The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines patterns for processing external data (e.g., query filters, search strings, and record attributes) into database operations, creating a potential surface for indirect injection if data is not handled correctly. However, the skill explicitly provides security mitigations for this surface.
Ingestion points: Input parameters in search_images_query(query_string), images_query(filters), and changeset(image, attrs) in SKILL.md and assets/changeset_snippets.ex are points where untrusted data enters the context.
Boundary markers: The instructions mandate the use of the Ecto pin operator (^) to ensure all user input is parameterized and never interpolated directly into SQL fragments.
Capability inventory: The skill utilizes Ecto.Repo for database read, write, and modification operations.
Sanitization: The skill provides extensive templates for using Ecto.Changeset validations (e.g., cast/3, validate_required/2, validate_inclusion/3) and database constraints to filter and sanitize incoming data.

ecto-essentials