The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze, write, and review Infrastructure as Code (HCL, TypeScript, Python). This creates an attack surface where malicious instructions could be embedded in infrastructure code or documentation provided by a user.
Ingestion points: Processes HCL files, Pulumi programs, and CI/CD workflow YAMLs via user input or file system access.
Boundary markers: None explicitly defined in the instructions to separate code from commands.
Capability inventory: The skill provides patterns for executing shell commands (tofu apply, pulumi up), running remote provisioners (remote-exec), and managing cloud resources via providers.
Sanitization: Standard HCL/Pulumi syntax validation is recommended, but the LLM must exercise caution when interpreting instructions found within user-provided code comments or string literals.
[DATA_EXPOSURE_&_EXFILTRATION]: The skill correctly identifies and mitigates data exposure risks.
The documentation explicitly warns against hardcoding credentials and provides patterns for using sensitive = true variables, environment variables, and OIDC (role-to-assume).
It details the usage of OpenTofu's state encryption feature to protect sensitive data at rest in state files.
[EXTERNAL_DOWNLOADS]: The skill references standard tools and providers from trusted sources.
Downloads providers from the official OpenTofu and Pulumi registries.
CI/CD examples use official GitHub Actions (e.g., actions/checkout, opentofu/setup-opentofu). Note: Some version numbers in examples (e.g., actions/checkout@v6) are forward-looking placeholders as v4 is current, but they target the official actions/ namespace.

opentofu