sql-backup-restore-to-azure-sql-vm-migration
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- Command Execution Framework: The skill utilizes command-line tools such as
sqlcmd,AzCopy, and the Azure CLI to perform discovery and data movement tasks. These operations are restricted to using Windows Integrated or Microsoft Entra authentication, explicitly forbidding the use of hardcoded SQL passwords or interactive password prompts. - Dynamic Script Generation: The skill assembles T-SQL batches and PowerShell scripts at runtime to perform backups and restores. To mitigate injection risks, the instructions mandate the use of escaping for SQL identifiers (e.g., doubling brackets) and literals (e.g., doubling single quotes) before interpolation.
- Input Sanitization and Path Safety: To prevent directory traversal and path injection, user-supplied database names are hashed using SHA256 to derive deterministic filenames. Furthermore, generated Blob URLs are validated to ensure they are secure HTTPS endpoints without query strings or fragments.
- Identity and Access Management: The workflow manages Azure RBAC by assigning the
Storage Blob Data Contributorrole to the target's managed identity. This process is scoped specifically to the identified storage account, following the principle of least privilege for the migration task. - Indirect Prompt Injection Surface: The skill processes external data (database names, server addresses) which could potentially contain instructions intended to influence the agent.
- Ingestion points: Input is collected via the consolidated intake parameters in
SKILL.mdand through SQL discovery probes. - Boundary markers: The skill employs deterministic hashing for file naming and T-SQL variable parameterization to separate data from instructions.
- Capability inventory: Capabilities include file system writes (
BACKUP TO DISK), network transfers (azcopy), and permission modifications (az role assignment create). - Sanitization: Inputs are sanitized via SHA256 hashing for filenames, T-SQL character replacement, and URI encoding for Blob path segments.
Audit Metadata