Problem

When OpenClaw attempts to send a Telegram message (sendMessage, sendChatAction) and the network request fails transiently, the message is permanently lost with no retry.

From gateway.err.log:

[telegram] sendMessage failed: Network request for 'sendMessage' failed!
[telegram] final reply failed: HttpError: Network request for 'sendMessage' failed!

Over 16 days of logs, there were 9 confirmed lost outbound messages out of ~1,732 total fetch failures (most being benign long-poll reconnections). The lost messages are replies to the user that simply never arrive — no notification, no queue, no retry.

Impact

• User sends a message, gets no reply, has to ask again
• Cron job results can be silently lost
• Sub-agent completion notifications disappear
• No way to know a message was dropped unless the user notices

Suggested Behavior

1. On transient network failure (TypeError: fetch failed, 5xx, timeout), retry the send with exponential backoff (e.g., 1s, 3s, 10s — 3 attempts max)
2. If all retries fail, queue the message to a local file (e.g., pending-messages.json) for delivery on next successful connection
3. Optionally log dropped messages so the agent can detect and re-send on heartbeat

Environment

• OpenClaw running as gateway daemon on macOS
• Telegram channel via long-polling (not webhook)
• Network is generally stable (Wi-Fi, caffeinate active) but has brief transient drops (~1 per 2 days that hits an outbound send)