Bug Report: Internal Metadata Leaking to AI Context

Summary

OpenClaw is passing internal routing metadata to the AI assistant context that should be filtered out for privacy/security reasons.

Environment

• OpenClaw Version: 2026.4.1 (after update from earlier version)
• OS: macOS (Darwin 24.6.0 arm64)
• Interfaces Tested: TUI (openclaw-tui), Telegram

Issue Details

The AI assistant receives JSON metadata blocks containing internal routing information that should not be visible in the AI context:

TUI Interface Leak

{
  "label": "openclaw-tui",
  "id": "openclaw-tui", 
  "name": "openclaw-tui",
  "username": "openclaw-tui"
}

Telegram Interface Leak

{
  "label": "Amanda Schwartz Ramirez (1906591603)",
  "id": "1906591603",
  "name": "Amanda Schwartz Ramirez", 
  "username": "djbabypanda"
}

Expected Behavior

• Internal routing metadata should be stripped before reaching AI context
• AI should only receive the actual user message content
• User identity/routing info should be handled internally

Security/Privacy Concerns

• Real names, usernames, and IDs are exposed in AI context
• Internal system architecture details visible to AI
• Potential for sensitive routing information to leak

Reproduction

1. Send any message through TUI or Telegram interface
2. AI receives the message with JSON metadata headers
3. User cannot see the metadata (display issue is separate)
4. Persists after gateway restart and system update

Additional Notes

• User cannot see the metadata leak in their interface
• Issue affects multiple communication channels
• Metadata format suggests it's added at the message processing pipeline level