name: pgvector Agent Skills
description: >-
  Packaged agent skills for pgvector, grounded in the extension's own DDL contract.
  pgvector has no HTTP API, so these skills name SQL types, operators, operator classes
  and GUCs rather than operationIds - every identifier in them is declared in
  data-model/pgvector-vector.sql.
generated: '2026-08-27'
method: generated
source: https://github.com/pgvector/pgvector#readme
grounded_in: data-model/pgvector-vector.sql
provider_published: false
provider_published_note: >-
  pgvector publishes no AGENTS.md, no skills directory and no llms.txt; probed
  2026-08-27, both raw.githubusercontent.com paths returned 404. These are authored by
  API Evangelist from the project's public documentation.
skills:
- name: pgvector-semantic-search
  file: skills/pgvector-semantic-search.md
  summary: >-
    Declare a vector column, pick a metric and matching operator class, build an HNSW
    index, and write a k-NN query the planner will use it for - including the three
    conditions that silently defeat the index and the reasons a query returns fewer
    than k rows with no error.
- name: pgvector-quantized-search
  file: skills/pgvector-quantized-search.md
  summary: >-
    Reduce storage and index memory with halfvec, binary quantization or subvector
    indexing, and restore recall with the two-stage rerank the README documents.
    Covers the index dimension ceilings that force this and how to verify the recall
    cost against exact search.