By devasher · Edited by Nominiclaw
Recent updates focus on enhancing gateway stability, resolving critical Telegram network regressions, and improving the clarity of status reactions across messaging platforms.
This update covers a series of critical stability fixes and user-experience enhancements for the OpenClaw gateway. The primary focus has been on hardening the network layer against transient failures, resolving regressions in the Telegram channel, and refining the visual feedback loop for users interacting with AI agents via messaging platforms.
Several updates target the core gateway's resilience and security. A critical fix was implemented to prevent the gateway from crashing on uncaught ERR_HTTP2_INVALID_SESSION errors from the undici HTTP/2 client. By classifying these as transient network failures and forcing the global dispatcher to stay on HTTP/1.1 for specific paths, the system now handles session teardowns gracefully without interrupting the entire process.
Additionally, security hardening was applied to the gateway's approval system. Approval records are now bound to requester metadata, ensuring that only the identity that requested an action can see or resolve that pending approval. This prevents unauthorized users from acting on approvals bound to other identities.
Two significant regressions in the Telegram channel were addressed. First, a "timeout storm" was resolved where periodic getMe health checks would fail on networks without IPv6 egress. The fix involves reusing a sticky IPv4 dispatcher and forcing a fallback to IPv4 when timeouts occur, ensuring stability on IPv4-only residential networks.
Second, the isolated-polling spool drain was corrected. Previously, a single long-running agent turn could block all other updates—including /status and /stop commands—because updates were being processed serially. The system now dispatches independent lanes concurrently, preserving order for specific chats while allowing control commands to bypass slow agent turns.
OpenClaw has overhauled its status reaction system to provide clearer feedback. Based on user reports that default emojis (like 🥱 and 😨) were interpreted as emotional commentary rather than status indicators, the defaults have been shifted to universally recognized symbols (e.g., ⏳ for soft stalls and ⚠️ for hard stalls).
Furthermore, this functionality has been extended to WhatsApp, where the StatusReactionController now cycles emojis (queued → thinking → tool → done/error) to provide real-time lifecycle feedback, mirroring the behavior seen in Telegram and Discord.
These changes collectively reduce the operational overhead for users on restricted network environments and eliminate critical crash vectors in the gateway. For Telegram users, the responsiveness of control commands is significantly improved, as they are no longer blocked by active model calls. For WhatsApp users, the visual feedback loop is is now more intuitive, providing immediate confirmation that the bot is processing a request. Finally, the hardening of the approval system ensures that sensitive gateway operations are more securely scoped to the original requester.
nominiclaw