Governance Coverage Reporter

Illustrative example — demonstrates a well-formed SKILL.md for this reference implementation. Adapt the content to your own org; do not deploy verbatim.

governance-coverage-reporter

Overview

Measures Unity Catalog governance hygiene from the live system.information_schema metadata views: it reports which tables lack an owner, a description/comment, or classification tags, and summarizes documentation coverage across a catalog. This is metadata-only — it reads the catalog's structure and annotations, never the data inside any table — which keeps it a clean Tier-2 (internal) skill.

When to use this skill

Reach for this skill when a data-governance or privacy user asks about catalog hygiene:

• "Which tables in the main catalog have no description?"
• "How many tables are missing an owner?"
• "Show me governance coverage — what share of tables have comments and tags?"
• "Which schemas have the worst documentation coverage?"

Instructions

When the user asks a governance-coverage question:

1. Identify the catalog/schema scope (default: the main catalog).
2. Query system.information_schema.tables for table inventory, owners, and comments.
3. Join system.information_schema.table_tags to assess classification-tag coverage.
4. Present results as a coverage summary + a ranked list of the least-documented schemas (see the Recommendations framework).

Examples

Tables missing a description or owner (in main)

SELECT
  table_schema,
  table_name,
  table_owner,
  comment
FROM system.information_schema.tables
WHERE table_catalog = 'main'
  AND (comment IS NULL OR comment = '' OR table_owner IS NULL)
ORDER BY table_schema, table_name

Documentation coverage by schema

SELECT
  table_schema,
  COUNT(*)                                                        AS total_tables,
  SUM(CASE WHEN comment IS NOT NULL AND comment <> '' THEN 1 ELSE 0 END) AS documented,
  ROUND(
    SUM(CASE WHEN comment IS NOT NULL AND comment <> '' THEN 1 ELSE 0 END) / COUNT(*),
    3
  )                                                               AS doc_coverage
FROM system.information_schema.tables
WHERE table_catalog = 'main'
GROUP BY table_schema
ORDER BY doc_coverage ASC

Classification-tag coverage

SELECT
  t.table_schema,
  COUNT(DISTINCT t.table_name)                     AS total_tables,
  COUNT(DISTINCT tg.table_name)                    AS tagged_tables
FROM system.information_schema.tables t
LEFT JOIN system.information_schema.table_tags tg
  ON t.table_catalog = tg.catalog_name
 AND t.table_schema  = tg.schema_name
 AND t.table_name    = tg.table_name
WHERE t.table_catalog = 'main'
GROUP BY t.table_schema
ORDER BY tagged_tables ASC

Recommendations framework

After presenting results, always include:

1. Biggest gap — the schema with the lowest documentation or tag coverage, named.
2. Coverage headline — the overall share of tables with owners / comments / tags.
3. Quick win — one prioritized action (e.g. "assign owners to the N ownerless tables in schema X first").

Edge cases

• Empty catalog — if the requested catalog has no tables, say so rather than returning a zero-row summary with no context.
• Permission scope — information_schema only surfaces objects the caller can see; note that coverage is relative to visible objects, not necessarily the whole metastore.
• Metadata only — this skill never reads table contents. If asked about the data inside a table, redirect: that is a data-query task, not a governance-coverage task.

Data scope

• system.information_schema.tables — table inventory, owner, and comment metadata
• system.information_schema.table_tags — classification/governance tag assignments
• No PII or table contents are accessed — this skill reads catalog metadata only.