OpenClaw fixes rejected messages and repair checks that falsely signaled success

OpenClaw has released v2026.11.1 to fix failures on both sides of an agent conversation: accepted messages could be refused, while failed maintenance inspections could appear successful. The patch adds no new features. Its release notes instead describe repairs to message handling, stuck settings changes and recovery paths that preserve conversations and credentials.

Accepted messages, rejected replies

The messaging fixes distinguish accepting a request from actually processing and answering it. OpenClaw says Telegram messages could receive a false restart refusal or lose a fallback reply during processing. Discord messages could also be refused after acceptance, while they were still waiting to be processed.

A separate Telegram correction addresses queued replies that failed after an earlier conversation turn ended. Other connection repairs let Slack relay connections and stalled Tlon streams reconnect. Channels that stop unexpectedly can retry without restarting channels the user deliberately stopped.

The same distinction reaches the apps. On Android phones connected to a compatible Gateway, ending Talk now lets accepted work finish and puts its answer in chat. In the web interface, a visible Control UI tab reconnects when its Gateway connection stops responding.

A failed check must look like failure

Doctor, OpenClaw’s inspection and repair tool, now returns a nonzero exit code when an inspection is blocked or repairs remain incomplete. That gives scripts a failure signal rather than a misleading success result. OpenClaw specifically warns that scripts using doctor --json must handle the changed exit status.

Unattended Doctor runs now report missing plugins without installing them unless explicitly authorized with --fix, --repair or --yes. Repair also preserves empty or inherited group allowlists, rather than changing them, and avoids unnecessary configuration conflicts when only plugin migration guidance has changed.

Migration handling favors keeping recoverable data over importing altered content. Doctor skips legacy text containing malformed UTF-8, an invalid text encoding, preserves its original bytes and continues independent migrations. Invalid legacy heartbeat hours are also preserved, with guidance identifying the value to correct before retrying.

Recovery details that need attention

• For conversation stores split by older relative paths, stop the Gateway before running openclaw doctor --fix.
• Updates from versions 2026.10.7 through 2026.10.9 can use a fresh Doctor if migration fails during recovery. A manual Gateway start may still be necessary.
• Doctor can repair an unsupported Node runtime for services it owns without overwriting unrelated service settings.

Keeping settings and saved conversations intact

The Gateway fixes address settings changes that became stuck while waiting to apply, alongside more reliable channel restarts. In the Mac app, failed Gateway-backed Browser or Tailscale settings saves no longer overwrite stored credentials with redaction placeholders—the masked values shown in place of secrets.

Startup and storage changes address a different source of interruption. Gateway startup retries temporary SQLite database locks instead of immediately stopping. Older shared-state databases retaining a worker index can complete migration rather than repeatedly failing at startup.

OpenClaw also says large conversation histories and archives can be read and indexed with less working memory. During large exports and memory preparation, other database work can spend less time waiting. If a transcript changes during a read, OpenClaw falls back to a consistent snapshot.

Usage and cost reports can now read verified archived conversations without restoring them from cold storage. Here, the release notes specify a deliberate failure condition: missing or corrupt archives stop the refresh rather than allowing it to produce incomplete results.