The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The BaseRepository implementation in SKILL.md contains an update pattern that is vulnerable to unauthorized data modification if used directly with external input.
Ingestion points: The update method in the BaseRepository class accepts **kwargs that are iterated and applied to the model instance.
Boundary markers: The code snippet lacks explicit boundary markers or instructions to ignore embedded instructions in the input data.
Capability inventory: The skill demonstrates the ability to perform database writes and commit transactions via self.db.commit().
Sanitization: The provided pattern uses setattr(obj, key, value) without a whitelist or validation of allowed fields, creating a mass assignment surface where untrusted data could overwrite sensitive model attributes (e.g., is_active or id) if the kwargs originate from user-controlled sources.

sqlalchemy-patterns