Merge pull request #3729 from curious-rabbit/sanitize

add --sanitize option

CHANGELOG.md

@@ -8,6 +8,7 @@
 
 ## Features
 
+- Add a `--sanitize=<auto|always|never>` flag for safe display of untrusted input. It implies `--strip-ansi` at the same value and additionally substitutes terminal-active control bytes (cursor moves, charset switches, beep, etc.) and Unicode bidi / zero-width formatting characters with the Unicode replacement character (U+FFFD). Mitigates Trojan-Source-style spoofing (CVE-2021-42574). See #3729 (@curious-rabbit)
 - Map justfile, Justfile, .justfile, and *.justfile to Makefile syntax highlighting, see #3623 (@zachvalenta)
 - Preserve `--diff` change markers and snip separators when `--plain` is set. Closes #3630, see #3643 (@mvanhorn)
 - Added support for `hidden_file_extensions` from `.sublime-syntax` files, see #3613 (@Matei02355)
@@ -22,6 +23,7 @@
 - Syntax highlighting for Python files using uv as script runner in shebang #3689 (@janlarres)
 
 ## Bugfixes
+- `--strip-ansi`: also strip 8-bit C1 introducers (U+0090, U+0098, U+009B, U+009D, U+009E, U+009F) and DCS/SOS/PM/APC sequence bodies, which previously passed through. See #3729 (@curious-rabbit)
 - Fix `--ignored-suffix` not falling back to first-line/shebang detection when the ignored suffix is also a registered extension (e.g. `--ignored-suffix .txt` on a shebang script), see #2745 and #3816 (@adnrivera)
 - Fix `capacity overflow` panic when printing a snip separator at `--terminal-width=1` with multiple line ranges. Closes #3803, see #3804 (@leeewee)
 - Pass `--no-paging` to `bat` invocations inside the bash / zsh / fish / PowerShell shell completion scripts so that shell-level pager wiring (e.g. `LESSOPEN='|-bat -f -pp %s'`) cannot inject ANSI escape sequences into the completion candidates. Closes #3760 (@mvanhorn)

README.md

@@ -843,6 +843,10 @@ If your version of `bat` supports the `--strip-ansi=auto` option, it can be used
 before syntax highlighting. Alternatively, you may disable both syntax highlighting and wrapping by
 passing the `--color=never --wrap=never` options to `bat`.
 
+For untrusted input, the `--sanitize=auto|always|never` option additionally replaces terminal-active
+control bytes and Unicode bidi / zero-width formatting characters with the Unicode replacement
+character. It implies `--strip-ansi` at the same value.
+
 > [!NOTE]
 > The `auto` option of `--strip-ansi` avoids removing escape sequences when the syntax is plain text.
 

assets/completions/_bat.ps1.in

@@ -100,6 +100,11 @@ Register-ArgumentCompleter -Native -CommandName '{{PROJECT_EXECUTABLE}}' -Script
             ForEach-Object {[System.Management.Automation.CompletionResult]::new($_, $_, [CompletionResultType]::ParameterValue, $_)}
             break
         }
+        '*;--sanitize' {
+            $ArrayWhen |
+            ForEach-Object {[System.Management.Automation.CompletionResult]::new($_, $_, [CompletionResultType]::ParameterValue, $_)}
+            break
+        }
         '*;--strip-ansi' {
             $ArrayWhen |
             ForEach-Object {[System.Management.Automation.CompletionResult]::new($_, $_, [CompletionResultType]::ParameterValue, $_)}
@@ -158,6 +163,7 @@ Register-ArgumentCompleter -Native -CommandName '{{PROJECT_EXECUTABLE}}' -Script
             [CompletionResult]::new('--ignored-suffix'        , 'ignored-suffix'        , [CompletionResultType]::ParameterName, 'Ignore extension. For example: ''bat --ignored-suffix ".dev" my_file.json.dev'' will use JSON syntax, and ignore ''.dev''')
             [CompletionResult]::new('--squeeze-blank'         , 'squeeze-blank'         , [CompletionResultType]::ParameterName, 'Squeeze consecutive empty lines into a single empty line.')
             [CompletionResult]::new('--squeeze-limit'         , 'squeeze-limit'         , [CompletionResultType]::ParameterName, 'Set the maximum number of consecutive empty lines to be printed.')
+            [CompletionResult]::new('--sanitize'              , 'sanitize'              , [CompletionResultType]::ParameterName, 'Specify when to sanitize untrusted input for safe display. Implies --strip-ansi and also replaces terminal-active and bidi / zero-width bytes. (auto, always, *never*).')
             [CompletionResult]::new('--strip-ansi'            , 'strip-ansi'            , [CompletionResultType]::ParameterName, 'Specify when to strip ANSI escape sequences from the input. The automatic mode will remove escape sequences unless the syntax highlighting language is plain text. (auto, always, *never*).')
         #   [CompletionResult]::new('-p'                      , 'p'                     , [CompletionResultType]::ParameterName, 'Show plain style (alias for ''--style=plain'').')
             [CompletionResult]::new('--plain'                 , 'plain'                 , [CompletionResultType]::ParameterName, 'Show plain style (alias for ''--style=plain'').')

assets/completions/bat.bash.in

@@ -132,6 +132,10 @@ _bat() {
 		COMPREPLY=($(compgen -W "auto never always" -- "$cur"))
 		return 0
 		;;
+	--sanitize)
+		COMPREPLY=($(compgen -W "auto never always" -- "$cur"))
+		return 0
+		;;
 	--completion)
 		COMPREPLY=($(compgen -W "bash fish zsh ps1" -- "$cur"))
 		return 0
@@ -221,6 +225,7 @@ _bat() {
 			--list-themes
 			--squeeze-blank
 			--squeeze-limit
+			--sanitize
 			--strip-ansi
 			--style
 			--line-range

assets/completions/bat.fish.in

@@ -221,6 +221,7 @@ complete -c $bat -s s -l squeeze-blank -d "Squeeze consecutive empty lines into
 complete -c $bat -l squeeze-limit -x -d "Set the maximum number of consecutive empty lines to be printed" -n __bat_no_excl_args
 
 complete -c $bat -l strip-ansi -x -a "auto never always" -d "Specify when to strip ANSI escape sequences from the input" -n __bat_no_excl_args
+complete -c $bat -l sanitize -x -a "auto never always" -d "Specify when to sanitize untrusted input for safe display" -n __bat_no_excl_args
 
 complete -c $bat -s p -l plain -d "Disable decorations" -n __bat_no_excl_args
 

assets/completions/bat.zsh.in

@@ -54,6 +54,7 @@ _{{PROJECT_EXECUTABLE}}_main() {
         --squeeze-blank'[squeeze consecutive empty lines into a single empty line]'
         --squeeze-limit='[set the maximum number of consecutive empty lines]:limit:'
         --strip-ansi='[specify when to strip ANSI escape sequences]:when:(auto never always)'
+        --sanitize='[specify when to sanitize untrusted input for safe display]:when:(auto never always)'
         --style='[comma-separated list of style elements to display]: : _values "style [default]"
             default auto full plain changes header header-filename header-filesize grid rule numbers snip'
         \*{-r+,--line-range=}'[only print the specified line range]:start\:end'

doc/long-help.txt

@@ -164,6 +164,14 @@ Options:
           escape sequences unless the syntax highlighting language is plain text. Possible values:
           auto, always, *never*.
 
+      --sanitize <when>
+          Specify when to sanitize input bytes for safe terminal display. Implies --strip-ansi to
+          the same value, and additionally substitutes terminal-active control bytes (cursor moves,
+          charset switches, beep, etc.) and Unicode bidi / zero-width formatting characters with the
+          Unicode replacement character (U+FFFD). Tab, LF, FF, and CRLF pass through. Useful for
+          displaying untrusted file content (e.g. file-manager preview panes). Possible values:
+          auto, always, *never*.
+
       --style <components>
           Configure which elements (line numbers, file headers, grid borders, Git modifications, ..)
           to display in addition to the file contents. The argument is a comma-separated list of

src/bin/bat/app.rs

@@ -38,6 +38,15 @@ pub fn env_no_color() -> bool {
     env::var_os("NO_COLOR").is_some_and(|x| !x.is_empty())
 }
 
+fn parse_strip_ansi_value(raw: Option<&str>, flag_name: &str) -> StripAnsiMode {
+    match raw {
+        Some("never") | None => StripAnsiMode::Never,
+        Some("always") => StripAnsiMode::Always,
+        Some("auto") => StripAnsiMode::Auto,
+        _ => unreachable!("other values for {flag_name} are not allowed"),
+    }
+}
+
 enum HelpType {
     Short,
     Long,
@@ -458,16 +467,32 @@ impl App {
                         4
                     },
                 ),
-            strip_ansi: match self
+            strip_ansi: {
-                .matches
+                let sanitize = parse_strip_ansi_value(
-                .get_one::<String>("strip-ansi")
+                    self.matches
-                .map(|s| s.as_str())
+                        .get_one::<String>("sanitize")
-            {
+                        .map(|s| s.as_str()),
-                Some("never") => StripAnsiMode::Never,
+                    "--sanitize",
-                Some("always") => StripAnsiMode::Always,
+                );
-                Some("auto") => StripAnsiMode::Auto,
+                let strip_ansi = parse_strip_ansi_value(
-                _ => unreachable!("other values for --strip-ansi are not allowed"),
+                    self.matches
+                        .get_one::<String>("strip-ansi")
+                        .map(|s| s.as_str()),
+                    "--strip-ansi",
+                );
+                // --sanitize implies --strip-ansi to the same value.
+                if sanitize != StripAnsiMode::Never {
+                    sanitize
+                } else {
+                    strip_ansi
+                }
             },
+            sanitize: parse_strip_ansi_value(
+                self.matches
+                    .get_one::<String>("sanitize")
+                    .map(|s| s.as_str()),
+                "--sanitize",
+            ),
             quiet_empty: self.matches.get_flag("quiet-empty"),
             unbuffered: self.matches.get_flag("unbuffered"),
             theme: theme(self.theme_options()).to_string(),

src/bin/bat/clap_app.rs

@@ -480,6 +480,24 @@ pub fn build_app(interactive_output: bool) -> Command {
                 language is plain text. Possible values: auto, always, *never*.")
                 .hide_short_help(true)
         )
+        .arg(
+            Arg::new("sanitize")
+                .long("sanitize")
+                .overrides_with("sanitize")
+                .value_name("when")
+                .value_parser(["auto", "always", "never"])
+                .default_value("never")
+                .hide_default_value(true)
+                .help("Sanitize untrusted input for safe display (auto, always, *never*)")
+                .long_help("Specify when to sanitize input bytes for safe terminal display. \
+                Implies --strip-ansi to the same value, and additionally substitutes \
+                terminal-active control bytes (cursor moves, charset switches, beep, etc.) \
+                and Unicode bidi / zero-width formatting characters with the Unicode \
+                replacement character (U+FFFD). Tab, LF, FF, and CRLF pass through. Useful \
+                for displaying untrusted file content (e.g. file-manager preview panes). \
+                Possible values: auto, always, *never*.")
+                .hide_short_help(true)
+        )
         .arg(
             Arg::new("style")
                 .long("style")

src/config.rs

@@ -111,6 +111,9 @@ pub struct Config<'a> {
     // Whether or not to strip ANSI escape codes from the input
     pub strip_ansi: StripAnsiMode,
 
+    // Substitute terminal-active and spoofing-relevant bytes; implies strip_ansi.
+    pub sanitize: StripAnsiMode,
+
     /// Whether or not to produce no output when input is empty
     pub quiet_empty: bool,
 

src/preprocessor.rs

@@ -139,16 +139,104 @@ pub fn replace_nonprintable(
 /// Strips ANSI escape sequences from the input.
 pub fn strip_ansi(line: &str) -> String {
     let mut buffer = String::with_capacity(line.len());
-
     for seq in EscapeSequenceOffsetsIterator::new(line) {
         if let EscapeSequenceOffsets::Text { .. } = seq {
             buffer.push_str(&line[seq.index_of_start()..seq.index_past_end()]);
         }
     }
-
     buffer
 }
 
+/// Strips ANSI escape sequences and substitutes terminal-active control bytes
+/// and visual-spoofing Unicode codepoints (bidi, zero-width) with U+FFFD.
+pub fn sanitize(line: &str) -> String {
+    let stripped = strip_ansi(line);
+    let mut buffer = String::with_capacity(stripped.len());
+    let bytes = stripped.as_bytes();
+    let mut start = 0;
+    let mut i = 0;
+    // Skip directly to the next trigger byte instead of testing each one.
+    while let Some(off) = bytes[i..].iter().position(|&b| is_sanitize_trigger(b)) {
+        i += off;
+        let len = sanitize_at(bytes, i, &stripped, &mut buffer, &mut start);
+        i += len;
+    }
+    buffer.push_str(&stripped[start..]);
+    buffer
+}
+
+#[inline]
+fn is_sanitize_trigger(b: u8) -> bool {
+    // C0 controls minus \t \n \f; DEL; UTF-8 leads with dangerous codepoints.
+    matches!(b, 0x00..=0x08 | 0x0B | 0x0D..=0x1F | 0x7F | 0xC2 | 0xE2 | 0xEF)
+}
+
+/// Substitutes the byte/sequence at `bytes[i]` (or passes it through on
+/// false-alarm trigger), flushing the prefix from `start`. Returns bytes consumed.
+fn sanitize_at(
+    bytes: &[u8],
+    i: usize,
+    full: &str,
+    buffer: &mut String,
+    start: &mut usize,
+) -> usize {
+    buffer.push_str(&full[*start..i]);
+    let consumed = match bytes[i] {
+        b'\r' if bytes.get(i + 1) == Some(&b'\n') => {
+            buffer.push_str("\r\n");
+            2
+        }
+        // 0xC2 leads U+0080..U+00FF; filter the C1 range.
+        0xC2 if matches!(bytes.get(i + 1), Some(0x80..=0x9F)) => {
+            buffer.push('\u{FFFD}');
+            2
+        }
+        0xE2 if is_dangerous_e2(bytes, i) => {
+            buffer.push('\u{FFFD}');
+            3
+        }
+        // 0xEF 0xBB 0xBF = U+FEFF (BOM / zero-width no-break space).
+        0xEF if bytes.get(i + 1) == Some(&0xBB) && bytes.get(i + 2) == Some(&0xBF) => {
+            buffer.push('\u{FFFD}');
+            3
+        }
+        // False-alarm trigger: pass the full UTF-8 sequence through.
+        lead @ (0xC2 | 0xE2 | 0xEF) => {
+            let n = utf8_len_from_lead(lead);
+            buffer.push_str(&full[i..i + n]);
+            n
+        }
+        _ => {
+            buffer.push('\u{FFFD}');
+            1
+        }
+    };
+    *start = i + consumed;
+    consumed
+}
+
+#[inline]
+fn is_dangerous_e2(bytes: &[u8], i: usize) -> bool {
+    // U+200B..D (zero-width), U+202A..E (bidi controls), U+2066..9 (isolates).
+    matches!(
+        (bytes.get(i + 1), bytes.get(i + 2)),
+        (Some(0x80), Some(0x8B..=0x8D | 0xAA..=0xAE)) | (Some(0x81), Some(0xA6..=0xA9))
+    )
+}
+
+#[inline]
+fn utf8_len_from_lead(lead: u8) -> usize {
+    if lead < 0x80 {
+        1
+    } else if lead < 0xE0 {
+        2
+    } else if lead < 0xF0 {
+        3
+    } else {
+        4
+    }
+}
+
 /// Escape C0, DEL, and C1 control characters so a string from an untrusted
 /// filename or path can be safely written to the terminal.
 pub fn sanitize_for_terminal(input: &str) -> String {
@@ -270,6 +358,85 @@ fn test_strip_ansi() {
     );
 }
 
+#[test]
+fn test_strip_ansi_8bit_c1_introducers() {
+    assert_eq!(strip_ansi("a\u{9B}31mRED\u{9B}0mb"), "aREDb");
+    assert_eq!(strip_ansi("a\x1bP1;0|payload\x1b\\b"), "ab");
+    assert_eq!(strip_ansi("a\u{90}body\u{9C}b"), "ab");
+}
+
+#[test]
+fn test_strip_ansi_single_char_esc() {
+    // strip_ansi must consume both bytes of single-byte ESC sequences (RIS, DECSC, keypad, VT52).
+    assert_eq!(strip_ansi("a\x1bcb"), "ab");
+    assert_eq!(strip_ansi("a\x1b7b\x1b8c"), "abc");
+    assert_eq!(strip_ansi("a\x1b=b\x1b>c"), "abc");
+    assert_eq!(strip_ansi("a\x1bZb"), "ab");
+}
+
+#[test]
+fn test_strip_ansi_preserves_control_bytes() {
+    // strip_ansi removes only ANSI escape sequences; control bytes pass through.
+    assert_eq!(strip_ansi("safe\rEVIL"), "safe\rEVIL");
+    assert_eq!(strip_ansi("a\x08b\x07c\x0E\x0Fd"), "a\x08b\x07c\x0E\x0Fd");
+}
+
+#[test]
+fn test_sanitize_substitutes_dangerous_bytes() {
+    let r = '\u{FFFD}';
+    assert_eq!(sanitize("safe\rEVIL"), format!("safe{r}EVIL"));
+    assert_eq!(sanitize("a\x08b"), format!("a{r}b"));
+    assert_eq!(sanitize("a\x07b"), format!("a{r}b"));
+    assert_eq!(sanitize("a\x0Bb"), format!("a{r}b"));
+    assert_eq!(sanitize("a\x0Eb\x0Fc"), format!("a{r}b{r}c"));
+    assert_eq!(sanitize("a\u{8D}b"), format!("a{r}b"));
+    assert_eq!(sanitize("a\u{85}b"), format!("a{r}b"));
+    assert_eq!(sanitize("trailing\r"), format!("trailing{r}"));
+    assert_eq!(sanitize("a\x7Fb"), format!("a{r}b"));
+}
+
+#[test]
+fn test_sanitize_substitutes_bidi_and_zero_width() {
+    let r = '\u{FFFD}';
+    // Trojan-Source bidi formatting: U+202A..U+202E
+    assert_eq!(sanitize("a\u{202A}b"), format!("a{r}b"));
+    assert_eq!(sanitize("a\u{202E}b"), format!("a{r}b"));
+    // Bidi isolates: U+2066..U+2069
+    assert_eq!(sanitize("a\u{2066}b"), format!("a{r}b"));
+    assert_eq!(sanitize("a\u{2069}b"), format!("a{r}b"));
+    // Zero-width: U+200B..U+200D
+    assert_eq!(sanitize("a\u{200B}b"), format!("a{r}b"));
+    assert_eq!(sanitize("a\u{200D}b"), format!("a{r}b"));
+    // BOM in middle of file: U+FEFF
+    assert_eq!(sanitize("a\u{FEFF}b"), format!("a{r}b"));
+}
+
+#[test]
+fn test_sanitize_preserves_legitimate_bytes() {
+    assert_eq!(sanitize("crlf\r\nline\r\n"), "crlf\r\nline\r\n");
+    assert_eq!(sanitize("a\tb\nc"), "a\tb\nc");
+    assert_eq!(sanitize("plain ascii"), "plain ascii");
+    assert_eq!(sanitize("üñíçödé"), "üñíçödé");
+    // FF (U+000C) passes through; section separator in C source / Emacs Lisp.
+    assert_eq!(sanitize("section1\x0Csection2"), "section1\x0Csection2");
+    // Common Unicode that shares a UTF-8 lead byte with dangerous codepoints
+    // must pass through unchanged.
+    assert_eq!(sanitize("snowman ☃ moon ☾"), "snowman ☃ moon ☾");
+    assert_eq!(sanitize("emoji 🎉 ☃"), "emoji 🎉 ☃");
+    assert_eq!(sanitize("0xC2 lead: ÿ ñ ç"), "0xC2 lead: ÿ ñ ç");
+    assert_eq!(sanitize("CJK 漢字 emoji 🦀"), "CJK 漢字 emoji 🦀");
+}
+
+#[test]
+fn test_sanitize_strips_ansi() {
+    let r = '\u{FFFD}';
+    // sanitize is a strict superset of strip_ansi.
+    assert_eq!(sanitize("a\x1B[31mb\x1B[0mc"), "abc");
+    assert_eq!(sanitize("a\u{9B}31mb\u{9B}0mc"), "abc");
+    // ANSI then dangerous byte: ANSI gone, byte substituted.
+    assert_eq!(sanitize("\x1B[31mhello\rEVIL"), format!("hello{r}EVIL"));
+}
+
 #[test]
 fn test_strip_overstrike() {
     // Bold: X\x08X (same char repeated)

src/pretty_printer.rs

@@ -192,6 +192,15 @@ impl<'a> PrettyPrinter<'a> {
         self
     }
 
+    /// Whether to sanitize untrusted input for safe display (default: never)
+    ///
+    /// Strips ANSI escape sequences and additionally substitutes terminal-active
+    /// control bytes and bidi / zero-width codepoints with U+FFFD.
+    pub fn sanitize(&mut self, mode: StripAnsiMode) -> &mut Self {
+        self.config.sanitize = mode;
+        self
+    }
+
     /// Text wrapping mode (default: do not wrap)
     pub fn wrapping_mode(&mut self, mode: WrappingMode) -> &mut Self {
         self.config.wrapping_mode = mode;

src/printer.rs

@@ -30,7 +30,8 @@ use crate::input::OpenedInput;
 use crate::line_range::{MaxBufferedLineNumber, RangeCheckResult};
 use crate::output::OutputHandle;
 use crate::preprocessor::{
-    expand_tabs, replace_nonprintable, sanitize_for_terminal, strip_ansi, strip_overstrike,
+    expand_tabs, replace_nonprintable, sanitize, sanitize_for_terminal, strip_ansi,
+    strip_overstrike,
 };
 use crate::style::StyleComponent;
 use crate::terminal::{as_terminal_escaped, to_ansi_color};
@@ -210,6 +211,7 @@ pub(crate) struct InteractivePrinter<'a> {
     background_color_highlight: Option<Color>,
     consecutive_empty_lines: usize,
     strip_ansi: bool,
+    sanitize: bool,
     strip_overstrike: bool,
 }
 
@@ -273,7 +275,9 @@ impl<'a> InteractivePrinter<'a> {
 
         let needs_to_match_syntax = (!is_printing_binary
             || matches!(config.binary, BinaryBehavior::AsText))
-            && (config.colored_output || config.strip_ansi == StripAnsiMode::Auto);
+            && (config.colored_output
+                || config.strip_ansi == StripAnsiMode::Auto
+                || config.sanitize == StripAnsiMode::Auto);
 
         let (is_plain_text, strip_overstrike, highlighter_from_set) = if needs_to_match_syntax {
             // Determine the type of syntax for highlighting
@@ -319,6 +323,14 @@ impl<'a> InteractivePrinter<'a> {
             _ => false,
         };
 
+        let sanitize = match config.sanitize {
+            _ if config.show_nonprintable => false,
+            StripAnsiMode::Always => true,
+            StripAnsiMode::Auto if is_plain_text => false,
+            StripAnsiMode::Auto => true,
+            _ => false,
+        };
+
         Ok(InteractivePrinter {
             panel_width,
             colors,
@@ -332,6 +344,7 @@ impl<'a> InteractivePrinter<'a> {
             background_color_highlight,
             consecutive_empty_lines: 0,
             strip_ansi,
+            sanitize,
             strip_overstrike,
         })
     }
@@ -675,8 +688,10 @@ impl Printer for InteractivePrinter<'_> {
                 }
             }
 
-            // If ANSI escape sequences are supposed to be stripped, do it before syntax highlighting.
+            // Sanitize is the strict superset; otherwise strip-ansi alone.
-            if self.strip_ansi {
+            if self.sanitize {
+                line = sanitize(&line).into()
+            } else if self.strip_ansi {
                 line = strip_ansi(&line).into()
             }
 

src/vscreen.rs

@@ -386,12 +386,23 @@ impl<'a> EscapeSequenceOffsetsIterator<'a> {
     }
 
     fn next_text(&mut self) -> Option<EscapeSequenceOffsets> {
-        self.chars_take_while(|c| c != '\x1B')
+        self.chars_take_while(|c| !is_sequence_introducer(c))
             .map(|(start, end)| EscapeSequenceOffsets::Text { start, end })
     }
 
     fn next_sequence(&mut self) -> Option<EscapeSequenceOffsets> {
         let (start_sequence, c) = self.chars.next().expect("to not be finished");
+
+        // Handle 8-bit C1 introducers as their 7-bit `ESC <x>` equivalents.
+        match c {
+            '\u{9B}' => return self.next_csi_body(start_sequence),
+            '\u{9D}' => return self.next_osc_body(start_sequence),
+            '\u{90}' | '\u{98}' | '\u{9E}' | '\u{9F}' => {
+                return self.next_string_terminated_body(start_sequence)
+            }
+            _ => {}
+        }
+
         match self.chars.peek() {
             None => Some(EscapeSequenceOffsets::Unknown {
                 start: start_sequence,
@@ -400,12 +411,24 @@ impl<'a> EscapeSequenceOffsetsIterator<'a> {
 
             Some((_, ']')) => self.next_osc(start_sequence),
             Some((_, '[')) => self.next_csi(start_sequence),
-            Some((i, c)) => match c {
+            // 7-bit DCS/SOS/PM/APC: `ESC P/X/^/_` introduces a string body.
+            Some((_, 'P' | 'X' | '^' | '_')) => {
+                self.chars.next();
+                self.next_string_terminated_body(start_sequence)
+            }
+            Some(&(i, c)) => match c {
                 '\x20'..='\x2F' => self.next_nf(start_sequence),
-                c => Some(EscapeSequenceOffsets::Unknown {
+                c => {
-                    start: start_sequence,
+                    // Single-byte ESC sequence (RIS, DECSC/DECRC, keypad, VT52 etc.).
-                    end: i + c.len_utf8(),
+                    let end = match self.chars.next() {
-                }),
+                        Some((j, fc)) => j + fc.len_utf8(),
+                        None => i + c.len_utf8(),
+                    };
+                    Some(EscapeSequenceOffsets::Unknown {
+                        start: start_sequence,
+                        end,
+                    })
+                }
             },
         }
     }
@@ -413,12 +436,27 @@ impl<'a> EscapeSequenceOffsetsIterator<'a> {
     fn next_osc(&mut self, start_sequence: usize) -> Option<EscapeSequenceOffsets> {
         let (osc_open_index, osc_open_char) = self.chars.next().expect("to not be finished");
         debug_assert_eq!(osc_open_char, ']');
+        let start_command = osc_open_index + osc_open_char.len_utf8();
+        Some(self.read_osc_body(start_sequence, start_command))
+    }
 
+    /// OSC body parser entered after the 8-bit introducer U+009D was consumed.
+    fn next_osc_body(&mut self, start_sequence: usize) -> Option<EscapeSequenceOffsets> {
+        let start_command = start_sequence + '\u{9D}'.len_utf8();
+        Some(self.read_osc_body(start_sequence, start_command))
+    }
+
+    fn read_osc_body(
+        &mut self,
+        start_sequence: usize,
+        start_command: usize,
+    ) -> EscapeSequenceOffsets {
         let mut start_terminator: usize;
         let mut end_sequence: usize;
 
         loop {
-            match self.chars_take_while(|c| !matches!(c, '\x07' | '\x1B')) {
+            // ST is BEL, ESC `\\`, or U+009C.
+            match self.chars_take_while(|c| !matches!(c, '\x07' | '\x1B' | '\u{9C}')) {
                 None => {
                     start_terminator = self.text.len();
                     end_sequence = start_terminator;
@@ -437,6 +475,11 @@ impl<'a> EscapeSequenceOffsetsIterator<'a> {
                     break;
                 }
 
+                Some((ti, '\u{9C}')) => {
+                    end_sequence = ti + '\u{9C}'.len_utf8();
+                    break;
+                }
+
                 Some((ti, '\x1B')) => {
                     match self.chars.next() {
                         Some((i, '\\')) => {
@@ -466,20 +509,82 @@ impl<'a> EscapeSequenceOffsetsIterator<'a> {
             }
         }
 
-        Some(EscapeSequenceOffsets::OSC {
+        EscapeSequenceOffsets::OSC {
             start_sequence,
-            start_command: osc_open_index + osc_open_char.len_utf8(),
+            start_command,
             start_terminator,
             end: end_sequence,
+        }
+    }
+
+    /// DCS/SOS/PM/APC body parser. Emitted as `Unknown` so the body is stripped.
+    fn next_string_terminated_body(
+        &mut self,
+        start_sequence: usize,
+    ) -> Option<EscapeSequenceOffsets> {
+        let mut end_sequence: usize;
+
+        loop {
+            match self.chars_take_while(|c| !matches!(c, '\x07' | '\x1B' | '\u{9C}')) {
+                None => {
+                    end_sequence = self.text.len();
+                    break;
+                }
+                Some((_, end)) => {
+                    end_sequence = end;
+                }
+            }
+
+            match self.chars.next() {
+                Some((ti, '\x07')) => {
+                    end_sequence = ti + '\x07'.len_utf8();
+                    break;
+                }
+                Some((ti, '\u{9C}')) => {
+                    end_sequence = ti + '\u{9C}'.len_utf8();
+                    break;
+                }
+                Some((ti, '\x1B')) => match self.chars.next() {
+                    Some((i, '\\')) => {
+                        end_sequence = i + '\\'.len_utf8();
+                        break;
+                    }
+                    None => {
+                        end_sequence = ti + '\x1B'.len_utf8();
+                        break;
+                    }
+                    _ => {}
+                },
+                None => break,
+                Some((_, tc)) => {
+                    panic!("this should not be reached: char {tc:?}")
+                }
+            }
+        }
+
+        Some(EscapeSequenceOffsets::Unknown {
+            start: start_sequence,
+            end: end_sequence,
         })
     }
 
     fn next_csi(&mut self, start_sequence: usize) -> Option<EscapeSequenceOffsets> {
         let (csi_open_index, csi_open_char) = self.chars.next().expect("to not be finished");
         debug_assert_eq!(csi_open_char, '[');
+        Some(self.read_csi_body(start_sequence, csi_open_index + csi_open_char.len_utf8()))
+    }
 
-        let start_parameters: usize = csi_open_index + csi_open_char.len_utf8();
+    /// CSI body parser entered after the 8-bit introducer U+009B was consumed.
+    fn next_csi_body(&mut self, start_sequence: usize) -> Option<EscapeSequenceOffsets> {
+        let start_parameters = start_sequence + '\u{9B}'.len_utf8();
+        Some(self.read_csi_body(start_sequence, start_parameters))
+    }
 
+    fn read_csi_body(
+        &mut self,
+        start_sequence: usize,
+        start_parameters: usize,
+    ) -> EscapeSequenceOffsets {
         // Keep iterating while within the range of `0x30-0x3F`.
         let mut start_intermediates: usize = start_parameters;
         if let Some((_, end)) = self.chars_take_while(|c| matches!(c, '\x30'..='\x3F')) {
@@ -498,13 +603,13 @@ impl<'a> EscapeSequenceOffsetsIterator<'a> {
             Some((i, c)) => i + c.len_utf8(),
         };
 
-        Some(EscapeSequenceOffsets::CSI {
+        EscapeSequenceOffsets::CSI {
             start_sequence,
             start_parameters,
             start_intermediates,
             start_final_byte,
             end: end_of_sequence,
-        })
+        }
     }
 
     fn next_nf(&mut self, start_sequence: usize) -> Option<EscapeSequenceOffsets> {
@@ -543,13 +648,25 @@ impl Iterator for EscapeSequenceOffsetsIterator<'_> {
     type Item = EscapeSequenceOffsets;
     fn next(&mut self) -> Option<Self::Item> {
         match self.chars.peek() {
-            Some((_, '\x1B')) => self.next_sequence(),
+            Some((_, c)) if is_sequence_introducer(*c) => self.next_sequence(),
             Some((_, _)) => self.next_text(),
             None => None,
         }
     }
 }
 
+/// True for ESC and the 8-bit C1 sequence introducers (DCS/SOS/CSI/OSC/PM/APC).
+#[inline]
+fn is_sequence_introducer(c: char) -> bool {
+    if (c as u32) < 0x80 {
+        return c == '\x1B';
+    }
+    matches!(
+        c,
+        '\u{90}' | '\u{98}' | '\u{9B}' | '\u{9D}' | '\u{9E}' | '\u{9F}'
+    )
+}
+
 /// An iterator over ANSI/VT escape sequences within a string.
 ///
 /// ## Example
@@ -717,6 +834,79 @@ mod tests {
         );
     }
 
+    #[test]
+    fn test_escape_sequence_offsets_iterator_parses_8bit_csi() {
+        let mut iter = EscapeSequenceOffsetsIterator::new("\u{9B}31m");
+        assert_eq!(
+            iter.next(),
+            Some(EscapeSequenceOffsets::CSI {
+                start_sequence: 0,
+                start_parameters: 2,
+                start_intermediates: 4,
+                start_final_byte: 4,
+                end: 5,
+            })
+        );
+        assert_eq!(iter.next(), None);
+    }
+
+    #[test]
+    fn test_escape_sequence_offsets_iterator_parses_8bit_osc_with_8bit_st() {
+        let mut iter = EscapeSequenceOffsetsIterator::new("\u{9D}0;title\u{9C}");
+        assert_eq!(
+            iter.next(),
+            Some(EscapeSequenceOffsets::OSC {
+                start_sequence: 0,
+                start_command: 2,
+                start_terminator: 9,
+                end: 11,
+            })
+        );
+        assert_eq!(iter.next(), None);
+    }
+
+    #[test]
+    fn test_escape_sequence_offsets_iterator_parses_7bit_dcs_consumes_body() {
+        let mut iter = EscapeSequenceOffsetsIterator::new("\x1BP1;0;|payload\x1B\\rest");
+        assert_eq!(
+            iter.next(),
+            Some(EscapeSequenceOffsets::Unknown { start: 0, end: 16 })
+        );
+        assert_eq!(
+            iter.next(),
+            Some(EscapeSequenceOffsets::Text { start: 16, end: 20 })
+        );
+        assert_eq!(iter.next(), None);
+    }
+
+    #[test]
+    fn test_escape_sequence_offsets_iterator_8bit_dcs_consumes_body() {
+        let mut iter = EscapeSequenceOffsetsIterator::new("\u{90}body\u{9C}rest");
+        assert_eq!(
+            iter.next(),
+            Some(EscapeSequenceOffsets::Unknown { start: 0, end: 8 })
+        );
+        assert_eq!(
+            iter.next(),
+            Some(EscapeSequenceOffsets::Text { start: 8, end: 12 })
+        );
+    }
+
+    #[test]
+    fn test_escape_sequence_offsets_iterator_truncated_dcs_consumes_to_eof() {
+        // Unterminated DCS must still be consumed, not emitted as `Text`.
+        let input = "\x1BPno-terminator";
+        let mut iter = EscapeSequenceOffsetsIterator::new(input);
+        assert_eq!(
+            iter.next(),
+            Some(EscapeSequenceOffsets::Unknown {
+                start: 0,
+                end: input.len(),
+            })
+        );
+        assert_eq!(iter.next(), None);
+    }
+
     #[test]
     fn test_escape_sequence_offsets_iterator_parses_csi() {
         let mut iter = EscapeSequenceOffsetsIterator::new("\x1B[m");

tests/integration_tests.rs

@@ -3844,6 +3844,158 @@ fn strip_ansi_auto_does_not_strip_ansi_when_plain_text_by_option() {
     assert!(output.contains("\x1B[33mYellow"))
 }
 
+#[test]
+fn sanitize_implies_strip_ansi() {
+    bat()
+        .arg("--style=plain")
+        .arg("--decorations=always")
+        .arg("--color=never")
+        .arg("--sanitize=always")
+        .write_stdin("\x1B[33mYellow\x1B[m")
+        .assert()
+        .success()
+        .stdout("Yellow");
+}
+
+#[test]
+fn sanitize_strips_osc_clipboard_hijack() {
+    // OSC 52 sets the system clipboard. A file containing this would silently
+    // overwrite the user's clipboard if displayed unfiltered.
+    bat()
+        .arg("--style=plain")
+        .arg("--decorations=always")
+        .arg("--color=never")
+        .arg("--sanitize=always")
+        .write_stdin("safe\x1B]52;c;cm0=\x07payload")
+        .assert()
+        .success()
+        .stdout("safepayload");
+}
+
+#[test]
+fn sanitize_strips_osc_8_hyperlink_spoof() {
+    // OSC 8 hyperlinks let displayed text point to an arbitrary URL.
+    bat()
+        .arg("--style=plain")
+        .arg("--decorations=always")
+        .arg("--color=never")
+        .arg("--sanitize=always")
+        .write_stdin("\x1B]8;;https://evil.example\x07click here\x1B]8;;\x07")
+        .assert()
+        .success()
+        .stdout("click here");
+}
+
+#[test]
+fn sanitize_strips_window_title_injection() {
+    // OSC 0/1/2 set the terminal window title.
+    bat()
+        .arg("--style=plain")
+        .arg("--decorations=always")
+        .arg("--color=never")
+        .arg("--sanitize=always")
+        .write_stdin("hello\x1B]0;evil-title\x07world")
+        .assert()
+        .success()
+        .stdout("helloworld");
+}
+
+#[test]
+fn sanitize_strips_8bit_csi() {
+    // 8-bit CSI introducer (U+009B) is the single-codepoint equivalent of ESC [.
+    bat()
+        .arg("--style=plain")
+        .arg("--decorations=always")
+        .arg("--color=never")
+        .arg("--sanitize=always")
+        .write_stdin("a\u{9B}31mRED\u{9B}0mb")
+        .assert()
+        .success()
+        .stdout("aREDb");
+}
+
+#[test]
+fn sanitize_substitutes_bare_cr() {
+    // Bare CR (not part of CRLF) is the line-overwrite forgery vector.
+    bat()
+        .arg("--style=plain")
+        .arg("--decorations=always")
+        .arg("--color=never")
+        .arg("--sanitize=always")
+        .write_stdin("safe\rEVIL")
+        .assert()
+        .success()
+        .stdout("safe\u{FFFD}EVIL");
+}
+
+#[test]
+fn sanitize_preserves_crlf() {
+    bat()
+        .arg("--style=plain")
+        .arg("--decorations=always")
+        .arg("--color=never")
+        .arg("--sanitize=always")
+        .write_stdin("line1\r\nline2\r\n")
+        .assert()
+        .success()
+        .stdout("line1\r\nline2\r\n");
+}
+
+#[test]
+fn sanitize_substitutes_bidi_controls() {
+    // Trojan-Source attack (CVE-2021-42574): U+202E (RLO) reorders display.
+    bat()
+        .arg("--style=plain")
+        .arg("--decorations=always")
+        .arg("--color=never")
+        .arg("--sanitize=always")
+        .write_stdin("admin\u{202E}check")
+        .assert()
+        .success()
+        .stdout("admin\u{FFFD}check");
+}
+
+#[test]
+fn sanitize_substitutes_zero_width() {
+    // Zero-width chars allow invisible content / identifier confusion.
+    bat()
+        .arg("--style=plain")
+        .arg("--decorations=always")
+        .arg("--color=never")
+        .arg("--sanitize=always")
+        .write_stdin("ad\u{200B}min")
+        .assert()
+        .success()
+        .stdout("ad\u{FFFD}min");
+}
+
+#[test]
+fn sanitize_preserves_form_feed_in_source() {
+    // FF (U+000C) is used as a section separator in C source and Emacs Lisp.
+    bat()
+        .arg("--style=plain")
+        .arg("--decorations=always")
+        .arg("--color=never")
+        .arg("--sanitize=always")
+        .write_stdin("section1\x0Csection2")
+        .assert()
+        .success()
+        .stdout("section1\x0Csection2");
+}
+
+#[test]
+fn sanitize_preserves_unicode_text() {
+    bat()
+        .arg("--style=plain")
+        .arg("--decorations=always")
+        .arg("--color=never")
+        .arg("--sanitize=always")
+        .write_stdin("snowman ☃ CJK 漢字 emoji 🦀")
+        .assert()
+        .success()
+        .stdout("snowman ☃ CJK 漢字 emoji 🦀");
+}
+
 // Tests that style components can be removed with `-component`.
 #[test]
 fn style_components_can_be_removed() {