The Agent Skills Directory

Structured Configuration Management: The skill provides a helper script, settings_upsert.py, to modify the user's VS Code settings.json. This is a sensitive operation, but the skill mitigates risk by enforcing a strict schema, performing atomic swaps to prevent file corruption, and maintaining the last five backups. The agent is required to show a diff and obtain explicit user consent before any write occurs.
Telemetry Data Privacy: The skill authors explicitly acknowledge that LLM telemetry can contain sensitive prompt content. To address this, the provided OpenTelemetry Collector configurations (local and Azure) utilize a fail-closed allow-list. This ensures that any telemetry attributes not explicitly recognized as safe are dropped before storage, significantly reducing the risk of accidental prompt exfiltration.
Container Image Supply Chain Integrity: The local capture stack uses Docker images for the OTel Collector and Grafana. These images are referenced using immutable manifest digests (e.g., sha256:...) rather than mutable tags. This prevents image substitution attacks where a tag could be repointed to a malicious image.
Network and Ingress Containment: Local services such as Prometheus, Tempo, and the Collector are configured to bind exclusively to the loopback interface (127.0.0.1). This prevents the telemetry store and its unauthenticated query APIs from being exposed to the local network.
Clear Execution Boundaries: The skill follows the principle of least privilege by having the agent generate configuration files and templates (Bicep, Terraform, Shell scripts) while leaving the actual execution of those files to the user. This ensures that the agent cannot autonomously provision billable cloud resources or start host-level services.

copilot-otel-metrics