The Agent Skills Directory

[METADATA_POISONING]: The SKILL.md file identifies 'Microsoft' as the author, which contradicts the provided developer context 'midudev'. This discrepancy can mislead users regarding the provenance and safety of the skill.\n- [PRIVILEGE_ESCALATION]: The run-ig.sh script uses kubectl debug --profile=sysadmin to execute Inspektor Gadget, which grants the resulting pod full administrative access to the underlying cluster node.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted logs and traces from Azure services, which could contain malicious instructions designed to influence agent behavior. Ingestion points: kubectl logs, az containerapp logs show, az webapp log download. Boundary markers: None. Capability inventory: Extensive command execution via Azure CLI and kubectl, including privileged pod creation. Sanitization: None.\n- [EXTERNAL_DOWNLOADS]: Diagnostic scripts pull container images from the trusted Microsoft Container Registry (mcr.microsoft.com).\n- [COMMAND_EXECUTION]: The skill executes a wide range of Azure CLI and kubectl commands, often using local scripts to aggregate diagnostic data.

azure-diagnostics