The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file contains deceptive metadata, claiming 'author: Microsoft' while the skill is provided by 'kinnekate-hub'. This discrepancy can lead to misjudgment of the skill's origin and safety.
[CREDENTIALS_UNSAFE]: The file references/functions/README.md includes commands to retrieve sensitive configuration data such as APPLICATIONINSIGHTS_CONNECTION_STRING and APPINSIGHTS_INSTRUMENTATIONKEY using az functionapp config appsettings list. This exposes secrets to the agent context.
[COMMAND_EXECUTION]: The skill facilitates high-privilege operations in aks-troubleshooting/node-issues.md via kubectl debug node, which provides interactive shell access to cluster nodes using a privileged container. It also includes potentially destructive commands like kubectl delete node and az aks nodepool scale.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core functionality of ingesting and analyzing untrusted data.
Ingestion points: Reads application logs via az monitor log-analytics query (in references/functions/README.md) and pod logs via kubectl logs (in aks-troubleshooting/pod-failures.md). It also reads cluster events via kubectl get events (in aks-troubleshooting/general-diagnostics.md).
Boundary markers: None identified. There are no instructions for the agent to ignore or delimit potentially malicious instructions within logs or events.
Capability inventory: Significant shell execution capabilities via az and kubectl tools for resource modification, pod deployment, and cluster management.
Sanitization: No sanitization or validation of the ingested log or event data is performed before processing.
[DATA_EXFILTRATION]: By querying App Insights and Log Analytics (references/kql-queries.md), the skill can access and potentially expose PII or other sensitive application data contained within logs.

azure-diagnostics