The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (HIGH): In the file rules/language-linting.md, the skill instructs the agent to install TFLint using a curl | bash command targeting a remote script. This pattern executes unverified code and is a major security risk.
[PROMPT_INJECTION] (HIGH): Indirect Prompt Injection vulnerability. The skill is designed to process external Terraform code for reviews and refactoring. This is a Tier 1 surface (high capability) that lacks boundary markers or instructions to ignore embedded instructions, making the agent susceptible to manipulation by malicious code comments.
[COMMAND_EXECUTION] (MEDIUM): Multiple rules including rules/resource-immutable.md and rules/org-audit-logging.md suggest the use of local-exec provisioners to execute shell commands like docker build, docker push, and file system operations.
[EXTERNAL_DOWNLOADS] (MEDIUM): The documentation refers to several non-whitelisted external repositories and packages, such as antonbabenko/pre-commit-terraform and tflint-ruleset-aws, which would be downloaded if the agent follows the guidelines.

terraform-best-practices