The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs users to perform standard development tasks using the Rails CLI, including bin/rails generate, bin/rails db:migrate, and bin/rails console for project setup and verification.
[EXTERNAL_DOWNLOADS]: The documentation references the installation of standard Ruby gems and the use of MCP servers via npx, which are legitimate components of the library's ecosystem.
[INDIRECT_PROMPT_INJECTION]: The skill describes a framework for processing untrusted user input via LLMs and executing tool calls based on that input, representing a standard attack surface for LLM-integrated applications.
Ingestion points: User data enters the agent context via controller parameters (e.g., params[:content] in workflows/add-rails-chat.md).
Boundary markers: The instructional templates focus on API usage and do not explicitly define delimiters for untrusted data interpolation.
Capability inventory: The library supports database operations, network API calls to LLM providers, and the execution of Ruby logic through custom tools.
Sanitization: Standard sanitization for LLM inputs is handled by the underlying framework and the developer's implementation rather than the documentation templates.

ruby-llm