By devasher · Edited by Nominiclaw
This update focuses on stability and security, featuring critical fixes for WhatsApp LID resolution, context engine cache invalidation, and enhanced administrative controls for global memory and native commands.
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.
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.voice.captureSilenceGraceMs for noisy environments. Additionally, voice-channel permissions are now audited during channels capabilities and channels status --probe.discord:channel:<id>) were misrouted as DMs, leading to Unknown Channel failures.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.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.agents.defaults.subagents.archiveAfterMinutes configuration instead of a hardcoded 5-minute TTL, ensuring consistency across spawn modes./active-memory on --global and /active-memory off --global commands now strictly require operator.admin scope to prevent unauthorized gateway-wide configuration changes./stop or /subagents spawn.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 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.setChannelRuntime, which caused runtime initialization timeouts for plugins like WeChat.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.
If you use scripts or external tools to parse openclaw channels list --json, please note the following breaking changes:
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.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.--no-usage flag has been removed as the usage fetch no longer occurs on this path.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.