openclaw - 💡(How to fix) Fix [Bug]: 2026.5.22 Telegram direct gets generic /new fallback after successful tool turn

Summary

On OpenClaw 2026.5.22 (a374c3a), a Telegram direct session can complete a tool-heavy/read-only turn successfully, but the user receives only the generic fallback message instead of the assistant's final reply:

⚠️ Something went wrong while processing your request. Please try again, or use /new to start a fresh session.

In the observed case, the underlying tool work had completed and the session stayed usable afterward. This looks like a turn finalization / payload delivery failure rather than a gateway-wide crash.

Environment

• OpenClaw: 2026.5.22 (a374c3a)
• Channel: Telegram direct
• Model: openai-codex/gpt-5.5
• Auth: Codex/ChatGPT subscription OAuth profile
• Runtime shown by /status: docker/all, Think xhigh
• Session at time of report: ~114k / 272k context, Compactions: 0
• Gateway uptime: ~13h

What happened

1. User asked in Telegram: “晚上 x list 有什么新消息？”

2. The assistant pulled current context and ran a read-only scan flow (signals brief + X list feed scan).

3. The tool run completed successfully. The feed output returned 25 tweets over a ~5.9h span and exited with code 0.

4. Instead of receiving the assistant's final summary, the Telegram user saw only:

   ⚠️ Something went wrong while processing your request. Please try again, or use /new to start a fresh session.

5. The next user message was processed normally, so the session/gateway was not permanently dead.

Observed tool completion evidence

The scan process completed successfully and returned normal output, including:

📡 X List Feed — market (6h window, CLI-first / API fallback)
──────────────────────────────────────────────────
  📡 25 tweets (5.9h span, 2 dupes removed)
...
⏱️  12.2s | 25 tweets | 2 dupes removed
Process exited with code 0

So the failure appears to happen after tool completion, when OpenClaw should deliver the assistant response to Telegram.

Expected behavior

If the tool turn succeeds, Telegram should receive the assistant's final answer.

If finalization fails, the user-facing error should preserve the actual cause where possible (e.g. incomplete turn, missing payload, delivery failure, timeout, tool-result/session-history mismatch) rather than collapsing into the generic /new fallback.

Actual behavior

Telegram receives the generic fallback message. The user loses the completed turn output/summary, but the session itself remains usable.

Why this may be related to existing issues

This may overlap with, but is not exactly the same as:

• #74822 — recurring Telegram Something went wrong fallback, but that report required manual gateway restart; this case recovered on the next message.
• #79637 — long/tool-heavy sessions crashing with Something went wrong after missing tool-result/session-history warnings; this case also happened after a tool-heavy turn.
• #67750 — generic /new fallback hides a more specific failure chain.
• #68076 — incomplete turn / payloads=0 surfacing an error to user.

I am filing this separately because the distinctive symptom here is: successful tool completion + no compaction + Telegram direct + generic fallback instead of final assistant payload + session remains usable.

Logs

I do not yet have a gateway log excerpt for the exact failing request. If maintainers can point to the most useful log subsystem/path for Telegram turn finalization in 2026.5.22, I can add a redacted excerpt.