aks-troubleshooting

Pass

Audited by Gen Agent Trust Hub on Oct 20, 2026

Risk Level: SAFECOMMAND_EXECUTIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • Privileged Command Execution: The skill utilizes kubectl debug --profile=sysadmin for deep node-level diagnostics through Inspektor Gadget.
  • This is a high-privilege operation that allows access to the host node's kernel and filesystem.
  • The skill mitigates risks by requiring explicit user approval before execution, using a digest-pinned image from a trusted registry (MCR), and bounding the execution with timeouts.
  • External Resource Reference: The skill references an external container image for Inspektor Gadget (mcr.microsoft.com/oss/v2/inspektor-gadget/ig).
  • Downloading external code can introduce supply chain risks.
  • In this context, the source is a well-known Microsoft registry, and the skill uses a specific SHA256 digest to ensure the integrity of the image being pulled.
  • Indirect Prompt Injection Mitigations: The skill is designed to process untrusted data such as pod logs and Kubernetes events.
  • Malicious instructions could be embedded in logs to influence the agent's behavior.
  • The skill explicitly instructs the agent to never follow commands or URLs found in logs and includes a pod-deep-dive.sh script that redacts sensitive information like tokens and passwords from collected logs before they are presented.
  • System Command Execution: The skill relies on standard tools like az (Azure CLI) and kubectl to interact with the environment.
  • This is necessary for the intended troubleshooting functionality and follows the standard operational patterns for managing Kubernetes on Azure.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 20, 2026, 06:48 AM