Fix characters (#372)

* Fix cat command

* Fix text rendering crash from undefined code points in bash output

* Revert unintentional model parameter changes from fix cat command commit

packages/coding-agent/src/modes/interactive/components/tool-execution.ts

@@ -13,6 +13,7 @@ import {
 import stripAnsi from "strip-ansi";
 import type { CustomTool } from "../../../core/custom-tools/types.js";
 import { DEFAULT_MAX_BYTES, DEFAULT_MAX_LINES, formatSize } from "../../../core/tools/truncate.js";
+import { sanitizeBinaryOutput } from "../../../utils/shell.js";
 import { getLanguageFromPath, highlightCode, theme } from "../theme/theme.js";
 import { renderDiff } from "./diff.js";
 import { truncateToVisualLines } from "./visual-truncate.js";
@@ -295,10 +296,8 @@ export class ToolExecutionComponent extends Container {
 
 		let output = textBlocks
 			.map((c: any) => {
-				let text = stripAnsi(c.text || "").replace(/\r/g, "");
-				text = text.replace(/\x1b./g, "");
-				text = text.replace(/[\x00-\x08\x0b\x0c\x0e-\x1f\x7f-\x9f]/g, "");
-				return text;
+				// Use sanitizeBinaryOutput to handle binary data that crashes string-width
+				return sanitizeBinaryOutput(stripAnsi(c.text || "")).replace(/\r/g, "");
 			})
 			.join("\n");
 

packages/coding-agent/src/utils/shell.ts

@@ -100,13 +100,38 @@ export function getShellConfig(): { shell: string; args: string[] } {
  * - Control characters (except tab, newline, carriage return)
  * - Lone surrogates
  * - Unicode Format characters (crash string-width due to a bug)
+ * - Characters with undefined code points
  */
 export function sanitizeBinaryOutput(str: string): string {
-	// Fast path: use regex to remove problematic characters
-	// - \p{Format}: Unicode format chars like \u0601 that crash string-width
-	// - \p{Surrogate}: Lone surrogates from invalid UTF-8
-	// - Control chars except \t \n \r
-	return str.replace(/[\p{Format}\p{Surrogate}]/gu, "").replace(/[\x00-\x08\x0B\x0C\x0E-\x1F]/g, "");
+	// Use Array.from to properly iterate over code points (not code units)
+	// This handles surrogate pairs correctly and catches edge cases where
+	// codePointAt() might return undefined
+	return Array.from(str)
+		.filter((char) => {
+			// Filter out characters that cause string-width to crash
+			// This includes:
+			// - Unicode format characters
+			// - Lone surrogates (already filtered by Array.from)
+			// - Control chars except \t \n \r
+			// - Characters with undefined code points
+
+			const code = char.codePointAt(0);
+
+			// Skip if code point is undefined (edge case with invalid strings)
+			if (code === undefined) return false;
+
+			// Allow tab, newline, carriage return
+			if (code === 0x09 || code === 0x0a || code === 0x0d) return true;
+
+			// Filter out control characters (0x00-0x1F, except 0x09, 0x0a, 0x0x0d)
+			if (code <= 0x1f) return false;
+
+			// Filter out Unicode format characters
+			if (code >= 0xfff9 && code <= 0xfffb) return false;
+
+			return true;
+		})
+		.join("");
 }
 
 /**

packages/tui/src/utils.ts

@@ -4,6 +4,7 @@ import stringWidth from "string-width";
  * Calculate the visible width of a string in terminal columns.
  */
 export function visibleWidth(str: string): number {
+	if (!str) return 0;
 	const normalized = str.replace(/\t/g, "   ");
 	return stringWidth(normalized);
 }
@@ -472,6 +473,9 @@ function breakLongWord(word: string, width: number, tracker: AnsiCodeTracker): s
 		}
 
 		const grapheme = seg.value;
+		// Skip empty graphemes to avoid issues with string-width calculation
+		if (!grapheme) continue;
+
 		const graphemeWidth = visibleWidth(grapheme);
 
 		if (currentWidth + graphemeWidth > width) {
@@ -576,6 +580,9 @@ export function truncateToWidth(text: string, maxWidth: number, ellipsis: string
 		}
 
 		const grapheme = seg.value;
+		// Skip empty graphemes to avoid issues with string-width calculation
+		if (!grapheme) continue;
+
 		const graphemeWidth = visibleWidth(grapheme);
 
 		if (currentWidth + graphemeWidth > targetWidth) {