Buy-Box

Create or update the buy box for a category. Any field names are allowed. Call get_buy_box first and REUSE the field names already in the box. - criteria: fields to set; each value you send replaces that field's value. A value of null deletes that field. - add: {field: [items]} appends items to any list field (created if missing, duplicates skipped). Use this to add one item without resending the whole list. - remove: {field: [items]} removes items from a list field (case-insensitive for text). - remove_fields: [field, ...] deletes whole fields. - replace=true: start from an empty box instead of merging. Order within one call: remove_fields, then criteria, then add, then remove. Server fields (_meta, requires_confirmation, last_updated) are ignored in criteria. Field names are normalized to snake_case. If a new field looks like an existing one (e.g. max_miles vs max_mileage), it is saved and _meta.warnings names the existing field: tell the user and fix it with remove_fields. Do not store health details, account numbers, or passwords. Returns the saved box.

set_buy_box

Remove the buy box for a category.

delete_buy_box

Return a buy box as readable Markdown the user can read or save.

export_buy_box

Retrieve the saved buy-box criteria for a category (homes, cars, or any freeform name). Call this before searching or filtering. _meta.requires_confirmation is always true: surface the criteria and get explicit user confirmation before acting on results.

get_buy_box

List all categories that have a buy box defined.

list_categories

Check a list of listings against a saved buy box and rank them. Pass each listing as an object with whatever fields you scraped (price, mileage, year, seats, a description, etc.). - Numeric rules (max_X, X_max, min_X, X_min) exclude listings that break them. - dealbreakers found in the listing text exclude it. - must_haves, nice_to_haves, preferred_brands_or_models and target_X / X_target score the listing. - Anything the listing doesn't state is 'unknown', not a pass: verify those with the seller's data before recommending. Excluded listings are returned last with the rule they broke. Show the user the ranked results and get confirmation before acting.

score_candidates