The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an inherent surface for indirect prompt injection as it is designed to ingest and analyze potentially untrusted infrastructure-as-code (IaC) files. This is addressed by a robust 'Prompt Injection Safety Notice' in SKILL.md that instructs the agent to treat all file contents as data rather than instructions.
Ingestion points: Infrastructure definitions (Terraform, Bicep, ARM templates) located via Step 1 of the process in SKILL.md.
Boundary markers: The skill includes specific instructions in the safety notice to disregard directives embedded in configuration files.
Capability inventory: The skill is restricted to Read, Grep, and Glob tools. It lacks network access, file-write capabilities, or shell execution tools.
Sanitization: The skill implements instructional sanitization by directing the agent to ignore non-technical claims within the analyzed files.
[EXTERNAL_DOWNLOADS]: The skill contains references to official security documentation and technical registries, including cisecurity.org, learn.microsoft.com, and registry.terraform.io. These are recognized as well-known and trusted sources for security benchmarks and infrastructure-as-code providers.

azure-review