Description

In a self-chat scenario (Mac's iMessage handle matches the DM contact), the echo cache fails to detect reflected bot messages due to a Unicode replacement character (U+FFFD / �) prepended during attributedBody text extraction.

Steps to Reproduce

1. Set up iMessage channel where the Mac's Apple ID phone number matches the DM contact's number (genuine self-chat)
2. Send a message to the bot
3. Bot replies via AppleScript → message stored in chat.db with text=NULL, content only in attributedBody
4. imsg extracts text from attributedBody, which prepends U+FFFD (\xEF\xBF\xBD)
5. Echo cache stored "Hello" but reflected message arrives as "\uFFFD Hello" → no match → message processed again → infinite echo loop

Evidence

-- Bot messages have text=NULL, content in attributedBody only
SELECT ROWID, is_from_me, text IS NOT NULL as has_text, attributedBody IS NOT NULL as has_body
FROM message WHERE ROWID >= 2665;

-- is_from_me=true (bot):  has_text=0, has_body=1
-- is_from_me=false (user): has_text=1, has_body=1

The normalizeEchoTextKey() function in monitor-provider only does .replace(/\r\n?/g, "\n").trim() — it does not strip U+FFFD characters.

Proposed Fix

In normalizeEchoTextKey(), strip U+FFFD before comparison:

function normalizeEchoTextKey(text: string): string | null {
  if (!text) return null;
  const normalized = text.replace(/\uFFFD/g, '').replace(/\r\n?/g, '\n').trim();
  return normalized || null;
}

Alternatively, the inbound message text extraction in imsg could strip U+FFFD when parsing attributedBody.

Environment

• OpenClaw: 2026.4.2
• imsg: 0.4.0
• macOS (Darwin 24.3.0 arm64)
• iMessage channel with dmPolicy: pairing

Workaround Attempted

Added responsePrefix: "🐙" to iMessage config — does not help because the U+FFFD breaks text matching regardless of prefix content.