The Agent Skills Directory

Indirect Prompt Injection Surface: The skill processes external engineering data through the documents.import function, which represents an ingestion point for potentially untrusted content (SKILL.md). This creates a surface area where instructions embedded within CAD files could attempt to influence the agent. The skill instructions mitigate this by explicitly noting that these imports are not sandboxed and recommending a qualified environment for suspicious inputs. Relevant capabilities include artifact preparation and generation (SKILL.md), with verification steps involving numerical measurements and structural checks.
Active Content Management: The skill classifies generated HTML setup sheets as 'quarantined active content' and prohibits their automatic execution or publication. This approach helps protect against the execution of scripts that might be included in exported documentation.
Input Validation Practices: The instructions mandate the use of root aliases instead of raw file paths and prohibit overwriting existing files. These safety guidelines are intended to maintain predictable file system behavior and prevent unauthorized data modification during the deliverable stage.

create-fusion-deliverables