openclaw - 💡(How to fix) Fix [Bug]: Two silent message-drop modes in v2026.5.7: inbound-during-busy-run and outbound-without-telegram-send [2 comments, 3 participants]

Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

In v2026.5.7 with Telegram long-polling, two distinct silent drop modes observed: (1) a message arriving during an active agent run is phantom-completed and never delivered to the agent; (2) a completed agent turn occasionally fails to dispatch its reply to Telegram with no retry.

Steps to reproduce

Issue #1 (Inbound drop during active run):

1. Start OpenClaw 2026.5.7 with Telegram long-polling enabled.
2. Send a message that triggers a long agent turn (>5 seconds).
3. While the agent is still processing, send a second message from the same user.
4. Observe: second message is never processed, agent never replies.

Issue #2 (Outbound silent drop):

1. Start OpenClaw 2026.5.7 with Telegram long-polling enabled.
2. Send a message that triggers a long agent turn.
3. Wait for turn to complete (outcome=completed in logs).
4. Observe: no message appears in Telegram despite successful turn completion.

Expected behavior

Issue #1: New messages should be queued while agent is busy and processed after the current turn completes (or user notified of the delay).

Issue #2: The outbound response should be delivered to Telegram. On failure, gateway should retry with backoff, or at minimum log the payload to allow manual recovery.

Actual behavior

Issue #1: Log shows queue message failed: reason=no_active_run followed by [turn] outcome=completed duration=385ms (phantom completion in ~385ms, physically impossible for a real turn). Agent never sees the message, never replies.

Issue #2: Log shows telegram/silent-reply-dispatch: queuedFinal=false sentFallback=false and [telegram] turn ended without visible final response. No telegram sendMessage ok line. No retry. Response exists in agent transcript but user receives nothing.

OpenClaw version

2026.5.7

Operating system

Debian-based Linux (Docker container)

Install method

docker (ghcr.io/openclaw/openclaw:latest, custom image with @anthropic-ai/claude-code added)

Model

anthropic/claude-sonnet-4-6

Provider / routing chain

openclaw -> anthropic (agentRuntime: claude-cli, persistent OAuth credentials)

Additional provider/model setup details

No response

Logs, screenshots, and evidence

**Issue #1 log signature** (from `/tmp/openclaw/openclaw-2026-05-10.log`):

[gateway] inbound: tg msg <id> from <user>
queue message failed: reason=no_active_run
[turn] outcome=completed duration=385ms


**Issue #2 log signature**:

[turn] outcome=completed
telegram/silent-reply-dispatch: queuedFinal=false sentFallback=false
[telegram] turn ended without visible final response

Impact and severity

Affected: All users communicating via Telegram channel in active conversations. Severity: Issue #1 medium-high (frequent in any active multi-message conversation); Issue #2 high (irreversible loss of agent reply with no indication of failure). Frequency: Issue #1 reproduced at least twice in one session; Issue #2 at least once. Consequence: Silent UX failures — user sends a message or waits for a reply and gets nothing. Both agent and system believe everything succeeded.

Additional information

Network: HTTP proxy via HTTPS_PROXY=http://192.168.2.1:7890 (upstream VPN). No direct network issues observed independently. Workaround for Issue #1: training users not to send rapid follow-ups during long turns. Issue #2 has no workaround — when triggered, requires manual log inspection and re-send via UI. Switching to a different channel (Discord, Signal) likely does not help — same gateway code paths.