By devasher · Edited by Nominiclaw
This update introduces advanced QA automation for Telegram, real-time voice diagnostics for Discord, and critical fixes for agent process persistence and model compatibility.
The v2026.5.10-beta.2 release focuses on enhancing the reliability of agent-tool interactions, expanding the capabilities of voice and chat integrations, and hardening the security of the Gateway's logging and configuration systems. Key highlights include new automation tools for Telegram QA, real-time voice diagnostics for Discord, and a significant fix for subagent process persistence during context compaction.
nominiclawopusscript decoder while providing an opt-in native @discordjs/opus install for high-performance lanes.action: "reply") now support attachments via the imsg send-rich --file flag, provided the installed imsg build supports it. The system now automatically probes for this capability during startup.group_topic sessions were posting to the group root; they are now correctly auto-threaded within the topic.exec processes during context compaction. Active background sessions are now preserved across compaction and after-turn runtime contexts (#79284).talk.realtime.instructions while maintaining built-in agent-consult guidance (#79081).openrouter/openai/gpt-5.4-mini).vscode-chat, ensuring image-capable models are correctly routed and not restricted to the copilot-language-server scope (#79946)./think levels (low/medium/high) for reasoning-capable Grok models.skills.install.allowUploadedArchives (#74430).config set and config patch would silently discard values that matched runtime defaults (#79856).exec tool failure instead of reporting the error (#60497).thread/start responses were rejected if they omitted sessionId (#80124).contextWindow (#79428).exec.approval.list do not trigger heavy dependency loads (#76943).imsg that supports the --file flag (refer to openclaw/imsg#114).4000 token entries to a safer default of 128000 tokens.