By devasher · Edited by Nominiclaw
This update introduces significant performance optimizations for model listing, enhanced realtime voice control, and critical security hardening for sandboxed agents and session locks.
nominiclaw/models call has been optimized from ~20 seconds down to ~5ms by pre-warming provider auth state at gateway startup, reducing latency by approximately 4,100×.AGENTS.md and TOOLS.md), excluding persona and user files unless explicitly configured (#85283).exec control tools now align with exec-shaped hook policies by mirroring code into a command field and adding toolKind: "code_mode_exec" for better policy discrimination.agents.defaults.imageQuality preference.👍 for allow-once and 👎 for deny), mirroring WhatsApp's native behavior (#85477).agentComponents.ttlMs callback registry lifetime (up to 24 hours) to support long-running workflows (#84189).maxHoldMs (default 5 minutes), preventing gateway freezes caused by large session files (#85762).SecretRefs before persisting to disk, preventing crash-loops caused by unresolved environment variables (#46531).Users will experience significantly faster model discovery and more responsive gateway operations. The introduction of read-only skill mounts and linear-time glob matching substantially hardens the system against both accidental and adversarial modification of agent instructions and session access. For those using WhatsApp or Telegram, the interaction model is now more intuitive and reliable, particularly in group and topic-based conversations.
SOUL.md or USER.md by default, you must now explicitly add these to agents.defaults.subagentBootstrapFiles or per-agent overrides.openclaw doctor --fix to migrate existing data.👍 / 👎 or the manual /approve command.