azure-resource-manager-cosmosdb-dotnet

Fail

Audited by Gen Agent Trust Hub on Feb 14, 2026

Risk Level: HIGHCOMMAND_EXECUTIONCREDENTIALS_UNSAFEDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION] (HIGH): The skill enables full management plane operations (Create, Update, Delete) on Azure Cosmos DB resources. This allows an agent to modify cloud infrastructure, which is a high-privilege capability.
  • [CREDENTIALS_UNSAFE] (HIGH): The skill includes explicit patterns for retrieving master account keys (GetKeysAsync) and connection strings (GetConnectionStringsAsync). These are 'root' credentials for the database accounts.
  • [DATA_EXFILTRATION] (HIGH): Combined with network capabilities or standard output, the ability to fetch master keys allows an agent to exfiltrate full access credentials for the Azure environment.
  • [Indirect Prompt Injection] (HIGH):
  • Ingestion points: Resource names, configuration payloads, and metadata fields across all management operations.
  • Boundary markers: None are present in the provided examples to separate instructions from resource data.
  • Capability inventory: Full infrastructure CRUD (Create, Read, Update, Delete) and master key retrieval across SKILL.md, references/account-management.md, and references/throughput.md.
  • Sanitization: No input validation or sanitization logic is provided to ensure that parameters like database names or configuration strings do not contain malicious instructions.
  • [EXTERNAL_DOWNLOADS] (LOW): Downloads official Microsoft packages (Azure.ResourceManager.CosmosDB, Azure.Identity) from NuGet. These are from a trusted organization (Microsoft), which downgrades the download risk per [TRUST-SCOPE-RULE].
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 14, 2026, 11:11 AM