The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a wide range of Azure CLI (az) and guest OS commands (bash, PowerShell) designed for system administration and troubleshooting. These include commands to reset passwords (az vm user update), modify firewall rules (netsh, iptables, ufw), and execute scripts via the Azure VM Agent (az vm run-command invoke). These capabilities are transparently documented as templates for the user or agent to use for problem resolution.
[EXTERNAL_DOWNLOADS]: The skill fetches real-time data from well-known and official Microsoft services. It queries the Azure Retail Prices API (prices.azure.com) for unauthenticated pricing information and uses web_fetch/fetch_webpage to retrieve the latest technical documentation from learn.microsoft.com. These interactions are restricted to trusted domains and are central to the skill's functionality.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface through its documentation-fetching workflow in vm-recommender.md and vm-troubleshooter.md.
Ingestion points: External content is fetched from learn.microsoft.com URLs.
Boundary markers: None explicitly defined in the prompts.
Capability inventory: The skill can execute Azure CLI commands and system-level scripts via az vm run-command.
Sanitization: Content is processed by the agent to summarize troubleshooting steps.
Conclusion: While the attack surface exists, the reliance on a highly trusted source (official Microsoft documentation) makes this a safe implementation of dynamic documentation retrieval.

azure-compute