The Agent Skills Directory

Indirect Prompt Injection Surface: The skill processes external CAD documents and operation schemas. While these could potentially contain untrusted instructions if sourced from outside the user's control, the skill includes several safeguards. It requires inputs to be part of a 'reviewed request,' uses 'unchanged hashes' to ensure integrity, and mandates verification of 'meaningful numerical postconditions' after execution.
Permission Boundary Management: The instructions explicitly reference 'scoped grants' and 'broad grants,' demonstrating a structured approach to permission management. This ensures that operations are performed within defined authorization limits provided by the platform.
Integrity and Idempotency: The use of hashes to verify inputs and idempotency keys to prevent accidental re-execution are positive security patterns that protect the integrity of the design data.

edit-fusion-design