Skip to main content

avd-alert-kql-validation

Create and review KQL for the FederalAVD AVD Alerts add-on. Use when adding or changing scheduled query alerts, comparing alert logic with Microsoft AVD Insights workbooks, investigating query failures, or documenting intentional workbook deviations.

Source facts

Repository
Azure/FederalAVD
Last source activity
August 18, 2026 at 18:24
Detected SKILL.md language
English
Stars
18
Forks
19

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

File Explorer
3 files

Showing SKILL.md

SKILL.md
Source instructions · Read-only preview
name
avd-alert-kql-validation
description
Create and review KQL for the FederalAVD AVD Alerts add-on. Use when adding or changing scheduled query alerts, comparing alert logic with Microsoft AVD Insights workbooks, investigating query failures, or documenting intentional workbook deviations.
argument-hint
[alert resource, KQL condition, or module]
# AVD Alert KQL Validation Use this workflow for log-based alerts under `deployments/add-ons/avdAlerts/`. ## Procedure 1. Read `deployments/add-ons/avdAlerts/README.md`, especially **Query Design and Validation**, and locate the corresponding operator response in `ALERT-RESPONSE.md`. 2. Identify the authoritative AVD Insights workbook for the condition. Use the official Microsoft workbook sources listed in [workbook sources](./references/workbook-sources.md). 3. Extract and structurally validate the repository queries: ```powershell & .github/skills/avd-alert-kql-validation/scripts/Test-AvdAlertQueries.ps1 ` -ModulePath deployments/add-ons/avdAlerts/modules ``` 4. To inspect standalone KQL, add `-OutputDirectory <temporary-folder>`. Do not commit generated extraction output. 5. Compare table selection, joins, event names or IDs, status filters, casts, deduplication, and aggregation with the workbook query covering the same condition. 6. Adapt interactive workbook queries for alert semantics: remove `render`, return rows suitable for the configured aggregation and threshold, and use a bounded time window. 7. Verify `overrideQueryTimeRange` is at least `windowSize` multiplied by `numberOfEvaluationPeriods` and that persistent-event behavior matches `autoMitigate`. 8. Preserve cloud compatibility. Do not introduce workspace-specific functions or preview-only KQL without confirming support in every target cloud. 9. Document every intentional deviation in the add-on README. The reason must describe alert semantics, noise suppression, cloud compatibility, or source-data behavior. 10. Build the affected Bicep entry template and update `ALERT-RESPONSE.md` when operator action, thresholds, or dimensions change.
View on GitHub