The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes the application's codebase to make configuration recommendations, which creates a surface for indirect prompt injection if malicious instructions are embedded in the code being scanned.\n
Ingestion points: The agent is instructed to run grep -r and find on the user's codebase (e.g., app/models/**/*.rb, app/views/**/*) and process the output.\n
Boundary markers: The instructions do not specify the use of delimiters or 'ignore embedded instructions' warnings when the agent processes code snippets.\n
Capability inventory: The skill has the ability to read files (grep, find) and write to configuration files (config/initializers/*.rb, config/application.rb).\n
Sanitization: Code content extracted from the application is analyzed directly without sanitization or escaping.\n- [OBFUSCATION]: A static analysis detector flagged potential Unicode steganography in templates/new_framework_defaults_8_1.rb.\n
Evidence: The file contains literal U+2028 (Line Separator) and U+2029 (Paragraph Separator) characters within a comment block.\n
Context: These characters are the specific subject of the escape_json_responses and escape_js_separators_in_json configuration settings introduced in Rails 8.1. Their inclusion in the file is for documentation and illustrative purposes, representing legitimate use in the context of a Rails upgrade guide.

rails-load-defaults