hermes - 💡(How to fix) Fix Feishu gateway: detect and annotate delayed/stale event callbacks [1 pull requests]

Problem

When network connectivity is lost, Feishu's event webhook retries delivery of pending messages after the connection is restored. The gateway receives these as new messages with fresh message_ids, so there's no way to distinguish them from real-time messages. This causes the agent to process and respond to old messages as if they were just sent, leading to confusing out-of-context replies.

Proposed Solution

Instead of skipping delayed messages (which may need handling), the gateway should:

1. Detect stale messages by comparing the create_time in the Feishu event payload against the current time
2. Annotate them in the conversation context with metadata like:
   • This is a delayed message
   • Approximately how long ago it was originally sent (e.g., "2 hours ago")
   • How many messages have passed since then (if trackable)
3. Let the agent decide how to respond — the agent can naturally handle delayed messages once it knows they're stale

This approach is better than silently dropping messages because:

• The user may still expect a response to that message
• The agent can intelligently decide whether to respond, ask for clarification, or acknowledge the delay
• No messages are lost

Example

Without fix:

User (10:00): "check the network"  ← delayed, arrives at 14:00
Agent (14:00): "network is fine, ping 25ms"  ← confusing, context moved on

With fix:

User (10:00): "[DELAYED MESSAGE, sent ~4 hours ago] check the network"
Agent (14:00): "这条是4小时前发的，当时网络确实有问题，现在已经修好了..."

Reproduction

1. Start Hermes gateway connected to Feishu
2. Send a message via Feishu
3. Disconnect network before gateway processes it
4. Restore network → Feishu retries the webhook → gateway processes the stale message as if it were new

Environment

• Hermes Agent v0.13.0+
• Platform: Feishu/Lark