRAM_TOKEN\")!;\nconst GROUP_CHAT_ID = Deno.env.get(\"GROUP_CHAT_ID\")!;\nconst TZ = \"America/New_York\";\n\n// Format a UTC ISO instant as a local time string for display in the message\nfunction formatLocalTime(iso: string): string {\n return new Date(iso).toLocaleTimeString(\"en-US\", {\n timeZone: TZ,\n hour: \"numeric\",\n minute: \"2-digit\",\n weekday: \"short\",\n month: \"short\",\n day: \"numeric\",\n });\n}\n\n// Build a Telegram inline mention: [Name](tg://user?id=ID)\n// Falls back to plain name when no ID is found.\nfunction buildMention(owner: string | null, users: Record[]): string {\n if (!owner) return \"\";\n if (owner === \"both\") {\n const mentions = users\n .map((u) => u.telegram_id ? `[${u.first_name}](tg://user?id=${u.telegram_id})` : u.first_name)\n .join(\" and \");\n return mentions || \"both\";\n }\n const lc = owner.toLowerCase();\n const match = users.find((u) => u.first_name?.toLowerCase() === lc || u.username?.toLowerCase() === lc);\n if (match?.telegram_id) return `[${match.first_name ?? owner}](tg://user?id=${match.telegram_id})`;\n return owner; // plain name fallback until user has messaged the bot\n}\n\nexport default async function () {\n if (!GROUP_CHAT_ID) return;\n\n // Ensure schema exists — defensive, in case reminder-cron runs before main.ts\n await initDb();\n\n const nowIso = new Date().toISOString();\n\n // Fetch due, un-fired reminders\n const dueResult = await sqlite.execute({\n sql: `SELECT id, text, owner, remind_at FROM fm_tasks\n WHERE completed = 0\n AND reminded = 0\n AND remind_at IS NOT NULL\n AND remind_at <= ?`,\n args: [nowIso],\n });\n const dueTasks = rowsToObjects(dueResult);\n if (!dueTasks.length) {\n console.log(\"No reminders due.\");\n return;\n }\n\n // Load known users for @mention resolution\n const usersResult = await sqlite.execute(\"SELECT telegram_id, first_name, username FROM fm_users\");\n const users = rowsToObjects(usersResult);\n\n const firedIds: number[] = [];\n\n for (const task of dueTasks) {\n const mention = buildMention(task.owner as string | null, users);\n const timeStr = formatLocalTime(task.remind_at as string);\n let msgText: string;\n if (mention) {\n msgText = `⏰ Reminder for ${mention}: *${task.text}*\\n_Scheduled for ${timeStr}_`;\n } else {\n msgText = `⏰ Reminder: *${task.text}*\\n_Scheduled for ${timeStr}_`;\n }\n\n try {\n const resp = await fetch(`https://api.telegram.org/bot${TELEGRAM_TOKEN}/sendMessage`, {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({\n chat_id: GROUP_CHAT_ID,\n text: msgText,\n parse_mode: \"Markdown\",\n }),\n });\n const data = await resp.json();\n if (data.ok) {\n firedIds.push(task.id as number);\n console.log(`Reminder sent for task ${task.id}: ${task.text}`);\n } else {\n console.error(`Failed to send reminder for task ${task.id}:`, data.description);\n }\n } catch (err) {\n console.error(`Error sending reminder for task ${task.id}:`, err);\n }\n }\n\n // Mark fired reminders so they don't re-send\n if (firedIds.length) {\n await sqlite.execute({\n sql: `UPDATE fm_tasks SET reminded = 1 WHERE id IN (${firedIds.map(() => \"?\").join(\",\")})`,\n args: firedIds,\n });\n console.log(`Marked ${firedIds.length} reminder(s) as sent.`);\n }\n}\n",[167],"b58bc3b5-1b3d-4c1c-a242-d6d063567740","morning-cron.ts","fef68e46-ab0f-11f1-a7cf-1607ee4eb77e",107,["D",1788970656966],4,"// Family Manager Bot — Daily Morning Briefing\n// Cron: 0 11 * * *\n// Runs every morning at 7am EDT (11:00 UTC). Adjust to 0 12 * * * for EST (winter).\n\nimport { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { createMessage } from \"./llm.ts\";\nimport { getGoogleAccessToken } from \"./google.ts\";\nimport { fetchForecast, wmoDesc } from \"./weather.ts\";\nimport { rowsToObjects } from \"./db.ts\";\n\nconst TELEGRAM_TOKEN = Deno.env.get(\"TELEGRAM_TOKEN\")!;\nconst GROUP_CHAT_ID = Deno.env.get(\"GROUP_CHAT_ID\")!;\nconst HOME_LOCATION = Deno.env.get(\"HOME_LOCATION\") || \"Boston, MA\";\nconst TZ = \"America/New_York\";\n\nasync function fetchTodayEvents(token: string): Promise {\n const calendarId = encodeURIComponent(Deno.env.get(\"CALENDAR_ID\") || \"primary\");\n const now = new Date();\n // Fetch from now through end of day (next 18 hours covers 7am → 1am)\n const endOfDay = new Date(now.getTime() + 18 * 3600 * 1000);\n const resp = await fetch(\n `https://www.googleapis.com/calendar/v3/calendars/${calendarId}/events?timeMin=${now.toISOString()}&timeMax=${endOfDay.toISOString()}&maxResults=20&singleEvents=true&orderBy=startTime`,\n { headers: { Authorization: `Bearer ${token}` } },\n );\n const data = await resp.json();\n return data.items ?? [];\n}\n\nasync function fetchDueTasks(): Promise {\n try {\n const today = new Date().toISOString().split(\"T\")[0];\n const result = await sqlite.execute({\n sql: \"SELECT id, text, due_date FROM fm_tasks WHERE completed = 0 ORDER BY CASE WHEN due_date IS NULL THEN 1 ELSE 0 END, due_date ASC LIMIT 10\",\n args: [],\n });\n const tasks = rowsToObjects(result);\n if (!tasks.length) return \"None.\";\n return tasks.map(t => {\n const isOverdue = t.due_date && t.due_date < today;\n const dueStr = t.due_date ? ` (${isOverdue ? \"OVERDUE: \" : \"due: \"}${t.due_date})` : \"\";\n return `• [${t.id}] ${t.text}${dueStr}`;\n }).join(\"\\n\");\n } catch { return \"Could not fetch tasks.\"; }\n}\n\nasync function fetchTodayWeather(location: string): Promise {\n const result = await fetchForecast(location, 1);\n if (!result || !result.daily.length) return \"\";\n const d = result.daily[0];\n const condition = wmoDesc(d.weathercode);\n return `${condition}, ${d.hi}°/${d.lo}°F${d.precip > 0.05 ? `, ${d.precip.toFixed(1)}\" rain expected` : \"\"}`;\n}\n\nexport default async function () {\n if (!GROUP_CHAT_ID) return;\n\n const token = await getGoogleAccessToken();\n if (!token) { console.error(\"No Google access token\"); return; }\n\n const [todayEvents, dueTasks, weather] = await Promise.all([\n fetchTodayEvents(token),\n fetchDueTasks(),\n fetchTodayWeather(HOME_LOCATION),\n ]);\n\n const now = new Date();\n const dateLabel = now.toLocaleDateString(\"en-US\", { weekday: \"long\", month: \"long\", day: \"numeric\", timeZone: TZ });\n\n const eventsText = todayEvents.length === 0\n ? \"Nothing scheduled today.\"\n : todayEvents.map((e: any) => {\n const start = e.start?.dateTime || e.start?.date;\n const timeStr = e.start?.dateTime\n ? new Date(start).toLocaleTimeString(\"en-US\", { hour: \"numeric\", minute: \"2-digit\", timeZone: TZ })\n : \"all day\";\n return `• ${e.summary}${e.start?.dateTime ? ` @ ${timeStr}` : \"\"}`;\n }).join(\"\\n\");\n\n const prompt = `You are a warm family assistant bot sending a morning briefing. Today is ${dateLabel}.\n\nWrite a short, practical morning message — something a busy parent will read in 30 seconds over coffee. Flag anything that needs attention today, note the weather if it affects plans, and mention any urgent tasks. Keep it to 4–6 lines.\n\nTODAY'S CALENDAR:\n${eventsText}\n\nOPEN TASKS:\n${dueTasks}\n\n${weather ? `WEATHER TODAY (${HOME_LOCATION}): ${weather}` : \"\"}\n\nUse Telegram Markdown. Start with \"Good morning!\" and the date. Be warm, not robotic. If there's nothing notable, keep it cheerful and brief.`;\n\n const message = await createMessage({ prompt, maxTokens: 250 }) || `Good morning! Today is ${dateLabel}.`;\n\n await fetch(`https://api.telegram.org/bot${TELEGRAM_TOKEN}/sendMessage`, {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({ chat_id: GROUP_CHAT_ID, text: message, parse_mode: \"Markdown\" }),\n });\n\n console.log(\"Morning briefing sent\");\n}\n",[176],"3995ff85-11cf-4716-b39a-e8c2a1263129","main.ts",115,["D",1788988957949],54,"http","// Family Manager Telegram Bot — HTTP Val\n// Deploy this as an HTTP val on Val.town. Set the webhook URL in Telegram after deploying.\n\nimport { initDb, listTasks, completeTask, getInfo, saveMessage, getRecentMessages, upsertUser } from \"./db.ts\";\nimport { listCalendars } from \"./google.ts\";\nimport { runAgent, AgentContent } from \"./llm.ts\";\nimport { TOOLS, executeTool, ToolCtx } from \"./tools.ts\";\n\n// ── Changelog ─────────────────────────────────────────────────────────────────\n\nconst CHANGELOG_LATEST = `🖍️ *Changelog*\n\n*Cerulean (v2.2) — Quieter acknowledgments*\nThe bot now reacts with an emoji instead of writing a full reply when the action was simple and unambiguous.\n• \"remind me at 7:30am to put out jackets\" → 👍 reaction on your message\n• Questions, calendar lookups, and anything with useful info still get a text reply\n\n*Navy Blue (v2.1) — Reminders that ping you*\nTasks now have an owner and an optional reminder time.\n• \"remind Marisa to call the dentist at 2pm\" → pings @Marisa at 2pm\n• Date-only tasks send a reminder at 8am on the due date\n• Owners get @mentioned so the right person's phone notifies them\n\n*Blue (v2.0) — Proactive Era*\nThe bot now shows up before you ask.\n• Daily 7am briefing: schedule, weather, open tasks\n• Sunday brief: Claude asks \"what does this week require?\"\n\n*Scarlet (v1.2)* — Send a photo, the bot sees it\n*Brick Red (v1.1)* — Weather anywhere, anytime\n*Red (v1.0)* — Tasks, calendar, info storage`;\n\n// ── Telegram ──────────────────────────────────────────────────────────────────\n\nconst TELEGRAM_TOKEN = Deno.env.get(\"TELEGRAM_TOKEN\")!;\nconst TG_API = `https://api.telegram.org/bot${TELEGRAM_TOKEN}`;\n\nasync function sendMessage(chatId: number, text: string, parseMode = \"Markdown\") {\n await fetch(`${TG_API}/sendMessage`, {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({ chat_id: chatId, text, parse_mode: parseMode }),\n });\n}\n\nasync function sendChatAction(chatId: number, action = \"typing\") {\n await fetch(`${TG_API}/sendChatAction`, {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({ chat_id: chatId, action }),\n });\n}\n\n/**\n * Attach an emoji reaction to a specific message.\n * Returns true if Telegram accepted it, false otherwise (caller should fall back\n * to sending the emoji as a plain message).\n */\nasync function setMessageReaction(chatId: number, messageId: number, emoji: string): Promise {\n try {\n const resp = await fetch(`${TG_API}/setMessageReaction`, {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({\n chat_id: chatId,\n message_id: messageId,\n reaction: [{ type: \"emoji\", emoji }],\n }),\n });\n const data = await resp.json();\n const ok = !!data.result;\n console.log(`setMessageReaction emoji=${emoji} ok=${ok}`);\n return ok;\n } catch (err) {\n console.error(\"setMessageReaction error:\", err);\n return false;\n }\n}\n\nasync function downloadTelegramPhoto(fileId: string): Promise<{ base64: string; mediaType: string }> {\n const getFileResp = await fetch(`${TG_API}/getFile?file_id=${encodeURIComponent(fileId)}`);\n const fileData = await getFileResp.json();\n const filePath: string = fileData.result?.file_path;\n if (!filePath) throw new Error(\"Could not get file path from Telegram\");\n\n const fileResp = await fetch(`https://api.telegram.org/file/bot${TELEGRAM_TOKEN}/${filePath}`);\n const arrayBuffer = await fileResp.arrayBuffer();\n const uint8 = new Uint8Array(arrayBuffer);\n let binary = \"\";\n for (let i = 0; i < uint8.byteLength; i++) binary += String.fromCharCode(uint8[i]);\n const base64 = btoa(binary);\n\n const ext = filePath.split(\".\").pop()?.toLowerCase() ?? \"jpg\";\n const mediaType = ext === \"png\" ? \"image/png\" : ext === \"gif\" ? \"image/gif\" : ext === \"webp\" ? \"image/webp\" : \"image/jpeg\";\n return { base64, mediaType };\n}\n\nasync function getBotUsername(): Promise {\n const envUsername = Deno.env.get(\"BOT_USERNAME\");\n if (envUsername) return envUsername;\n const resp = await fetch(`${TG_API}/getMe`);\n const data = await resp.json();\n return data.result?.username ?? \"\";\n}\n\n// ── Agent ──────────────────────────────────────────────────────────────────────\n\nasync function processMessage(\n text: string,\n chatId: number,\n messageId: number,\n userName: string,\n imageData?: { base64: string; mediaType: string } | null,\n): Promise<{ reply: string; reacted: boolean }> {\n const familyName = Deno.env.get(\"FAMILY_NAME\") || \"the family\";\n const systemPrompt = `You are the shared family operations manager for ${familyName}. You live in their Telegram group chat where both Jeff and Marisa message you.\n\nYour core job is removing the small coordination burden between them: remembering, reconciling, planning, and following up so things don't fall through the cracks.\n\nWhen someone shares anything — an event, a message, a screenshot, a reminder, a piece of news — your instinct is to ask: \"What does this require us to do?\" Then act on it: add the task, note the date, store the info, flag the follow-up. Don't wait to be told explicitly.\n\nEither of them should be able to throw essentially anything into this chat and trust: \"If our family needs to remember, decide, prepare for, or do something about this, the bot will make sure it doesn't get dropped.\"\n\nCapabilities: shared task list (assign to a person if specified), Google Calendar, and storing household information (doctor contacts, account numbers, passwords, important dates, anything worth keeping).\n\nTasks support an **owner** (jeff/marisa/both) and an optional **reminder time**. Set the owner when a person is named (\"Marisa should call the dentist\"). Use a full ISO 8601 UTC datetime in due_date for a timed reminder — that exact time triggers a proactive ping. A date-only due_date triggers a reminder at 8am ET. Use update_task to assign or reschedule existing tasks.\n\n**Response style:** After completing a simple, unambiguous action — adding a task, setting a reminder, adding or updating a calendar event, storing info — call the \\`react\\` tool (👍 for completions, 🎉 for scheduled events, ❤ for stored info) and produce NO text reply. The reaction is the acknowledgment; do not also write a sentence.\n\nReply with text when: the request was ambiguous or needs a clarifying question; the user asked a question or needs information back (weather, calendar events, task list, stored info); something failed; or the situation otherwise warrants a conversational response.\n\nBe concise and warm. Current UTC time: ${new Date().toISOString()}.\n\nFor calendar event times, use ISO 8601 Eastern Time WITHOUT a UTC offset — e.g. \"2026-09-09T15:00:00\" means 3pm ET. Never convert calendar times to UTC. For task due_dates with exact times, use full ISO 8601 UTC (e.g. 2026-09-09T19:00:00Z). For relative times (\"in 2 hours\"), calculate from current UTC time.`;\n\n const history = await getRecentMessages(chatId, 20);\n const savedText = imageData ? `${userName}: [photo] ${text}` : `${userName}: ${text}`;\n await saveMessage(chatId, \"user\", savedText, userName);\n\n const userMessage = `${userName}: ${text}`;\n\n // Build the current user message content (may include an image block for vision)\n const currentContent: AgentContent = imageData\n ? [\n { type: \"image\", base64: imageData.base64, mediaType: imageData.mediaType },\n { type: \"text\", text: userMessage },\n ]\n : userMessage;\n\n const messages = [\n ...history.map((m) => ({ role: m.role as \"user\" | \"assistant\", content: m.content })),\n { role: \"user\" as const, content: currentContent },\n ];\n\n // Mutable context shared between this function and the react tool case.\n // The react callback fires setMessageReaction; on failure it falls back to\n // sending the emoji as a plain message.\n let reactedEmoji = \"\";\n const ctx: ToolCtx = {\n userName,\n react: async (emoji: string) => {\n reactedEmoji = emoji;\n const ok = await setMessageReaction(chatId, messageId, emoji);\n if (!ok) {\n // Telegram rejected the reaction (e.g. old message, group settings) —\n // send as a plain message so the acknowledgment still reaches the family.\n await sendMessage(chatId, emoji);\n }\n return ok;\n },\n };\n\n const reply = await runAgent({\n system: systemPrompt,\n messages,\n tools: TOOLS,\n executeTool: (name, input) => executeTool(name, input, ctx),\n onAssistantText: (text) => saveMessage(chatId, \"assistant\", text),\n });\n\n const reacted = ctx.reacted ?? false;\n\n // When the bot reacted and produced no meaningful text, save a synthetic\n // history entry so the next turn's context isn't missing the bot's response.\n if (reacted && (!reply || reply === \"Done.\" || reply.trim() === \"\")) {\n await saveMessage(chatId, \"assistant\", `[reacted ${reactedEmoji}]`);\n }\n\n return { reply, reacted };\n}\n\n// ── Main Handler ──────────────────────────────────────────────────────────────\n\nexport default async function (req: Request): Promise {\n if (req.method !== \"POST\") return new Response(\"Family Bot OK\");\n\n try {\n await initDb();\n } catch (err) {\n console.error(\"initDb failed:\", err);\n }\n\n let update: any;\n try {\n update = await req.json();\n } catch {\n return new Response(\"Bad Request\", { status: 400 });\n }\n\n const message = update.message || update.edited_message;\n const hasText = !!message?.text;\n const hasPhoto = !!message?.photo;\n if (!hasText && !hasPhoto) return new Response(\"OK\");\n\n const chatId: number = message.chat.id;\n const messageId: number = message.message_id;\n const chatType: string = message.chat.type;\n const text: string = message.text || message.caption || \"\";\n const userName: string = message.from?.first_name || \"Someone\";\n const respondToAll = Deno.env.get(\"RESPOND_TO_ALL\") === \"true\";\n console.log(`message: \"${text}\" | type: ${chatType} | respondToAll: ${respondToAll}`);\n\n // Auto-learn user identity for @mention reminders\n if (message.from?.id) {\n try {\n await upsertUser(message.from.id, message.from.first_name ?? null, message.from.username ?? null);\n } catch { /* non-fatal — don't break the message handler */ }\n }\n\n if (text.startsWith(\"/start\")) {\n await sendMessage(\n chatId,\n \"👋 Hi! I'm your family manager bot.\\n\\nTalk to me naturally:\\n• \\\"Add a task to call the plumber\\\"\\n• \\\"Remind me to take out the trash at 8pm\\\"\\n• \\\"What's on the calendar this week?\\\"\\n• \\\"Remember the WiFi password is xyz123\\\"\\n\\nCommands: /tasks · /done · /info · /help\",\n );\n return new Response(\"OK\");\n }\n\n if (text.startsWith(\"/calendars\")) {\n const list = await listCalendars();\n await sendMessage(chatId, `*Your calendars:*\\n${list}`);\n return new Response(\"OK\");\n }\n\n if (text.startsWith(\"/tasks\")) {\n const tasks = await listTasks();\n if (!tasks.length) {\n await sendMessage(chatId, \"No pending tasks! 🎉\");\n } else {\n const lines = tasks.map((t) => {\n const ownerStr = t.owner ? ` — _${t.owner}_` : \"\";\n const dueStr = t.due_date ? ` _(due ${t.due_date})_` : \"\";\n return `[${t.id}] ${t.text}${ownerStr}${dueStr}`;\n });\n await sendMessage(chatId, `*Pending tasks:*\\n${lines.join(\"\\n\")}`);\n }\n return new Response(\"OK\");\n }\n\n if (text.startsWith(\"/done\")) {\n const parts = text.trim().split(/\\s+/);\n const id = parseInt(parts[1]);\n if (isNaN(id)) {\n await sendMessage(chatId, \"Usage: /done \");\n } else {\n const ok = await completeTask(id);\n await sendMessage(chatId, ok ? `✅ Task ${id} marked complete.` : `No incomplete task with ID ${id}.`);\n }\n return new Response(\"OK\");\n }\n\n if (text.startsWith(\"/info\")) {\n const query = text.replace(\"/info\", \"\").trim() || undefined;\n const items = await getInfo(query);\n if (!items.length) {\n await sendMessage(chatId, query ? `No info found for \"${query}\".` : \"No stored information yet.\");\n } else {\n await sendMessage(chatId, items.map((i) => `*${i.key}:* ${i.value}`).join(\"\\n\"));\n }\n return new Response(\"OK\");\n }\n\n if (text.startsWith(\"/changelog\")) {\n await sendMessage(chatId, CHANGELOG_LATEST);\n return new Response(\"OK\");\n }\n\n if (text.startsWith(\"/help\")) {\n await sendMessage(\n chatId,\n \"*Family Manager Bot*\\n\\nJust talk to me! Examples:\\n• \\\"Add a task to buy groceries\\\"\\n• \\\"Remind me at 5pm to pick up the kids\\\"\\n• \\\"What's on the calendar this week?\\\"\\n• \\\"Remember that Dr. Smith's number is 555-1234\\\"\\n\\n*Commands:*\\n/tasks — Pending tasks\\n/done — Complete a task\\n/info [search] — Stored information\\n/changelog — What's new\\n/help — This message\",\n );\n return new Response(\"OK\");\n }\n\n // In groups: only respond to @mentions unless RESPOND_TO_ALL\n if (chatType !== \"private\" && !respondToAll) {\n const botUsername = await getBotUsername();\n if (!text.includes(`@${botUsername}`)) return new Response(\"OK\");\n }\n\n // Strip @mention from message text\n let messageText = text;\n const botUsername = await getBotUsername();\n messageText = messageText.replace(`@${botUsername}`, \"\").trim();\n\n await sendChatAction(chatId, \"typing\");\n\n let imageData: { base64: string; mediaType: string } | null = null;\n if (hasPhoto) {\n try {\n // Telegram sends multiple sizes; last entry is the largest\n const photos: any[] = message.photo;\n const largest = photos[photos.length - 1];\n imageData = await downloadTelegramPhoto(largest.file_id);\n } catch (err) {\n console.error(\"Photo download failed:\", err);\n await sendMessage(chatId, \"Sorry, I couldn't download that photo — please try again.\");\n return new Response(\"OK\");\n }\n }\n\n try {\n const result = await processMessage(messageText, chatId, messageId, userName, imageData);\n\n // When the bot reacted with an emoji, suppress the text reply unless the\n // model also produced something meaningful (e.g. a warning or clarification).\n const isMeaningfulReply = result.reply && result.reply !== \"Done.\" && result.reply.trim() !== \"\";\n if (!result.reacted || isMeaningfulReply) {\n await sendMessage(chatId, result.reply);\n }\n } catch (err) {\n console.error(\"processMessage error:\", err);\n await sendMessage(chatId, \"Something went wrong — please try again.\");\n }\n\n return new Response(\"OK\");\n}\n",[184],"218fcda1-6cff-4e69-b7d8-3af030b5fe9e","llm.ts","e4a87a9e-ab92-11f1-bee4-1607ee4eb77e",103,["D",1788970373165],"import Anthropic from \"npm:@anthropic-ai/sdk\";\nimport OpenAI from \"npm:openai\";\n\nexport const MODEL = Deno.env.get(\"MODEL\") ?? \"claude-sonnet-4-6\";\nexport const anthropic = new Anthropic({ apiKey: Deno.env.get(\"ANTHROPIC_API_KEY\")! });\nexport const openai = new OpenAI({ apiKey: Deno.env.get(\"OPENAI_API_KEY\") ?? \"not-set\" });\n\nexport const isGptModel = (model: string) => model.startsWith(\"gpt-\");\n\n// \"low\" gives some reasoning for tool routing without medium/high latency; \"none\" disables it entirely\nexport const REASONING_EFFORT = \"low\" as const;\n\n// One-shot message (no tools, no history) — used by cron jobs and simple callers\nexport async function createMessage(opts: {\n prompt: string;\n system?: string;\n maxTokens?: number;\n}): Promise {\n const { prompt, system = \"\", maxTokens = 512 } = opts;\n\n if (isGptModel(MODEL)) {\n const params: any = {\n model: MODEL,\n input: [{ role: \"user\", content: prompt }],\n max_output_tokens: maxTokens,\n reasoning: { effort: REASONING_EFFORT },\n };\n if (system) params.instructions = system;\n\n const response = await (openai as any).responses.create(params);\n const msgItem = (response.output as any[]).find((o: any) => o.type === \"message\");\n return msgItem?.content?.find((c: any) => c.type === \"output_text\")?.text ?? response.output_text ?? \"\";\n }\n\n const response = await anthropic.messages.create({\n model: MODEL,\n max_tokens: maxTokens,\n ...(system ? { system } : {}),\n messages: [{ role: \"user\", content: prompt }],\n });\n return response.content.find((b: any) => b.type === \"text\")?.text ?? \"\";\n}\n\n// ── Agentic loop ───────────────────────────────────────────────────────────────\n\n// Provider-neutral content block types. History messages from the DB are plain\n// strings; the current user message may include an image block for vision.\nexport type AgentContentBlock =\n | { type: \"text\"; text: string }\n | { type: \"image\"; base64: string; mediaType: string };\n\nexport type AgentContent = string | AgentContentBlock[];\n\nexport interface AgentMessage {\n role: \"user\" | \"assistant\";\n content: AgentContent;\n}\n\n// Single dual-provider agentic tool-use loop. Tool definitions use Anthropic's\n// schema (the canonical format); the GPT-path conversion happens internally.\n// Each caller passes its own system prompt, messages, and executeTool callback.\nexport async function runAgent(opts: {\n system: string;\n messages: AgentMessage[];\n tools: Anthropic.Tool[];\n executeTool: (name: string, input: any) => Promise;\n maxTokens?: number;\n maxIterations?: number;\n onAssistantText?: (text: string) => Promise;\n}): Promise {\n const {\n system,\n messages,\n tools,\n executeTool,\n maxTokens = 1024,\n maxIterations = 10,\n onAssistantText,\n } = opts;\n\n if (isGptModel(MODEL)) {\n return _runAgentGpt({ system, messages, tools, executeTool, maxTokens, maxIterations, onAssistantText });\n }\n return _runAgentAnthropic({ system, messages, tools, executeTool, maxTokens, maxIterations, onAssistantText });\n}\n\n// ── GPT Responses API path ─────────────────────────────────────────────────────\n\nfunction _toGptContent(content: AgentContent): any {\n if (typeof content === \"string\") return content;\n return content.map((block) => {\n if (block.type === \"text\") return { type: \"input_text\", text: block.text };\n return { type: \"input_image\", image_url: `data:${block.mediaType};base64,${block.base64}` };\n });\n}\n\nasync function _runAgentGpt(opts: {\n system: string;\n messages: AgentMessage[];\n tools: Anthropic.Tool[];\n executeTool: (name: string, input: any) => Promise;\n maxTokens: number;\n maxIterations: number;\n onAssistantText?: (text: string) => Promise;\n}): Promise {\n const { system, messages, tools, executeTool, maxTokens, maxIterations, onAssistantText } = opts;\n\n const oaiTools = tools.map((t) => ({\n type: \"function\" as const,\n name: t.name,\n description: t.description,\n parameters: t.input_schema as any,\n }));\n\n const initialInput = messages.map((m) => ({\n role: m.role,\n content: _toGptContent(m.content),\n }));\n\n let responseId: string | null = null;\n let toolResults: any[] = [];\n\n for (let i = 0; i < maxIterations; i++) {\n const params: any = {\n model: MODEL,\n input: i === 0 ? initialInput : toolResults,\n tools: oaiTools,\n max_output_tokens: maxTokens,\n reasoning: { effort: REASONING_EFFORT },\n };\n if (i === 0) {\n params.instructions = system;\n } else {\n params.previous_response_id = responseId;\n }\n\n const response = await (openai as any).responses.create(params);\n responseId = response.id;\n\n const toolCalls = (response.output as any[]).filter((o: any) => o.type === \"function_call\");\n\n if (!toolCalls.length) {\n const msgItem = (response.output as any[]).find((o: any) => o.type === \"message\");\n const reply = msgItem?.content?.find((c: any) => c.type === \"output_text\")?.text ?? response.output_text ?? \"Done.\";\n if (onAssistantText) await onAssistantText(reply);\n return reply;\n }\n\n toolResults = [];\n for (const tc of toolCalls) {\n const result = await executeTool(tc.name, JSON.parse(tc.arguments));\n toolResults.push({ type: \"function_call_output\", call_id: tc.call_id, output: result });\n }\n }\n\n return \"Sorry, I couldn't complete that.\";\n}\n\n// ── Anthropic path ─────────────────────────────────────────────────────────────\n\nfunction _toAnthropicContent(content: AgentContent): Anthropic.ContentBlockParam[] | string {\n if (typeof content === \"string\") return content;\n return content.map((block): Anthropic.ContentBlockParam => {\n if (block.type === \"text\") return { type: \"text\", text: block.text };\n return {\n type: \"image\",\n source: {\n type: \"base64\",\n media_type: block.mediaType as \"image/jpeg\" | \"image/png\" | \"image/gif\" | \"image/webp\",\n data: block.base64,\n },\n };\n });\n}\n\nasync function _runAgentAnthropic(opts: {\n system: string;\n messages: AgentMessage[];\n tools: Anthropic.Tool[];\n executeTool: (name: string, input: any) => Promise;\n maxTokens: number;\n maxIterations: number;\n onAssistantText?: (text: string) => Promise;\n}): Promise {\n const { system, messages, tools, executeTool, maxTokens, maxIterations, onAssistantText } = opts;\n\n const anthropicMessages: Anthropic.MessageParam[] = messages.map((m) => ({\n role: m.role,\n content: _toAnthropicContent(m.content) as any,\n }));\n\n for (let i = 0; i < maxIterations; i++) {\n const response = await anthropic.messages.create({\n model: MODEL,\n max_tokens: maxTokens,\n system,\n tools,\n messages: anthropicMessages,\n });\n\n if (response.stop_reason === \"end_turn\") {\n for (const block of response.content) {\n if (block.type === \"text\") {\n if (onAssistantText) await onAssistantText(block.text);\n return block.text;\n }\n }\n return \"Done.\";\n }\n\n if (response.stop_reason === \"tool_use\") {\n anthropicMessages.push({ role: \"assistant\", content: response.content });\n const results: Anthropic.ToolResultBlockParam[] = [];\n for (const block of response.content) {\n if (block.type === \"tool_use\") {\n const result = await executeTool(block.name, block.input as any);\n results.push({ type: \"tool_result\", tool_use_id: block.id, content: result });\n }\n }\n anthropicMessages.push({ role: \"user\", content: results });\n } else {\n break;\n }\n }\n\n return \"Sorry, I couldn't complete that.\";\n}\n",[192],"2511ecec-8255-4d02-a1a0-a68166e41ebf","google.ts","1cee0a82-ac69-11f1-9b74-1607ee4eb77e",100,["D",1788970288200],"// Shared Google Calendar helpers for mibbs-bot.\n// Single copy of auth (with token refresh + blob cache) and all calendar operations.\n\nimport { blob } from \"https://esm.town/v/std/blob\";\n\n// ── Auth ──────────────────────────────────────────────────────────────────────\n\nexport async function getGoogleAccessToken(): Promise {\n let tokenData: any = await blob.getJSON(\"gcalToken\").catch(() => null);\n\n if (!tokenData) {\n const tokenJson = Deno.env.get(\"GOOGLE_TOKEN_JSON\");\n if (!tokenJson) return null;\n try {\n const raw = JSON.parse(atob(tokenJson));\n if (raw.accounts) {\n const accountId = Deno.env.get(\"GOOGLE_CALENDAR_ACCOUNT\") || raw.defaultAccount || Object.keys(raw.accounts)[0];\n tokenData = raw.accounts[accountId];\n } else {\n tokenData = raw;\n }\n } catch (err) {\n console.error(\"Failed to parse GOOGLE_TOKEN_JSON:\", err);\n return null;\n }\n }\n\n const accessToken: string | null = tokenData.access_token ?? tokenData.token ?? null;\n const refreshToken: string | null = tokenData.refresh_token ?? null;\n const expiryMs: number = tokenData.expiry_date\n ? (typeof tokenData.expiry_date === \"number\" ? tokenData.expiry_date : new Date(tokenData.expiry_date).getTime())\n : tokenData.expiry\n ? new Date(tokenData.expiry).getTime()\n : 0;\n\n if (expiryMs - Date.now() < 5 * 60 * 1000) {\n const credsJson = Deno.env.get(\"GOOGLE_CREDENTIALS_JSON\");\n if (!credsJson || !refreshToken) return accessToken;\n try {\n const creds = JSON.parse(atob(credsJson));\n const client = creds.installed || creds.web;\n const resp = await fetch(\"https://oauth2.googleapis.com/token\", {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/x-www-form-urlencoded\" },\n body: new URLSearchParams({\n client_id: client.client_id,\n client_secret: client.client_secret,\n refresh_token: refreshToken,\n grant_type: \"refresh_token\",\n }),\n });\n const newToken = await resp.json();\n if (newToken.error) {\n console.error(\"Token refresh error:\", newToken.error, newToken.error_description);\n return accessToken;\n }\n const refreshed = {\n ...tokenData,\n access_token: newToken.access_token,\n token: newToken.access_token,\n expiry_date: Date.now() + (newToken.expires_in || 3600) * 1000,\n };\n await blob.setJSON(\"gcalToken\", refreshed);\n return newToken.access_token;\n } catch (err) {\n console.error(\"Token refresh failed:\", err);\n return accessToken;\n }\n }\n\n return accessToken;\n}\n\n// ── Low-level fetch ────────────────────────────────────────────────────────────\n\n// Returns raw event items — crons use this and format their own way.\nexport async function fetchCalendarEvents(token: string, daysAhead: number): Promise {\n const calendarId = encodeURIComponent(Deno.env.get(\"CALENDAR_ID\") || \"primary\");\n const now = new Date().toISOString();\n const end = new Date(Date.now() + daysAhead * 86400000).toISOString();\n const resp = await fetch(\n `https://www.googleapis.com/calendar/v3/calendars/${calendarId}/events?timeMin=${now}&timeMax=${end}&maxResults=50&singleEvents=true&orderBy=startTime`,\n { headers: { Authorization: `Bearer ${token}` } },\n );\n const data = await resp.json();\n return data.items ?? [];\n}\n\n// ── High-level helpers (used by tools + slash commands) ───────────────────────\n\nexport async function listCalendars(): Promise {\n const token = await getGoogleAccessToken();\n if (!token) return \"Calendar not configured.\";\n const resp = await fetch(\"https://www.googleapis.com/calendar/v3/users/me/calendarList\", {\n headers: { Authorization: `Bearer ${token}` },\n });\n const data = await resp.json();\n if (!data.items?.length) return \"No calendars found.\";\n return data.items.map((c: any) => `• ${c.summary}\\n ID: \\`${c.id}\\``).join(\"\\n\");\n}\n\nexport async function getCalendarEvents(daysAhead = 7): Promise {\n const token = await getGoogleAccessToken();\n if (!token) return \"Calendar not configured.\";\n const events = await fetchCalendarEvents(token, daysAhead);\n if (!events.length) return \"No upcoming events.\";\n return events.map((e: any) => `• ${e.summary} — ${e.start?.dateTime || e.start?.date}`).join(\"\\n\");\n}\n\nexport async function createCalendarEvent(summary: string, startDt: string, endDt: string, description = \"\"): Promise {\n const token = await getGoogleAccessToken();\n if (!token) return \"Calendar not configured.\";\n const calendarId = encodeURIComponent(Deno.env.get(\"CALENDAR_ID\") || \"primary\");\n const timezone = Deno.env.get(\"TIMEZONE\") || \"America/New_York\";\n const resp = await fetch(\n `https://www.googleapis.com/calendar/v3/calendars/${calendarId}/events`,\n {\n method: \"POST\",\n headers: { Authorization: `Bearer ${token}`, \"Content-Type\": \"application/json\" },\n body: JSON.stringify({\n summary,\n description,\n start: { dateTime: startDt, timeZone: timezone },\n end: { dateTime: endDt, timeZone: timezone },\n }),\n },\n );\n const data = await resp.json();\n return data.id ? `Event \"${summary}\" added to calendar.` : \"Failed to add event.\";\n}\n\nexport async function findCalendarEventByQuery(query: string): Promise<{ id: string; summary: string; start: string; end: string } | null> {\n const token = await getGoogleAccessToken();\n if (!token) return null;\n const calendarId = encodeURIComponent(Deno.env.get(\"CALENDAR_ID\") || \"primary\");\n const now = new Date().toISOString();\n const end = new Date(Date.now() + 60 * 86400000).toISOString();\n const resp = await fetch(\n `https://www.googleapis.com/calendar/v3/calendars/${calendarId}/events?timeMin=${now}&timeMax=${end}&maxResults=50&singleEvents=true&orderBy=startTime`,\n { headers: { Authorization: `Bearer ${token}` } },\n );\n const data = await resp.json();\n if (!data.items?.length) return null;\n const q = query.toLowerCase();\n const match = data.items.find((e: any) => e.summary?.toLowerCase().includes(q));\n if (!match) return null;\n return {\n id: match.id,\n summary: match.summary,\n start: match.start?.dateTime || match.start?.date,\n end: match.end?.dateTime || match.end?.date,\n };\n}\n\nexport async function updateCalendarEvent(\n eventQuery: string,\n updates: { summary?: string; start_dt?: string; end_dt?: string; description?: string },\n): Promise {\n const token = await getGoogleAccessToken();\n if (!token) return \"Calendar not configured.\";\n const event = await findCalendarEventByQuery(eventQuery);\n if (!event) return `No upcoming event found matching \"${eventQuery}\".`;\n const calendarId = encodeURIComponent(Deno.env.get(\"CALENDAR_ID\") || \"primary\");\n const timezone = Deno.env.get(\"TIMEZONE\") || \"America/New_York\";\n const patch: any = {};\n if (updates.summary !== undefined) patch.summary = updates.summary;\n if (updates.description !== undefined) patch.description = updates.description;\n if (updates.start_dt !== undefined) patch.start = { dateTime: updates.start_dt, timeZone: timezone };\n if (updates.end_dt !== undefined) patch.end = { dateTime: updates.end_dt, timeZone: timezone };\n // Preserve the unchanged side of a time-range update\n if (updates.start_dt && !updates.end_dt) patch.end = { dateTime: event.end, timeZone: timezone };\n if (updates.end_dt && !updates.start_dt) patch.start = { dateTime: event.start, timeZone: timezone };\n const resp = await fetch(\n `https://www.googleapis.com/calendar/v3/calendars/${calendarId}/events/${encodeURIComponent(event.id)}`,\n {\n method: \"PATCH\",\n headers: { Authorization: `Bearer ${token}`, \"Content-Type\": \"application/json\" },\n body: JSON.stringify(patch),\n },\n );\n const data = await resp.json();\n if (data.id) {\n const changed = Object.keys(updates).filter((k) => updates[k as keyof typeof updates] !== undefined).join(\", \");\n return `Updated \"${event.summary}\": ${changed} changed.`;\n }\n return `Failed to update event: ${data.error?.message || \"unknown error\"}`;\n}\n",[199],"3a7bf455-1eb0-40b8-9498-6251fed057f2","email.ts","460e9734-ab19-11f1-951d-1607ee4eb77e",105,["D",1788970605248],11,"email","// Family Manager Bot — Email Handler\n// Receives forwarded emails; Claude processes them and takes action.\n// Forward any email to this val's address. Add a note at the top with instructions if you want.\n\nimport { email as sendEmail } from \"https://esm.town/v/std/email\";\nimport { initDb } from \"./db.ts\";\nimport { runAgent } from \"./llm.ts\";\nimport { TOOLS, executeTool } from \"./tools.ts\";\n\n// ── Email Parsing ─────────────────────────────────────────────────────────────\n\n// Splits a forwarded email into the forwarder's added note vs. the original content.\n// If no forward marker is found, returns the full body as emailContent.\nfunction parseForwardedEmail(text: string): { instructions: string; emailContent: string } {\n const forwardMarkers = [\n /^-{3,}\\s*forwarded message\\s*-{3,}/im,\n /^begin forwarded message:/im,\n /^on .+wrote:/im,\n /^>{0,2}\\s*from:.+\\n.*to:.+\\n.*subject:/im,\n ];\n for (const marker of forwardMarkers) {\n const match = text.search(marker);\n if (match > 20) {\n return {\n instructions: text.slice(0, match).trim(),\n emailContent: text.slice(match).trim(),\n };\n }\n }\n return { instructions: \"\", emailContent: text };\n}\n\nfunction extractEmailAddress(from: string): string {\n const match = from.match(/<([^>]+)>/);\n return match ? match[1] : from.trim();\n}\n\nfunction extractSenderName(from: string): string {\n const match = from.match(/^(.+?)\\s* {\n const token = Deno.env.get(\"TELEGRAM_TOKEN\");\n const chatId = Deno.env.get(\"GROUP_CHAT_ID\");\n if (!token || !chatId) { console.warn(\"Telegram not configured — skipping notification\"); return; }\n await fetch(`https://api.telegram.org/bot${token}/sendMessage`, {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({ chat_id: chatId, text, parse_mode: \"Markdown\" }),\n });\n}\n\n// ── Main Handler ──────────────────────────────────────────────────────────────\n\nexport default async function (incomingEmail: {\n from: string;\n to: string[];\n subject?: string;\n text?: string;\n html?: string;\n}): Promise {\n console.log(`Email received from: ${incomingEmail.from} | subject: ${incomingEmail.subject}`);\n\n await initDb();\n\n const subject = incomingEmail.subject || \"(no subject)\";\n const body = incomingEmail.text || \"\";\n const senderAddress = extractEmailAddress(incomingEmail.from || \"\");\n const senderName = extractSenderName(incomingEmail.from || \"\") || \"Someone\";\n\n if (!body.trim()) {\n console.log(\"Empty email body — skipping\");\n return;\n }\n\n const { instructions, emailContent } = parseForwardedEmail(body);\n\n const familyName = Deno.env.get(\"FAMILY_NAME\") || \"the family\";\n const systemPrompt = `You are the shared family operations manager for ${familyName}. A family member has forwarded you an email for processing.\n\nYour job is to read the email and take whatever actions it requires:\n- If it contains an event, appointment, class, or activity with a date → add it to the calendar\n- If it requires follow-up action → add a task\n- If it contains useful reference info (a contact, address, price, deadline, account detail) → store it\n- If it contains multiple actionable items → handle all of them\n- If the forwarder gave specific instructions, follow them\n\nBe decisive. You don't need to ask for confirmation — just act on what the email clearly requires. After taking actions, write a brief summary of what you did.\n\nCurrent UTC time: ${new Date().toISOString()}.\nFor event times, convert to ISO 8601 UTC. When a year isn't specified, assume the next upcoming occurrence.`;\n\n const parts: string[] = [];\n if (instructions) parts.push(`Instructions from ${senderName}: \"${instructions}\"\\n`);\n parts.push(`Subject: ${subject}`);\n parts.push(`From: ${senderName}`);\n parts.push(`\\n${emailContent}`);\n const userMessage = parts.join(\"\\n\");\n\n const finalSummary = await runAgent({\n system: systemPrompt,\n messages: [{ role: \"user\", content: userMessage }],\n tools: TOOLS,\n executeTool: (name, input) => {\n console.log(`Calling tool: ${name}`);\n return executeTool(name, input, { userName: senderName }).then((result) => {\n console.log(`Tool result: ${result}`);\n return result;\n });\n },\n maxTokens: 1024,\n });\n\n // Post to Telegram\n const tgMessage = `📧 *Email processed*\\n*Subject:* ${subject}\\n*From:* ${senderName}\\n\\n${finalSummary}`;\n await sendTelegramMessage(tgMessage);\n\n // Reply to sender\n if (senderAddress) {\n try {\n await sendEmail({\n to: [senderAddress],\n subject: `Re: ${subject}`,\n text: `${finalSummary}\\n\\n— mibbs-bot`,\n });\n } catch (err) {\n console.error(\"Failed to send email reply:\", err);\n }\n }\n\n console.log(\"Email processed successfully\");\n}\n","mibbsy",[206],"0f23e8de-5118-457c-bc99-dd3becaee793","db.ts","0b7fd604-ac69-11f1-8efb-1607ee4eb77e",99,["D",1788970258940],"// Shared database layer for mibbs-bot.\n// Single source of truth for SQLite schema, migrations, and all DB operations.\n\nimport { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\n\nconst TZ = \"America/New_York\";\n\n// ── Utilities ──────────────────────────────────────────────────────────────────\n\nexport function rowsToObjects(result: { rows: any[][]; columns: string[] }): Record[] {\n return result.rows.map((row) => {\n const obj: Record = {};\n result.columns.forEach((col, i) => { obj[col] = row[i]; });\n return obj;\n });\n}\n\n// Convert a date-only string (YYYY-MM-DD) + a local hour to a UTC ISO 8601 instant.\n// Two-pass Intl adjustment handles DST transitions correctly.\nexport function localToUtcIso(dateStr: string, hour: number, tz = TZ): string {\n let utc = Date.UTC(+dateStr.slice(0, 4), +dateStr.slice(5, 7) - 1, +dateStr.slice(8, 10), hour, 0, 0);\n for (let i = 0; i < 2; i++) {\n const parts = new Intl.DateTimeFormat(\"en-US\", {\n timeZone: tz, hour12: false,\n year: \"numeric\", month: \"2-digit\", day: \"2-digit\",\n hour: \"2-digit\", minute: \"2-digit\", second: \"2-digit\",\n }).formatToParts(new Date(utc)).reduce((a, x) => { (a as any)[x.type] = x.value; return a; }, {} as Record);\n const asIfUtc = Date.UTC(+parts.year, +parts.month - 1, +parts.day, +parts.hour, +parts.minute, +parts.second);\n utc -= asIfUtc - utc;\n }\n return new Date(utc).toISOString();\n}\n\nexport async function ensureColumns(table: string, cols: Record): Promise {\n const info = await sqlite.execute(`PRAGMA table_info(${table})`);\n const existing = new Set(rowsToObjects(info).map((r) => r.name as string));\n for (const [name, def] of Object.entries(cols)) {\n if (!existing.has(name)) {\n await sqlite.execute(`ALTER TABLE ${table} ADD COLUMN ${name} ${def}`);\n }\n }\n}\n\n// ── Schema ─────────────────────────────────────────────────────────────────────\n\n// Superset initDb — safe to call from any file; all statements are idempotent.\nexport async function initDb(): Promise {\n await sqlite.execute(`CREATE TABLE IF NOT EXISTS fm_tasks (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n text TEXT NOT NULL,\n created_by TEXT,\n created_at TEXT NOT NULL,\n due_date TEXT,\n completed INTEGER NOT NULL DEFAULT 0,\n completed_at TEXT\n )`);\n await sqlite.execute(`CREATE TABLE IF NOT EXISTS fm_info (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n key TEXT NOT NULL,\n value TEXT NOT NULL,\n created_by TEXT,\n created_at TEXT NOT NULL\n )`);\n await sqlite.execute(`CREATE TABLE IF NOT EXISTS fm_messages (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n chat_id INTEGER NOT NULL,\n role TEXT NOT NULL,\n content TEXT NOT NULL,\n user_name TEXT,\n created_at TEXT NOT NULL\n )`);\n await sqlite.execute(`CREATE TABLE IF NOT EXISTS fm_users (\n telegram_id INTEGER PRIMARY KEY,\n first_name TEXT,\n username TEXT,\n updated_at TEXT NOT NULL\n )`);\n // Additive migrations — safe to run against existing prod table\n await ensureColumns(\"fm_tasks\", {\n owner: \"TEXT\",\n remind_at: \"TEXT\",\n reminded: \"INTEGER NOT NULL DEFAULT 0\",\n });\n}\n\n// ── Tasks ─────────────────────────────────────────────────────────────────────\n\nexport async function addTask(text: string, createdBy: string, dueDate?: string, owner?: string): Promise {\n let remindAt: string | null = null;\n if (dueDate) {\n remindAt = dueDate.includes(\"T\") ? dueDate : localToUtcIso(dueDate, 8);\n }\n await sqlite.execute({\n sql: \"INSERT INTO fm_tasks (text, created_by, created_at, due_date, owner, remind_at, reminded) VALUES (?, ?, ?, ?, ?, ?, 0)\",\n args: [text, createdBy, new Date().toISOString(), dueDate ?? null, owner ?? null, remindAt],\n });\n}\n\nexport async function listTasks(): Promise<{ id: number; text: string; created_by: string; due_date: string | null; owner: string | null; remind_at: string | null }[]> {\n const result = await sqlite.execute(\"SELECT id, text, created_by, due_date, owner, remind_at FROM fm_tasks WHERE completed = 0 ORDER BY created_at DESC\");\n return rowsToObjects(result) as any;\n}\n\nexport async function completeTask(id: number): Promise {\n const result = await sqlite.execute({\n sql: \"UPDATE fm_tasks SET completed = 1, completed_at = ? WHERE id = ? AND completed = 0\",\n args: [new Date().toISOString(), id],\n });\n return (result.rowsAffected ?? 0) > 0;\n}\n\nexport async function updateTask(id: number, owner?: string, dueDate?: string): Promise {\n const fields: string[] = [];\n const args: any[] = [];\n if (owner !== undefined) {\n fields.push(\"owner = ?\");\n args.push(owner);\n }\n if (dueDate !== undefined) {\n const remindAt = dueDate ? (dueDate.includes(\"T\") ? dueDate : localToUtcIso(dueDate, 8)) : null;\n fields.push(\"due_date = ?\", \"remind_at = ?\", \"reminded = 0\");\n args.push(dueDate || null, remindAt);\n }\n if (!fields.length) return false;\n args.push(id);\n const result = await sqlite.execute({\n sql: `UPDATE fm_tasks SET ${fields.join(\", \")} WHERE id = ? AND completed = 0`,\n args,\n });\n return (result.rowsAffected ?? 0) > 0;\n}\n\n// ── Info ──────────────────────────────────────────────────────────────────────\n\nexport async function storeInfo(key: string, value: string, createdBy: string): Promise {\n await sqlite.execute({\n sql: \"INSERT INTO fm_info (key, value, created_by, created_at) VALUES (?, ?, ?, ?)\",\n args: [key, value, createdBy, new Date().toISOString()],\n });\n}\n\nexport async function getInfo(query?: string): Promise<{ key: string; value: string }[]> {\n let result;\n if (query) {\n result = await sqlite.execute({\n sql: \"SELECT key, value FROM fm_info WHERE key LIKE ? OR value LIKE ? ORDER BY created_at DESC\",\n args: [`%${query}%`, `%${query}%`],\n });\n } else {\n result = await sqlite.execute(\"SELECT key, value FROM fm_info ORDER BY created_at DESC\");\n }\n return rowsToObjects(result) as any;\n}\n\n// ── Messages ──────────────────────────────────────────────────────────────────\n\nexport async function saveMessage(chatId: number, role: \"user\" | \"assistant\", content: string, userName?: string): Promise {\n try {\n await sqlite.execute({\n sql: \"INSERT INTO fm_messages (chat_id, role, content, user_name, created_at) VALUES (?, ?, ?, ?, ?)\",\n args: [chatId, role, content, userName ?? null, new Date().toISOString()],\n });\n } catch (err) {\n console.error(\"saveMessage: INSERT failed:\", err);\n }\n}\n\nexport async function getRecentMessages(chatId: number, limit = 20): Promise<{ role: string; content: string }[]> {\n const result = await sqlite.execute({\n sql: \"SELECT role, content FROM fm_messages WHERE chat_id = ? ORDER BY id DESC LIMIT ?\",\n args: [chatId, limit],\n });\n return rowsToObjects(result).reverse() as any;\n}\n\n// ── Users ─────────────────────────────────────────────────────────────────────\n\nexport async function upsertUser(telegramId: number, firstName: string | null, username: string | null): Promise {\n await sqlite.execute({\n sql: `INSERT INTO fm_users (telegram_id, first_name, username, updated_at)\n VALUES (?, ?, ?, ?)\n ON CONFLICT(telegram_id) DO UPDATE SET\n first_name = excluded.first_name,\n username = excluded.username,\n updated_at = excluded.updated_at`,\n args: [telegramId, firstName, username, new Date().toISOString()],\n });\n}\n",[216],"cc4805be-bf2a-484a-9131-84f29f41ff62","cron.ts","bbeca6ca-aa52-11f1-b0ae-1607ee4eb77e",106,["D",1788970637948],8,"// Family Manager Bot — Weekly Briefing Cron\n// Cron: 0 23 * * 0\n// Runs every Sunday at 7pm ET (23:00 UTC). Claude synthesizes a \"what does this week require?\" briefing.\n\nimport { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { createMessage } from \"./llm.ts\";\nimport { getGoogleAccessToken, fetchCalendarEvents } from \"./google.ts\";\nimport { fetchForecast, wmoDesc } from \"./weather.ts\";\nimport { rowsToObjects } from \"./db.ts\";\n\nconst TELEGRAM_TOKEN = Deno.env.get(\"TELEGRAM_TOKEN\")!;\nconst GROUP_CHAT_ID = Deno.env.get(\"GROUP_CHAT_ID\")!;\nconst HOME_LOCATION = Deno.env.get(\"HOME_LOCATION\") || \"Boston, MA\";\nconst TZ = \"America/New_York\";\n\nfunction formatEventList(events: any[]): string {\n if (!events.length) return \"None.\";\n return events.map((e: any) => {\n const start = e.start?.dateTime || e.start?.date;\n const dateStr = new Date(start).toLocaleDateString(\"en-US\", { weekday: \"short\", month: \"short\", day: \"numeric\", timeZone: TZ });\n const timeStr = e.start?.dateTime\n ? new Date(start).toLocaleTimeString(\"en-US\", { hour: \"numeric\", minute: \"2-digit\", timeZone: TZ })\n : \"all day\";\n return `• ${e.summary} — ${dateStr}${e.start?.dateTime ? ` @ ${timeStr}` : \"\"}`;\n }).join(\"\\n\");\n}\n\nasync function fetchOpenTasks(): Promise {\n try {\n const result = await sqlite.execute(\n \"SELECT id, text, due_date FROM fm_tasks WHERE completed = 0 ORDER BY due_date ASC NULLS LAST, created_at DESC LIMIT 15\",\n );\n const tasks = rowsToObjects(result);\n if (!tasks.length) return \"None.\";\n return tasks.map(t => `• [${t.id}] ${t.text}${t.due_date ? ` (due: ${t.due_date})` : \"\"}`).join(\"\\n\");\n } catch { return \"Could not fetch tasks.\"; }\n}\n\nasync function fetchFamilyInfo(): Promise {\n try {\n const result = await sqlite.execute(\n \"SELECT key, value FROM fm_info ORDER BY created_at DESC LIMIT 15\",\n );\n const items = rowsToObjects(result);\n if (!items.length) return \"\";\n return items.map(i => `${i.key}: ${i.value}`).join(\"\\n\");\n } catch { return \"\"; }\n}\n\nasync function fetchWeekForecastText(location: string): Promise {\n const result = await fetchForecast(location, 7);\n if (!result) return \"\";\n return result.daily.map((d) => {\n const label = new Date(d.date + \"T12:00:00\").toLocaleDateString(\"en-US\", { weekday: \"short\", month: \"short\", day: \"numeric\" });\n return `${label}: ${wmoDesc(d.weathercode)}, ${d.hi}°/${d.lo}°F${d.precip > 0.05 ? ` (${d.precip.toFixed(1)}\" rain)` : \"\"}`;\n }).join(\"\\n\");\n}\n\nexport default async function () {\n if (!GROUP_CHAT_ID) return;\n\n const token = await getGoogleAccessToken();\n if (!token) { console.error(\"No Google access token\"); return; }\n\n const weekCutoff = Date.now() + 7 * 86400000;\n const [allTwoWeeks, openTasks, familyInfo, weatherForecast] = await Promise.all([\n fetchCalendarEvents(token, 14),\n fetchOpenTasks(),\n fetchFamilyInfo(),\n fetchWeekForecastText(HOME_LOCATION),\n ]);\n\n const thisWeekEvents = allTwoWeeks.filter((e: any) =>\n new Date(e.start?.dateTime || e.start?.date).getTime() < weekCutoff,\n );\n const nextWeekEvents = allTwoWeeks.filter((e: any) =>\n new Date(e.start?.dateTime || e.start?.date).getTime() >= weekCutoff,\n );\n\n const now = new Date();\n const dateLabel = now.toLocaleDateString(\"en-US\", { weekday: \"long\", month: \"long\", day: \"numeric\", timeZone: TZ });\n\n const prompt = `You are the family's Sunday evening weekly briefer. Today is ${dateLabel}.\n\nYour job is NOT just to list what's on the calendar. Answer: \"What does this week require us to do?\"\n\nLook for things like:\n- Events needing preparation (gifts to buy, forms to fill, logistics to arrange, packing)\n- Potential conflicts (both parents busy at the same time, solo bedtime coverage needed)\n- Overdue or urgent tasks\n- Weather that affects plans (rainy weekend = indoor backup needed, etc.)\n- Things that are unresolved or need a decision this week\n\nCALENDAR THIS WEEK:\n${formatEventList(thisWeekEvents)}\n\n${nextWeekEvents.length ? `COMING UP NEXT WEEK (heads up):\\n${formatEventList(nextWeekEvents)}\\n` : \"\"}OPEN TASKS:\n${openTasks}\n\n${weatherForecast ? `WEATHER FORECAST (${HOME_LOCATION}):\\n${weatherForecast}\\n` : \"\"}${familyInfo ? `STORED FAMILY INFO:\\n${familyInfo}` : \"\"}\n\nWrite the weekly briefing. Be warm but practical. Use Telegram Markdown (*bold* for emphasis). Lead with a one-line vibe for the week. Use short bullets for the key \"to-dos this week\" items. End with a brief encouraging line. Aim for ~200 words — something a tired parent will actually read on Sunday evening.`;\n\n const message = await createMessage({ prompt, maxTokens: 600 }) || \"Have a great week! 🌟\";\n\n await fetch(`https://api.telegram.org/bot${TELEGRAM_TOKEN}/sendMessage`, {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({ chat_id: GROUP_CHAT_ID, text: message, parse_mode: \"Markdown\" }),\n });\n\n console.log(\"Weekly briefing sent\");\n}\n",[223],"8080f8a0-b62d-4224-98c4-39bf02e18271","changelog.md","b5f8e242-ab10-11f1-b36b-1607ee4eb77e",116,["D",1788988981710],"# mibbs-bot Changelog\n\nEvery release gets a crayon name.\n**Major versions** debut a new base color — a meaningful new capability era.\n**Minor versions** are shades of that color — additions and polish within the era.\n\n---\n\n## 🖍️ Cerulean (v2.2) — Quieter acknowledgments\n*September 2026*\n\nThe bot now reacts with an emoji instead of writing a full reply when the action was simple and unambiguous. Less noise, same reliability.\n\n**Emoji reactions** — \"remind me at 7:30am to put out jackets\" → 👍 on your message. No sentence needed.\n**Smart fallback** — if Telegram rejects the reaction (group settings, old message), the emoji is sent as a plain message instead.\n**Text replies preserved** — questions, lookups, ambiguous requests, and anything with useful info to read still get a full response.\n\n---\n\n## 🐛 v2.1.1 — Timezone fix\n*September 2026*\n\nCalendar events now create in Eastern Time by default. Previously, the bot was converting times to UTC before passing them to Google Calendar, causing events to land in the wrong timezone (PST). Fixed by passing datetimes in local ET format so Google Calendar's timezone field is respected.\n\n---\n\n## 🖍️ Navy Blue (v2.1) — Reminders that ping you\n*September 2026*\n\nTasks now know who owns them and when to fire. The bot sends a proactive reminder — with a real @mention — so the right person's phone notifies them, not just the group.\n\n**Task owners** — \"remind Marisa to call the dentist\" sets her as owner and @mentions her in the ping.\n**Timed reminders** — a time in the due date pings at that exact moment; a date-only due date pings at 8am ET.\n**Auto-learned identity** — no setup needed; the bot learns each user's Telegram ID from their messages automatically.\n\n---\n\n## 🖍️ Blue (v2.0) — Proactive Era\n*September 2026*\n\nThe bot no longer waits to be asked. It shows up.\n\n**Daily 7am briefing** — Every morning: today's schedule, weather, and open tasks in one message.\n**Sunday week-ahead** — Claude synthesizes the week, answering \"what does this week *require* of you?\" — not just lists events.\n\n---\n\n## 🖍️ Scarlet (v1.2) — Eyes\n*September 2026*\n\nSend a photo to the chat and the bot can see it. Works for anything: whiteboards, handwritten lists, screenshots, receipts, forms with due dates. Claude describes what it sees and can act on it (add a task, save info, etc.).\n\n---\n\n## 🖍️ Brick Red (v1.1) — Weather\n*September 2026*\n\nAsk about the weather anywhere, anytime. \"What's the weather in Boston this weekend?\" — uses Open-Meteo for free, no API key needed.\n\n---\n\n## 🖍️ Red (v1.0) — First Coat\n*September 2026*\n\nThe foundation: shared task list, Google Calendar (read + write), and a key/value info store (\"remember the wifi password is…\"). Natural language throughout — no slash commands required.\n",[231],"1b4c56ac-d913-47be-bb4a-96a908149a9f","announceRelease.ts","ce8043f0-ab10-11f1-be1f-1607ee4eb77e",49,["D",1788822541740],1,"// Run this script after shipping a major release to announce it in the family chat.\n// Update VERSION and MESSAGE below before running.\n// Usage: run this file via Val Town's run_file tool or the \"Run\" button in the editor.\n\nconst TELEGRAM_TOKEN = Deno.env.get(\"TELEGRAM_TOKEN\")!;\nconst GROUP_CHAT_ID = Deno.env.get(\"GROUP_CHAT_ID\")!;\n\n// ── Update these two constants for each major release ──────────────────────\n\nconst VERSION = \"Blue (v2.0)\";\n\nconst MESSAGE = `🖍️ *mibbs-bot — ${VERSION}*\n\nThe bot is now proactive. It shows up before you ask.\n\n*What's new:*\n• *Daily 7am briefing* — Every morning: today's events, weather, and open tasks\n• *Sunday night brief* — Claude reads the whole week and asks \"what does this require of you?\" — not just lists events\n\n*Ideas to try:*\n• Add tasks with due dates and they'll surface in tomorrow's morning message\n• Tell me about an upcoming birthday party — the Sunday brief will notice if a gift isn't sorted\n• Watch the Sunday brief flag conflicts you hadn't noticed yet`;\n\n// ───────────────────────────────────────────────────────────────────────────\n\nasync function announce() {\n if (!GROUP_CHAT_ID) {\n console.error(\"GROUP_CHAT_ID not set\");\n return;\n }\n\n const resp = await fetch(`https://api.telegram.org/bot${TELEGRAM_TOKEN}/sendMessage`, {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({ chat_id: GROUP_CHAT_ID, text: MESSAGE, parse_mode: \"Markdown\" }),\n });\n\n const result = await resp.json();\n if (result.ok) {\n console.log(`✅ Announced ${VERSION} to the family chat`);\n } else {\n console.error(\"Failed to send announcement:\", result);\n }\n}\n\nexport default announce;\n\nawait announce();\n",[238],"19b95ced-e3ea-41eb-b599-2f760c3c20ce","CLAUDE.md","c5d57522-ab10-11f1-9d03-1607ee4eb77e",109,["D",1788970714248],"# mibbs-bot — Project Guide for Claude\n\nThis is a family manager Telegram bot running on Val Town. It uses Claude (Anthropic) for natural language understanding, Google Calendar, and SQLite for storage.\n\n## Files\n\n| File | Type | Purpose |\n|---|---|---|\n| `main.ts` | HTTP | Telegram webhook handler, system prompt, agent call |\n| `email.ts` | Email | Email forwarding handler, system prompt, agent call |\n| `cron.ts` | Interval (Sun 7pm ET) | Weekly briefing — Claude synthesizes \"what does this week require?\" |\n| `morning-cron.ts` | Interval (daily 7am ET) | Daily morning briefing with today's events, weather, tasks |\n| `reminder-cron.ts` | Interval (every 15 min) | Proactive reminder pings — fires when a task's `remind_at` is reached |\n| `llm.ts` | Script | **Shared LLM layer**: model config, `createMessage()` (one-shot), `runAgent()` (agentic tool-use loop — both Anthropic and GPT paths) |\n| `tools.ts` | Script | **Shared toolset**: `TOOLS` array (10 tools) + `executeTool()` switch |\n| `db.ts` | Script | **Shared DB layer**: `initDb()`, all task/info/message/user functions, `rowsToObjects`, `localToUtcIso` |\n| `google.ts` | Script | **Shared Google Calendar layer**: auth + token refresh, `fetchCalendarEvents()`, all CRUD helpers |\n| `weather.ts` | Script | **Shared weather layer**: Open-Meteo, `fetchForecast()` (raw), `getWeather()` (formatted) |\n| `changelog.md` | File | Release history (crayon versioning — see below) |\n| `updates.md` | File | Feature backlog and implementation plans |\n| `CLAUDE.md` | File | This file |\n\n## Architecture\n\nEach deployable val (`main.ts`, `email.ts`, crons) handles its own I/O surface and passes context to the shared layer:\n\n```\n┌─────────────────┐ ┌──────────────────┐ ┌────────────────────┐\n│ main.ts │ │ email.ts │ │ cron.ts / │\n│ (Telegram I/O) │ │ (Email I/O) │ │ morning-cron.ts │\n└────────┬────────┘ └────────┬─────────┘ └─────────┬──────────┘\n │ │ │\n ▼ ▼ ▼\n runAgent() runAgent() createMessage()\n └──────────────────────┘ │\n │ │\n llm.ts (shared) llm.ts (shared)\n │\n ┌─────────┴─────────┐\n ▼ ▼\n tools.ts db.ts / google.ts / weather.ts\n (TOOLS + executeTool) (all shared helpers)\n```\n\n## Env Vars (set in Val Town)\n\n- `TELEGRAM_TOKEN` — Telegram bot token\n- `GROUP_CHAT_ID` — Family group chat ID\n- `ANTHROPIC_API_KEY` — Anthropic API key\n- `GOOGLE_TOKEN_JSON` — Base64-encoded Google OAuth token (Calendar)\n- `GOOGLE_CREDENTIALS_JSON` — Base64-encoded Google OAuth client credentials\n- `CALENDAR_ID` — Google Calendar ID (or \"primary\")\n- `HOME_LOCATION` — Home city for weather (e.g. \"Newton, MA\")\n- `MODEL` — Claude model override (default: claude-sonnet-4-6)\n- `RESPOND_TO_ALL` — \"true\" to respond to all group messages (default: @mention only)\n- `BOT_USERNAME` — Telegram bot username (optional, avoids a getMe API call)\n\n## SQLite Tables\n\n- `fm_tasks` — household task list (id, text, created_by, due_date, completed, owner, remind_at, reminded)\n- `fm_info` — key/value store for household info\n- `fm_messages` — conversation history per chat_id\n- `fm_users` — Telegram user identity map (telegram_id, first_name, username) — auto-populated from inbound messages, used to resolve owner names to @mentions\n\n## Adding New Features\n\n**Adding a new tool** (e.g. `add_note`):\n1. Add DB helpers (if needed) to `db.ts`\n2. Add the `Anthropic.Tool` definition to `TOOLS` in `tools.ts`\n3. Add the case to `executeTool` in `tools.ts`\n4. That's it — `main.ts` and `email.ts` both pick it up automatically\n\n**Adding a new cron** (e.g. Friday outing suggestions):\n1. Create a new `*.ts` interval file\n2. Import `createMessage` from `./llm.ts`, data helpers from `./db.ts`, `./google.ts`, `./weather.ts`\n3. Build the prompt from live data and call `createMessage({ prompt, maxTokens })`\n\n**Adding a new DB table**:\n1. Add the `CREATE TABLE IF NOT EXISTS` + any migrations to `initDb()` in `db.ts`\n2. Add the CRUD functions in `db.ts`\n3. Import them in `tools.ts` and/or whichever val needs them\n\n## Changelog — Crayon Versioning\n\nEvery release gets a **crayon name**. This is the versioning system:\n\n- **Major version** = new base color (Red → Blue → Green → Orange…) — a new capability era\n- **Minor version** = shade of that color — additions and polish within the era\n\n### Color sequence (in order)\n1. 🔴 Red family: Red, Brick Red, Scarlet, Crimson, Mahogany, Maroon\n2. 🔵 Blue family: Blue, Navy Blue, Cerulean, Cornflower, Denim, Cadet Blue\n3. 🟢 Green family: Green, Forest Green, Jungle Green, Fern, Shamrock, Pine Green\n4. 🟠 Orange family: Orange, Burnt Orange, Peach, Apricot, Mango Tango\n5. 🟡 Yellow family: Yellow, Goldenrod, Dandelion, Banana Mania, Sunglow\n6. 🟣 Purple family: Purple, Violet, Plum, Mulberry, Orchid, Lavender\n7. ⚫ Gray/Black family: Silver, Gray, Charcoal, Black, White (reserved for infrastructure-only releases)\n\nCurrent version: **Navy Blue (v2.1)**. Next minor would be **Cerulean (v2.2)**. Next major would be **Green (v3.0)**.\n\n### When to bump versions\n\n**Minor release** — new tool, new capability added to existing era (e.g. web search, open loops, family memory):\n1. Update `changelog.md` — prepend a new `## 🖍️ [Color] (vX.Y) — [Short title]` section\n2. Update `/changelog` response text in `main.ts` (the `CHANGELOG_LATEST` constant)\n3. No Telegram announcement needed for minor releases\n\n**Major release** — new capability era (e.g. proactive behavior, email integration, trip planning):\n1. Update `changelog.md` — prepend a new `## 🖍️ [Color] (vX.0) — [Era name]` section\n2. Update `CHANGELOG_LATEST` in `main.ts`\n3. Update the message in `announceRelease.ts` with the new version notes\n4. Run `announceRelease.ts` (use Val Town's run_file tool) to post the announcement to the family chat\n\n### Changelog entry format\n\n```markdown\n## 🖍️ [Color] (vX.Y) — [Punchy era/feature name]\n*Month Year*\n\nOne-sentence summary of what changed and why it matters.\n\n**Feature 1** — What it does, concretely.\n**Feature 2** — What it does, concretely.\n```\n\nKeep entries short. Focus on what the family can *do* with each release, not implementation details.\n",[246],"pulls","branches","forks","envVars",{"_259":261,"_262":62},"connections",{"_259":62,"_260":62},"count","unfilledCount",15,"emptyCount",{"_42":245,"_32":246,"_30":247,"_146":-5,"_40":41,"_39":248,"_148":25,"_37":317,"_13":180,"_33":20,"_34":250,"_24":25,"_152":-5,"_278":318,"_27":319},{"_42":237,"_32":238,"_30":239,"_146":-5,"_40":41,"_39":240,"_148":58,"_37":314,"_13":242,"_33":150,"_34":243,"_24":25,"_152":-5,"_278":315,"_27":316},{"_42":230,"_32":231,"_30":232,"_146":-5,"_40":41,"_39":233,"_148":25,"_37":311,"_13":163,"_33":20,"_34":235,"_24":25,"_152":-5,"_278":312,"_27":313},{"_42":222,"_32":223,"_30":224,"_146":-5,"_40":41,"_39":225,"_148":58,"_37":308,"_13":227,"_33":172,"_34":228,"_24":25,"_152":-5,"_278":309,"_27":310},{"_42":215,"_32":216,"_30":217,"_146":-5,"_40":41,"_39":218,"_148":58,"_37":305,"_13":62,"_33":150,"_34":220,"_24":25,"_152":-5,"_278":306,"_27":307},{"_42":205,"_32":206,"_30":207,"_146":-5,"_40":41,"_39":208,"_148":58,"_37":302,"_13":210,"_33":211,"_34":212,"_24":25,"_152":213,"_278":303,"_27":304},{"_42":198,"_32":199,"_30":200,"_146":-5,"_40":41,"_39":201,"_148":25,"_37":299,"_13":62,"_33":150,"_34":203,"_24":25,"_152":-5,"_278":300,"_27":301},{"_42":191,"_32":192,"_30":193,"_146":-5,"_40":41,"_39":194,"_148":58,"_37":296,"_13":171,"_33":150,"_34":196,"_24":25,"_152":-5,"_278":297,"_27":298},{"_42":183,"_32":184,"_30":56,"_146":-5,"_40":41,"_39":185,"_148":25,"_37":293,"_13":187,"_33":188,"_34":189,"_24":25,"_152":-5,"_278":294,"_27":295},{"_42":175,"_32":176,"_30":177,"_146":-5,"_40":41,"_39":178,"_148":58,"_37":290,"_13":180,"_33":172,"_34":181,"_24":25,"_152":-5,"_278":291,"_27":292},{"_42":166,"_32":167,"_30":168,"_146":-5,"_40":41,"_39":169,"_148":58,"_37":287,"_13":171,"_33":172,"_34":173,"_24":25,"_152":-5,"_278":288,"_27":289},{"_42":158,"_32":159,"_30":160,"_146":-5,"_40":41,"_39":161,"_148":25,"_37":284,"_13":163,"_33":150,"_34":164,"_24":25,"_152":-5,"_278":285,"_27":286},{"_42":154,"_32":28,"_30":31,"_146":-5,"_40":41,"_39":14,"_148":25,"_37":281,"_13":156,"_33":20,"_34":36,"_24":25,"_152":-5,"_278":282,"_27":283},{"_42":143,"_32":144,"_30":145,"_146":-5,"_40":41,"_39":147,"_148":58,"_37":277,"_13":62,"_33":150,"_34":151,"_24":25,"_152":-5,"_278":279,"_27":280},["D",1788970302763],"children",[],[144],["D",1788988992452],[],[28],["D",1788988887845],[],[159],["D",1788970674252],[],[167],["D",1788970656966],[],[176],["D",1788988957949],[],[184],["D",1788970373165],[],[192],["D",1788970288200],[],[199],["D",1788970605248],[],[206],["D",1788970258940],[],[216],["D",1788970637948],[],[223],["D",1788988981710],[],[231],["D",1788822541740],[],[238],["D",1788970714248],[],[246],["D",1788733325122],["D",1789522252250],{"_99":98,"_106":102},{"_99":98,"_100":-5,"_101":102,"_18":54,"_103":104,"_105":25},{"_13":88,"_32":89,"_42":41,"_90":43,"_91":-5,"_92":-5,"_93":-5,"_94":-5,"_95":-5,"_96":44,"_97":98},"ctx",{"_81":-5,"_126":358,"_359":360,"_361":362,"_363":25,"_364":25,"_365":25,"_366":58,"_367":-5,"_368":-7,"_369":-7,"_370":62,"_371":25,"_372":25,"_373":-5,"_374":25,"_375":25,"_376":25,"_377":-5},"ENV",{"_329":330,"_331":332,"_333":89,"_334":335,"_336":337,"_338":339,"_340":341,"_342":343,"_344":345,"_346":347,"_348":349,"_350":351,"_352":353,"_354":355,"_356":357},"REACT_APP_SENTRY_DSN","https://586020d86f05489db63a554690acf1e9@o1398359.ingest.sentry.io/6724499","RENDER_GIT_COMMIT","e87eb04e707cf64e7e532c8e8ec395f209e17f15","RENDER_GIT_BRANCH","POSTHOG_API_TOKEN","phc_nmoyuAkStEZuUp69rYlg1DZgrIt19Gx0Y4Hwf8tGgJP","POSTHOG_API_HOST","https://wart.val.town","WEB_API_TEMPLATE","https://{val}.web.val.run","APP_URL","https://www.val.town","MODULE_URL","https://esm.town","VANITY_WEB_API_TEMPLATE","https://{label}.val.run/","DENOLS_URL","https://lsp.val.town","NODE_ENV","production","CLOUDFLARE_TURNSTILE_SITE_KEY","0x4AAAAAACtaIZJH7s9rBMZ6","GOOGLE_CLIENT_ID","875989696899-jtsqkhlbdv5jrlsk8631p7ipjg3c6o62.apps.googleusercontent.com","STRIPE_PUBLISHABLE_KEY","pk_live_51N2zvWEpZTTLipIkVEwXrUG9AT0yeOOEYxszarZT1jpsZV58Ly6wlg3SltZpHN26CUalOBmUSY7adQRQiSkCZz2500qSXNkf8Y","API_URL","https://api.val.town",[],"preferencesKeybindings","default","preferencesTheme","light","preferencesFixers","preferencesCopilot","preferencesSelectionAi","preferencesCloseBrackets","preferencesChatPanelState","preferencesDprint","preferencesDenofmt","notificationCount","flagDenols","isBusinessOrgMember","businessOrgHandle","isBusinessOrgAdmin","isHigherTownieTier","navCollapsed","currentOrgId"]

Diff · v117