The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of AI agents that ingest untrusted data from multiple sources, creating a surface for indirect prompt injection.\n
Ingestion points: RubyLLM.chat user input, with_file_store attachments, and with_mcp server tool outputs (SKILL.md).\n
Boundary markers: Absent. The instructions do not specify the use of delimiters or warnings to ignore instructions within processed data.\n
Capability inventory: Network requests to AI providers and MCP servers, file system operations via directory stores, and shell command execution via MCP configurations or Rails generators (SKILL.md).\n
Sanitization: The instructions recommend browser-side sanitization for streaming UI content but do not detail sanitization for prompt construction.\n- [COMMAND_EXECUTION]: The instructions involve executing shell commands for version verification (bundle exec), application setup (bin/rails generate), and MCP server execution. These are standard operations for a Ruby on Rails development environment.\n- [EXTERNAL_DOWNLOADS]: The skill describes functionality for downloading model catalogs and data from remote sources using RubyLLM.models.refresh and bin/rails ruby_llm:load_models.

rubyllm