Read receipts

Set native.readPolicy to choose when a chat is marked read:

readPolicy	Marks the chat read
on_inbound (default)	After OpenClaw has handled the message: it passed the access checks and the agent ran, or a built-in command answered it
never	Never. Chats stay unread on your phone.
on_model_dispatch	Only after confirming the model actually received the message. Stricter, and needs extra setup (below).

Marking read covers the whole chat, up to the newest message. WeChat is told, so your phone and other devices can clear the unread badge too.

"native": {
  "readPolicy": "on_inbound"
}

Strict mode

on_model_dispatch waits for proof that the exact message, and any images, reached the model. It needs this permission:

"plugins": {
  "entries": {
    "agent-wechat": { "hooks": { "allowConversationAccess": true } }
  }
}

Without it, messages stay unread and the channel status shows handoffUnconfirmed. Some providers’ history formats prevent confirmation even when the agent replied. For a normal bot, use on_inbound.

Good to know

• Ignored messages stay unread, along with messages whose media is still downloading or whose handling failed.
• Earlier messages are included. Marking a chat read also clears older unread messages the agent never saw.
• Read doesn’t mean understood. A chat can be marked read after the agent handled a file it couldn’t open.
• No double notifications. If WeChat’s answer to a read notice is lost, it’s marked needs_review and not resent. Pending reads survive a Gateway restart.
• The token needs the read-state scope unless readPolicy is never.