openclaw - 💡(How to fix) Fix WebChat message tool replies are not rendered after result summary [2 pull requests]

Summary

On current main, WebChat replies sent through the internal message tool can complete successfully but never appear in the chat UI. The model calls message(action=send) with the visible reply text, the tool result succeeds, and the run ends cleanly, but no assistant response is rendered because the persisted tool result no longer carries the structured source-reply payload needed by the chat mirror path.

Repro

1. Run current main runtime.
2. Open Control UI / WebChat.
3. Send a normal prompt such as Hey.
4. Let the model respond through the message tool.

Observed

The latest run completed successfully:

• message tool call arguments included the actual visible reply:
  • message: "Hey Shakker. I’m here."
• The tool result persisted only:
  • Sent visible reply to the current webchat conversation via internal-ui.
• No sourceReply / message details were present on the tool result.
• No follow-up delivery-mirror assistant message was written to the transcript.
• The UI showed the user message but no assistant reply.

This makes the reply effectively disappear even though the model did respond.

Expected

A successful internal WebChat message tool send should preserve machine-readable reply details for downstream chat rendering, while still returning a concise model-facing summary if desired.

The transcript/chat path should receive enough structured data to mirror the visible assistant reply exactly once.

Regression Point

This appears to be caused by:

83c2e96a16 fix: summarize internal webchat message tool results

That change summarized the internal WebChat message tool result, but the summary-only result dropped the structured details that the app-server/chat mirror path depends on.

Impact

This breaks visible replies for WebChat/Control UI when the model uses the message tool as instructed by the WebChat prompt. The run looks successful in logs and transcript telemetry, but the user sees no assistant response.

Notes

A fix should preserve the short model-visible tool result content, but also keep structured details such as sourceReply, message, and any media fields on the tool result details so existing transcript/chat mirroring can work without duplicating replies.