Problem

When the Matrix homeserver returns M_LIMIT_EXCEEDED with retry_after_ms, OpenClaw does not retry the send and the message is permanently lost.

Reproduction

1. Self-hosted Synapse with default rate limiting
2. Agent processes a long-running request (10+ minutes due to API timeouts/retries)
3. Agent generates reply and attempts to send
4. Synapse returns M_LIMIT_EXCEEDED with retry_after_ms: 1120 (1.1 seconds)
5. OpenClaw logs [matrix] final reply failed: Error: M_LIMIT_EXCEEDED: Too Many Requests and drops the message

Observed logs

2026-03-05T08:56:31.667Z [diagnostic] lane wait exceeded: lane=session:agent:code:matrix:direct:@vitaliy:m.neomentor.tech waitedMs=743990 queueAhead=0
2026-03-05T11:56:36.733+03:00 MatrixHttpClient (REQ-2382) {
  errcode: 'M_LIMIT_EXCEEDED',
  error: 'Too Many Requests',
  retry_after_ms: 1120
}
2026-03-05T08:56:36.746Z [matrix] final reply failed: Error: M_LIMIT_EXCEEDED: Too Many Requests

Expected behavior

When the homeserver returns M_LIMIT_EXCEEDED with retry_after_ms, OpenClaw should wait the specified time and retry (at least once). The retry_after_ms value is typically small (1-2 seconds), so this would not noticeably delay delivery.

Environment

• OpenClaw: 2026.3.2
• Matrix plugin: @openclaw/matrix 2026.3.2
• Homeserver: Synapse (self-hosted)
• E2EE: enabled

Related

• Plugin/extension channel delivery failures are silent and unrecoverable #29126 (general plugin delivery failures are silent/unrecoverable)
• The built-in Telegram channel likely handles rate limits internally; Matrix should follow the same pattern.