AI Data and Knowledge-Base Spec Skill

Overview

The data-and-knowledge artefact that retrieval-augmented and fine-tuned features absolutely need but the generic Database Design skill does not produce. Captures lineage, retention, freshness, and the cross-tenant control story.

Core Instructions

Step 1: Knowledge source inventory

For every source feeding an AI feature: name, type (document store / data warehouse / external API / customer-uploaded), owner, classification (public / internal / confidential / PII / SPI), volumes, refresh cadence.

Step 2: Per-source scope

For each source state whether it is shared across tenants (e.g. a public-document corpus) or per-tenant. Shared sources MUST also state the licence / copyright / opt-out posture.

Step 3: Ingestion pipeline

For each source: ingestion mode (push / pull / event), schedule, transformation, chunking strategy (size, overlap), embedding model, vector store, index segmentation rule (per-tenant index, namespace, metadata filter).