OpenClaw v2026.5.7 Release Summary

OpenClaw v2026.5.7 introduces a series of targeted improvements to channel reliability, agent memory management, and administrative security. This release is particularly critical for users of WhatsApp and those utilizing high-context Anthropic models, addressing several edge cases that previously led to silent delivery failures or API rejections.

Key Changes

Channel & Integration Enhancements

• WhatsApp LID Resolution: Fixed a critical bug where proactive messages to phone numbers were creating "ghost chats" (sender-only threads) instead of reaching the recipient. The system now correctly resolves phone numbers to LIDs via the account auth directory forward mapping.
• WhatsApp Media Deduplication: Resolved an issue where captioned MEDIA: directives were sent twice—once as an empty media message and once as a captioned reply. The dispatcher now buffers interim payloads to ensure only the final captioned version is delivered.
• Discord Voice Improvements: Voice capture is now less choppy thanks to an extended post-speech silence grace period (2.5s) and the addition of voice.captureSilenceGraceMs for noisy environments. Additionally, voice-channel permissions are now audited during channels capabilities and channels status --probe.
• Telegram & Discord Routing: Fixed routing errors where provider-prefixed targets (e.g., discord:channel:<id>) were misrouted as DMs, leading to Unknown Channel failures.

Agent & Context Engine Fixes

• Anthropic Compaction Fix: Resolved a failure where compaction for Anthropic models with 1M context windows would request max_tokens exceeding the model's output limit (e.g., requesting 240k when the limit is 128k). The system now clamps reserveTokens based on the provider registry's maxTokens.
• Context Engine Cache Invalidation: Fixed a high-severity bug where the installContextEngineLoopHook cached assembled views indefinitely. The cache is now invalidated when source history shrinks (e.g., during a /reset) or when assembly fails, preventing stale pre-reset history from leaking into new sessions.
• Subagent Retention: Completed session-mode subagent registry rows now honor the agents.defaults.subagents.archiveAfterMinutes configuration instead of a hardcoded 5-minute TTL, ensuring consistency across spawn modes.

Security & Administrative Controls

• Global Memory Guardrails: The /active-memory on --global and /active-memory off --global commands now strictly require operator.admin scope to prevent unauthorized gateway-wide configuration changes.
• Native Command Enforcement: Native slash commands now honor owner enforcement. If a plugin enforces owner-only commands, non-owner authorized senders will be rejected for native commands like /stop or /subagents spawn.
• Tool Dispatch Gating: Inline skill tool dispatch is now gated through before-tool-call authorization hooks, ensuring that security policies are applied consistently across all tool invocation surfaces.

CLI & Developer Experience

• channels list Refactor: The openclaw channels list command has been streamlined. Model-provider auth and usage snapshots have been moved to openclaw models auth list and openclaw status / openclaw models list respectively. A new --all flag now allows discovery of all bundled and catalog channels, including those installed but not yet configured.
• Cron JSON Output: cron list --json and cron show --json now include a computed status field (disabled, running, ok, error, skipped, idle), removing the need for external tooling to reimplement status derivation logic.
• Plugin Setup: Fixed a regression where non-bundled external plugin setup entries were dropping setChannelRuntime, which caused runtime initialization timeouts for plugins like WeChat.

Impact

This release significantly improves the reliability of outbound messaging on WhatsApp and the stability of long-term agent sessions. By fixing the context engine cache, users will no longer experience "memory leaks" where agents remember conversations from before a session reset. The security updates ensure that administrative functions are properly gated, reducing the risk of accidental or unauthorized global configuration changes.

Upgrade Guide

CLI Tooling Migration

If you use scripts or external tools to parse openclaw channels list --json, please note the following breaking changes:

• Auth & Usage: The auth and usage fields have been removed. Use openclaw models auth list --json for authentication profiles and openclaw status --json or openclaw models list --json for usage data.
• Channel Shape: The chat.<id> field is no longer a simple string[] of account IDs. It is now an object: { accounts: string[], installed: boolean, origin: "configured" | "available" | "installable" }. Update your parsers to access chat.<id>.accounts.
• Flags: The --no-usage flag has been removed as the usage fetch no longer occurs on this path.

Administrative Access

Users who previously relied on operator.write scope to toggle global Active Memory will now need operator.admin scope to execute /active-memory on --global or /active-memory off --global.