Michaela Links GitHub

SRE

Investigates operational issues across the platform and proposes fixes through GitHub pull requests.

Request flow

  1. Readiness event

    Argo Events watches the target preview workload.

  2. Broker Job

    A short-lived Job collects workload state and recent events in AKS.

  3. Foundry agent

    The broker submits evidence to a pinned, Entra-authenticated agent version.

  4. Diagnosis

    The broker records the structured response and trace ID.

Components

  • Hosted runtime

    A versioned Microsoft Foundry container serves the Invocations API. The Python adapter delegates to the shared agent runtime.

    azure.yaml
  • Model access

    The bounded-agent policy uses scoped reasoning, coding, and embedding connections through the shared LiteLLM gateway.

    config/agent.yaml
  • Monitoring and evidence

    The EventSource and Sensor launch a broker Job with its own workload identity. The broker reads the target workload and calls the hosted agent.

    tools/broker.py
  • Notes and knowledge

    Caller-scoped notes and checkpoints support ongoing investigations. Reviewed runbooks provide context for retrieval.

    knowledge/
  • Telemetry

    OpenTelemetry can export investigation traces with model and tool spans to the shared collector.

    telemetry/tracing.py
  • Delivery and evaluation

    The shared CI workflow builds and scans the image. Native Foundry deployment, hosted evaluations, and broker verification gate the monitoring integration.

    evaluation/