The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches Terraform modules from public GitHub repositories, such as those under the cloudposse-terraform-components organization, during component provisioning operations.
[REMOTE_CODE_EXECUTION]: The Atmos toolchain downloads and installs specific versions of Terraform and OpenTofu binaries from official distribution points to ensure environment consistency across development and CI/CD systems.
[COMMAND_EXECUTION]: The skill wraps the Terraform and OpenTofu CLI binaries to execute infrastructure lifecycle commands, including plan, apply, deploy, and destroy across one or more stacks.
[DYNAMIC_EXECUTION]: The skill processes Go templates and YAML functions within stack configuration manifests to dynamically generate backend.tf.json and terraform.tfvars.json files before binary execution.
[INDIRECT_PROMPT_INJECTION]: The skill processes external configuration data from stack manifests, Git commit history, and remote component source URLs, which represents a potential attack surface for indirect influence on agent behavior.
Ingestion points: Stack manifests (atmos.yaml and stack YAML files), Git repository state (used for identifying affected components), and remote module source URLs.
Boundary markers: No explicit delimiters or instructional constraints are defined to isolate untrusted configuration data from the execution logic during template interpolation.
Capability inventory: The skill performs file system writes (SKILL.md), shell command execution (SKILL.md), and network operations for module and binary acquisition (SKILL.md, references/toolchain-pinning.md).
Sanitization: No explicit sanitization, validation, or escaping logic for variables interpolated into configuration files is described.

atmos-terraform