decision-labs/geobase-skills/geobase-embeddings-create-via-workers

Create Embeddings via Workers

When to use this skill

• User wants a new embeddings table or pipeline output in geoembeddings schema
• Before @geobase-worker-srai-embeddings or @geobase-worker-geoai-embeddings
• Prevent direct SQL/RPC table creation shortcuts

Do not create embeddings tables through ad-hoc direct RPC flows. Use worker job APIs.

Core rules

• Worker jobs are project-scoped and run asynchronously (pending/running/success/failed).
• Use worker job routes (project host /worker/jobs/*) or Studio project job proxies.
• Treat table creation as a create-operation with strict name validation/collision checks.

Task-specific guides

• @geobase-worker-srai-embeddings — SRAI GeoEmbeddings: OSM features → H3 regions → CountEmbedder / Hex2Vec / GeoVex → geoembeddings.* table (createSraiEmbeddings)
• @geobase-worker-geoai-embeddings — GeoAI / imagery: GeoTIFF tiles → vision models (e.g. DINOv) → `geoembeddings. (createGeoEmbeddings`)
• @geobase-worker-osm-import (context only; OSM import is not itself an embeddings creation flow, but can be upstream data prep).

Required inputs

• authenticated CLI session
• project_ref
• job payload matching the selected worker task
• `GEOBASE_SERVICE_ROLE_KEY` in user-provided `.env.secrets` for worker HTTP (human in the loop; see @geobase)

Verification

• confirm job moved to terminal state
• confirm expected embeddings table/catalogue entry exists
• record job id + status for traceability