Function isVisible

coderd/x/chatd/sanitize.go:51–130

isVisible reports whether r is a visible Unicode character that should be preserved in prompt text. Each invisible range is documented with its Unicode name and rationale.

(r rune)

// should be preserved in prompt text. Each invisible range is
// documented with its Unicode name and rationale.
func isVisible(r rune) bool {
	switch {
	// Soft hyphen — invisible in most renderers, used to hide
	// content boundaries.
	case r == 0x00AD:
		return false

	// Combining grapheme joiner — invisible, no legitimate
	// prompt use.
	case r == 0x034F:
		return false

	// Arabic letter mark — bidi control, invisible.
	case r == 0x061C:
		return false

	// Mongolian vowel separator — invisible spacing character.
	case r == 0x180E:
		return false

	// Zero-width space (U+200B).
	case r == 0x200B:
		return false

	// U+200C (ZWNJ) is deliberately NOT stripped. It is
	// required for correct rendering of Persian, Urdu, and
	// Kurdish scripts where it controls cursive joining.
	// Stripping ZWS (U+200B) and ZWJ (U+200D) already breaks
	// zero-width steganography encodings regardless of whether
	// ZWNJ survives.

	// Zero-width joiner (U+200D) — also used in compound emoji,
	// but actively exploited in steganography. See
	// SanitizePromptText doc comment.
	case r == 0x200D:
		return false

	// Left-to-right mark (U+200E).
	case r == 0x200E:
		return false

	// Right-to-left mark (U+200F).
	case r == 0x200F:
		return false

	// Bidi embedding and override controls (U+202A–U+202E):
	// LRE, RLE, PDF, LRO, RLO.
	case r >= 0x202A && r <= 0x202E:
		return false

	// Word joiner and invisible operators (U+2060–U+2064):
	// word joiner, function application, invisible times,
	// invisible separator, invisible plus.
	case r >= 0x2060 && r <= 0x2064:
		return false

	// Bidi isolate controls (U+2066–U+2069):
	// LRI, RLI, FSI, PDI.