The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is primarily designed to execute shell commands and subprocesses via its native_api backend.
Evidence: The documentation explicitly lists the native_api backend as being triggered by the run_command action for subprocess and shell command execution.
Evidence: The macro YAML schema example shows the execution of arbitrary commands: command: [my-app, --export, "${output}"].
[INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect injection by interpolating user-supplied parameters into command templates within macro definitions.
Ingestion points: Untrusted data enters the context through the --param flag in the macro run command and via YAML files stored in cli_anything/macrocli/macro_definitions/.
Boundary markers: The skill does not document any boundary markers or escaping mechanisms for the ${variable} interpolation syntax.
Capability inventory: The skill possesses capabilities for subprocess execution, file writing/transformation, and GUI interaction.
Sanitization: There is no documented evidence of input validation or sanitization for parameters before they are executed in a shell context.
[DYNAMIC_EXECUTION]: The gui_macro backend executes precompiled files, which may involve runtime binary or script execution.
Evidence: The documentation identifies gui_macro as a backend for "Precompiled coordinate replay".
[EXTERNAL_DOWNLOADS]: The installation process involves installing Python dependencies from a local harness.
Evidence: Installation instructions require pip install -e . from the macrocli/agent-harness directory, which pulls in the PyYAML, click, and prompt-toolkit packages.

cli-anything-macrocli