feature: improve postgres best practices and add evals

[Rodriguespn]

Summary

This PR introduces PostgreSQL version awareness and a comprehensive evaluation framework to the postgres-best-practices skill. The changes ensure AI agents only recommend rules compatible with the target PostgreSQL version and available extensions.

Changes

1. Rule Version & Extension Metadata

Added version constraints and extension requirements to rules that depend on specific PostgreSQL features:

Rule	Added Metadata	Requirement
query-covering-indexes.md	minVersion: "11"	INCLUDE clause requires PG11+
schema-partitioning.md	minVersion: "10"	Declarative partitioning requires PG10+
data-upsert.md	minVersion: "9.5"	ON CONFLICT requires PG9.5+
lock-skip-locked.md	minVersion: "9.5"	SKIP LOCKED requires PG9.5+
advanced-jsonb-indexing.md	minVersion: "9.4"	JSONB type requires PG9.4+
monitor-pg-stat-statements.md	extensions: pg_stat_statements	Requires extension availability

2. Profile System

Introduced deployment profiles to customize rule filtering based on environment constraints:

Profiles Added:

• supabase.json: Supabase platform (PG15+, rich extension support including pg_stat_statements, pgjwt, pg_graphql)
• aurora.json: AWS Aurora PostgreSQL (PG13-16, limited extensions, pg_cron/timescaledb unavailable)
• self-hosted.json: Generic self-hosted (PG12+, all extensions installable)

Each profile specifies:

• Minimum/maximum PostgreSQL version
• Available, installable, and unavailable extensions
• Environment-specific notes

3. Build System Enhancements

• New profiles.ts module with version comparison and compatibility checking
• isRuleCompatibleWithProfile() function filters rules based on version and extension requirements
• Build now generates profile-specific outputs: AGENTS.supabase.md, AGENTS.aurora.md, AGENTS.self-hosted.md

4. SKILL.md Documentation

Added runtime detection guidance for agents:

• SQL queries to detect PostgreSQL version
• Extension availability checking via pg_available_extensions
• Configuration verification via pg_settings
• Version compatibility matrix

Evaluation Scenarios

A comprehensive eval framework was added to test agent behavior with version and extension constraints.

Framework

• Test Runner: Vitest + Vercel AI SDK (Claude integration)
• Location: skills/postgres-best-practices/evals/
• Run: npm run eval (requires ANTHROPIC_API_KEY)

Scenarios by Category

Category 1: Core Query Patterns

ID	Name	Difficulty	What it tests
missing-index-detection	Missing Index	Basic	Agent identifies missing indexes on filtered columns
n-plus-one-detection	N+1 Query	Intermediate	Agent recognizes N+1 pattern and recommends JOINs
covering-index-suggestion	Covering Index	Intermediate	Agent suggests INCLUDE clause for covering indexes

Category 2: Version Constraints

ID	Name	Difficulty	What it tests
version-constraint-pg10-no-covering	PG10 No Covering Index	Intermediate	Agent does NOT recommend INCLUDE on PG10 (requires PG11+)
version-constraint-pg93-no-upsert	PG9.3 No UPSERT	Intermediate	Agent does NOT recommend ON CONFLICT on PG9.3 (requires PG9.5+)

Category 3: Extension Requirements

ID	Name	Difficulty	What it tests
extension-available-pg-stat-statements	Extension Available	Basic	Agent recommends pg_stat_statements when available
extension-unavailable-no-pg-stat-statements	Extension Unavailable	Intermediate	Agent suggests alternatives when extension unavailable