Skip to content

architecture.md

Latest commit

 

History

History
59 lines (46 loc) · 2.2 KB

architecture.md

File metadata and controls

59 lines (46 loc) · 2.2 KB
Raw

Architecture Decisions

Design choices and rationale behind the skills pack.

Tools & Safety

  • Tools used: RunAzCliReadCommands + execute_kusto_query — read-only by design for safety
  • No write operations: Skills analyze and report; remediation commands are suggested but not auto-executed
  • Remediation guidance: Each skill includes inline 📖 Microsoft Learn links and instructs the agent to validate CLI commands via GetAzCliHelp before suggesting

Implementation Details

  • Date handling: Uses Linux shell $(date -u -d '...' +%Y-%m-%dT%H:%M:%SZ) syntax (SRE Agent sandbox is Linux)
  • Cost data: Uses az costmanagement query (stable) instead of deprecated az consumption commands
  • Quota increases: Documents REST API approach (az rest) since az quota CLI is preview/unstable
  • Resource Graph: Used as fallback when az cli commands require --resource-group but subscription-wide scan is needed
  • Kusto queries: Skills 07 and 08 use execute_kusto_query for Log Analytics data (429 throttling, token trends)

Skill Structure

Each SKILL.md follows a consistent structure:

---
name: skill-name
description: >
  Clear description with trigger keywords for intent matching.
tools:
  - RunAzCliReadCommands
  - execute_kusto_query  (if Kusto queries needed)
---

# Skill Title
## Purpose
## When to use this skill
## Pre-check
## Assessment procedure (with checks, commands, scoring tables)
## Scoring
## Accepted exceptions (optional)
## Expected output
## Sample output
## Remediation guidance
## References

Scoring Philosophy

  • Every skill produces a numeric score (0–100) with maturity levels
  • Scores are designed for trend tracking — run monthly and compare
  • Maturity labels describe Azure governance state, not business maturity
  • Each check has explicit pass/partial/fail criteria with point values

Safety Model

Skills operate in a suggest, don't execute model:

  1. Skills use RunAzCliReadCommands — read-only commands only
  2. Remediation commands are printed as suggestions in the output
  3. The agent validates command syntax via GetAzCliHelp before suggesting
  4. No changes are made to the Azure environment automatically