What problem does it solve?
Developers wiring their own LLM app or agent to Grafana Agent Observability often end up with silent failures — generations never arrive, metrics vanish into a no-op OTel provider, or a misconfigured .env ships data to localhost — with no error signal to diagnose. This Skill audits the real code against a fixed gap checklist and verifies through gcx that data actually lands in the stack.
Core Features & Use Cases
- Gap checklist audit: Classifies instrumentation state (none / partial / broken) and checks 12 known failure points with file:line evidence, headed by the missing TracerProvider/MeterProvider silent failure.
- Verified instrument→run→verify loop: Applies minimal confirmed diffs, runs the app, and uses
gcx agento11y agents|conversations|generations to confirm generations arrive, iterating up to 3-4 times.
- Two-channel verification: Separately validates generation ingest (via gcx) and OTel traces/metrics (by construction, runtime OTLP POSTs, or the Performance/Tempo UI), since gcx cannot see the OTel channel.
- Use Case: A developer says "my generations aren't showing up" — the Skill reads the app, finds the .env is overridden by a library's import-time defaults, fixes the load order, and confirms via gcx that the agent's generation count climbs.
Quick Start
Ask the agent to instrument my LLM app with Grafana Agent Observability and verify that generations actually land in my Grafana Cloud stack.
