The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to define and execute arbitrary shell commands through the steps attribute in the atmos.yaml configuration. Examples demonstrate the execution of system tools such as terraform, ansible-playbook, aws eks, and tflint.
[EXTERNAL_DOWNLOADS]: The dependencies section enables the automatic download and installation of CLI tools from a toolchain registry. While this is a core feature of the vendor's tool, it represents a mechanism for fetching and executing external binaries.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it uses Go templates to interpolate user-provided arguments and flags directly into shell commands. An attacker providing a malicious value for an argument could potentially execute unauthorized shell commands.
Ingestion points: Positional arguments ({{ .Arguments.<name> }}), named flags ({{ .Flags.<name> }}), and trailing arguments ({{ .TrailingArgs }}) are the primary entry points for external data as described in SKILL.md and references/command-syntax.md.
Boundary markers: The skill does not implement any boundary markers or instructions to the agent to ignore potentially malicious content within these variables.
Capability inventory: The skill possesses the capability to execute sequences of shell commands, set environment variables, and change the working directory (including using !repo-root).
Sanitization: No sanitization, escaping logic, or input validation is described or required by the command schema when processing template variables for shell execution.

atmos-custom-commands