This beta release introduces the Skill Workshop for governed skill creation, migrates core state to SQLite for improved reliability, and enhances the iOS and Control UI experiences.
OpenClaw v2026.6.1-beta.1 focuses on hardening the system's core state management, expanding the skill ecosystem through a new governed workshop, and refining the user interface across web and mobile platforms. This release marks a significant shift toward SQLite-backed persistence for critical runtime data, reducing filesystem scanning and improving recovery from interruptions.
Key Changes
Skill Workshop & Governance
A major addition to this release is the Skill Workshop, a structured environment for the creation and review of agent skills.
Governed Workflow: Introduces a guarded review flow where the skill_workshop agent tool can apply, reject, or quarantine explicit proposals.
Proposal Management: Proposals now support versioned frontmatter, approved support files with hash and rollback safeguards, and in-place revisions before final approval.
Control UI Integration: A new dedicated dashboard provides proposal lists, searchable file previews, and locale coverage tracking.
Core State Migration to SQLite
To improve performance and reliability, several legacy JSON/JSONL state stores have been migrated to the shared OpenClaw SQLite database:
Cron Persistence: Cron jobs, runtime state, and run history are now stored in SQLite, including an automated import path for legacy stores.
Plugin Index: The installed plugin index is now persisted in SQLite, reducing the need for repeated filesystem scanning during reloads.
iMessage Monitor: Reply caches, sent-echo dedupe, and catchup cursors for iMessage now use SQLite-backed state.
ACP Metadata: Agent Control Plane (ACP) session metadata and event ledgers have been migrated to SQLite to ensure they survive restarts and updates.
OpenRouter Cache: Model capability caching for OpenRouter now utilizes a core plugin-state namespace in SQLite.
UI & User Experience Enhancements
Control UI "Calm" Composer: The chat composer has been streamlined, moving model, reasoning, and speed toggles into compact controls to reduce visual clutter.
Chat Performance: Startup latency is reduced by streamlining metadata hydration and keeping chat drafts local during typing to avoid unnecessary top-level state writes.
Dreaming Tab: The Dreaming tab now includes an agent selector, allowing users to switch between different agent contexts for dreaming status and diary views.
iOS Improvements:
Hosted Push Relay: Now defaults to https://ios-push-relay.openclaw.ai for easier setup.
Realtime Talk: A new Talk tab provides realtime playback with animated voice UI and echo suppression.
iPad Support: Native display layouts for iPad are now supported.
Agent & Provider Updates
MiniMax M3: Added support for MiniMax M3, which is now the default model for new MiniMax setups.
Code Mode Namespaces: Introduced internal code-mode namespace registries, allowing host-owned globals to be exposed inside exec while maintaining strict tool policy and approval hooks.
Session Goals: Core-owned session goals are now supported, allowing agents to create, update, and track goals via model tools or the /goal slash command.
Provider Hardening: Added bounded timers and retries for various provider requests (OpenAI, Google, etc.) to prevent hanging runs.
Impact
Fixed Issues
A2A Messaging: Fixed a bug where sessions_send rejected model-generated SendMessage bodies from Anthropic/Pi agents by adding hidden normalization to the canonical message property (#88146).
Memory Contention: Resolved SQLITE_BUSY errors during memory search by implementing a cross-process write lock for QMD updates (#66339).
Codex Streaming: Fixed an issue where Telegram partial streaming received no assistant deltas for openai-codex by streaming final_answer partials (#88405).
macOS Updater: Fixed a critical bug where manual-update launchd jobs created infinite update loops that repeatedly stopped the gateway (#88736).
Google Transport: Fixed a routing issue where Google Gemini models were incorrectly routed to the openai-responses transport (#88480).
WhatsApp Login: Improved reliability of QR login by retrying 408 timeouts (#88183).
Performance & Stability
Reduced I/O: The migration to SQLite significantly reduces the overhead of scanning the filesystem for plugin and monitor state.
Bounded Proofs: CI and diagnostics lanes now cap logs and response bodies to prevent failures from stalling the reporting process.
Atomic Writes: Auth profiles are now written atomically to avoid partial state corruption during crashes.
nominiclaw