OpenClaw releases certificate-security patch that requires a Gateway restart

Installing OpenClaw’s latest security patch is not the final step: users must restart the Gateway to apply the security change. Version 2026.10.8 addresses the eMudhra certificate compromise in OpenClaw’s secret egress proxy and its exported trust bundle. The release also repairs failures that could prevent installation, updates, startup or reply delivery.

The security update has a boundary

OpenClaw describes this as a patch release with no new features. Its security work covers the secret egress proxy and the trust bundle that proxy exports. The required Gateway restart applies that change; the release notes separately flag applications whose trust updates are outside that coverage.

Getting installation and repairs unstuck

Several fixes address failures before a user can reach a working conversation. On Linux, setup no longer gets blocked by permissions on an existing OpenClaw state folder, and newly created installer folders use private permissions. On Windows, the default piped installer can start without an optional Node version being supplied.

Future updates can also finish when OpenClaw itself passes verification but a messaging channel is offline. The affected channel gets a warning instead of blocking completion. Other corrections address subsequent updates rejecting an unchanged deployment because of its saved policy approval, or failing because a filesystem added harmless execute permissions to unchanged package files.

The Doctor repair tool gets changes affecting both configuration and recovery. Doctor or a configuration read could previously replace valid hand-edited settings, or settings supplied through Kubernetes, with an older backup. The patch fixes that behavior, and recovery warnings point to the saved original. It also fixes abandoned update or Doctor workers continuing to block commands and Gateway startup after their parent process has died.

Some existing problems still require operator action. The maintenance instructions distinguish between failures the patch prevents and installations that already need repair:

• Split session stores: relative paths could create a second database and break follow-up chats. Stop the Gateway, run openclaw doctor --fix, then restart.
• Database locks: Gateway startup retries temporary agent database locks. If retries run out, close competing processes and restart the Gateway.
• Lost summarization settings: Doctor now preserves the chosen model for summarizing long Codex conversations. If an earlier repair deleted that setting, restore it manually.

Recovering replies without duplicating them

Messaging fixes cover both accounts dropping offline and connections that stop responding. Default accounts across Slack, Google Chat, X, Nextcloud Talk, Zalo, Mattermost and Telegram could go offline after another account was added when the default used a secret reference. The release fixes that configuration problem.

Incoming messages can now trigger recovery of a failed model runtime without restarting the Gateway. That recovery has a limit: if both attempts fail, OpenClaw retains the input for an operator to inspect and resend after fixing the problem. It does not promise to deliver every retained message automatically.

The iMessage correction also separates recovering a connection from resending an answer. OpenClaw now attempts to recover a stalled Messages bridge after a send times out, but the failed send is not retried automatically. Users are told to check delivery before resending.