The Agent Skills Directory

[Remote Command Execution]: The skill enables the agent to define and execute shell commands on remote Azure Batch compute nodes (e.g., using BatchTaskCreateParameters in SKILL.md). This is the core functionality of the Azure Batch service, providing a capability for remote execution. A consideration for developers is to ensure that command strings are not directly interpolated from untrusted user input without validation to maintain control over the commands executed on the remote infrastructure.
[External Resource Downloads]: The skill facilitates downloading files from remote URLs to compute nodes using the ResourceFile API as shown in references/examples.md. This is a standard feature for staging application code and data. Verifying the integrity and source of these files is a recommended practice when using SAS tokens or public URLs.
[Credential Management Considerations]: The documentation references environment variables such as AZURE_BATCH_ACCESS_KEY and AZURE_CLIENT_SECRET. The skill correctly encourages the use of environment variables and Microsoft Entra ID for secure authentication, supporting standard security practices for managing cloud credentials.
[Indirect Prompt Injection Surface]: The skill's ability to create remote tasks based on provided instructions represents a potential surface for indirect prompt injection.
Ingestion points: Data describing task parameters may enter the agent context through user prompts or processed files (referenced in SKILL.md and references/examples.md).
Boundary markers: The provided examples do not include explicit delimiters to separate untrusted data from instructions.
Capability inventory: The skill provides functionality to execute commands and download files on remote nodes (SKILL.md and references/examples.md).
Sanitization: Standard validation of external content before task creation is a recommended architectural consideration.

azure-compute-batch-java