T2: pgvector image embeddings — plugin, schema, seed, hook, tests
- docker-compose: swap postgres image to pgvector/pgvector:pg16
- postgres/init.sql: CREATE EXTENSION vector; image_embedding table
- plugins/embeddings.py: embed_images + search_images_semantic
(sentence-transformers all-MiniLM-L6-v2, lazy-loaded, pgvector <=> cosine)
- plugins/images.py: register_image kicks off background embed worker
- seed.py: seed_embeddings writes 4 embeddings for the mock images
- README: semantic image search section + T3 note
- 11 tests across 4 files, all green:
test_embeddings_plugin.py (4): schema, ordering, idempotency, stub
test_embeddings_real_model.py (3): real MiniLM, acceptance queries
test_register_image_hook.py (2): manifest row, end-to-end hook
test_seed_embeddings.py (2): writes 4, idempotent
- Includes T3 consistency plugin skeleton (4 stub tools)
add264eb04