feat: enhance ANSI code stripping in ClaudeUsageService

- Improved the stripAnsiCodes method to handle various ANSI escape sequences, including CSI, OSC, and single-character sequences.
- Added logic to manage backspaces and explicitly strip known "Synchronized Output" and "Window Title" garbage.
- Updated tests to cover new functionality, ensuring robust handling of complex terminal outputs and control characters.

This enhancement improves the reliability of text processing in terminal environments.

apps/server/src/services/claude-usage-service.ts

@@ -468,10 +468,41 @@ export class ClaudeUsageService {
 
   /**
    * Strip ANSI escape codes from text
+   * Handles CSI, OSC, and other common ANSI sequences
    */
   private stripAnsiCodes(text: string): string {
+    // First strip ANSI sequences (colors, etc) and handle CR
     // eslint-disable-next-line no-control-regex
-    return text.replace(/\x1B\[[0-9;]*[A-Za-z]/g, '');
+    let clean = text
+      // CSI sequences: ESC [ ... (letter or @)
+      .replace(/\x1B\[[0-9;?]*[A-Za-z@]/g, '')
+      // OSC sequences: ESC ] ... terminated by BEL, ST, or another ESC
+      .replace(/\x1B\][^\x07\x1B]*(?:\x07|\x1B\\)?/g, '')
+      // Other ESC sequences: ESC (letter)
+      .replace(/\x1B[A-Za-z]/g, '')
+      // Carriage returns: replace with newline to avoid concatenation
+      .replace(/\r\n/g, '\n')
+      .replace(/\r/g, '\n');
+
+    // Handle backspaces (\x08) by applying them
+    // If we encounter a backspace, remove the character before it
+    while (clean.includes('\x08')) {
+      clean = clean.replace(/[^\x08]\x08/, '');
+      clean = clean.replace(/^\x08+/, '');
+    }
+
+    // Explicitly strip known "Synchronized Output" and "Window Title" garbage
+    // even if ESC is missing (seen in some environments)
+    clean = clean
+      .replace(/\[\?2026[hl]/g, '') // CSI ? 2026 h/l
+      .replace(/\]0;[^\x07]*\x07/g, '') // OSC 0; Title BEL
+      .replace(/\]0;.*?(\[\?|$)/g, ''); // OSC 0; Title ... (unterminated or hit next sequence)
+
+    // Strip remaining non-printable control characters (except newline \n)
+    // ASCII 0-8, 11-31, 127
+    clean = clean.replace(/[\x00-\x08\x0B-\x1F\x7F]/g, '');
+
+    return clean;
   }
 
   /**
@@ -550,7 +581,7 @@ export class ClaudeUsageService {
     sectionLabel: string,
     type: string
   ): { percentage: number; resetTime: string; resetText: string } {
-    let percentage = 0;
+    let percentage: number | null = null;
     let resetTime = this.getDefaultResetTime(type);
     let resetText = '';
 
@@ -564,7 +595,7 @@ export class ClaudeUsageService {
     }
 
     if (sectionIndex === -1) {
-      return { percentage, resetTime, resetText };
+      return { percentage: 0, resetTime, resetText };
     }
 
     // Look at the lines following the section header (within a window of 5 lines)
@@ -572,7 +603,8 @@ export class ClaudeUsageService {
 
     for (const line of searchWindow) {
       // Extract percentage - only take the first match (avoid picking up next section's data)
-      if (percentage === 0) {
+      // Use null to track "not found" since 0% is a valid percentage (100% left = 0% used)
+      if (percentage === null) {
         const percentMatch = line.match(/(\d{1,3})\s*%\s*(left|used|remaining)/i);
         if (percentMatch) {
           const value = parseInt(percentMatch[1], 10);
@@ -584,18 +616,31 @@ export class ClaudeUsageService {
 
       // Extract reset time - only take the first match
       if (!resetText && line.toLowerCase().includes('reset')) {
-        resetText = line;
+        // Only extract the part starting from "Resets" (or "Reset") to avoid garbage prefixes
+        const match = line.match(/(Resets?.*)$/i);
+        // If regex fails despite 'includes', likely a complex string issues - verify match before using line
+        // Only fallback to line if it's reasonably short/clean, otherwise skip it to avoid showing garbage
+        if (match) {
+          resetText = match[1];
+        }
       }
     }
 
     // Parse the reset time if we found one
     if (resetText) {
+      // Clean up resetText: remove percentage info if it was matched on the same line
+      // e.g. "46%used Resets5:59pm" -> " Resets5:59pm"
+      resetText = resetText.replace(/(\d{1,3})\s*%\s*(left|used|remaining)/i, '').trim();
+
+      // Ensure space after "Resets" if missing (e.g. "Resets5:59pm" -> "Resets 5:59pm")
+      resetText = resetText.replace(/(resets?)(\d)/i, '$1 $2');
+
       resetTime = this.parseResetTime(resetText, type);
       // Strip timezone like "(Asia/Dubai)" from the display text
       resetText = resetText.replace(/\s*\([A-Za-z_\/]+\)\s*$/, '').trim();
     }
 
-    return { percentage, resetTime, resetText };
+    return { percentage: percentage ?? 0, resetTime, resetText };
   }
 
   /**
@@ -624,7 +669,7 @@ export class ClaudeUsageService {
     }
 
     // Try to parse simple time-only format: "Resets 11am" or "Resets 3pm"
-    const simpleTimeMatch = text.match(/resets\s+(\d{1,2})(?::(\d{2}))?\s*(am|pm)/i);
+    const simpleTimeMatch = text.match(/resets\s*(\d{1,2})(?::(\d{2}))?\s*(am|pm)/i);
     if (simpleTimeMatch) {
       let hours = parseInt(simpleTimeMatch[1], 10);
       const minutes = simpleTimeMatch[2] ? parseInt(simpleTimeMatch[2], 10) : 0;
@@ -649,8 +694,11 @@ export class ClaudeUsageService {
     }
 
     // Try to parse date format: "Resets Dec 22 at 8pm" or "Resets Jan 15, 3:30pm"
+    // The regex explicitly matches only valid 3-letter month abbreviations to avoid
+    // matching words like "Resets" when there's no space separator.
+    // Optional "resets\s*" prefix handles cases with or without space after "Resets"
     const dateMatch = text.match(
-      /([A-Za-z]{3,})\s+(\d{1,2})(?:\s+at\s+|\s*,?\s*)(\d{1,2})(?::(\d{2}))?\s*(am|pm)/i
+      /(?:resets\s*)?(Jan|Feb|Mar|Apr|May|Jun|Jul|Aug|Sep|Oct|Nov|Dec)\s+(\d{1,2})(?:\s+at\s+|\s*,?\s*)(\d{1,2})(?::(\d{2}))?\s*(am|pm)/i
     );
     if (dateMatch) {
       const monthName = dateMatch[1];

apps/server/tests/unit/services/claude-usage-service.test.ts

@@ -124,6 +124,59 @@ describe('claude-usage-service.ts', () => {
 
       expect(result).toBe('Plain text');
     });
+
+    it('should strip OSC sequences (window title, etc.)', () => {
+      const service = new ClaudeUsageService();
+      // OSC sequence to set window title: ESC ] 0 ; title BEL
+      const input = '\x1B]0;Claude Code\x07Regular text';
+      // @ts-expect-error - accessing private method for testing
+      const result = service.stripAnsiCodes(input);
+
+      expect(result).toBe('Regular text');
+    });
+
+    it('should strip DEC private mode sequences', () => {
+      const service = new ClaudeUsageService();
+      // DEC private mode sequences like ESC[?2026h and ESC[?2026l
+      const input = '\x1B[?2026lClaude Code\x1B[?2026h more text';
+      // @ts-expect-error - accessing private method for testing
+      const result = service.stripAnsiCodes(input);
+
+      expect(result).toBe('Claude Code more text');
+    });
+
+    it('should handle complex terminal output with mixed escape sequences', () => {
+      const service = new ClaudeUsageService();
+      // Simulate the garbled output seen in the bug: "[?2026l ]0;❇ Claude Code [?2026h"
+      // This contains OSC (set title) and DEC private mode sequences
+      const input =
+        '\x1B[?2026l\x1B]0;❇ Claude Code\x07\x1B[?2026hCurrent session 0%used Resets3am';
+      // @ts-expect-error - accessing private method for testing
+      const result = service.stripAnsiCodes(input);
+
+      expect(result).toBe('Current session 0%used Resets3am');
+    });
+
+    it('should strip single character escape sequences', () => {
+      const service = new ClaudeUsageService();
+      // ESC c is the reset terminal command
+      const input = '\x1BcReset text';
+      // @ts-expect-error - accessing private method for testing
+      const result = service.stripAnsiCodes(input);
+
+      expect(result).toBe('Reset text');
+    });
+
+    it('should remove control characters but preserve newlines and tabs', () => {
+      const service = new ClaudeUsageService();
+      // BEL character (\x07) should be stripped, but the word "Bell" is regular text
+      const input = 'Line 1\nLine 2\tTabbed\x07 with bell';
+      // @ts-expect-error - accessing private method for testing
+      const result = service.stripAnsiCodes(input);
+
+      // BEL is stripped, newlines and tabs preserved
+      expect(result).toBe('Line 1\nLine 2\tTabbed with bell');
+    });
   });
 
   describe('parseResetTime', () => {