wanderlust

name	wanderlust
description	Place discovery within walking distance (Google Places + OSM + Wikipedia + Reddit + Atlas Obscura). Powers (1) the scheduled trip-city daily pings — weather + 3 picks to the trip's iMessage chat, and (2) ad-hoc COFFEE SHOP recommendations in conversation. For ANY message asking about coffee shops anchored to a place ("coffee shop we can walk to from the aquarium", "good coffee near Pike Place", "where can we grab coffee in Capitol Hill") — call goat BEFORE reaching for web_search / tavily / firecrawl. Other place categories (restaurants, sights, events) are NOT in conversational scope yet — use web search for those. Spratt suggests, never books.
homepage	https://github.com/mvanhorn/printing-press-library
metadata	{"clawdbot":{"emoji":"🧭","requires":{"bins":"[Truncated]"}}}

Wanderlust (trip-city discovery)

CLI-driven place discovery for travel. Same engine for ad-hoc "what's near me" queries and the scheduled trip-city digest that pings active-trip iMessage chats with weather + 3 picks each morning.

When to use in conversation

ONE conversational trigger today: coffee shop recommendations anchored to a place.

Examples that MUST route here, not web_search / tavily / firecrawl:

• "coffee shop we can walk to from the Seattle Aquarium"
• "good coffee near Pike Place"
• "where should I grab espresso in Capitol Hill"
• "best third-wave coffee near our hotel"

Command: wanderlust-goat-pp-cli goat "<anchor>" --criteria "third-wave coffee" --minutes 15 --top 5 --agent

Other place categories (restaurants, museums, photo spots, walking routes, etc.) are NOT yet in conversational scope — keep using web_search / tavily / firecrawl for those until Manan expands this list.

Boundary

• Suggest, don't book. Wanderlust is read-only. Reservations live in the table-reservation skill (which has its own booking gate).
• Trip-city messages are non-personalized. They land in the trip's existing iMessage chat (trips.group_chat_guid — phone number or chat GUID), not in Manan/Harshita's per-person briefings. That keeps the briefings focused on home + today, and lets travelers see the same content together.

Prerequisites

• Binary: ~/go/bin/wanderlust-goat-pp-cli
• wanderlust-goat-pp-cli doctor → Auth: not required, GOOGLE_PLACES_API_KEY: set, API: reachable.
• ANTHROPIC_API_KEY enables --llm criteria matching on near. The goat subcommand is deterministic — no LLM, uses static keyword tables. Use goat for cron/automation; use near only when judgment matters.

Command surface

All commands accept --agent for JSON + non-interactive mode.

Command	Purpose
doctor	Config + auth + API reachability.
goat <anchor> --criteria <free-text> --minutes N --top K	Deterministic top-K picks within K-minute walk of anchor. Use this in cron.
near <anchor> --criteria ... --llm	Same fanout as goat with LLM criteria-match. Interactive only.
crossover --anchor ...	Food-near-culture pairings (high-trust restaurant within 200m of Wikipedia-notable site or Atlas Obscura entry).
golden-hour <anchor>	Local sunrise/sunset/blue-hour + photographer viewpoints.
route-view <from> <to>	Walking polyline + interesting things along the path.
quiet-hour <anchor> --at HH:MM	Places locals describe as quiet at the requested time.
reddit-quotes <place>	Verbatim Reddit comments mentioning a place (no LLM summarization).
sync-city <city>	Pre-cache editorial best-of, Reddit, Wikipedia, OSM POIs, Atlas Obscura for offline use.
coverage <city>	Per-tier source coverage report for a synced city.
why <place-id>	Score breakdown — which sources mentioned it, country boost, walking time, criteria match.

Trip-city digest (the scheduled flow)

A daily launchd job runs ~/.config/spratt/skills/wanderlust/scripts/trip-city-digest.py for every status IN ('active','upcoming') trip in trips.sqlite. For each trip:

1. Geocode the destination text via weather-goat-pp-cli config set-location --dry-run (same path the briefings use — keeps geocoding logic in one place).
2. Weather via weather-goat-pp-cli forecast --lat/--lon --agent (current conditions + today's hi/lo).
3. Top 3 picks via wanderlust-goat-pp-cli goat <lat>,<lng> --criteria "<seed>" --minutes 15 --top 3 --agent. Criteria seed rotates day-of-week to keep it fresh (Mon=coffee, Tue=market, Wed=museum, Thu=street food, Fri=bookstore, Sat=viewpoint, Sun=quiet park).
4. Compose a 4-line plain-text message — no markdown, no emojis other than 🌤 and 📍.
5. Skip list: ~/.config/spratt/skills/wanderlust/state/skip-trips.json is a JSON array of trip ids. Any trip whose id is in that list is silently skipped — used to opt a specific trip out without changing code. Currently contains 2026-05-india-dad (Mom's solo May trip).
6. Schedule via outbox.py schedule --to <recipient>:
   • If trips.group_chat_guid is set → use that (works for chat GUIDs OR phone numbers; the outbox accepts both).
   • Else → skip. We don't fan out to individual travelers without an explicit chat target on the trip row.
7. Source tag: wanderlust:<trip-id>:<YYYY-MM-DD> so the outbox dedupes if the job runs twice in one day.

The job is deterministic — no LLM in the runtime path. goat uses static keyword tables, weather is a forecast API, the message template is fixed. Runs as exec-style launchd job.

Workflow — interactive "what should we do near X"

1. wanderlust-goat-pp-cli goat "<anchor>" --criteria "<what>" --minutes 15 --top 5 --agent
2. Present top 3 by name + walking minutes + score's why field.
3. If Manan picks one and it's a restaurant, hand off to table-reservation skill via goat <name> --location <city> --party N to check availability.

Safety

• Never book from this skill. Route booking through table-reservation.
• Never sync-city a destination automatically — that's a deliberate pre-trip step Manan triggers. Cron only runs goat, which works without sync (live Google Places + Nominatim).
• Never post to a trip chat if group_chat_guid is empty — skip and surface to Manan via outbox (source: wanderlust:warning).

Open items

• Sync-city automation — run sync-city <destination> 48h before each trip's start_date. Deferred until the digest has run for a couple of trips.
• Criteria seed tuning — the day-of-week rotation is a guess. Adjust after a week of real data.