The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to ingest and analyze application logs, telemetry, and event data (e.g., via kubectl logs, az webapp log tail, and KQL queries against AppServiceConsoleLogs). If these logs contain malicious instructions, they could potentially influence the agent's behavior. The skill provides remediation guidance and emphasizes focused diagnostic flows to mitigate accidental execution of embedded instructions.
High-Privilege Command Execution: Certain diagnostic workflows, particularly for Azure Kubernetes Service (AKS), involve the use of privileged debug containers (kubectl debug --profile=sysadmin) and Inspektor Gadget. These tools provide deep visibility into the host node's network and process state but represent a significant privilege level. The instructions correctly mandate explicit user consent and pre-flight safety checks before these operations are initiated.
Execution of Diagnostic Scripts: The skill includes several shell and PowerShell scripts (e.g., aks-baseline.sh, pod-evidence.sh, run-ig.sh) to automate data collection. While these are intended for troubleshooting, they execute a series of Azure CLI and Kubernetes commands. This functionality requires the agent to have sufficient permissions within the target Azure environment to perform these reads and potentially influence resource state if remediation is requested.

azure-diagnostics