The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides functionality to create Azure Batch tasks that execute arbitrary shell commands on compute nodes. This represents a vulnerability surface for indirect prompt injection if untrusted user input is directly interpolated into these commands.
Ingestion Points: Methods such as createTask, createTasks, and createTaskCollection in SKILL.md accept command-line strings.
Boundary Markers: The skill includes specific security guidance: "Never interpolate untrusted task input into shell text" and "use the appropriate explicit shell only for reviewed fixed commands."
Capability Inventory: The skill enables full management of Azure Batch resources, including pool creation, job scheduling, and remote node management.
Sanitization: The skill relies on the implementing agent to follow the provided security warnings as no automatic sanitization logic is included in the snippets.
[EXTERNAL_DOWNLOADS]: The skill references official dependencies and documentation from Microsoft's trusted infrastructure.
Evidence: Maven dependency com.azure:azure-compute-batch and GitHub repository https://github.com/Azure/azure-sdk-for-java are official Azure resources.
Context: These references are standard for Azure Batch development and originate from well-known, trusted sources.

azure-compute-batch-java