foundry-observability

What problem does it solve?

This Skill fixes a silent telemetry gap where Foundry pilots deploy successfully but App Insights receives no traces, logs, or metrics due to missing or miswired observability connections across Bicep, hosted agents, and ACA workloads.

Core Features & Use Cases

• End-to-end observability wiring: Provisions a single Log Analytics workspace and workspace-bound App Insights, then ensures every workload (MCP servers, bot service, workspace UI, ACA jobs, and cron-triggered containers) uses the same App Insights connection string.
• Hosted-agent connection and RBAC setup: Creates the required account-level Foundry App Insights connection and applies the Monitoring Metrics Publisher RBAC to the platform-managed agent identities.
• Operational diagnostics ready to use: Ships KQL queries to verify that any traces are arriving, to break down MCP tool calls, to debug silent cron failures, and to smoke-test first trace ingestion.
• Silent-failure guardrails: Includes guidance and workarounds for real-world cases where platform auto-injection fails or returns credentials:null, requiring defensively guarded telemetry initialization.

Quick Start

Deploy the skill’s Bicep substrate and run the postprovision hook so the Foundry account gets an account-level App Insights connection and all hosted and ACA components start emitting telemetry.