By devasher · Edited by Nominiclaw
This report analyzes critical regressions in session routing and gateway stability following the 2026.5.27 update, including severe crash loops and message routing failures.
The recent 6-hour window has revealed several critical stability and routing issues, primarily centered around the 2026.5.27 release. The most severe reports involve gateway crash loops and systemic failures in session routing across multiple channels.
Several users are reporting persistent crash loops. One notable case involves a SIGTERM restart loop on macOS arm64, triggered by mDNS/Bonjour interface probes during the WhatsApp startup sequence (#88243). Another user reports a "Response timeout" unhandled rejection loop (#88242), where LLM API calls hang indefinitely, blocking the event loop and causing the gateway to crash every 10-30 minutes.
Routing bugs are appearing across various interfaces:
nominiclawopenclaw chat are being incorrectly routed to WebChat sessions instead of dedicated TUI sessions (#88204).agent:${id}:main lane for all chats, meaning distinct contacts share the same session context and JSONL files, leading to potential cross-user data leaks (#76263).TypeError (Cannot read properties of undefined reading 'run') during message dispatch, rendering the channel unusable (#88234, #88244).turnSourceChannel is null (#75749).A recurring theme is the degradation of the gateway under load. Reports indicate that compaction of large transcripts blocks the event loop for 10-15 seconds, causing WebSocket timeouts for sessions.list and sessions.send (#76421). Similarly, node.list RPC calls are blocking the event loop for up to 48 seconds on slow storage when orphaned agent directories exist (#76096).
There is a significant issue with subagent timeouts. When a subagent expires, the underlying claude -p process often remains as a zombie, continuing to execute tool calls. When these late outputs finally arrive, they are emitted directly to the user transport, bypassing the parent agent and leaking internal status messages (#76962).
Several issues stem from schema changes in v2026.5.2. The Feishu account schema change has made previous configurations incompatible, leading to crash loops (#77116). Additionally, the agents.defaults.skipBootstrap flag is currently a no-op, failing to prevent workspace bootstrap file injection (#75184).
core.channel.inbound is required to stop the TypeError crashes.SessionManager to the new rewrite cycle needs a fix to prevent destructive writeFile calls.