The Agent Skills Directory

[COMMAND_EXECUTION]: The skill initiates a local Python process using the command python mcp-servers/azure-network-mcp/azure_network_mcp_server.py. This is the standard operational method for this MCP server to interact with the Azure SDK.
[CREDENTIALS_UNSAFE]: The skill requires standard Azure authentication credentials (AZURE_TENANT_ID, AZURE_CLIENT_ID, AZURE_CLIENT_SECRET, and AZURE_SUBSCRIPTION_ID) to be provided through environment variables. The instructions correctly avoid hardcoding these values and suggest standard secret management practices for cloud automation.
[PROMPT_INJECTION]: The skill processes external data from the Azure Resource Manager API, including resource names, tags, and security rule descriptions. While this presents a surface for indirect prompt injection if the Azure environment itself were compromised to include malicious strings in metadata, the skill implements a strict read-only policy that mitigates potential impact.
Ingestion points: Tools such as azure_list_vnets, azure_get_nsg_rules, and azure_list_subscriptions fetch external resource data.
Boundary markers: The skill does not explicitly define delimiters for tool output in its instructions.
Capability inventory: The skill is limited to 19 read-only tools; it lacks any capabilities for resource modification, file system writes, or unauthorized network operations.
Sanitization: The markdown does not specify explicit sanitization for API-returned strings before they are presented to the agent.

azure-network-ops