FIG-001/ AGENT

agent output.

A simulated agent streams a PR-review response mixing legitimate rich HTML (verdict box, findings table, real link) with deliberate XSS attempts — javascript: URLs, onerror / onclick / onsubmit handlers, a vbscript: protocol, a form posting to javascript:. Watch the source, the sanitized render, and the audit log fill in together.

  • Sanitization happens at the Parser layer — custom renderers and snippets cannot bypass it.
  • The URL allowlist covers javascript:, vbscript:, data: (where unsafe), and other dangerous protocols.
  • Inline event handlers (onclick, onerror, onsubmit, …) and srcdoc are stripped before render.
  • Streaming-aware — every chunk is sanitized as it arrives, not after the document completes.
↩ all examples
sanitize · url + attrs mode · live running source
file · agent-output.svelte chunks 0/— events 0 ○ IDLE
SRC / AGENT STREAM raw text in
// agent response will arrive here word by word.
OUT / RENDERED sanitized live

click "start" to render the simulated agent response.

BLOCKED / SANITIZATION LOG wraps defaultSanitizeUrl & defaultSanitizeAttributes URL ATTR

nothing blocked yet — the log fills in as malicious payloads stream in.

category · agent
sheet · sheet 01 / 01
⟳ to re-run

AI Agent Output