The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides patterns for Terragrunt before_hook, after_hook, and error_hook within references/common-patterns.md and the template assets. These blocks allow the generated HCL to execute arbitrary shell commands, such as bash -c "curl ...", on the host system when Terragrunt is executed.
[REMOTE_CODE_EXECUTION]: The skill includes usage patterns for the run_cmd function in references/common-patterns.md, which facilitates the execution of local binaries like git and aws. While standard for Terragrunt, this mechanism enables local command execution during HCL evaluation.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download of infrastructure components from external repositories, including modules from the Terraform Registry (tfr:///), Git repositories, and OpenTofu engines from GitHub. These sources are well-known and standard in the DevOps ecosystem.
[PROMPT_INJECTION]: The skill contains an attack surface for indirect prompt injection (Category 8). It takes user-provided metadata (e.g., module names, project identifiers, and environment variables) and interpolates them into generated HCL files. If these strings are not sanitized, they could potentially inject malicious shell commands into the generated hooks or run_cmd blocks.
Ingestion points: User input for project-specific variables and module configuration details as described in the SKILL.md generation workflow.
Boundary markers: No explicit delimiters are suggested to isolate user input from the surrounding HCL code or shell command strings.
Capability inventory: The generated configurations possess the capability to execute shell commands and perform network operations via curl.
Sanitization: The skill lacks explicit instructions or logic for the agent to escape or validate user-provided data before interpolating it into executable shell arrays.

terragrunt-generator