The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest user-defined code and feature requirements, which are then processed by the agent using shell commands like bin/rails test, ruby -c, and bundle exec rubocop. This pattern creates an attack surface where maliciously crafted instructions within the user's codebase or documentation could attempt to influence the agent's behavior during command execution.
Ingestion points: User-provided feature descriptions in SKILL.md and data processing examples (e.g., CSV imports) in references/models.md and references/background-jobs.md.
Boundary markers: The instructions do not explicitly define delimiters to separate system prompts from untrusted content in the code being analyzed.
Capability inventory: The skill facilitates local command execution for syntax checking and testing within the verification_loop of SKILL.md.
Sanitization: There is no explicit requirement for the agent to sanitize file paths or command-line arguments derived from user input.
[EXTERNAL_DOWNLOADS]: The skill documentation in references/validations-callbacks.md suggests using the bundle add command to install the validate_url gem. This is a standard procedure for managing dependencies in Ruby on Rails projects and utilizes a well-known community gem.

ruby-on-rails