According to Inquisitive Beating monitoring, the open-source autonomous agent framework OpenClaw has released version v2026.6.1. Addressing the issues of the previous version, which lacked a skill management portal and the agent's inability to persist experience, the new version introduces the "Skill Workshop" proposal mechanism for the first time. The agent can no longer directly modify local skill files; instead, it must first submit a proposal file (PROPOSAL.md) containing execution steps and metadata. Users must review the proposal in the control interface or command line (supporting approval, rejection, or isolation) before it takes effect. The workshop includes a static scanner, hash locking to prevent editing conflicts, and version rollback metadata, providing a secure governance foundation for the agent to accumulate reusable experiences.
In addition to the workshop, the new version has upgraded various core orchestrations and multi-platform support. The Windows platform has officially launched a signed Windows Hub native desktop client (supporting x64 and ARM64 architectures), with support for system tray management and WSL clipboard pass-through, along with a convenient non-administrator installation process. The Workboard has complemented multi-agent planning capabilities, supporting task board-level execution tracking; the built-in output compression tool Tokenjuice and GitHub Copilot have been decoupled as official standalone plugins to streamline the core. To reduce redundant filesystem scans during restarts, iMessage listening status, plugin installation index, and inbound queue have been migrated to an SQLite database. The iOS mobile end has added managed push default configuration and real-time voice playback, with support for iPad native layout. In terms of model adaptation, native support for MiniMax M3 has been added.
