OpenClaw 2026.6.1-beta.3: Skill Workshop and Workboard prerelease
Third beta of the 2026.6.1 line, published with the aggregate 2026.6.1 release notes rather than beta-specific ones. The line adds Skill Workshop (a guarded propose/review/rollback flow for skills), Workboard multi-agent orchestration primitives, SQLite-backed plugin and channel state, and MiniMax M3 model support.
PUBLISHED2026-06-03
OBSERVED2026-08-11
AGE2mo
SOURCES1
- Skills: Skill Workshop with pending proposals, CLI/Gateway review actions, rollback metadata and the
skill_workshopagent tool; proposals can carry approved support files with scanner/hash/rollback safeguards and be revised in place before approval - Control UI: Skill Workshop dashboard, proposal view, revision dialog, file preview modal; Dreaming-tab agent selector (#78748); calmer composer controls, local draft state, first-output latency instrumentation
- Plugins: Tokenjuice and the GitHub Copilot agent runtime externalised as
@openclaw/tokenjuiceand@openclaw/copilot; SecretRef provider integration manifest contract (#82326); install index persisted in SQLite (#88794); typed presentation command actions and a boundedresolve_exec_envhook (#88721) - Workboard: orchestration primitives and agent coordination tools, task-backed board runs (#87469)
- Providers: MiniMax M3 model support (#88860)
- Channels: inbound queues and iMessage monitor state moved to SQLite (#88797)
- Gateway: Tailscale Serve service names for local service routing
- Code mode: internal namespaces for scoped agent/global sessions, exact namespace tool dispatch (#88043)
- Mobile: iOS hosted push relay defaults, realtime Talk playback, guarded WebSocket ping, native iPad layouts; Android installed-app inspection and notification picker helpers
- Doctor: disk space health checks
- Fixes across interrupted tool calls, stale session bindings, compaction handoffs, auth-profile failover, WhatsApp/iMessage/Discord/QQBot delivery, and Gateway node drain revisions
COMMUNITY
No curated reactions recorded for this event. Facts and takes are kept in separate layers — community context is added by hand, never blended into the record above.