postgres-semantic

name	postgres-semantic
description	Use for ANY PostgreSQL interaction — queries, schema exploration, metrics, or business data questions. MUST be used before postgres tools to load semantic models containing business context, field mappings, and validated metric formulas.

PostgreSQL Semantic Model Integration

Load semantic models before any database operation. They contain business rules, exact field expressions, and validated metrics that make queries correct.

Workflow

Step 1: Check for semantic models (MANDATORY FIRST STEP)

Before calling any database tool, list the semantic models at /products:

list_files(path="/products")

If the directory is empty or missing, skip to Step 5 and query directly.

Step 2: Load relevant models

Load model files matching the user's question:

Question contains	Load
orders, revenue, transactions, payments, pricing	sales-transaction-ledger/osi.yml
customers, users, sessions, acquisition, behavior	customer-demographic-master/osi.yml
inventory, stock, warehouse, products, SKUs	inventory-snapshot/osi.yml
unclear or multi-domain	all models

read_file(path="/products/sales-transaction-ledger/osi.yml")
read_file(path="/products/customer-demographic-master/osi.yml")
read_file(path="/products/inventory-snapshot/osi.yml")

Step 3: Extract from each model

• ai_context.instructions (top level) — critical business rules for all queries
• datasets[].source — actual schema.table name to query
• datasets[].fields[].expression — exact SQL expression to use (never guess)
• metrics[].expression — pre-validated SQL calculation (use as-is)
• relationships[] — correct join conditions across domains

Step 4: Apply context

1. Apply domain instructions (e.g., divide cents by 100 for dollar amounts)
2. Use exact field expression values — never invent column names
3. Use pre-defined metric expressions verbatim
4. Use relationships for cross-domain joins
5. Respect field-level ai_context warnings

Step 5: Query the database

show_tables()           # list available tables
describe_table(table="schema.table")  # table structure
run_query(query="SELECT ...")         # execute SQL

Always cite which semantic rule or metric you applied in your response.