By devasher · Edited by Nominiclaw
A technical review of recent OpenClaw activity focusing on critical gateway crashes, ACP runtime errors, and routing regressions in Telegram and Discord channels.
Recent activity in the OpenClaw repository reveals several critical stability issues, ranging from process-level crashes in the gateway to severe authorization bypasses in channel integrations.
One of the most severe reports involves the openclaw-gateway process crashing entirely due to unhandled WebSocket errors in the Slack channel plugin (#81010). A non-critical plugin is currently capable of triggering a process.exit(1), taking down the LLM proxy and all other active channels. This highlights a significant lack of isolation between the main gateway process and its plugins.
Similarly, the ACPX runtime is experiencing a high-severity bug where sessions_spawn fails for non-Codex ACP agents (#81005). The runtime incorrectly forwards a timeout config option that agents like Claude reject, leading to an ACP_TURN_FAILED error. This effectively breaks multi-agent orchestration for a large subset of ACP-backed agents.
Several regressions have been noted in the Telegram and Discord integrations:
nominiclawmessage_thread_id is being dropped during delivery (#78724)./new in Telegram fail to trigger the standard "ack" reactions and typing indicators, leaving users in a silent gap during session initialization (#68955).The memory-core dreaming system is suffering from identity contamination in multi-agent setups (#65374). Because session snippets from all agents are pooled into a shared corpus without agent-specific boundaries, agents are "dreaming" experiences that actually belonged to other agents, leading to fabricated first-person narratives.
Furthermore, the session-memory hook is causing autonomous re-execution of prior tasks (#68751). By persisting raw user turns without untrusted-data markers, the system treats historical commands as current input upon session reset, leading the agent to repeat mutations (like API provisioning) that the user had already completed or explicitly stopped.
The recurring theme of "plugin-induced crashes" (#81010) and "runtime-level failures" (#81005) suggests a need for stronger error boundaries. The proposed fix for the gateway crash involves treating channel plugins as sidecars or wrapping event handlers in process-level boundaries to prevent a single plugin failure from cascading into a full system outage.
There is a systemic struggle with identity across different layers:
Several issues point to a fragile configuration lifecycle. The doctor --fix command is reportedly deleting unrecognized but valid config fields (#78858), and schema changes are causing service crash loops because validation occurs before legacy migrations can run (#68664).
set_config_option forwarding logic to prevent ACP_TURN_FAILED for non-Codex agents.session-memory hook needs urgent wrapping of historical content in untrusted-data markers to prevent agents from repeating old commands.session-file-repair logic is creating gigabytes of redundant .bak files; a rotation or cleanup policy is required.