On this page
On this page
Release notes
v2026.10.8
v2026.10.8
AI agents and tools can read these release notes as plain Markdown.
Security release: v2026.10.8 adds remediations for the eMudhra certificate compromise in the secret egress proxy and its exported trust bundle. Restart the Gateway to apply the change. Apps with independent trust stores and bypass tunnels need their own updates.
This is a patch release with no new features. It fixes problems that could stop OpenClaw from installing, updating, starting or delivering replies, alongside smaller app, channel and provider fixes.
Installation and Onboarding
-
Fixes fresh provider setup accidentally hiding other models in the picker after you select a model.
More details
#175945. Thanks to @matthematics1137, @obviyus, @jb-dev1.
-
Fixes Linux setup being blocked by the permissions on an existing OpenClaw state folder. New installer folders are also created with private permissions.
More details
#179538. Thanks to @steipete, @xiaopan-xydt, @XiaoPengYouCode.
-
Fixes the default piped Windows installer failing to start when no optional Node version is supplied.
Web UI
-
Deleting several conversations in the Control UI no longer repeatedly starts the same background archive process.
-
The Control UI now reconnects when a visible tab has lost contact but still shows Connected.
-
Fixes a missing-reply retry failing in the Control UI when the original turn also had permission to create a personal skill.
More details
#182219. Thanks to @sukhdeepjohar.
-
Fixes conversation links opening a 404 page when you reload or restore them.
More details
#183148. Thanks to @xiaopan-xydt, @steipete, @2991495215.
-
Fixes settings refusing to save when a plugin's saved secret reference is fully hidden in the editor.
More details
#183571. Thanks to @lonexreb, @steipete, @Valkster70.
-
Fixes follow-up messages being rejected in ordinary chats when an agent uses a separate session store alongside older stored sessions.
More details
-
Chat recovery notices no longer say a Gateway restart interrupted a turn when another cause is possible.
More details
Updates and Maintenance
-
Future updates can finish when OpenClaw itself passes verification but a channel is offline, with a warning for the affected channel.
More details
#179471. Thanks to @steipete, @duyufansh2008.
-
Fixes subsequent updates rejecting an unchanged deployment because of its saved policy approval.
More details
#180505. Thanks to @steipete, @waynegault.
-
Fixes subsequent npm updates failing on filesystems that add harmless execute permissions to unchanged package files.
More details
#181141. Thanks to @steipete, @hannesrudolph.
-
Fixes debug logging breaking update validation and Doctor's recovery data capture.
-
Fixes subsequent pnpm 10 updates using the wrong pnpm installation or missing its global commands.
-
Fixes subsequent Windows pnpm updates failing on a missing link to an optional platform package.
-
Fixes subsequent unattended Windows updates locking themselves out when the temporary folder uses a short-name path.
-
Fixes Doctor refusing retained conversations because their incoming-message headings were repaired.
More details
#179089. Thanks to @steipete, @LiuwqGit, @mrsavingsblogspot-rgb.
-
Fixes Doctor stack overflows while processing very large retained session histories and recovery diagnostics.
-
Doctor uses less retained memory while checking large fleets of agents.
-
Doctor now repairs stale plugin dependency links that could block a repair or update, including dangling Windows junctions.
More details
#179549. Thanks to @steipete, @JIRBOY, @Dembe-agent.
-
A missing legacy OAuth directory alone no longer blocks update readiness. Doctor still warns and offers to repair it.
More details
#180174. Thanks to @steipete, @VN77-77VN.
-
Doctor now keeps the model you chose for summarizing long Codex conversations. If an earlier repair deleted that setting, restore it manually.
More details
#179914. Thanks to @steipete, @scotthuang.
-
Doctor now moves the configured owner's login into shared storage when that owner is not the main agent and the main store is empty. Conflicting credential stores still need to be consolidated before retrying.
More details
#179726. Thanks to @steipete, @sukhdeepjohar, @bdjben.
-
Fixes read-only database checks interrupting a stopped-state backup during a subsequent Update or Doctor run.
-
A subsequent same-version repair can restart the managed Gateway after a post-plugin Doctor failure when it is safe to do so, including restoring Windows autostart. The failed repair is still reported.
More details
#180668. Thanks to @steipete, @alwynispat, @wilsonthesphere.
-
On Linux, Doctor can repair a managed Gateway that is using an unsupported Node version when a supported replacement is available. Explicitly pinned runtimes still need manual selection.
More details
#181618. Thanks to @steipete, @samueltardieu.
-
Future macOS updates wait for the serving Gateway to exit and release its databases before running Doctor.
More details
#182055. Thanks to @steipete, @RockhopperBot.
-
When a terminal hangup interrupts Doctor repair, Doctor attempts to restore the managed Gateway it stopped before exiting.
-
Standalone Doctor starts cancelling prolonged original-state capture after ten minutes and gives Gateway status and start commands if restoration fails. Work already underway may take longer to settle.
More details
#181941. Thanks to @steipete, @kenrohde, @kqtxmaster-boop.
-
Fixes an abandoned update or Doctor worker continuing to block CLI commands and Gateway startup after its parent process has died.
More details
#181695. Thanks to @steipete, @PonySlayer28.
-
Gateway startup retries transient agent database locks. If retries are exhausted, close competing processes and restart the Gateway.
More details
#182446. Thanks to @steipete, @Oldrich333.
-
Doctor preserves specific recovery errors for agents affected by a missing deletion journal, and can continue other repairs when a memory-index import is deferred.
More details
#181804. Thanks to @steipete, @4140leal-cw.
-
Fixes relative session-store paths creating a second database and breaking follow-up chats. If you already have split stores, stop the Gateway, run
openclaw doctor --fix, then restart.More details
#182444. Thanks to @steipete, @erickj92, @kqtxmaster-boop.
-
Fixes trusted reverse proxies receiving a 403 when their own health checks omit forwarded client headers.
More details
#182823. Thanks to @ericcaiwx-star, @steipete, @treez-admin.
-
The background session-archive process now releases its memory after five idle minutes and starts again when needed.
-
Fixes Doctor or a configuration read replacing valid hand-edited or Kubernetes-supplied settings with an older backup. Recovery warnings also point to the saved original.
More details
#183703. Thanks to @steipete, @csabatatrai, @yetval.
Messaging
-
Fixes default Slack, Google Chat, X, Nextcloud Talk, Zalo, Mattermost and Telegram accounts going offline after a separately configured account is added when the default uses a secret reference.
More details
#176267, #180450, #182608, #182706, #183618. Thanks to @555aaditya, @steipete, @progerjkd, @masatohoshino, @wangmiao0668000666, @rodricksz4h5.
-
Feishu quick-action and confirmation cards can be delivered with their buttons again.
More details
#79832. Thanks to @xiabee, @fuller-stack-dev, @steipete.
-
Feishu streaming cards can keep the text written before a tool call when the answer continues afterward.
-
Deeply nested Tlon messages no longer hold up later group messages and DMs. Excessively nested text can be truncated, and events too deeply nested to store are dropped rather than replayed.
More details
#177536. Thanks to @eleqtrizit.
-
Tlon can reconnect after a silent stream interruption and replay unacknowledged messages.
More details
#178358. Thanks to @rodricksz4h5, @steipete.
-
iMessage now attempts to recover a stalled Messages bridge after a send times out. The failed send is not automatically retried, so check delivery before resending.
-
iMessage self-chat replies are less likely to trigger repeated agent turns. A genuine self-chat message with the same text as a pending reply can still be suppressed for up to about three minutes.
More details
#183383. Thanks to @sukhdeepjohar, @steipete, @kunalchaudhary82.
-
Slack Socket Mode can reconnect after an isolated authentication error, although repeated failures can still block the channel.
More details
#181056. Thanks to @mushuiyu886, @marns.
-
An unresponsive Slack relay connection can reconnect so incoming events can resume.
-
An unexpectedly stopped channel can restart even if its previous run is cancelled while waiting to retry.
-
Fixes Synology Chat reporting an unconfirmed send even though it returned a valid message ID for the recipient.
More details
#177654. Thanks to @HootRock, @steipete, @thegrizzliestcat.
-
Twitch replies now resolve supported channel address forms to the intended channel.
More details
#178480. Thanks to @HeZ2z, @mweiland23-ux.
-
Zalo can process text retained in unsupported incoming events, including URL previews. Events without usable text need to be resent.
More details
#178938. Thanks to @zzrs-xxxr, @steipete, @namchamvinhcuu.
-
Discord realtime voice consultations can reach the agent and return an answer in a fresh session.
More details
#179462. Thanks to @shoaibulhaque, @yagizkus.
-
Fixes authorized Discord slash commands failing because their source conversation was archived; eligible conversations are restored in place with their history.
More details
Commit dcbe91c. Thanks to @PollyBot13.
-
Ordinary Matrix room messages now follow the configured reply visibility and silence policies.
More details
#179506. Thanks to @xiaopan-xydt, @obviyus, @doomy.
-
Fixes CLI-backed group agents losing conversation context when switching between room events and direct mentions with automatic replies enabled.
More details
#182273. Thanks to @xiaopan-xydt, @goslingmanagment, @steipete.
-
Incoming messages can now trigger recovery of a failed model runtime without a Gateway restart. If both recovery attempts fail, the input is retained for an operator to inspect and resend after fixing the problem.
More details
#181300. Thanks to @steipete, @handcraftrc-pixel.
-
Google Chat now delivers remote attachments as web links, including replies containing only media. This does not add local-file or native attachment uploads.
-
A channel image that was temporarily unreadable can be retried on a later conversation turn once it becomes readable. Images already marked permanently suppressed are not repaired.
More details
#183390. Thanks to @masatohoshino, @steipete, @yetval.
-
X accounts with capitalized configuration keys can be selected and managed in the Control UI, including their allowlists and guest settings.
-
Fixes GitHub collaborator verification and token refresh for X accounts whose configuration names contain capitals. Affected existing accounts may need to reauthorize X once.
-
Malformed Microsoft Teams message HTML no longer causes the parsing stalls that could delay other Gateway sessions.
More details
#183438. Thanks to @LeizhenZhang, @steipete.
Memory
-
Fixes Memory Core's nightly dreaming failing to write diary entries after delayed startup.
More details
#179031. Thanks to @steipete, @rebarshop-24.
-
Memory Wiki now preserves your handwritten Notes when an imported OKF concept disappears from a later import.
More details
#179657. Thanks to @steipete, @YangManBOBO.
-
Fixes overlapping session resets overwriting each other's saved memory files.
More details
#181081. Thanks to @jason-allen-oneal, @steipete.
Native Apps
-
Fixes macOS Gateway service reinstall and uninstall failing after the saved LaunchAgent file has been deleted.
More details
#180390. Thanks to @steipete, @avnishkirnalli.
-
Fixes the Tauri desktop app crashing while opening a local dashboard on macOS.
-
Talk calls using the Gateway relay on iOS and macOS can stay open through brief recoverable stalls. Audio may be dropped during the stall; iOS asks you to repeat it, while macOS has no matching notice.
-
GPT-Live Talk sessions can deliver speech or transcripts when the provider responds before microphone input, instead of closing at startup.
More details
#182919. Thanks to @anmol0b, @StatsR4Losers, @ulfschnabel, @siraustin, @steipete.
-
iOS Talk recovery guidance is localized in 21 non-English locales when microphone input recovers or realtime audio falls behind.
More details
-
Oversized macOS camera clips and screen recordings now return an error instead of disconnecting the node. The recording is not delivered, so retry with a shorter duration or, for screen recordings, a lower frame rate.
-
Fixes Android Chat failing after restart when your role allows an agent other than the Gateway's default.
More details
#182767. Thanks to @xiaopan-xydt, @steipete, @dawidkulpa.
Models and Providers
-
Fixes Gemini and Vertex requests waiting after a compatible proxy signals the end of a response. Gemini can also try another configured AI Studio key after an eligible initial quota error.
-
Oversized model-provider responses now report their streaming error promptly instead of leaving the request waiting.
More details
Landed commit. Thanks to @steipete.
-
Fixes cases of empty titles and background summaries caused by optional model thinking consuming the reply budget.
More details
#169920. Thanks to @jacobtomlinson and @dwalthour.
-
Session titles can complete on affected custom Zhipu and Z.AI GLM endpoints that reject a request to turn thinking off. Title failures also produce a clearer warning.
More details
#172580. Thanks to @steipete and @TommyLei666.
-
Claude CLI answers no longer wait for stopped tasks, work from an earlier input, or background agents and workflows started inside a subagent. Active top-level work still holds its reply until it finishes.
More details
#170322 and #182267. Thanks to @cassthebandit, @obviyus, @steipete and @goslingmanagment.
-
Fixes GPT-Live Talk and OpenAI realtime voice connections bypassing the configured managed proxy on the affected Gateway-relay and Bun paths. A refused proxy connection does not fall back to a direct connection.
More details
#175886 and #176567. Thanks to @MertBasar0, @jalehman, @steipete and @CheatosBu.
-
Fixes ChatGPT subscription sharing being blocked by the default managed proxy's loopback settings.
-
Fixes Anthropic PDF analysis rejecting saved setup-token or OAuth credentials that already work for chat.
More details
#178420. Thanks to @sunlit-deng, @steipete and @khensthoth.
-
Configured Claude CLI models remain selectable with eligible saved Anthropic subscription credentials, and model switches keep the configured CLI runtime.
More details
#179203. Thanks to @steipete and @josh-cornelius.
-
Agents can refresh shared OAuth credentials when a deleted agent left an empty or unrelated credential store behind.
-
Local session, transcript and MCP cleanup failures now report their own error instead of silently switching the turn to another model or provider.
More details
#179347, #180042 and #180075. Thanks to @steipete, @davidcittadini, @nikolay-bratanov, @sunlit-deng, @danielsiemers and @wr-web.
-
Rate-limited profiles now follow the provider's retry timing instead of accumulating long local lockouts. A rate limit without retry timing leaves the profile available for the next turn.
More details
#180280. Thanks to @fuller-stack-dev.
-
Fixes OpenAI models refusing existing credentials because stale catalog data points to a custom endpoint you did not configure.
More details
#181356. Thanks to @sasan1200.
-
Fixes temporary environment settings from a running skill making the model catalog unavailable.
-
Fixes a delayed authentication refresh after startup leaving model lists, conversation lists and agent requests unavailable until a restart.
More details
#182615. Thanks to @steipete, @VictorSesma and @hoangtohl.
-
Fixes Codex losing its final answer after a delivery retry or a temporary provider error when the completed tool results can be verified. Those tools are not run again.
More details
#181779 and #183077. Thanks to @sukhdeepjohar, @dthvt, @steipete, @gloss-dandelion-cult and @Lidashi1025.
-
Native Codex turns no longer receive a false stall interruption while a host tool call or output handling is still in progress.
More details
#181839. Thanks to @sunlit-deng and @robertmsale.
-
Fixes turns timing out immediately after a core upgrade when an older installed Codex plugin omits its start timestamp.
More details
#182995. Thanks to @RomneyDa, @kip-claw, @cjn119-ui, @fray-ai and @sukhdeepjohar.
-
Fixes Kimi K3.1 Pro and Highspeed using the wrong thinking controls for their provider. Thinking remains enabled on Kimi Code; the direct Moonshot route also handles native video and reasoning history for these variants.
-
Fixes automatically discovered GPT models on Bedrock Mantle using an incompatible API route.
-
Stopping the first Claude CLI chat turn can preserve its native conversation for the next turn when its transcript has been saved.
More details
#181966. Thanks to @holgergruenhagen.
-
Resumed Claude CLI and Gemini CLI sessions can receive recent conversation updates they missed while cron or Talk added activity to the same conversation.
More details
#183059. Thanks to @steipete, @scarabone and @atomsmasher81.
-
Fixes a queued Codex chat staying stuck after a plugin reload when another chat still needs the old process.
More details
#182104. Thanks to @jjjhenriksen and @RomneyDa.
-
Fixes Codex conversations resetting because their configured tool list changed when a completion arrived.
More details
#183286. Thanks to @steipete, @Marvinthebored and @RomneyDa.
-
Codex now gives submitted follow-ups and child results a short chance to be acknowledged before a tool ends the turn, reducing uncertain delivery. Unconfirmed input is not replayed automatically.
-
Long native conversations can use an eligible configured fallback model to summarize their history when the primary summary fails. Sessions locked to one model still refuse cross-model fallback.
More details
#183282. Thanks to @pstarkgit and @steipete.
-
Claude CLI operators can again opt in to Agent and Workflow native tools for eligible managed turns, including per-agent selection. The default exposes no native tools, and restricted turns remain excluded.
More details
Set
plugins.entries.anthropic.config.managedNativeToolsto select the tools. #183765. Thanks to @VACInc and @holgergruenhagen. -
Fixes Claude CLI turns waiting about 15 minutes before failing after their output limit is exceeded. The turn now stops promptly with the actual parsing or size error.
More details
Landed commit. Thanks to @Terwox.
Automations and Scheduling
-
Fixes Agents API scheduled jobs stopping immediately after sending to their own delivery target when the send was not explicitly final.
More details
#177518. Thanks to @masatohoshino.
-
Cron scripts can use the current plugin hooks on their next run after a plugin is replaced, without restarting the Gateway.
More details
#177825. Thanks to @jlapenna, @obviyus, @superandylin.
-
Fixes scheduled cleanup removing a cron session while it still has active child conversations. This does not restore parents deleted before the fix.
More details
#179057. Thanks to @Oldrich333, @steipete.
-
Background command completions now wait for the heartbeat that started them, preventing the completion from interrupting the heartbeat and repeatedly waking the same cron work.
-
An accepted heartbeat response now ends its tool turn instead of continuing into duplicate calls or a timeout.
More details
Commit 35bdf391. Thanks to @teddashh.
-
Conversation history can load after switching providers when a retained Claude CLI binding and a forwarded cron message previously caused it to fail.
More details
#181311, with the original repair proposed in #166980. Thanks to @jaxonparrott, @steipete.
Browser and Computer Use
-
Fixes successful computer actions on older paired Macs being reported as failed, which could cause the agent to repeat an action.
-
Fixes browser shutdown getting stuck after a navigation or interaction times out during a DNS check.
-
First browser use with trusted-proxy authentication no longer writes a Gateway password or restarts the Gateway. Out-of-process clients of the optional standalone loopback browser API using trusted-proxy auth need an explicitly configured
gateway.auth.password.More details
#181620. Thanks to @fuller-stack-dev.
Plugins and Integrations
-
Fixes older plugins failing to process incoming messages after an OpenClaw upgrade.
-
Fixes scheduled plugin work stopping after startup or reload. Changes to external credential files also continue to refresh Gateway authentication.
-
A disconnected FaceTime call releases the bridge so another authorized call can come through without restarting the Gateway.
More details
#154230. Thanks to @Solvely-Colin, @beeven, @omarshahine.
-
Long meeting talk-back replies can play in order when a provider sends audio faster than playback consumes it.
More details
#178254. Thanks to @steipete, @sukhdeepjohar.
-
Voice-call consultations and classic responses now use the call agent model when
responseModelis unset. An explicit override still takes precedence.More details
#181451. Thanks to @BeyonceBot, @potoior.
-
Fixes temporary Codex check processes being left running after their checks finish or fail.
More details
#167390. Thanks to @maxdjohnson, @steipete.
-
Fixes repeated tool-list checks loading unchanged npm-installed plugins again and adding to memory use in multi-agent setups.
-
Fixes undeliverable Codex child results repeatedly returning after a restart. Doctor can also repair existing duplicate or obsolete assignments when it can verify them.
-
Fixes already completed nested Codex work producing the same warning after each Gateway restart.
More details
#179050. Thanks to @mmartoccia, @steipete.
-
Codex parent conversations can resume after all native children finish, including their follow-ups, even while their results await delivery.
-
A Codex parent conversation can continue after an interrupted native child stops, without discarding the resumable assignment.
More details
#179099. Thanks to @obviyus, @wtfichigo.
-
Fixes Codex child agents receiving an authorization error on model requests after startup.
-
Doctor can load affected native plugins when the plugin files and OpenClaw state are on different filesystems.
-
Fixes native plugins failing to load on Windows when endpoint security changes file metadata without changing the verified contents.
More details
#179949. Thanks to @coaiMax, @paulcam206.
-
Gateway startup repairs stale managed plugin links after an external OpenClaw reinstall, allowing affected plugins to load again.
-
Agents sharing a valid MCP OAuth login can make concurrent tool requests instead of waiting for another agent's credential lock.
More details
#179997. Thanks to @danielsiemers, @maliv2, @steipete.
-
A failed remote MCP session deletion no longer blocks the next CLI-backend turn when local cleanup succeeds.
More details
#182189. Thanks to @aniketkrs, @goslingmanagment, @steipete.
-
Tailscale Serve and Funnel access can recover automatically after the managed route is lost, including after a Tailscale restart.
-
Fixes 1Password-backed credentials being rejected on group-writable OpenClaw installations. World-writable resolver paths remain refused.
-
Gateway shutdown can finish after a timed-out plugin retirement once the plugin work and cleanup have settled.
Security and Privacy
-
Adds remediations for the eMudhra certificate compromise in the secret egress proxy and its exported trust bundle. Restart the Gateway to apply the change. Connections relying only on the excluded root will fail; apps with independent trust stores and bypass tunnels need their own updates.
More details
PR #179249. Thanks to @joshavant.
-
Fixes sandboxed child agents being able to launch a CLI model outside their sandbox. Use an API-backed child model, or explicitly change the sandbox policy if you need a host CLI.
More details
PR #178742. Thanks to @eleqtrizit.
-
Fixes agents bypassing browser-tool restrictions by invoking the paired node's browser proxy through the generic Nodes tool.
More details
PR #179880. Thanks to @eleqtrizit.
-
POSIX exec allowlists no longer automatically trust
test -vor[ -vpredicates. Legitimate uses may need an allowlist entry or approval, and an approved externaltestmay behave differently from the Bash builtin.More details
PR #180880. Thanks to @eleqtrizit.
-
Fixes Fish Audio and Gradium errors exposing an active API key when the provider echoes it back.
More details
PR #178877. Thanks to @steipete.
-
Silent maintenance turns now suppress attachments and other rich content as well as text, while still delivering errors, voice messages and notices marked for delivery.
More details
PR #129244. Thanks to @steipete.
-
Fixes some cases where a model's thinking appeared in streamed answers despite reasoning visibility being turned off.
More details
PR #182454. Thanks to @steipete, and to @zax0rz for the report.
Other Bug Fixes
-
Fixes replies failing in an already configured workspace because packaged templates are missing. Your existing files are left unchanged.
More details
PR #120943. Thanks to @steipete.
-
Fixes accepted messages failing before the agent starts while their conversation history is being rebuilt.
More details
PR #164843. Thanks to @PollyBot13 and @steipete.
-
Fixes recovered turns disappearing from conversation context after oversized tool results are shortened.
More details
PR #182666. Thanks to @MSNexploder and @foiegrastyle.
-
Fixes parent agents being stopped as idle while they are still waiting for active delegated work.
More details
PR #172959. Thanks to @johan-eilertsen and @steipete.
-
Delayed
sessions_sendreplies can recover the full available answer instead of silently delivering only a preview. If the full answer is unavailable, the preview includes guidance for getting the rest.More details
PR #178271. Thanks to @rjdotdev, @steipete and @seriouslysean.
-
Fixes tool calls losing their results when they are batched with a handoff to wait for child agents.
More details
PR #184021. Thanks to @steipete, and to @jailbirt for the report.
-
Fixes long streamed HTML and JSX answers stalling OpenClaw and delaying unrelated conversations.
More details
PR #175472. Thanks to @JasonBuildAI.
-
Fixes malformed replies that look like tool calls stalling the Gateway.
More details
PR #183252. Thanks to @LeizhenZhang and @steipete.
-
A new human message can continue an eligible interrupted child session after a Gateway restart. The interrupted task is not automatically replayed.
More details
PR #178888. Thanks to @jalehman.
-
Conversations blocked by a stale restart-recovery source can accept new messages without a reset, while retaining their history and showing an interruption notice.
More details
PR #180853. Thanks to @steipete, and to @Cloud-Ocean for the report.
-
A stopped conversation blocked by an orphaned restart-recovery record can accept another message without losing its transcript.
More details
PR #181992. Thanks to @steipete, and to @najef1979-code for the report.
-
Fixes a child agent accepted just before a restart remaining stuck as running even though it never started.
More details
PR #181871. Thanks to @steipete.
-
Fixes private child-agent results repeatedly retrying after restart recovery had already processed them.
More details
PR #184041. Thanks to @steipete, and to @aounakram for the related report.
-
After a restart, parent agents can receive saved child results even if that child conversation has since been reused.
More details
PR #184069. Thanks to @steipete.
-
Session deletion, transcript archiving and retained cron history work through symlinked state directories with SQLite and PostgreSQL.
More details
PR #181554. Thanks to @steipete.
-
Agent commands can run under an unattended Windows S4U Gateway when a Store PowerShell alias is inaccessible. If native PowerShell 7 is unavailable, OpenClaw falls back to Windows PowerShell 5.1. Install PowerShell 7 using its MSI or ZIP if your commands need PowerShell 7 syntax.
More details
PR #182898. Thanks to @RomneyDa, and to @tre1d for the report.
-
Fixes workspace synchronization failing to start on older Windows shared-host nodes.
More details
PR #183064. Thanks to @shakkernerd.
-
Fixes session objects staying in memory after agent and child-agent attempts finish.
More details
PR #183884. Thanks to @steipete.
