• Matrix replies contained internal-looking reasoning text.
• Some leaked text referenced whether to respond, whether a skill applied, and whether NO_REPLY should be emitted.
• Later artifacts in group-chat context looked like stale draft/outdated assessment leakage rather than a clean current reply.
• The symptom overlapped with the same incident window as replay/dedup and Matrix reply-path instability.

TL;DR

Inspect and adjust the reply assembly and filtering logic in the OpenClaw 2026.4.2 line to prevent internal-looking reasoning text and stale draft content from being surfaced to users.

Guidance

• Review the retry path and resend/recovery logic in the Matrix reply-path to ensure that the final answer buffer is being sent instead of intermediate/internal buffers.
• Verify that the internal reasoning metadata is being properly separated from user-visible output to prevent leakage of Reasoning: text and NO_REPLY deliberation.
• Investigate the possibility of stale draft buffers being emitted on replay/resend and implement measures to prevent this, such as clearing or updating the draft buffers before sending the final reply.
• Consider adding additional logging or debugging statements to help identify the root cause of the issue and verify the effectiveness of any fixes.

Example

No specific code snippet can be provided without more information about the implementation details of the OpenClaw system. However, the fix may involve modifying the reply assembly and filtering logic to include additional checks or buffers to prevent internal-looking reasoning text and stale draft content from being surfaced.

Notes

The exact fix may depend on the specific implementation details of the OpenClaw system, which are not provided in the issue description. Additional debugging and logging may be necessary to fully understand the root cause of the issue and develop an effective fix.

Recommendation

Apply a workaround to the reply assembly and filtering logic to prevent internal-looking reasoning text and stale draft content from being surfaced to users, as the root cause of the issue is not yet fully understood and a complete fix may require additional investigation and testing.