Strip invisible Unicode from content model at editor initialization

[romanisa]

Summary

Strips invisible Unicode characters (zero-width chars, bidirectional marks, Unicode Tags, etc.) from text and link hrefs as a defense-in-depth measure against hidden content injection via mailto: links.

Key design decision: Sanitization runs once at editor initialization rather than on every DOM-to-model conversion, avoiding performance overhead on the hot path.

Problem

Invisible Unicode characters embedded in mailto: links can bypass Human-in-the-Loop (HiTL) review ΓÇö users see a benign-looking email draft but hidden content may be present. This is a secondary safety net (the primary fix is upstream in Copilot/BizChat).

Changes

File	Purpose
stripInvisibleUnicode.ts	Core utility ΓÇö regex strips ~30 categories of invisible chars from raw strings
sanitizeInvisibleUnicode.ts	New ΓÇö walks the entire content model at init: sanitizes text segments, link hrefs, and Text nodes inside General segment elements
Editor.ts	Calls sanitizeInvisibleUnicode(initialModel) in constructor before setContentModel
checkXss.ts	Strips invisible Unicode before script: XSS check (covers insertLink API at runtime)

What was removed from earlier iterations

• linkProcessor.ts ΓÇö no longer strips on every DOM-to-model pass (hot path)
• linkFormatHandler.ts ΓÇö no longer strips on every format parse (hot path)
• sanitizeElement.ts ΓÇö no longer strips during HTML sanitization
• decodeURIComponent pre-processing ΓÇö removed per review feedback; percent-encoded content may be intentional

Characters stripped

• Zero-width characters (U+200B-U+200F)
• Bidirectional controls (U+202A-U+202E, U+2066-U+2069)
• Unicode Tags (U+E0001-U+E00FF)
• Soft hyphens, BOM, word joiners, Hangul fillers, Mongolian FVS, interlinear annotation anchors, and other invisible formatting chars

Test plan

• 33 unit tests for stripInvisibleUnicode (individual chars, combinations, supplementary plane, emoji preservation, visible Unicode preservation)
• 9 unit tests for sanitizeInvisibleUnicode (text segments, link hrefs, tables, General segments, nested block groups, empty models)
• 12 unit tests for checkXss (XSS detection with invisible Unicode bypass attempts)
• All 109 related tests pass on Chrome 145

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Strips invisible Unicode characters from link href values across sanitization, model conversion, and XSS checking to prevent hidden-content injection (notably via mailto:).

Changes:

• Introduces stripInvisibleUnicode() utility and exports it from roosterjs-content-model-dom.
• Applies stripping during DOM→model conversion (linkProcessor, linkFormatHandler) and HTML sanitization (sanitizeElement).
• Updates checkXss() to strip invisible Unicode before evaluating script: patterns and adds unit tests across layers.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
packages/roosterjs-content-model-dom/lib/domUtils/stripInvisibleUnicode.ts	Adds the core stripping utility via a consolidated regex.
packages/roosterjs-content-model-dom/lib/index.ts	Exposes stripInvisibleUnicode from the package entrypoint.
packages/roosterjs-content-model-dom/lib/formatHandlers/segment/linkFormatHandler.ts	Strips invisibles when reading link formats from DOM attributes.
packages/roosterjs-content-model-dom/lib/domToModel/processors/linkProcessor.ts	Strips invisibles while processing <a href> into the content model.
packages/roosterjs-content-model-core/lib/command/createModelFromHtml/sanitizeElement.ts	Strips invisibles when sanitizing href attributes.
packages/roosterjs-content-model-dom/test/domUtils/stripInvisibleUnicodeTest.ts	Adds focused unit coverage for stripping behavior across many code points.
packages/roosterjs-content-model-dom/test/domToModel/processors/linkProcessorTest.ts	Ensures DOM→model link processing strips invisible Unicode in href.
packages/roosterjs-content-model-core/test/command/createModelFromHtml/sanitizeElementTest.ts	Ensures sanitizer strips invisibles from href but not unrelated attributes.
packages/roosterjs-content-model-api/lib/publicApi/utils/checkXss.ts	Strips invisibles before XSS detection and returns sanitized links.
packages/roosterjs-content-model-api/test/publicApi/utils/checkXssTest.ts	Adds tests for invisible Unicode stripping + script: obfuscation detection.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.