Bug Report

Reported via: OpenClaw (automated issue via agent)

Summary

When the OpenClaw gateway is restarted, the Feishu channel re-delivers the most recently unacknowledged message, causing it to be processed again by the agent. The deduplication logic does not consistently prevent this.

Steps to Reproduce

1. Have a conversation with the OpenClaw agent via Feishu DM
2. Restart the gateway (openclaw gateway restart)
3. Observe that the last message from the previous session is re-delivered and processed again

Expected Behavior

After gateway restart, previously processed messages should be reliably deduplicated and not re-processed.

Actual Behavior

The last message is re-delivered by Feishu (at-least-once delivery) and processed again as if it were a new message. This was observed multiple times in the same session today.

Log Evidence

Gateway log shows deduplication sometimes works:

feishu: skipping duplicate message om_x100b55666a33c428c3b3791789633b8

But in other restart cycles, the same message was processed again without being skipped.

Environment

• OpenClaw version: 2026.2.26
• Channel: Feishu (DM)
• OS: macOS (Darwin 24.6.0 arm64)
• Node: v25.6.1
• Gateway mode: local (LaunchAgent)

Additional Context

Feishu uses an at-least-once delivery model for webhooks/long-polling. On gateway reconnect, Feishu re-pushes recent unacknowledged messages. The deduplication window or message ID tracking may not persist across gateway restarts, causing the duplicate processing.

A fix could involve persisting the set of recently processed message IDs to disk (or a short-lived cache) so they survive gateway restarts.