Adx Schema Check

ADX / Kusto Schema Check

Cross-checks code-side KQL queries against the source-of-truth schema, since table/column name typos or renames are a common source of silent or 400-level ADX query failures.

1. Find the source-of-truth schema

Look for the canonical schema definition, typically one or both of:

• A KQL setup script (e.g. scripts/adx-schema-setup.kql) with .create table

statements.

• A Terraform-managed schema (e.g. modules/data/adx_schema.tf using

azurerm_kusto_* or azapi resources with embedded KQL).

Extract from these: table names and, per table, column names + types.

2. Find code-side KQL usage

Grep the application/API code for KQL constructs:

• Table references (bare identifiers used as the first pipe-stage source, e.g.

TableName | where ...).

• Column references after where, project, extend, summarize, order by,

join on, etc.

• Look in API route handlers, query-builder helper functions, and any

.kql files.

3. Cross-check and report

For each query found:

• Does the table name exist in the schema? (Case-sensitive — Kusto identifiers

are case-sensitive.)

• Do referenced columns exist on that table with matching names?
• Report mismatches as file:line — schema says X, query uses Y.

Common failure pattern: a schema migration renames a table/column but one query site is missed. Search for all occurrences of the old and new names to confirm consistency project-wide, not just the file you started in.

4. Optional: live cluster drift check

If the project's conventions allow ADX access (via Key Vault + Managed Identity — never via connection strings in code), confirm the deployed schema matches the source files:

.show table  schema as json

If the live cluster differs from the schema source files, that's infra drift — flag it separately from code-side mismatches.

5. Fixing mismatches

• If the schema files are the intended source of truth, fix the query side to

match.

• If you change the schema (table/column names) instead, you must update every

query referencing the old names — re-run this whole check afterward to confirm nothing was missed, and check whether any project rules require keeping schema/query names in sync as a hard constraint.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: sergeyitaly
• Source: sergeyitaly/claude-skill-deployer
• License: MIT

Install and usage instructions live in the source repository linked above.