slide-dsl
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents a rendering environment that lacks markup sanitization, creating a surface for injection attacks.
- Ingestion points: The agent is instructed to write to various JSON fields such as
/content/canvas/elements/N/contentandtext.contentinSKILL.md. - Boundary markers: None specified; the manual states that its instructions are the "whole of what makes a page render correctly," potentially overriding other constraints.
- Capability inventory: The agent uses tools to perform
set,add_element, andstr_replaceoperations, which allow it to inject arbitrary HTML, including inline event handlers that the manual acknowledges "become live handlers." - Sanitization: The system explicitly performs no sanitization or filtering, stating that "Nothing filters your markup and nothing normalises your values."
Audit Metadata