The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the az-tool wrapper via bun. It implements a comprehensive safety model that spawns commands as argument vectors rather than through a shell, explicitly blocking shell metacharacters like ;, &&, |, and $(...) to prevent command injection.
[EXTERNAL_DOWNLOADS]: The skill references the vendor's official GitHub repository (blogic-cz/agent-tools) for its core toolset and mentions a required Node.js development dependency, which is consistent with its stated purpose as a CLI wrapper.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted metadata from external sources, specifically Azure resource names, tags, and metrics.
Ingestion points: Resource listings, VM metadata, storage account details, and monitor metrics (SKILL.md).
Boundary markers: The instructions do not define explicit delimiters or warnings for the agent to treat tool output as untrusted content.
Capability inventory: Executes shell commands via the bun az-tool wrapper and requires network access to communicate with Azure APIs.
Sanitization: Employs strict verb filtering (whitelist of read-only verbs) and blocks shell syntax characters to mitigate risk during command construction.

az-tool