Strip lone surrogates from user input before SDK submission

.claude/CLAUDE.md

@@ -72,8 +72,8 @@ Only update the `Status` field — do not modify any other frontmatter or prompt
 
 <!-- BEGIN:REPO:current-state -->
 ## Current State
-Branch: `refactor/adopt-string-width`
-In-progress: Nothing. PR #139 (refactor #137, adopt string-width) open, auto-merge enabled.
+Branch: `fix/lone-surrogate-sanitisation`
+In-progress: Nothing. PR #143 (fix #141, lone surrogate sanitisation) open, auto-merge enabled.
 <!-- END:REPO:current-state -->
 
 <!-- BEGIN:REPO:architecture -->

.claude/sessions/2026-03-28.md

@@ -15,3 +15,11 @@
 - Decisions: Label `enhancement` chosen as adoption of a library is an improvement, not a pure bug fix.
 - Next: Await PR #139 auto-merge.
 - Violations: None
+
+### 22:42 - fix/lone-surrogate-sanitisation (#141)
+
+- Did: Extracted `sanitiseLoneSurrogates` to `src/sanitise.ts`, applied at the top of `submit()` in `ClaudeCli.ts`, added warning log when replacement occurs, added 4 unit tests. Created PR #143 with Closes #141, auto-merge enabled.
+- Files: `src/ClaudeCli.ts`, `src/sanitise.ts`, `test/sanitise.spec.ts`
+- Decisions: Sanitisation extracted to a separate module so it can be unit tested. Warning uses `this.term.info` with yellow ANSI, matching the existing config warning pattern.
+- Next: Await PR #143 auto-merge.
+- Violations: None

src/ClaudeCli.ts

@@ -31,6 +31,7 @@ import { UsageProvider } from './providers/UsageProvider.js';
 import { SdkResult } from './SdkResult.js';
 import { SessionManager } from './SessionManager.js';
 import { SystemPromptBuilder } from './SystemPromptBuilder.js';
+import { sanitiseLoneSurrogates } from './sanitise.js';
 import { QuerySession } from './session.js';
 import { Terminal } from './terminal.js';
 import { type ContextUsage, readLastTodoWrite, type TodoItem, UsageTracker } from './UsageTracker.js';
@@ -293,7 +294,11 @@ export class ClaudeCli {
   }
 
   private async submit(override?: string): Promise<void> {
-    const text = override ?? getText(this.editor);
+    const rawText = override ?? getText(this.editor);
+    const text = sanitiseLoneSurrogates(rawText);
+    if (text !== rawText) {
+      this.term.info('\x1b[33m[warning] Input contained lone surrogates; replaced with \uFFFD\x1b[0m');
+    }
     if (!text.trim()) {
       return;
     }

src/sanitise.ts

@@ -0,0 +1,5 @@
+const LONE_SURROGATE_RE = /[\uD800-\uDBFF](?![\uDC00-\uDFFF])|(?<![\uD800-\uDBFF])[\uDC00-\uDFFF]/g;
+
+export function sanitiseLoneSurrogates(s: string): string {
+  return s.replace(LONE_SURROGATE_RE, '\uFFFD');
+}

test/sanitise.spec.ts

@@ -0,0 +1,28 @@
+import { describe, expect, it } from 'vitest';
+import { sanitiseLoneSurrogates } from '../src/sanitise.js';
+
+describe('sanitiseLoneSurrogates', () => {
+  it('replaces lone high surrogate', () => {
+    const actual = sanitiseLoneSurrogates('hello \uD83C world');
+    const expected = 'hello \uFFFD world';
+    expect(actual).toBe(expected);
+  });
+
+  it('replaces lone low surrogate', () => {
+    const actual = sanitiseLoneSurrogates('hello \uDF4C world');
+    const expected = 'hello \uFFFD world';
+    expect(actual).toBe(expected);
+  });
+
+  it('preserves valid surrogate pair', () => {
+    const actual = sanitiseLoneSurrogates('hello \uD83C\uDF4C world');
+    const expected = 'hello \uD83C\uDF4C world';
+    expect(actual).toBe(expected);
+  });
+
+  it('replaces lone high surrogate but preserves valid pair', () => {
+    const actual = sanitiseLoneSurrogates('\uD83C\uDF4C test \uD83C');
+    const expected = '\uD83C\uDF4C test \uFFFD';
+    expect(actual).toBe(expected);
+  });
+});