openclaw - 💡(How to fix) Fix [Bug]: WebChat / Control UI long replies still show `...(truncated)...` after v2026.5.15 and hide final conclusions

Summary

In WebChat / Control UI, long assistant replies are still being visibly truncated and replaced by ...(truncated)..., even when the model appears to have generated the full response. This makes the final conclusion and later sections unreadable.

Why this matters

This is not just a cosmetic issue. It causes practical message loss for:

• long task handoff prompts
• multi-part implementation plans
• final conclusions placed near the bottom of the reply
• long code / diff / structured instruction outputs

The visible symptom is that the response structure appears to finish generation, but the rendered message bubble replaces the tail with ...(truncated)....

Environment

• Surface: WebChat / Control UI direct chat
• Observed starting around: v2026.5.15
• Still observed on: v2026.5.19
• Related releases/issues checked:
  • v2026.5.18 had a fix for web chat live response truncation (#83658)
  • but truncation / long-message-reader issues still appear active (#84651, #84516)

Repro

1. Open a WebChat / Control UI direct conversation.
2. Ask for a long structured response, for example:
   • long handoff prompt
   • multi-section diagnosis with detailed conclusions
   • long code block or detailed step-by-step plan
3. Let the assistant finish generating.
4. Observe that the final rendered reply bubble shows ...(truncated)... and hides part of the content.

Expected

• The full reply should remain readable in WebChat.
• If the UI intentionally collapses long replies, there should be a built-in full-message reader / expand action.
• Final conclusions should not become inaccessible.

Actual

• The message visibly ends with ...(truncated)...
• The response appears generated, but the user cannot access the full visible text in the chat bubble
• This is especially painful when the important conclusion is near the end

Notes

This feels related to recent WebChat / Control UI long-reply regressions rather than a model-side brevity issue. The problem is reproducible enough that a practical workaround is to force manual chunking, which should not be necessary for normal usage.

Related issues

• #84651 WebChat: add a full-message reader for truncated or long messages
• #84516 long agent replies silently truncated at ~1000-1100 chars
• #83658 Web chat live response unconditionally truncated at ~4,000 chars

Request

Please either:

1. fully fix long reply rendering so content is not hidden, or
2. add a reliable built-in "expand full message" reader for truncated messages, and
3. document any current server/client reply length boundary that still applies in WebChat.