html-throwaway-editor

Pass

Audited by Gen Agent Trust Hub on May 18, 2026

Risk Level: SAFE
Full Analysis
  • [DATA_EXFILTRATION]: The skill implements a submission pipeline to return data from the generated HTML artifact to the agent. It uses a local HTTP listener (typically 127.0.0.1 via a companion tool) or the system clipboard. This is a documented architectural feature for facilitating interactive agent-user workflows.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and display user-provided data within the generated HTML. To mitigate XSS risks, the instructions explicitly mandate the use of safe DOM APIs (e.g., textContent, createElement) and strictly forbid the use of innerHTML with variable or untrusted content.
  • [EXTERNAL_DOWNLOADS]: Artifacts are required to be self-contained and are prohibited from loading external runtimes or libraries from CDNs or package registries, with a narrow exception for Google Fonts.
  • [SAFE]: The skill enforces a 'throwaway' architecture, specifically forbidding the use of localStorage, sessionStorage, or IndexedDB in the generated artifacts to ensure no data persists in the browser environment beyond the active session.
Audit Metadata
Risk Level
SAFE
Analyzed
May 18, 2026, 12:58 PM
Security Audit — agent-trust-hub — html-throwaway-editor