Add Quality Monitors reference to Unity Catalog skill

[CheeYuTan]

Summary

Documents the manage_uc_monitors MCP tool which had zero skill coverage. Adds 8-quality-monitors.md as a new reference file under the Unity Catalog skill.

What's in the file (384 lines)

• MCP tool reference — all 5 actions (create, get, run_refresh, list_refreshes, delete) with parameters
• Verified patterns — create with/without schedule, get, refresh, list refreshes, delete
• Output table schemas — all 28 profile metrics columns and 22 drift metrics columns
• SQL queries — discovering monitors, querying profiles, detecting drift, checking freshness
• Troubleshooting — 7 real error patterns with solutions

Test evidence — full lifecycle on live workspace

E2E monitor lifecycle

Step	Action	Input	Result	Status
1	Create monitor	table_name="main.pizza_llc.menu_items"	MONITOR_STATUS_ACTIVE, dashboard_id assigned	PASS
2	Get monitor	same table	Full config returned with output table names	PASS
3	Run refresh	same table	refresh_id returned, state=PENDING, trigger=MANUAL	PASS
4	List refreshes	same table	refreshes array with correct shape	PASS
5	Query profile metrics	SELECT * FROM ...menu_items_profile_metrics	All 28 columns present and queryable	PASS
6	Run 2nd refresh	same table	Triggered successfully	PASS
7	Query drift metrics	SELECT * FROM ...menu_items_drift_metrics	8 rows after 2nd refresh (0 after 1st — confirmed 2+ needed)	PASS
8	Delete monitor	same table	Returns {} (empty dict)	PASS

What testing found and fixed

Fix	Before (wrong)	After (correct)	How discovered
Delete response	{"status": "deleted", "table_name": "..."}	{} (empty dict)	Live delete returned {}
Profile metrics schema	19 columns listed	28 columns (added 9 missing)	DESCRIBE TABLE on real output table
Drift metrics schema	14 columns listed	22 columns (added 8 missing)	DESCRIBE TABLE on real output table
assets_dir behavior	"auto-generated"	Clarified MCP vs REST API difference	Tested both paths

Key verified claims

• No list action exists — GET /unity-catalog/monitors returns "No API found"
• Only snapshot monitors via MCP (TimeSeries requires Python SDK)
• Drift metrics require 2+ refreshes to populate (confirmed: 0 rows after 1st, 8 after 2nd)
• Output tables follow {table}_profile_metrics / {table}_drift_metrics naming

Test plan

• CI validation passes
• MCP tool schema verified against JSON descriptor
• Full lifecycle tested: create → get → refresh → query → delete
• All 28 profile metrics columns verified against real output
• All 22 drift metrics columns verified against real output
• All SQL query examples executed successfully
• Test monitors cleaned up after testing