mvp-list-curate

name	mvp-list-curate
description	Use when extracting candidate MVP List places from a URL, image, pasted itinerary, map link, menu link, or free-form note. Identifies restaurant and point_of_attraction places, resolves canonical references, and creates minimal schema-shaped Markdown candidates for later enrichment.

MVP List Curate

Curate turns raw discovery inputs into one or more candidate Place Markdown records. It does identity resolution and light normalization only; use mvp-list-enrich to fill hours, ratings, dishes, attractions, and family context.

Inputs

• Single place URL: Google Maps, Yelp, Apple Maps, official site, menu page.
• Image or screenshot: extract visible names, signs, addresses, menus, or landmarks first.
• Itinerary text: extract every public place, including alternate options; skip non-public stops such as home breakfast.

Workflow

1. Extract place candidates

   • Keep both restaurants and attractions.
   • Treat sub-items inside a venue as nested facts unless they are independently visitable public places.
   • For itinerary options, create separate candidates for each option.

2. Resolve identity

   • Cross-check name, city, address, coordinates, official site, or map URL.
   • Prefer one canonical reference: Google Maps stable URL/id, Yelp alias, Apple Maps URL/id, or official site.
   • If multiple same-name places remain plausible, stop and ask for disambiguation.

3. Classify

   • restaurant: food service is the primary reason to visit.
   • point_of_attraction: parks, gardens, museums, landmarks, trails, stores used as historic sites, photo spots.

4. Write candidate Markdown when requested

   • Use ../_shared/references/place-markdown-template.md as the shape.
   • Frontmatter must follow ../_shared/references/place.schema.json.
   • Keep uncertain fields as null; do not invent missing address, coordinates, hours, ratings, dishes, or attractions.
   • Set status: candidate.
   • Save into the requested Markdown repo, for example ./demo-md-repo or ./.tmp/gbrain.

Google Maps Parsing

• Decode /place/<name> into the name candidate, treating + as spaces.
• Extract !3d<lat>!4d<lon> as the preferred coordinates.
• Extract !1s<feature_id> or !16s<token> as Google reference hints.
• Ignore sharing parameters such as entry and g_ep for identity.

Output

Return a concise list of candidates:

- <name> — <restaurant | point_of_attraction> — <canonical_reference> — <candidate_file_or_needs_disambiguation>

Then hand off each candidate to mvp-list-enrich when the user wants fully searchable records.