The new version of OpenClaw optimizes sub-agent flow and certificate rotation, and fixes the Gemini context truncation bug.

According to Beating's monitoring, the personal AI agent project OpenClaw has officially released version v2026.5.19. This incremental update focuses on multi-model fault tolerance, secure mobile communication, and lifecycle synchronization of sub-agents. Regarding model ecosystem adaptation, the system has specifically fixed a serious interruption defect for Google Gemini. The new version will proactively discard Gemini thought signatures truncated by the system's compression mechanism before dialogue replay, preventing malformed Base64 data from causing the assistant's next response to terminate directly. For fault tolerance of domestic large models, the system has reclassified the HTTP 429 payload thrown when Moonshot (Kimi) runs out of balance. These errors are no longer crudely treated as general rate limits, but are accurately identified as billing blocks, thereby triggering correct billing guidance and backup model rollback. At the cross-device communication layer, the Android client introduces a smooth TLS certificate rotation mechanism. When the gateway TLS fingerprint changes, the system will display the old and new SHA-256 fingerprints for the user to verify and allow passage, completely replacing the previous hard crash caused by fixed public keys. The underlying sub-agent workflow has also undergone rigorous race condition fixes. The new version mandates that handover signals in the queue must be fully incorporated into the parent context before the system marks them as declared. This change, combined with the parent session's timeout protection mechanism, fundamentally eliminates the risk of losing sub-level execution results under extremely busy conditions. Furthermore, browser plugins now prioritize Chrome CDP-managed diagnostics, eliminating false alarms and failures caused by cold-start HTTP probe contention; WhatsApp channels have fully enabled mandatory document-level delivery of original images and videos. From eliminating cross-generational concurrency overhead to accurately identifying vendor error codes, OpenClaw is gradually solidifying the scheduling stability of multi-level agent systems under extremely complex networks and long sessions by refining the subtle boundaries of runtime.