The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents the use of Thor actions such as run, rails_command, rake, and git within generator scripts and application templates (e.g., in references/file-actions.md and references/multitenant-template.md). These methods allow for the execution of arbitrary shell commands on the host system where the generator is run.
[INDIRECT_PROMPT_INJECTION]: The skill creates a vulnerability surface by teaching the agent to construct code generators that directly interpolate user-supplied arguments (such as model names, attribute types, and options) into Ruby templates and shell commands. There is a lack of instructions for sanitizing this input, which could allow a malicious user to inject arbitrary code into the generated application files.
Ingestion points: User-provided arguments for generators and application templates (e.g., name, attributes, options used in SKILL.md and references/scaffold-generator.md).
Boundary markers: Absent; the skill does not instruct the agent to use delimiters or warnings for processed data.
Capability inventory: Access to run, rake, git, create_file, template, and inject_into_file across various generator examples.
Sanitization: Absent; no methods for escaping or validating user input are described before interpolation.
[DYNAMIC_EXECUTION]: The primary function of the skill is to facilitate dynamic code generation using ERB templates (.tt files) and Thor's file manipulation DSL. This behavior is inherent to the purpose of building code generators and application setup automation.
[REMOTE_CODE_EXECUTION]: The documentation includes patterns for loading Rails application templates from remote URLs, such as rails new myapp -m https://example.com/template.rb. While this is a native feature of Ruby on Rails, it represents a mechanism for downloading and executing external Ruby scripts.

rails-generators