@skeems/wikidata

Read-only WikiData (and any Wikibase instance) adapter for skeem.

Usage

npm install -g skeem @skeems/wikidata

adapter: wikidata
connection:
  url: https://www.wikidata.org   # or any Wikibase instance
  language: en                    # label/description language (default: en)

No authentication is required. The adapter talks to the MediaWiki Action API (/w/api.php) and sends a descriptive User-Agent per Wikimedia etiquette.

Tier: Read

WikiData is a knowledge graph, not a table store, so it maps to two synthetic, read-only collections:

• items — Q-id entities (people, places, works, concepts, …)
• properties — P-id properties (the predicates used in statements)

Each is keyed by id, with stable label, description, and aliases fields.

| Capability | Supported | | --- | --- | | ls, describe, discover | ✅ | | get <items\|properties> <id> | ✅ (full entity: simplified claims + externalIds) | | find <items\|properties> --where label=<text> | ✅ (search; --offset supported) | | find … --where id=<id> | ✅ (summary lookup) | | find items --where imdb=… / doi=… / P####=… | ✅ (reverse lookup by external id) | | count | returns null (the corpus is effectively unbounded) | | create / update / delete | ❌ read-only | | define / init / link | ❌ |

Examples

skeem get items Q42 --json                          # Douglas Adams: claims + externalIds
skeem find items --where label="Douglas Adams" --json
skeem find items --where imdb=nm0010930 --json      # reverse: external id -> Q42
skeem get properties P31 --json                     # the "instance of" property
skeem find properties --where label="instance of" --json

Claims and external identifiers

get splits an entity's statements into two objects:

• externalIds — statements whose property has the external-id datatype (Q42 carries 245 of them), keyed by the shared vocabulary in docs/reference/external-ids.md where curated (imdb, doi, musicbrainz_artist, …) and by the raw P#### otherwise. P856 (official website) and the language's Wikipedia sitelink are folded in as official_website / wikipedia. Values are always arrays of strings.
• claims — everything else, flattened to { propertyId: [value, ...] }. Entity references collapse to their id (e.g. "Q5"), times to their timestamp string, quantities to their amount, monolingual text to its string, and coordinates to { latitude, longitude }. Property ids and entity references are not resolved to labels — resolve them with another get if needed.

Changed in 0.6.0: external-id statements used to appear inside claims (e.g. claims.P345); they now live in externalIds under friendly keys.

Reverse lookup

find items --where <key>=<value> accepts one external-id filter — a curated key (imdb, doi, viaf, …) or any raw P#### — and resolves it via CirrusSearch haswbstatement. Matching is exact and case-sensitive (imdb values are lowercased for you); --limit is capped at 50; items only.

This simplification is intentionally lossy: it aims for a readable payload an agent can act on, not a faithful RDF reproduction. For full statement structure or graph queries, use the SPARQL endpoint directly.