html-comparison-matrix
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-defined strings (such as candidate names, selection criteria, and rationales) to generate a dynamic HTML artifact, creating an injection surface for potentially malicious data.
- Ingestion points: User-provided text from the conversation is interpolated into the rows, columns, and tooltips of the generated HTML matrix in
SKILL.md. - Boundary markers: The instructions include explicit directives to the agent to use programmatic DOM construction (e.g.,
textContentandcreateElement) rather than direct HTML string interpolation. - Capability inventory: The resulting HTML file can execute local JavaScript, interact with
localStorage, and perform network POST requests to a local listener or write to the clipboard via an internal submission handler. - Sanitization: The skill mandates strict anti-XSS measures by explicitly forbidding the use of
innerHTMLfor any content derived from user variables. - [SAFE]: The submission pipeline prioritizes a local loopback handshake initiated by a pre-flight tool call, ensuring that interactive data remains within the local environment.
- [SAFE]: External dependencies are restricted to internal plugin assets and well-known services like Google Fonts, minimizing external supply chain risks.
Audit Metadata