• Choose any named agent from either sidebar view while keeping main-chat status readable. #150587 Docs · Docs · Thanks @vyctorbrzezowski, @steipete.
• Subagents and Processes panels let users inspect delegated work and recent command output beside the parent chat, with targeted Stop controls. #164061 Docs · Docs · Docs · Thanks @shakkernerd, @steipete.
• Remove duplicate owner buttons from person-group headers; owner filters remain under Sessions → Filter & sort → Owners. #164608 Docs · Thanks @vyctorbrzezowski.
• Session browsing reuses verified row updates, downloads less list data and preserves loaded dashboard pages. #164764 Docs · Docs · Thanks @steipete.
• Finished commands start collapsed in the Processes panel while running work remains visible. #164776 Docs · Thanks @shakkernerd.
• Person cards keep showing accessible recent-session links when the sidebar is filtered to a person, owner or status. #165004 Docs · Thanks @vyctorbrzezowski.
• Pin, unpin, or confirm deletion of Workboard boards directly from the sidebar. #165020 Docs · Thanks @steipete.
• Pinning and renaming sessions avoids waiting for transcript-history loading. #165073 Docs · Docs · Thanks @steipete.
• Session-list refreshes avoid duplicate reads and decoding of fresh session details. #165096 Thanks @steipete.
• Equivalent session-list views share repeated work, and runner availability updates refresh affected sidebar rows without reloading the whole list. #165150 Thanks @steipete.
• Sidebar child-count badges exclude archived sessions after refresh and match the children users can expand. #165160 Docs · Thanks @obviyus, @tre1d.
• Sidebar pin editing and reorder controls no longer overlap and stay aligned on desktop and mobile without shifting activity indicators. #165199 Docs · Thanks @steipete.
• The default workspace name stays readable in narrow all-agent sidebars while adjacent controls retain their touch targets. #165272 Thanks @shakkernerd.
• Enter opens the typed folder path; a missing starting workspace falls back cleanly to home. #165446 Docs · Thanks @steipete.
• Collapsing the Online sidebar hides its filter button and preserves filter selections for reopening. #165449 Thanks @vyctorbrzezowski.
• Pin favorite agents from the sidebar's agent menu when at least four agents are available. #165640 Docs · Thanks @vyctorbrzezowski.
• Read longer agent lists in rows, distinguish repeated names by ID, and search names or IDs when more than six agents are available. #165641 Docs · Thanks @vyctorbrzezowski.
• Reading a parent conversation or Home clears unread dots from eligible hidden subagent runs. #165792 Docs · Thanks @holgergruenhagen, @obviyus.
• Session lists reuse unchanged rows while unrelated subagents run, keeping current status and permissions live. #165932 Thanks @steipete.
• Deleting a child during refresh no longer leaves a false list error or hides remaining sessions behind Retry. #166128 Docs · Thanks @shakkernerd.
• Mobile sidebar reorder controls no longer crowd session icons and color markers. #166287 Thanks @jacobtomlinson.
• Session-list loading reuses roster selection as display rows arrive. #166405 Thanks @steipete.
• Keeps valid session previews through renames and pinning without rereading unchanged transcripts. #166635 Thanks @steipete.
• Keeps session previews readable when Markdown links contain nested brackets or parentheses, preserving the following text. #166653 Thanks @jayzhou2309, @rudycelekli, @obviyus.
• Duplicate session names stay in the rename dialog with an inline explanation so they can be corrected without retyping. #168034 Thanks @romneyda, @vyctorbrzezowski.
• Delivered messages stop showing false sidebar and Inbox warnings when an earlier send remains unconfirmed. #168090 Docs · Thanks @steipete.

Existing Auto and unset automatic-learning settings are preserved. Legacy proposal modes move to Off, and a setting already changed to Off stays off until you change it. The library keeps up to ten saved versions; new sessions pick up skill edits. Supported unused-skill cleanup archives skills after 30 days and is disabled for Codex. Restoring pre-upgrade proposal history requires its matching backup; pending proposals are exported inactive.

• Skill Workshop saves learned procedures directly, with change history, saved versions and Undo. #161057 Docs · Docs · Docs · Docs · Docs · Docs · Thanks @obviyus, @steipete, @RomneyDa.
• Skill Workshop is guided to learn repeatable procedures and direct personal facts to memory and repository-specific knowledge to project docs. #167404 Thanks @obviyus.
• Workshop brings learned skills, their files and change history together, with clear ways to undo changes, restore versions and manage archived skills. #167440 Docs · Docs · Thanks @obviyus.
• Older saved skills with long names are visible and manageable again. #167639 Docs · Thanks @obviyus.
• The Workshop migration retires approval-first settings; its later correction preserves Auto while older proposal modes move to Off. #167641 Docs · Docs · Thanks @obviyus.
• Skill Workshop shows configuration-load errors with Retry while keeping the skill library available. #167656 Docs · Thanks @steipete.
• Doctor preserves Auto and unset automatic learning settings; legacy proposal mode and settings already changed to Off remain off. #167846 Docs · Thanks @obviyus.
• Saved Skill Workshop versions show their own descriptions and include metadata changes in comparisons. #168001 Thanks @steipete.

Deployments behind a login proxy need the public-reader route setup, with private app and API routes kept protected.

• Long shared-conversation titles and messages avoid broken characters where shortening crosses an emoji boundary. #141240 Thanks @hugenshen, @obviyus, @xialonglee.
• Opening session sharing no longer waits for unrelated session display refreshes after a model catalog change. #165344 Docs · Thanks @steipete.
• Published conversations use their normal chat address for anonymous read-only sharing, with optional publication of eligible X work sessions. #166049 Docs · Docs · Docs · Docs · Thanks @steipete.
• Signed-in token/password users can reload, bookmark and open chat links in new tabs without an extra login click, preserving drafts and presentation parameters. #166783 Docs · Docs · Thanks @steipete.
• Refreshing accessible conversation links shows loading while access is checked instead of briefly claiming the conversation is unavailable. #167097 Thanks @jacobtomlinson.
• Sharing requests recheck current permissions and conversation identity before returning a public link. #167504 Thanks @steipete.
• Session-link guidance explains which renames preserve bookmarks and when moving to another agent requires a new canonical link. #167586 Docs · Thanks @steipete.
• Public chat links reach normal session lookup and login handling after Gateway startup instead of a false temporarily-unavailable response. #167604 Thanks @steipete.
• Returning signed-in proxy users can open conversation links without a public-reader or login flash. #168007 Docs · Thanks @steipete.

• Streamed text avoids a tiny trailing message at an exact size boundary. 0427411.

• Supporting history integration keeps saved corrections from overwriting a newer edit or another message while preserving later additions. 1cd6492

• Conversation history avoids a read failure when replies start concurrently. 20ef748

• Transcript search can recover from failed cleanup without trusting checks from a database that has been replaced. 2839466

• Newly created databases remain ready through worker handoffs, and replacement checks retire obsolete copies reached through alternate paths. 400bab4.

• Long-running OpenClaw processes release memory held by completed database-opening requests, even when reusing database workers. 4db0224.

• A closed database deleted from disk can be recreated through a directory alias. 58e1c3f.

• Avoid duplicate remote-worker answers and accept follow-ups after cancellation 5c231631

• Rejected worker transcript saves stop with a visible error instead of repeatedly reconnecting without progress. 5dcf668 Docs

• Supporting conversation integration uses current saved history for replay and delivery and keeps meeting exports consistent during simultaneous edits. 66471ca

• Finding saved conversations avoids an extra database hold that could interrupt recovery reads. 7294c69

• Keep transcript and trace storage in the intended Gateway location 737560a5

  Thanks @jacobtomlinson.

• Supporting worker recovery lets eligible interrupted agent work resume when its worker disappears. 740564a

• Messages queued during worker setup can proceed when setup finishes or the preceding turn updates. 7ecc80a Docs

• Register replacement subagents when their preparation overlaps Gateway startup 873b0a59

• Transcript searches can recover after a shared SQLite worker fails without reusing its stale validation. 8a470f8

• Record reply cancellation before session deletion retires the reply, giving connected clients deletion as its stop reason. 9564061.

• Supporting workspace integration keeps inspection and cleanup with the recorded owner and lets temporary transcript-save refusals be retried. 96ccec4 Docs · Docs · Docs

• Pending subagent completion checks stop cleanly when their original database closes during shutdown. 9c057a3

• Restore authorized Incognito message sending that could fail with SESSION_MUTATION_AUTHORIZATION_CHANGED, while retaining current permission checks. a5a43a0

• Keep eligible read-only history recovery available while writable startup is pending a73ef55.

• Restored subagents wait for saved conversations to become readable after restart before making their first request. aa154c3

• Late subagent cleanup preserves unfinished delivery for restart recovery instead of using a database that has closed. ab29c88

• Avoid false history-read failures when queued writes already finished before the read began ab56112.

• Conversations can recover after a crash even when their database checks finish after startup. c1b9f4e

• Recent sharing-membership updates no longer cause unnecessary shared-session lookup failures. d511ab9

• API and plugin requests can continue after a completed Stop without a false session-work rejection. df8b0cb

• Worker turns can recover from temporary transcript-save interruptions without duplicating messages after an uncertain save. dfb2723

• Authorized users can send messages and attachments in affected existing global sessions. e6b93ae

• Missing images from older conversation history are described as unavailable context rather than a fresh image-input failure. eb1b48c

• Valid sharing access survives display-only updates when a waiting session request resumes. f6e6805

• Resumed delegated work avoids duplicate completion announcements from an earlier pause notice. f71437a Docs

• Worker turns can retry a temporary transcript-save conflict during database preparation. fbac0be Docs

• Session tail timestamps show a Z so readers can identify UTC. #148022 Thanks @TUARAN, @obviyus, @aniruddhaadak80.

• Replies keep useful text after internal tool blocks containing quoted closing markers. #149775 Thanks @bodapatisaikrishna, @steipete.

• Identity guidance explains which updates change agent configuration and which may update IDENTITY.md. #149992 Docs · Docs · Thanks @sunlit-deng, @pusherofbrooms, @holny, @obviyus.

• Final and streamed replies retain short corrections added after a message-tool send. #151921 Thanks @amittell, @obviyus.

• Interrupting or stopping a turn keeps canceled work out of restart recovery so the next message can proceed. #154175 Docs · Thanks @Grynn, @steipete, @hannesrudolph, @VACInc.

• Channel requests remain in conversation history when required compaction fails. #160462 Thanks @masatohoshino, @obviyus, @aacodex401.

• Background session summaries allow slower CLI-backed utility models up to 30 seconds to finish. #162067 Thanks @rogeriochaves, @llmsommelier, @obviyus.

• Replies continue after automatic dashboard session resets without title generation blocking them. #162689 Thanks @ericcaiwx-star, @dimonnld, @OmniScaleMatt, @gloria0021, @obviyus.

• An agent turn can continue without a resend when its database worker exits before dispatch. #163258 Thanks @goutamadwant, @steipete, @hawkit.

• Malformed file-edit requests receive a clear JSON correction before edits run. #163577 Thanks @GoldArowana, @aniruddhaadak80, @altaywtf.

• Multi-step agent work gets a bounded chance to continue after a later tool call is rejected as malformed. #164274 Thanks @fransqaas, @obviyus.

• Parent agents can finish work triggered by a subagent result without the delivery deadline cancelling work already started. #164282 Thanks @Oldrich333, @obviyus.

• Session operations avoid repeating already validated database schema checks. #164490 Docs · Thanks @steipete.

• Quoted replies load their previews with chat history, and links to older originals keep navigating when rendering is delayed. #164605 Thanks @steipete.

• Covered failed worktree setup can clean its empty branch so the same session can retry, and delayed heartbeat wakeups avoid false failures. #164615 Thanks @steipete.

• A queued message can continue after the covered interrupted Control UI run, while unsafe ownership conflicts show understandable retry guidance. #164656 Thanks @RomneyDa, @LivenseAI, @lucascmg-vx.

• Timed-out or killed child tasks keep their recorded status and return results from the right owner, even when agents reuse a session name. #164674 Thanks @steipete.

• Agents can send Ctrl+Space and modified F1–F12 correctly to interactive terminal programs. #164688 Thanks @yetval, @obviyus.

• Agent patches match en quad and em quad spaces without changing preserved file text. #164742 Thanks @ooiuuii, @obviyus.

• Durable conversation rewind and branch switching perform their main history mutation work in the database worker. #164753 Thanks @steipete.

• Managed worktree cleanup runs as a background job, with a configurable global checkout cap and eviction policy. #164758 Docs · Docs · Thanks @steipete.

• View-only subagent chats hide saved-draft recovery controls while parent drafts remain available. #164766 Docs · Thanks @shakkernerd.

• Commands containing NUL are rejected before helpers start or an existing scoped command is cancelled. #164770 Docs · Thanks @shakkernerd, @IWhatsskill, @chabal2020.

• Preparing and finishing conversations moves more saved-data reads into background workers. #164779 Thanks @steipete.

• At shutdown, OpenClaw waits for managed-worktree cleanup it has already accepted to finish before closing background workers. Saved cleanup data stays available for recovery if a write's outcome is uncertain. #164824 Docs · Thanks @steipete.

• Reopening or reloading an active Claude CLI session preserves the original elapsed Working time, including preparation before handoff. #164873 Thanks @rogeriochaves, @llmsommelier, @fuller-stack-dev.

• Session setup, deletion, and history changes move selected database work into the background. #164922 Docs · Thanks @steipete.

• Managed-worktree sessions archive promptly while checkout cleanup continues in the background; unarchiving waits for any in-flight snapshot and restores saved work. #164938 Docs · Docs · Thanks @steipete.

• Selected conversation-history checks run through the existing background database reader. #164989 Docs · Docs · Thanks @steipete.

• First-use session preparation leaves the Gateway available while session data is read. #165027 Thanks @steipete.

• Session checks do less blocking Gateway work while replies use current session and transcript state. #165028 Thanks @steipete.

• Artifact downloads in long conversations can start sooner without checking unrelated messages first. #165030 Docs · Thanks @steipete.

• Progress-card updates avoid repeated database setup work. #165081 Thanks @steipete.

• Preparing model context avoids repeatedly decompressing the same history messages. #165086 Thanks @steipete.

• Recovering older results and checking duplicate messages avoids unrelated compressed history reads. #165089 Thanks @steipete.

• Less database overhead when registering or settling groups of subagents. #165098 Thanks @steipete.

• Restoring a saved chat position avoids unnecessary pane redraws while messages finish laying out. #165101 Thanks @steipete, @Yuyorman.

• Completed resumed chats clear stuck working timers and Stop buttons automatically while keeping unsent drafts. #165108 Thanks @shakkernerd, @obviyus.

• A message arriving as the prior turn finishes avoids a spurious placement error and resend request. #165166 Thanks @steipete.

• Canceling one queued WebChat message no longer makes unrelated queued messages fail and retry. #165168 Thanks @steipete.

• Missing-summary recovery can return a useful task summary when automatic compaction previously caused it to reject its own result. #165179 Thanks @ericcurtin, @johan-eilertsen, @Patrick-Erichsen.

• Reply finalization and routine session checks avoid loading large saved prompts and diff data they do not need. #165181 Thanks @steipete.

• Routine session-start and completion saves run off the Gateway main thread while accepted writes settle during shutdown. #165193 Thanks @steipete.

• Queued messages from multiple clients route run timing, model information and live tool activity to the client that sent each message. #165194 Thanks @steipete.

• Chat turns no longer fail merely because unrelated sessions start or finish while maintenance runs. #165196 Thanks @steipete.

• Managed worktree cleanup uses incremental Git maintenance and suspends failing repositories so cleanup and other repositories can continue. #165204 Docs · Thanks @steipete.

• Status, Side chat, and progress reads run in the background, discard superseded results, and finish accepted summary writes during shutdown. #165205 Docs · Thanks @steipete.

• Publish PR cards clear stale worktree-owner errors after cleanup and show compact status labels with expandable recovery details. #165219 Thanks @steipete.

• Large-history attachment cleanup and context reports use smaller reads, reducing per-request memory spikes and time holding the history worker. #165221 Thanks @steipete.

• Progress images and attachments from session-bound recovered or internally dispatched agent runs survive scratch-file cleanup and chat reload. #165275 Docs · Thanks @steipete.

• Workboard opens its Board agent in a dedicated dock conversation without duplicating the ordinary session list or inflating people counts. #165319 Docs · Docs · Docs · Thanks @steipete.

• Eligible Control UI turns interrupted by a Gateway restart resume after startup, with repair for sessions left stuck running by this specific cleanup race. #165333 Docs · Thanks @steipete.

• Slack and Discord sessions no longer show header buttons that open their source conversation; sessions and transcripts remain readable. #165334 Docs · Docs · Thanks @Patrick-Erichsen.

• Outdated subagent records no longer block current siblings and nested children from settling after a restart so their parent can continue. #165335 Docs · Thanks @steipete.

• Completed chat steps stay counted when their only nested work is a hidden routine progress update. #165358 Docs · Thanks @obviyus, @shakkernerd.

• Starting ordinary chats moves selected-session metadata rereads off the Gateway main thread while retaining current permissions and stale-read checks. #165360 Docs · Thanks @steipete.

• Permission and skill checks run in the background while messages sent during active work keep their order and avoid duplicate submission after uncertain acceptance. #165370 Docs · Docs · Thanks @steipete.

• Chat keeps delegated-work waiting status beneath the reply, shows remaining subagents and preserves opened tool details through handoff refreshes. #165378 Docs · Thanks @shakkernerd, @steipete.

• Active tools retain live progress after chat history refresh instead of being replaced by a misleading unknown outcome. #165392 Docs · Thanks @steipete.

• Session setup and status recover missing start or completion times from saved history in the background without changing reset decisions. #165401 Docs · Docs · Thanks @steipete.

• Named Windows-1252 text uploads preserve accented words in model context when the sender omits a specific content type. #165405 Thanks @Chessing234, @obviyus.

• Opening a session from an excluded agent store promptly reports not found instead of repeatedly asking clients to reload. #165408 Thanks @obviyus, @Oldrich333.

• Chat activity outcome badges stay on one line while long command headlines truncate cleanly. #165415 Thanks @steipete.

• Completed chat steps keep their titles and open their details in one click. #165426 Thanks @shakkernerd.

• A steer sent before a reply target exists can join the first turn after workspace setup. #165462 Docs · Thanks @steipete.

• Busy session lists reuse unchanged rows during unrelated updates, reducing repeated allocation work. #165464 Thanks @steipete.

• The chat working line shows how many direct sub-agents remain running while the parent works. #165477 Docs · Thanks @shakkernerd.

• Reopening branch pickers in large worktree setups avoids repeated Git discovery; empty snapshot cleanup skips an unnecessary process. #165480 Thanks @steipete.

• Stopping a turn sends one cancelled completion without trailing lifecycle updates or duplicate finals. #165483 Docs · Thanks @steipete.

• Chats at the latest content keep following messages from other devices while readers who scroll up keep their place. #165484 Docs · Docs · Thanks @steipete.

• Large saved prompts cost less to validate, and eligible sub-agent starts avoid redundant parent-skill reads. #165495 Thanks @steipete.

• Parent conversations recover after restart without repeating the same child-completion continuation. #165496 Thanks @steipete, @Oldrich333.

• Pending child-agent results can recover after obsolete child records are retired on restart. #165499 Docs · Docs · Thanks @steipete.

• Saved unanswered instructions remain actionable when sub-agent work resumes the conversation after recovery. #165502 Docs · Thanks @steipete.

• Repeated base-branch suggestions return with less work in large repositories when refs are unchanged. #165513 Docs · Thanks @steipete.

• Steered instructions stay visible in conversation order, including saved Codex history after refresh. #165535 Docs · Thanks @steipete.

• Steer messages sent while a worktree or runtime is starting can guide that initial task once it is ready. #165537 Docs · Thanks @steipete.

• Saved-reply and attachment corrections move database work off the Gateway request thread. #165541 Thanks @steipete.

• Answers resumed after subagent work stay together, with timing and known parent-run usage covering the whole request. #165542 Docs · Thanks @shakkernerd.

• Injected reports and alerts appear once as standalone notes without interrupting an ongoing reply; intentionally identical notes remain separate. #165549 Docs · Thanks @obviyus, @legarconc.

• System-agent history and resets finish saving through shutdown, and configuration notices avoid outdated audit records. #165561 Thanks @steipete.

• Session suggestions load with less repeated preparation, improving first and subsequent requests in a large-session benchmark. #165570 Thanks @steipete.

• Admitted chat messages are acknowledged while skill preparation continues, with preparation failures visible on the run. #165608 Docs · Docs · Docs · Thanks @steipete.

• Parent agents retry completed child work after stale-result preparation or temporary worker-capacity failures instead of losing their continuation. #165610 Docs · Thanks @steipete.

• Saved-chat model and account information prepares in background workers while preserving current visibility and selected-account checks. #165614 Thanks @steipete.

• Long chat status and model labels avoid broken emoji at the shortening boundary. #165619 Thanks @nocodet888-arch, @obviyus.

• Required child-agent replies keep retrying through temporary cleanup failures within the existing delivery deadline. #165634 Docs · Thanks @steipete.

• Unchanged Workboard Sessions boards load again with much less repeated work. #165698 Docs · Thanks @steipete.

• Chat shows subagent names, current status, duration and direct session links, with separate subagent activity counts. #165725 Docs · Thanks @shakkernerd.

• Repeated branch suggestions in the managed-worktree picker can return faster for unchanged repositories. #165750 Docs · Thanks @steipete.

• Downloaded attachment names preserve apostrophes at the start or end. #165754 Thanks @LinzeShi, @Patrick-Erichsen.

• Repeated file listings from an unchanged conversation reuse their summaries instead of decoding the transcript again. #165759 Thanks @steipete.

• Queued follow-up messages wait for the original reply to finish when only its source operation owns the queue. #165788 Thanks @steipete.

• Parents receive complete answers from new ACP and CLI fallback subagent runs, even after the child has moved on to another turn. #165800 Docs · Thanks @yetval, @steipete, @holgergruenhagen, @potterdigital, @Leli1024, @zyz619963502zyz.

• Repeated searches and refreshes of large conversation lists reuse more prepared information, reducing processing in the measured workloads. #165834 Thanks @steipete.

• Shorthand GitHub references in chat open the matching repository linked in the same paragraph, with correct previews and hovercards; conflicting references stay unlinked. #165849 Thanks @jjjhenriksen, @obviyus.

• Session diffs and pull-request statistics avoid repeated slow Git work in large managed repositories, while limiting competing content reads. #165858 Docs · Thanks @steipete.

• Failed conversation setup can complete cleanup without a conflicting database permission read, including conversations with native plugin bindings. #165859 Docs · Thanks @steipete.

• Shared-conversation browsing, search and pagination use background-refreshed listings instead of waiting on slow paired nodes for each request. #165887 Docs · Docs · Thanks @steipete.

• Accepted conversation-history saves finish in order before the writer is released, so later writes and Gateway shutdown do not overtake them. #165894 Docs · Docs · Thanks @steipete.

• Sessions boards update changed conversation information without rebuilding the entire roster, keeping existing visibility and explicit stale-data warnings. #165896 Docs · Docs · Thanks @steipete.

• Chat history downloads smaller ordinary pages, and repeated delivery checks avoid downloading unchanged history again while older messages remain available through pagination. #165897 Docs · Docs · Thanks @steipete.

• Cancelling work still waiting for conversation admission releases its locks so reset, recovery and cleanup can proceed. #165898 Thanks @steipete.

• Sequential conversation saves receive their own commit confirmation, preventing a later save from appearing successful because an earlier save succeeded. #165904 Thanks @steipete.

• Repeated branch-list refreshes reuse validated summaries and background readers instead of rescanning long conversations after every small append. #165909 Docs · Thanks @steipete.

• Confirming composed text preserves the New Session draft until deliberate submission. #165919 Thanks @ly85206559, @obviyus, @Patrick-Erichsen.

• New conversations inherit the parent agent in explicit multi-agent fleets unless a child target is selected. #165929 Thanks @alkor2000, @steipete, @markstones78, @aldiz1711.

• New sessions avoid spurious worker-placement errors while other local conversations are active. #165984 Thanks @steipete.

• Completed delegated results reach the parent after the bounded wait for a stalled finished descendant expires. #165987 Thanks @Oldrich333, @obviyus, @LifeViwer, @vinsew, @VACInc, @mgr-nick.

• New Session returns a retry message and preserves the draft when model-catalog loading stalls. #165989 Thanks @steipete, @DonnieFi, @dprev.

• Switching among agents avoids repeated database startup delays by keeping recently used workers warm. #165993 Thanks @steipete.

• Queued messages stay pending until their replies finish, and clients track the result using the original returned ID. #165996 Docs · Thanks @steipete.

• Simultaneous first requests for a new agent share database creation instead of failing with ownership errors. #166003 Thanks @steipete.

• Sending a message no longer fails solely because the previous turn or activity summary updates the same session simultaneously. #166027 Thanks @steipete.

• Accepted first messages survive a Gateway restart during local managed-worktree preparation instead of leaving an empty unanswered session. #166042 Docs · Thanks @steipete.

• Creating and updating other sessions can continue while a model or native-runtime-consent change waits for preparation. #166059 Thanks @steipete.

• Several isolated session worktrees can prepare concurrently, with abandoned capacity recovered after a creator crash. #166060 Docs · Thanks @steipete.

• Continuing a catalog conversation fails with a retryable message when model loading stalls, rather than waiting indefinitely. #166089 Thanks @ericcaiwx-star.

• Background commands that redirect output to a file can continue the conversation and report the result after finishing. #166123 Docs · Thanks @obviyus, @gkoch02.

• Late completion receipts preserve a newer child answer, intentional silence, or empty reply. #166139 Docs · Thanks @Yun-0000, @steipete, @michaeljgood71.

• Follow-ups about earlier photos no longer receive missing-image instructions that undermine a previous answer. #166165 Thanks @jayzhou2309, @steipete, @nogikeyaksh.

• Cancellation confirmed during preparation consistently tells clients that provider execution never started. #166185 Thanks @shakkernerd.

• Managed-workspace checks do less database work on the Gateway; cross-process listings report cleanup candidates without removing them. #166236 Docs · Docs · Thanks @steipete.

• Completed subagent results can reach healthy requester chats when empty recovery metadata previously blocked admission. #166294 Docs · Thanks @sunlit-deng, @steipete, @Jackten.

• New dashboard conversation and Telegram DM-topic names focus on the supplied first message. #166308 Docs · Thanks @jjjhenriksen, @obviyus.

• Agents stored in included roster files can be deleted while preserving the root include. #166316 Docs · Thanks @jayzhou2309, @jaoa12or, @obviyus.

• Browser chats avoid a repeated final answer after background commands resume and delegate private child work. #166362 Thanks @shakkernerd.

• Sending a message can be stored and acknowledged while unrelated history reads are busy. #166365 Docs · Thanks @steipete.

• Unrelated session writes stop causing false worker workspace-recovery failures. #166375 Docs · Thanks @steipete, @Oldrich333.

• Session setting updates stop waiting indefinitely for a stalled model catalog and can be retried without a late change. #166376 Docs · Thanks @steipete, @dprev.

• Restart recovery hands history and search index repair to its host owner. #166395 Docs · Thanks @steipete.

• Session CLI archive and delete preserve the chosen agent, and trajectory exports resolve a default or sole agent. #166415 Docs · Thanks @steipete.

• Agent binding commands honor configured defaults and account casing, and report incomplete noninteractive creation as failure. #166426 Docs · Thanks @steipete.

• Linux worker hosts can recover container sessions owned by root and return useful sanitized failure details. #166441 Docs · Docs · Thanks @steipete.

• Stop can finish an already accepted abandoned Move, while model selection refuses a session that became locked or was replaced during preparation. #166445 Thanks @steipete.

• Recovered subagent failures and timeouts retain their recorded cause for the parent. #166454 Thanks @steipete.

• Equivalent macOS temporary paths no longer interrupt reply finalization and replay the same inbound message. #166462 Thanks @ericcaiwx-star.

• Subagents using plain session keys list correctly with metadata from their recorded owner. #166550 Thanks @steipete.

• Control UI conversations can resume authorized automation work after a Gateway restart while rechecking current permissions. #166553 Docs · Thanks @romneyda.

• Conversations can continue after a corrupt image instead of repeatedly failing, including after reopening. #166564 Thanks @jayzhou2309, @steipete, @mirasrael.

• Fitting emoji and combining accents stay together at hard cuts in streamed prose replies. #166567 Thanks @rudycelekli, @obviyus.

• Claude CLI-bound sessions can retrieve and page around messages from before a reset again. #166578 Thanks @VACInc.

• Lets follow-up instructions wait for an image completion and delivery instead of failing with a session-change error. #166597 Docs · Thanks @steipete.

• Stops superseded worker setup while retaining session resources until unfinished cleanup settles. #166609 Thanks @RomneyDa.

• Prepares session-owned worker workspaces and preserves their recorded destination during eligible recovery. #166616 Thanks @RomneyDa, @shakkernerd.

• WebChat keeps generated attachments and their reply text in the chat that requested them when sessions are shared with external channels. #166710 Docs · Thanks @wsk19831112-web, @steipete, @ing650423.

• Browser Stop and /stop cancel their request's ordinary commands even after they return a process handle, retaining output and reporting uncertain cleanup. #166753 Docs · Docs · Thanks @shakkernerd.

• Follow-up guidance no longer cancels running tools when its steering receipt is uncertain; input is retained without replay and uncertainty is reported. #166754 Docs · Thanks @RomneyDa.

• Channel /stop ends the selected request's ordinary commands and waits for cleanup while preserving later work and independent background services. #166764 Docs · Thanks @shakkernerd.

• Stop retires an ordinary command's queued or running follow-up so it cannot send a pending late reply after cancellation. #166775 Docs · Thanks @shakkernerd.

• Automatic conversation naming no longer keeps finished chats busy, and history edits wait for completed replies to finish saving. #166784 Docs · Thanks @steipete.

• Parent Stop reaches ordinary commands retained by completed native helpers and routed continuations, reporting cleanup failures while preserving completed results. #166832 Docs · Thanks @shakkernerd.

• Agents see command execution hosts permitted for their session, reducing failures from selecting an unavailable sandbox. #166835 Docs · Thanks @steipete.

• Subagent recovery and cleanup use the recorded child owner so same-named sessions belonging to another agent are preserved. #166845 Thanks @steipete.

• Parents receive interrupted child-task outcomes during Gateway startup, and native-child follow-ups check uncertain effects before retrying. #166846 Docs · Thanks @steipete.

• Compacting, forking or rewinding immediately after a completed reply waits for its remaining cleanup instead of requiring a retry. #166981 Docs · Thanks @steipete.

• Long-running agents count completed background tools as progress even when a later model response hits its output limit, avoiding false stalled-task cancellations. #166997 Docs · Thanks @SunnyShu0925, @steipete, @hannesrudolph, @hippoley.

• Cmd/Ctrl+Enter opens the new conversation; add Shift to start it in the background. #167019 Thanks @vincentkoc.

• Simultaneous conversation-history and model-context reads can wait less on hosts with spare CPU. #167029 Thanks @steipete.

• Large older-message pages avoid reading content that cannot fit, while pagination keeps older messages available. #167032 Thanks @steipete.

• Reloading or reconnecting to long active chats avoids repeated commentary paragraphs and keeps new text visible. #167054 Thanks @steipete, @RomneyDa.

• Independent prompt preparation and reads use bounded parallel work when CPU capacity is available. #167060 Thanks @steipete.

• Conversation searches can proceed alongside other searches while keeping indexing notices accurate. #167163 Thanks @shakkernerd.

• Complete conversation searches stop showing a false indexing notice in the command palette. #167174 Thanks @steipete.

• Chat turns reuse coherent conversation-history reads instead of repeatedly loading the same saved information. #167238 Docs · Thanks @steipete.

• Concurrent session creation in a new shared store reaches normal creation and duplicate-label checks instead of an intermittent discovery error. #167257 Thanks @steipete.

• Chat preparation repeats less database setup, while accepted Memory indexing retains its executor through provider waits and cleanup. #167263 Thanks @steipete, @shakkernerd.

• Saving conversations and checking session maintenance require fewer repeated database calls. #167291 Thanks @steipete.

• Later turns in image-heavy conversations avoid repeatedly decoding and resizing unchanged stored images. #167355 Thanks @steipete.

• Accepted new conversations open while local draft and saved checkout-name cleanup finishes, so the first turn is visible sooner. #167384 Thanks @steipete.

• Conversation branch lists avoid reading superseded message bodies when refreshing after message bursts. #167434 Thanks @steipete.

• Saved conversations stay readable when transcript search is waiting on a stalled database writer. #167446 Docs · Thanks @steipete.

• Opening chats with personal model accounts no longer waits behind unrelated database writes just to display account information. #167447 Thanks @steipete.

• Long conversations can continue with recent whole turns when automatic byte-limit compaction cannot recover, while saved history stays intact. #167505 Docs · Docs · Thanks @steipete, @jaikharbanda, @jlapenna.

• Sent prompts and images stay cleared when a newly confirmed conversation tab is closed immediately. #167507 Thanks @steipete.

• Simultaneous first conversation requests avoid intermittent creation failures on fresh Gateways. #167515 Thanks @steipete.

• Account labels and shared-conversation attribution require less repeated database work on the Gateway. #167520 Docs · Thanks @steipete.

• Session reset, deletion, cleanup and branch listing repeat less database work; protected-history cleanup avoids a false refresh notice. #167549 Docs · Docs · Thanks @steipete, @Patrick-Erichsen.

• Completed replies reach users instead of being replaced by a failure warning after a reasoning-only follow-up. #167567 Thanks @goutamadwant, @hannesrudolph, @obviyus.

• A saved session change stays successful when an ordinary change notification fails, and incomplete cached state is discarded. #167570 Docs · Thanks @steipete.

• Combined session repairs move database work into the background, keep Talk on the selected store, and retain saved-chat identity if display details cannot load. #167578 Docs · Docs · Thanks @steipete.

• Automatic session titles preserve quoted phrases and apostrophes, removing unmatched quote marks after truncation. #167584 Docs · Thanks @steipete.

• Gateway logs stop repeating warnings for old blocked context-engine turns on every later session run. #167595 Thanks @jalehman, @Clode2026, @steipete.

• Turns stop with recovery guidance after three consecutive identical tool errors. #167628 Docs · Docs · Thanks @steipete.

• Chat, history, and Board operations avoid deadlocks when database reads, writes, and usage refresh overlap. #167662 Thanks @steipete.

• Image bursts have more processing capacity, and dashboard asset reads avoid waiting behind unrelated compute work. #167676 Thanks @steipete.

• Preparing ordinary saved chat input avoids an unnecessary conversation-history read. #167682 Thanks @steipete.

• Finishing chat turns avoids repeated session-database preparation. #167685 Thanks @steipete.

• Ordinary chat preparation reuses current session facts to avoid repeated database lookups. #167686 Thanks @steipete.

• Saving ordinary chat turns avoids redundant database reads and nested savepoints. #167690 Thanks @steipete.

• Workspace publication and move recovery repeat fewer database reads while checking current ownership. #167721 Thanks @steipete.

• Diagnostic logging failures release queue slots so later turns can proceed and completed tasks retain their results. #167733 Docs · Thanks @steipete.

• Timed-out history, usage, and storage requests cancel waiting host work so expired requests do not remain queued. #167740 Thanks @steipete.

• An interrupted task's accepted completion reply recovers before later conversation input takes over, preventing silent reply loss. #167814 Docs · Thanks @goutamadwant, @steipete, @marknilsn.

• Typing /stop in a connected Control UI chat cancels the selected session's work and clears queued follow-ups, so they do not immediately restart it. #167860 Docs · Thanks @masatohoshino, @steipete.

• Session details and history reflect committed changes despite delayed or lost worker replies, while newer changes remain protected. #167875 Thanks @steipete.

• Tool-heavy chats download shorter output previews while keeping complete captured results available on demand. #167896 Docs · Thanks @steipete.

• Chat preparation and reply inspection avoid a transcript-read deadlock that could leave them waiting indefinitely. #167901 Thanks @steipete.

• Concurrent session activity avoids repeatedly copying the same configuration when preparing database access. #167915 Thanks @steipete.

• Preparing agent instructions creates less temporary data, including after a conversation is restored. #167922 Thanks @steipete.

• Repository-backed chats spend less time preparing a cold Git worker before their first turn. #167936 Thanks @steipete.

• One conversation's first-turn preparation no longer unnecessarily holds up unrelated conversations on the same gateway. #167959 Docs · Thanks @steipete.

• Required turns that exhaust retries without visible output show a saved try-again notice instead of ending silently. #167972 Docs · Thanks @steipete.

• Session cleanup holds the database writer for less time, reducing interference with creating and archiving conversations. #168013 Thanks @steipete.

• Worktree operations can recover a confirmed committed result when their worker reply is lost, avoiding a false failure. #168025 Thanks @steipete.

• Sending a follow-up while subagents work no longer prevents their accepted results from returning to the parent conversation. #168068 Docs · Thanks @steipete, @stackingrockss, @adamamzalag, @klabir, @a1fred-ai, @oldrich333.

• Messages sent during a subagent or command announcement remain queued and run automatically after its cleanup finishes. #168077 Docs · Thanks @steipete.

• Chats recover from confirmed rolled-back maintenance conflicts, and model fallback keeps the original request intact. #168078 Thanks @steipete, @jacobtomlinson.

• Messages that redirect an active agent stay before the later replies they prompt, including after reconnect or reload. #168099 Docs · Thanks @steipete.

• Parent chats show each unfinished subagent's name and current activity, with a direct link to its details. #168102 Docs · Thanks @steipete.

• Prevents stack-overflow errors when opening session stores for large agent fleets. #168354 Thanks @steipete, @609NFT.

• Keeps existing and new conversations working after restart when the state folder uses a symlink. #168361 Thanks @obviyus.

• Resumes eligible interrupted replies once session initialization or workspace preparation finishes. #168650 Docs · Thanks @steipete.

• Automatically resumes eligible interrupted main chats after restart even when shutdown missed their recovery marker. #168651 Docs · Thanks @steipete.

• Frees node-worker capacity after a forced stop by cleaning up surviving owned child processes. #168746 Thanks @steipete, @aniruddhaadak80, @linhongyu510.

• Waits for an agent to finish loading before starting its replies while ready agents continue serving. #168826 Docs · Thanks @steipete, @xiaozishan.

• Cross-session messages retain their sender labels; the later stable-message repair supersedes the original prompt-prefix tradeoff. #169130 Thanks @RomneyDa.

• Restart-interrupted delegated work returns privately to its original parent with saved context, while cancelled work stays cancelled. #169144 Docs · Thanks @steipete.

• Cancelling a waiting CLI-backed turn takes effect before earlier work finishes, without running the cancelled turn. #169303 Thanks @steipete, @changeroa.

• Session catalog loading times out and recovers when Unix login-shell startup hangs. #169507 Thanks @zhangguiping-xydt, @obviyus.

• Visible subagents launched during an active parent turn can reach their configured model and fallback sequence. #169984 Thanks @steipete.

• Independently launched agent turns stop inheriting another conversation's transcript writer and failing before a reply. #170077 Thanks @steipete, @mariorivas1, @ghbalf.

• Interrupted worker tasks recover before a continuation or queued follow-up starts after a Gateway restart. #170161 Docs · Thanks @steipete.

• Reply text already received from an interrupted native worker can remain in conversation history with a clearer failure reason. #170223 Thanks @steipete.

• Human conversations receiving another session's steering update can resume after restart when their original sender authority survives. #170252 Thanks @steipete.

• A follow-up message preserves an interrupted request's opportunity to resume before taking over the conversation. #170647 Thanks @steipete.

• Gateway state can stay available after sleep or backward clock changes when its original locks are verified. #170790 Docs · Thanks @steipete, @jjjhenriksen, @AnvinaLI, @sgh6688, @waynegault, @andersgreg, @shawnholt.

• Manual --max-lines trimming rebuilds retained search indexes before reporting success; removed transcript text is permanently deleted without an archive. #171134 Docs · Thanks @steipete.

• Long streamed answers that quote tool-call XML in code blocks avoid repeated processing that could delay replies and other sessions. #171573 Thanks @steipete, @boeingchoco.

• Manual compaction works with a small session token cap when the model has a larger native context window. #172457 Thanks @steipete.

• Long technical replies avoid repeated full-answer scans merely for mentioning tools or functions, reducing unnecessary streaming work. #172631 Thanks @JasonBuildAI, @steipete.

• Affected OpenAI live replies no longer repeat their opening because streamed snapshots share mutable text; intentional repetition is preserved. #172653 Thanks @Richard355168, @steipete, @Gavin-ifjc, @Lamer217.

• After a restart interrupts a quiet subagent, an affected parent can get a turn to reconcile unfinished work; this does not automatically relaunch the child. #175278 Thanks @RomneyDa.

• Record memory-forget requests when a reused worker has stale database information 8f7e75a.
• Agents retain readable memory and profile files when workspace directory listing fails. #107921 Docs · Thanks @tzy-17, @altaywtf.
• Post-compaction context estimates include the normal prompt and tool definitions. #141341 Docs · Thanks @ooiuuii, @obviyus.
• Skill Workshop reviews and other hidden helpers skip unnecessary failed Active Memory lookups. #143054 Docs · Thanks @RileyJJY, @steipete, @syfvb, @diegomarino.
• Session documentation clarifies that completed manual and automatic compactions share one counter. #149719 Docs · Thanks @harjothkhara, @ingenuity2k, @obviyus.
• Memory search finds visually identical accented and Korean words across composed and decomposed spellings. #150968 Docs · Thanks @Leon-SK668, @steipete, @Alix-007.
• Corrected embedding-provider settings refresh Memory Core search without restarting the Gateway or rebuilding indexes. #151263 Docs · Thanks @ruel225, @steipete, @Mirksen.
• Keyword memory search remains available during local embedding failures without falsely demanding an index rebuild. #152210 Docs · Thanks @synthalorian, @steipete, @clanksy-code.
• Memory searches keep their search-time allowance while managed Ollama or LM Studio embeddings start. #156483 Docs · Thanks @wangmiao0668000666, @obviyus.
• Memory searches through Tool Search accept their existing score and result-limit aliases. #158632 Thanks @masatohoshino, @obviyus, @he-yufeng.
• Long conversations are summarized before helper-completion and approval-follow-up replies while preserving the selected model. #163117 Thanks @SunnyShu0925, @steipete, @AXEG0, @obviyus.
• Long conversations can continue after an automatic summary timeout by keeping recent context and omitting older unsummarized details from active context. #164246 Docs · Thanks @obviyus.
• Compaction summaries can continue while producing useful output instead of timing out solely for taking longer than one window. #164263 Docs · Thanks @obviyus, @elharonyclaw, @liyangbing, @VACInc, @wynxo.
• Active Memory status separates the session toggle, trigger-recall configuration and cross-conversation memory, and logs missing-agent configuration. #164409 Docs · Thanks @Ayushdevo, @steipete, @davideditelabs.
• Large sessions on affected rollback-journal databases can finish memory indexing instead of repeatedly failing with database-lock errors. #164624 Thanks @triciah73, @steipete.
• Changing one agent memory limit preserves the other configured defaults, including the memory excerpt cap. #164633 Docs · Thanks @steipete.
• Equivalent Korean and Japanese spellings no longer crowd distinct memories out of limited search results. #164660 Docs · Thanks @ly85206559, @obviyus.
• Agents can search already-visible ACP conversation transcripts when the configured coding harness is absent from the native agent roster. #164897 Thanks @sunlit-deng, @techdou, @obviyus.
• Disk-budget cleanup preserves referenced conversation checkpoints whose filenames begin with two dots. #164920 Thanks @steipete.
• Accepted Memory indexing reuses database execution as the Gateway shuts down, preserving final writes with less repeated setup. #164962 Docs · Docs · Thanks @steipete.
• Memory indexing avoids maintaining a redundant index. #165075 Docs · Thanks @steipete.
• Conversation search avoids processing unrelated sessions and discarded excerpts. #165080 Thanks @steipete.
• Conversation-search readiness checks reuse current indexing status instead of repeatedly scanning already-indexed large histories. #165377 Docs · Thanks @steipete.
• New session writes avoid repeatedly storing the full accumulated tool-result trimming metadata, reducing future transcript growth. #165404 Docs · Thanks @steipete.
• Manual and automatic compaction prepare session metadata and transcript statistics through background database workers. #165406 Docs · Thanks @steipete.
• Memory indexing uses the embedding destination instead of inheriting a recognized chat-only subscription endpoint. #165514 Docs · Thanks @dhruv1220, @vit-pavlenko, @obviyus.
• Session keyword recall and memory indexing do less database work on the Gateway thread, with faster recall in a bounded benchmark. #165568 Thanks @steipete.
• Memory correctly identifies cron-generated sessions even when parent links form a cycle or the cron parent was pruned. #165601 Docs · Thanks @steipete.
• Memory index setup and rebuilding move schema checks off the Gateway thread; missing-index status avoids creating temporary stores. #165611 Thanks @steipete.
• Low-memory local-model setup now explains how to choose the existing embedding-only route for local memory search without replacing the chat model. #165832 Docs · Thanks @obviyus, @NovaUnboundAi.
• Conversation search no longer asks users to wait for indexing when messages are already indexed and only database cleanup invalidated the search snapshot. #165879 Thanks @steipete.
• Memory architecture documentation gives the existing automatic recall threshold correctly. #166079 Docs · Thanks @n0mad-ai, @obviyus.
• Tool-loop history refills leave room for system instructions and compaction, reducing overflow interruptions in long conversations. #166120 Thanks @ruel225, @steipete, @igs-rogenlo.
• Manual /compact lets an active answer finish for up to 60 seconds and explains when an interrupted request needs resending. #166163 Docs · Thanks @holny, @steipete, @alexph-dev.
• Wiki searches stop on turn cancellation or a 30-second deadline and report an explicit error. #166329 Docs · Thanks @obviyus, @etzelm.
• Multi-word Wiki searches avoid unrelated page reads when matching metadata supplies enough results. #166355 Thanks @obviyus, @etzelm.
• Scheduled jobs reusing a chat stop cutting existing subagents off from memory. #166417 Thanks @jalehman.
• Long agent turns keep memory tools available during unrelated session activity and ordinary metadata updates. #166419 Thanks @jalehman.
• Gateway warnings explain why a child lost memory after a parent reset and how to respawn it. #166422 Docs · Thanks @jalehman.
• Long conversations can summarize and continue through a configured healthy fallback while the primary model is cooling down. #166459 Thanks @steipete, @hawkit.
• Keeps post-compaction memory indexing from being canceled before it starts and reduces repeated database reads during chat recovery. #166587 Thanks @steipete.
• Prevents automatic memory-saving turns from regaining unrelated tools and changing project files. #166650 Thanks @jlapenna, @steipete.
• Memory tools can continue through routine session bookkeeping without transient access errors, while resets still revoke old access. #166758 Thanks @jalehman.
• Memory preparation repeats fewer eligibility checks, and selected saved skill resources can prepare away from the Gateway thread. #167254 Thanks @steipete, @Patrick-Erichsen.
• The System sessions filter now hides internal Active Memory search rows while keeping the requesting conversation visible. #167357 Thanks @obviyus, @tre1d.
• Compaction references explain that a summary timeout may remove older unsummarized facts from model context. #167640 Docs · Thanks @obviyus.
• Background memory indexing uses the current embedding provider after plugin reloads. #167679 Thanks @steipete, @service-cyber, @jonathanlindsay, @obviyus, @aoclaw-glitch, @holgergruenhagen, @alfred429, @ekinnee, @kringsc.
• Historical dreaming sessions clean up after the selected agent finishes startup preparation, avoiding misleading warnings. #167722 Docs · Thanks @steipete.
• Scheduled memory consolidation and accepted dream diary writes continue across plugin reloads. #167724 Docs · Thanks @steipete.
• New memory files remain searchable after plugin or embedding-provider reloads, and obsolete watchers stop. #167756 Docs · Thanks @steipete.
• Conversation compaction can reconcile recent corrections and completed checks into the main handoff instead of retaining stale current decisions. #167757 Docs · Thanks @jacobtomlinson.
• Slow embedding requests can finish within their existing budget, while deep memory checks stop within a bounded diagnostic attempt. #167891 Docs · Thanks @steipete, @holny, @fonoempresa, @KarbonXX.
• Memory Wiki status reuses a valid cache compiled by another process instead of rebuilding an unchanged vault. #167929 Docs · Thanks @obviyus, @etzelm.
• Memory indexing resumes after crash or upgrade preparation without needing a search to wake it, and dreaming cleanup waits for readiness. #167970 Docs · Docs · Thanks @steipete.
• Slow models can finish Dream Diary entries using the existing agent timeout instead of stopping after one minute. #167984 Docs · Thanks @steipete, @woshicby, @pfrederiksen, @paterskip, @Kaspnov, @SWDadmin.
• Local memory indexing uses less server memory and accepts ordinary chunks with custom and older embedding models. #168020 Docs · Thanks @steipete, @frankhli843, @spunzon, @Deregtx, @jj5638521, @Nehanth.
• Memory indexing explains invalid embedding responses and offers content-free shape diagnostics for remote services. #168035 Docs · Thanks @steipete, @jzmudzinski.
• EmbeddingGemma gets required query and document formatting, with a one-time rebuild of older vectors; remove any previous manual prefix workaround. #168058 Docs · Docs · Thanks @steipete, @polley-mister, @jared-rebel, @ndakota79, @h6y3.
• Daily memory-note saves reject copied blocks already on disk and can retry with only new facts. #168123 Docs · Thanks @steipete, @bit-777, @touraround.
• Keeps completed exchanges in durable conversation context after a follow-up steers a running reply. #168394 Docs · Thanks @steipete, @robertoronderosjr.
• Retains configured workspace rules during compaction even when the Gateway starts elsewhere. #168446 Docs · Thanks @steipete, @jakeaadams.
• Doctor preserves original memory files when invalid UTF-8 prevents a safe merge. #169029 Docs · Thanks @steipete, @wangmiao0668000666.
• Memory search and Forget workers retain less memory while preserving their existing indexing and selection behavior. #172392 Thanks @steipete, @RomneyDa.
• Memory indexes open when first needed instead of all at Gateway startup; populated indexes may refresh asynchronously, so the first search may precede the latest edits. #174718 Docs · Thanks @RomneyDa.

• Supporting worker-inference integration uses explicitly provisioned model endpoints while Gateway permission checks still govern each turn. 80d27bd
• Supporting sign-in integration makes account-specific models available to catalog workers already running at first login. 811d442 Docs
• Overlapping model or credential changes avoid a misleading Gateway restart requirement. ad332df
• Removing provider credentials or logging out avoids unnecessary Gateway restart recovery. b755f03
• Related model-discovery integration refreshes account-specific choices after first login without restarting existing discovery workers. bbdc68b
• OpenAI and Azure Responses turns avoid oversized-output-budget errors by using the model's declared limit. #128282 Thanks @parker-fawcett, @obviyus, @steipete.
• Together model selections using the togetherai prefix resolve through the existing provider. #144946 Docs · Thanks @ly85206559, @obviyus.
• StepFun and Step Plan model selections accept their models.dev provider names. #144951 Docs · Thanks @ly85206559, @obviyus.
• Kilo-prefixed model selections resolve through the Kilocode plugin. #144960 Docs · Thanks @ly85206559, @obviyus.
• Repeated usage reports no longer keep a stalled Chat Completions request waiting indefinitely. #145850 Docs · Thanks @RileyJJY, @steipete, @elharonyclaw, @anyech.
• Reasoning effort controls remain available after a model refresh fails when previously loaded information is still usable. #146155 Thanks @ngo-a, @obviyus.
• Provider model lists stay visible when working keys use protected secret references. #147374 Thanks @pheidon, @steipete, @kybrcore.
• Configured vLLM reasoning choices reach Qwen templates and DeepSeek V4 receives its expected template controls. #150149 Docs · Thanks @leapdragon, @steipete, @krell11.
• Google Vertex setup guidance explains ADC discovery, project/location settings and when a credential marker is needed. #150593 Docs · Docs · Thanks @jayswu, @jayden, @obviyus, @gadgrandez.
• Silent replies and quiet heartbeats contribute token usage and estimated cost to monitoring while staying silent. #152396 Thanks @ly85206559, @MutenCath, @wangmiao0668000666.
• Recognized provider keepalives prevent premature idle failures on managed model streams. #152705 Docs · Thanks @ZengWen-DT, @steipete, @alfredjbclaw.
• Claude CLI receives registered OpenClaw tool names and discovery guidance to avoid failed calls and missed tool-only replies. #153032 Docs · Docs · Thanks @kiranvk-2011, @RomneyDa, @lakemike, @philknows, @destinedenergy, @wbumgardner-max.
• Local chat and embeddings can restore a missing executable from the current managed llama.cpp installation. #153523 Docs · Thanks @samuelreyes982, @steipete, @dh-js.
• Long Claude CLI turns can deliver their final report when repeated streaming wrappers previously exhausted output limits. #153545 Docs · Thanks @zhangguiping-xydt, @methylester, @obviyus, @VACInc, @aaajiao, @telbus, @aoclaw-glitch, @aikohoshinoai-prog, @RTapiador.
• Provider sign-in errors show recovery details immediately and offer a copy button. #155429 Thanks @RomneyDa.
• Codex automation app-access checks keep their request timeout when the computer clock changes. #155855 Thanks @SunnyShu0925, @altaywtf.
• Local inference errors explain reasoning-only responses and suggest lower thinking when the output budget leaves no final answer. #156137 Docs · Thanks @taekop, @obviyus, @constant1n0, @wzg0911.
• Native Ollama tool narration no longer arrives as extra durable channel replies. #156389 Docs · Docs · Thanks @Olli0103, @Solvely-Colin, @steipete, @noamliav, @solutionsbyvance.
• Copilot execution can use accepted reported prompt capacity instead of a synthetic 128K estimate while preserving real limits and user caps. #156636 Docs · Thanks @saju01, @RomneyDa, @snotty, @vinaycodeaxe, @zhangguiping-xydt, @Yigtwxx.
• Gateway logs explain failed Claude Code version detection without exposing raw probe output. #157697 Docs · Thanks @Bruce-Yii, @PashaGanson, @geekoagent, @DaCal90, @altaywtf.
• Short native Ollama plugin calls can disable hidden thinking to avoid empty results when reasoning is off. #159380 Docs · Thanks @YangManBOBO, @steipete, @hsheresd, @sxh313.
• Recognized prompt-size rejections tell users to shorten the prompt or choose a larger-limit model. #159712 Thanks @sunlit-deng, @NovaUnboundAi, @altaywtf.
• Session and context reports retain the last verified model capacity after a qualifying turn. #159731 Docs · Thanks @Yigtwxx, @steipete, @nhhc-lxr, @vinaycodeaxe.
• Recognized retired Ollama models can reach an eligible configured fallback without repeated timeout retries. #160047 Thanks @Irish-Joseph, @benjaml4, @obviyus.
• Existing MiniMax M3.1 Flash Preview controls expose five supported reasoning efforts and avoid rejected disabled-thinking requests. #160089 Docs · Docs · Thanks @NianJiuZst, @RomneyDa, @steipete, @Takhoffman, @php737, @ns-spec, @Hansen1018, @olveww-dot.
• MiniMax OAuth sign-ins can renew expired access tokens in their issuing region. #160459 Docs · Docs · Thanks @masatohoshino, @RomneyDa, @rongtianhua, @mychataa.
• OpenCode Go image descriptions use the selected vision model without a MissingSessionID failure. #160604 Thanks @sxh313, @obviyus, @lian-hu-agent.
• Model scan help now explains that live probes replace configured fallbacks. #161413 Docs · Docs · Thanks @ericcurtin, @obviyus, @Truck0ff.
• Bedrock keeps equivalent tool lists in a stable order, removing one cause of prompt-cache misses. #161838 Thanks @ericcurtin, @steipete, @netbr3ak.
• Volcengine and BytePlus setup examples now match their bundled model catalogs. #161909 Docs · Thanks @Yigtwxx, @obviyus.
• Claude CLI long conversations leave automatic compaction to Claude Code, avoiding extra API-key errors and preserving its session. #162066 Docs · Thanks @rogeriochaves, @llmsommelier, @RomneyDa, @ajmtrz, @MaShAn1027, @Jeehut, @Oldrich333.
• Opening the model picker or sending a new message can restore a failed model catalog without restarting the Gateway. #162901 Thanks @scotthuang, @steipete.
• Managed llama.cpp setup explains unsupported older macOS before downloading models and offers compatible-server or remote-provider alternatives. #163358 Docs · Thanks @NovaUnboundAi, @xXG0DLessXx, @VACInc.
• Opted-in workers use their own provisioned model and credentials for Gateway-authorized requests. #163645 Thanks @RomneyDa.
• Explicitly configured paired workers can make model requests with their own settings while the Gateway retains permission checks and chat history. #163646 Docs · Thanks @RomneyDa.
• TUI model choices follow the current session, and Control UI shows Direct from worker without requiring duplicate Gateway credentials for configured-default worker sessions. #163647 Thanks @RomneyDa.
• CLI diagnostics identify whether background model calls came from titles, recaps or other helpers. #164140 Thanks @dasx, @obviyus.
• Authorized Claude CLI resumes retain announced subagent results so the next turn can use them. #164175 Docs · Thanks @etzelm, @steipete, @potterdigital, @synapse-homeroot, @cassthebandit, @holgergruenhagen.
• Custom Bedrock provider routes load their transport and use their own configured AWS region. #164406 Docs · Thanks @Ayushdevo, @obviyus, @kai-ocean-kurt.
• Fresh LM Studio discovery maps maximum reasoning to the highest advertised canonical effort regardless of option order. #164478 Docs · Thanks @ooiuuii, @steipete, @obviyus.
• Model lookups keep provider spellings separate, and Vertex uses refreshed Google Auth tokens without an extra cache delay. #164520 Thanks @steipete.
• Daybreak speed controls avoid unsupported tiers, and native OpenAI validation failures can recover safely or show clearer diagnostics. #164597 Docs · Docs · Docs · Thanks @VACInc, @steipete.
• Copilot plugin and MCP tools continue working after the first conversation turn. #164598 Thanks @ericcurtin, @takyyon.
• Personal model-account storage runs in background workers and accepted writes settle before Gateway shutdown. #164630 Docs · Thanks @steipete.
• Valid annotated OpenAI tool definitions retain strict validation, OpenRouter rejects invalid cache lifetimes, and retired trajectory sidecars no longer enter exports. #164738 Docs · Docs · Docs · Thanks @steipete.
• Completed Astra answers survive a silent async-tool follow-up and are not resent after streaming. #164774 Thanks @obviyus.
• TypeSafe request rejections prompt the agent to correct its question instead of reporting an unreachable provider. #164835 Docs · Thanks @danhayman, @obviyus.
• Kimi K3 is less likely to abandon Code Mode work because of malformed calls. #165022 Thanks @steipete.
• Interrupted Claude CLI replies no longer save standalone tool markup into chat history. #165031 Thanks @jayzhou2309, @steipete, @ndakota79.
• Model catalog refreshes no longer hold up Gateway callbacks while waiting for database writes. #165034 Docs · Thanks @steipete.
• Finished OpenAI Responses runs release conversation memory while keeping response reuse for follow-up turns. #165142 Thanks @steipete.
• Longer Code Mode tasks continue when providers reuse tool-call IDs; rejected reserved node commands give clearer tool guidance. #165210 Docs · Thanks @steipete.
• Repeated provider call IDs no longer cause false duplicate conflicts in later Code Mode, computer-use and mobile UI steps. #165212 Docs · Thanks @steipete.
• CLI-backed runs preserve their original failure through cleanup even when the rejection value is empty, null, zero or false. #165288 Thanks @steipete.
• Providers reusing tool IDs no longer cause false duplicate publications, missing tool-authored replies or overwritten client tool calls. #165297 Thanks @steipete.
• CLI-backed agents keep Gateway tools available on later turns after the first turn ends. #165428 Thanks @nano-richtera, @steipete.
• Later CLI turns use their own permissions so session, history, and admin tools remain available after the connection's starting task ends. #165432 Thanks @Oldrich333, @Leit-motif, @steipete, @XX441, @aaajiao, @alexph-dev, @Claynetic, @N4DH, @Jeehut, @S1gv4rd, @arifarai, @davidevelyn, @marl0n, @Mydudeclaps.
• A stopped Codex chat can take a follow-up without waiting for other active chats to finish. #165487 Docs · Thanks @steipete.
• Codex provider failures after tool work show their cause and recovery advice instead of appearing as a stopped turn. #165488 Docs · Thanks @steipete.
• Local conversation-save conflicts stop without repeating model requests or switching accounts and models. #165490 Docs · Thanks @steipete.
• Ordinary Codex threads are guided to call tools they already have without an extra discovery step. #165508 Docs · Thanks @mbelinky.
• Claude CLI replies finish after a background agent runs Bash instead of waiting for timeout and potentially repeating work through fallback. #165596 Docs · Thanks @adele-with-a-b, @steipete.
• Gateway-relayed realtime voice can connect to a complete custom compatible endpoint while keeping API keys on the Gateway. #165632 Docs · Thanks @jacobtomlinson, @starship863, @wangyan411303.
• Saving credentials causes fewer Gateway pauses on larger multiagent setups. #165644 Docs · Thanks @steipete.
• Personal OAuth refresh keeps credential persistence off the Gateway thread while preserving account selection and safe shutdown. #165658 Thanks @steipete.
• Large model lists need less repeated Gateway work when clients request them again. #165708 Thanks @steipete.
• Codex-backed chats exclude the bundled visualization instructions that can produce unsupported raw markers, while keeping OpenClaw widgets. #165728 Docs · Thanks @RomneyDa.
• Claude CLI failures show the process exit reason and redacted diagnostics instead of a misleading broken-pipe error. #165769 Docs · Thanks @steipete.
• Gemini and Vertex AI responses that disconnect inside a streamed event can use existing automatic retry and recovery instead of ending with partial text. #165870 Docs · Thanks @ericcurtin, @masatohoshino.
• Large Codex conversation histories take less local work to page and filter after activity, especially when matching titles are readily available. #165911 Docs · Thanks @steipete.
• Worker-local inference setup errors identify what to fix and which machine owns the configuration. #165960 Docs · Thanks @RomneyDa.
• Codex stays selectable after startup for eligible official OpenAI Responses API-key routes. #165979 Docs · Thanks @steipete.
• Empty model-stream keepalives no longer postpone normal timeout retry and fallback until the whole turn expires. #165983 Thanks @steipete, @Irish-Joseph, @cestercian, @rmyers64, @obviyus.
• Ultrafast remains selectable after OpenAI serves a slower tier, with temporary feedback explaining the mismatch. #165990 Docs · Thanks @steipete.
• Status and session reports follow discovered model and runtime context limits and show unknown capacity explicitly. #165991 Docs · Docs · Thanks @snotty, @steipete.
• Codex session-catalog decoding uses less worker memory. #166007 Thanks @steipete.
• CLI-backed turns can recover once when a prior reply updates saved history during preparation. #166009 Thanks @Oldrich333, @steipete.
• Unrelated session writes no longer abort valid Codex turns through a state-read race. #166010 Thanks @steipete.
• Saved-session model pickers respond faster and show recovered native runtime choices again. #166037 Docs · Thanks @steipete.
• Compatible Codex choices remain selectable after background agent database recovery. #166081 Thanks @steipete.
• Compatible providers that omit token usage can use approximate context estimates for active sessions and compaction. #166151 Docs · Thanks @fede-kamel, @steipete, @DigitalCharon.
• GPT-6 Astra receives the existing instructions for carrying out tasks and reporting results in OpenClaw-assembled prompts. #166152 Docs · Thanks @obviyus.
• Claude CLI plugin completions, including Memory Dreaming, honor the configured eligible account order. #166173 Docs · Thanks @sunlit-deng, @steipete, @Fealves0.
• Affected official OpenAI API-key tool turns retain reasoning by using the existing Responses transport. #166182 Docs · Thanks @obviyus, @astutoai.
• Claude Code login changes become available in model lists without restarting the Gateway. #166305 Docs · Thanks @obviyus.
• Claude CLI models stop appearing twice when both entries represent the same configured route. #166311 Docs · Thanks @obviyus.
• Claude Code setup guidance explains primary-runtime use and canonical model configuration. #166319 Docs · Docs · Docs · Thanks @obviyus.
• Configured models stay visible with sign-in guidance when their authentication is missing. #166338 Docs · Thanks @obviyus.
• Standalone refreshed model lists recognize configured Claude models with a valid Claude Code login. #166361 Thanks @obviyus, @hippoley.
• Background CLI sign-in checks avoid rebuilding the prepared model catalog while preserving fresh login status and saved-profile authority. #166434 Docs · Thanks @steipete.
• Preserves a valid native CLI conversation after a resumed reply sends progress and is stopped or times out. #166658 Thanks @VACInc.
• Accepts external ACP agent model selections in config commands while retaining native fallback and inherited-default validation. #166662 Docs · Thanks @jayzhou2309, @ftlc-ian, @obviyus.
• Allows ACP reset-target lookup to use the default account when optional channel configuration is absent. #166685 Thanks @steipete.
• Prevents closed Codex threads with background terminals from blocking subsequent messages, and makes stuck startup waits cancellable and bounded. #166693 Thanks @jalehman.
• Restores model-catalog generation when the upstream metadata feed exceeds its previous download limit. #166699 Thanks @obviyus.
• Honors configured OpenRouter thinking levels and model defaults on the first turn of a fresh process. #166705 Thanks @obviyus.
• Supporting Haiku 5.5 API and Claude CLI behavior covers thinking, conversation replay, and pricing for model access already available in 2026.10.1. #166706 Thanks @RomneyDa.
• Catalog recommendations are centrally curated; plugin manifest recommendation lists are now ignored. #166737 Docs · Thanks @obviyus.
• Codex forks and catalog continuations preserve native conversation links, and native web search respects persistent and temporary tool permissions. #166787 Thanks @RomneyDa.
• Reopening the same model picker preserves expanded provider groups while clearing search text. #166821 Docs · Docs · Thanks @patrick-erichsen.
• Long ChatGPT responses can continue beyond 16 MiB of total streaming traffic. #166853 Docs · Thanks @Jackten, @steipete.
• Codex overload retries retain the current request and completed work instead of answering an older task. #166868 Docs · Thanks @steipete, @Baumus.
• Astra can wait for parallel tool results before completing its answer, preserving the requested full response. #166876 Docs · Docs · Docs · Thanks @obviyus.
• Explicit configured model choices keep the requested model when an alias collides; unsupported agent-scoped model commands explain how to correct the command. #166983 Docs · Thanks @steipete.
• Chats keep an explicitly selected compatible provider account through temporary credential loss and restoration. #166985 Docs · Thanks @vincentkoc.
• OpenAI chats can reply with affected MCP or plugin tools enabled when their input patterns use lookarounds. #166999 Docs · Thanks @scs0209, @adamjgmiller, @obviyus.
• Model pickers reuse the completed catalog without repeated discovery and share the same model inventory used for execution. #167008 Docs · Thanks @steipete.
• Known account model choices and speed tiers remain visible during a pending refresh; failed production refreshes can still lose those choices. #167009 Docs · Thanks @steipete.
• Stalled provider reconnects reach retries promptly instead of holding a turn for minutes. #167013 Thanks @steipete, @sszyh2026, @Yigtwxx.
• Codex explains when failed history compaction prevented sending a message and tells users to start a new session and resend. #167016 Docs · Thanks @steipete, @jaikharbanda.
• Repeated authentication reads avoid unnecessary reader restarts when callers select the same state directory in equivalent ways. #167024 Thanks @steipete.
• ChatGPT subscription replies interrupted by connection expiry can reconnect and finish from saved conversation history. #167063 Thanks @steipete, @jackten.
• Refreshing model lists from several providers overlaps independent requests to shorten waiting. #167101 Docs · Thanks @steipete.
• Changing one provider's configuration no longer hides unrelated automatically discovered models. #167117 Thanks @obviyus.
• Model listing explains rejected or unavailable provider catalogs and points to recovery steps while preserving script output. #167128 Thanks @tianhaotian, @steipete, @ClawMoew.
• Model choices and prices stay available while a downloaded catalog finishes provider discovery before being adopted. #167178 Thanks @obviyus.
• Codex shows an identical ultrafast policy warning once per stored chat instead of repeating it across turns and reconnects. #167220 Thanks @jacobtomlinson.
• Model choices recover automatically after temporary provider-discovery failures, without a manual refresh. #167230 Thanks @obviyus.
• Claude CLI conversations keep their latest history when switching between channel messages and web chat. #167241 Docs · Thanks @steipete, @cmatrx0.
• Background model discovery avoids loading unnecessary native agent execution code while retaining model choices. #167282 Thanks @romneyda.
• Expected caller cancellations stop appearing as provider failure warnings while timeouts and genuine failures retain warnings. #167303 Thanks @steipete.
• Chats using strict model templates can continue after runtime notices without clearing their history or switching providers. #167309 Thanks @jlapenna, @steipete.
• Opening a personal-account model list uses published choices immediately; Refresh obtains account-specific catalog details. #167334 Thanks @steipete.
• Preparing selected plugin model catalogs avoids loading unrelated saved catalog payloads. #167359 Thanks @vincentkoc.
• Browser and terminal model pickers put current and recommended choices first, with All models and full-list search keeping other choices accessible. #167370 Thanks @obviyus.
• Codex connection-loss failures remain in conversation history after partial output, reloads and reconnects, with one consistent live error. #167383 Docs · Docs · Thanks @steipete.
• The model-discovery change that advertised the latest published Codex version is superseded by the bundled-version correction below. #167387 Thanks @obviyus.
• The model picker's All models count label gains translations in 20 languages. #167420
• Hosted-only Ollama setup now uses the dedicated Ollama Cloud choice, with old local-provider cloud URLs rejected before reset. #167433 Docs · Docs · Thanks @obviyus.
• Model choices appear under the right provider with clearer labels and predictable expansion. #167438 Thanks @obviyus.
• xAI models appear once in the model list while existing saved selections continue to work. #167444 Thanks @obviyus.
• Concurrent agent runs retain less duplicate model-catalog data. #167454 Thanks @steipete.
• Gateway restarts restore missing model choices through background discovery without requiring a manual Refresh. #167458 Thanks @obviyus.
• Known provider models appear immediately after sign-in and remain available while the first account listing is pending or fails. #167460 Thanks @obviyus.
• OpenAI and xAI discovery can show newly listed chat models without waiting for an OpenClaw catalog update. #167462 Docs · Docs · Docs · Thanks @obviyus.
• Newly pulled self-hosted Ollama chat models appear after Refresh while saved model settings stay intact. #167463 Docs · Thanks @obviyus.
• Catalog updates no longer interrupt an already admitted reply or scheduled turn before its first model request. #167492 Docs · Thanks @steipete, @jalehman.
• ChatGPT-backed tasks can recover unfinished responses when a connection closes normally before the answer finishes. #167512 Docs · Thanks @ericcurtin, @obviyus, @mightyx, @aleps001.
• Long image conversations preserve more reusable prompt content between image-cleanup batches. #167613 Docs · Thanks @steipete.
• Managed llama.cpp setup allows healthy Metal initialization to finish instead of timing out prematurely. #167630 Docs · Thanks @steipete.
• ChatGPT discovery uses OpenClaw's bundled Codex version, reversing earlier advertisement of a newer client that could expose unusable models. #167633 Docs · Thanks @obviyus.
• Direct Anthropic discovery can admit newly listed Claude models and shape thinking requests from readable advertised capabilities. #167658 Docs · Thanks @obviyus, @fuller-stack-dev, @NOVA-Openclaw.
• Eligible provider routes can reuse the original prompt prefix when skills, memory, or session instructions refresh. #167698 Docs · Docs · Thanks @steipete.
• A corrective retry after a long model response gets its recovery window measured from the retry's own start. #167708 Docs · Thanks @steipete.
• Resumed Claude CLI conversations receive updated instructions and changing plugin context while retaining their history. #167817 Docs · Thanks @yu-xin-c, @steipete, @robburke.
• Refreshed model catalogs remove retired models from pickers and recommendations while retaining exact configured identifiers. #167855 Docs · Thanks @obviyus.
• Oversized proxy requests recover or report a context error instead of returning an automatically squeezed one-token answer. #167878 Docs · Thanks @steipete, @jacobtomlinson, @amittell, @glebsergeevich, @dankarization.
• Gemma 4 can recover complete tool requests returned as plain text instead of showing their raw syntax. #167886 Docs · Thanks @steipete, @jlapenna.
• Prompt-cache warnings compare the right session even when background reviews use the same provider. #167890 Docs · Thanks @steipete.
• Subagent completion keeps ordinary conversation guidance stable so unchanged prompts can reuse more cached input. #167898 Docs · Thanks @steipete, @Eragim, @apalmiere.
• Small-context conversations can continue after automatic compaction, and messages too large to fit receive a useful size notice. #167900 Docs · Thanks @steipete.
• Supported macOS gateways can reclaim a precisely matched orphaned managed llama.cpp server and start an owned replacement. #167902 Thanks @steipete, @TomSchuman.
• Long conversations need less summarization work before continuing because oversized histories use a bounded sample. #167916 Docs · Docs · Thanks @obviyus, @elharonyclaw, @wynxo.
• Affected Ollama Qwen models can execute commands with Tool Search enabled instead of repeatedly misreading tool names. #167933 Docs · Thanks @steipete.
• Plugin-added prompt context stays with its original user turn on later requests and after reopening a conversation. #167938 Docs · Docs · Thanks @steipete.
• Listed Copilot models use supported request routes, and eligible newly listed Claude models receive compatible thinking settings. #167941 Docs · Thanks @obviyus.
• Claude CLI choices exclude API-only models that fail when selected through the CLI. #167942 Thanks @obviyus.
• Grok subscription thinking choices follow the account's advertised capabilities and send the corresponding effort. #167949 Docs · Thanks @obviyus.
• Claude CLI sign-in routes newly listed Claude IDs through Claude Code; existing sign-ins use the later update/Doctor routing migration. #167963 Docs · Thanks @obviyus.
• ChatGPT-only model pickers omit pro models missing from the selected account's successful listing. #167975 Docs · Thanks @obviyus.
• NVIDIA's featured models lead its hosted recommendations, followed by remaining OpenClaw suggestions. #167989 Docs · Docs · Thanks @obviyus.
• Configuration commands explain which newly entered model was rejected and why, while keeping the current configuration intact. #168009 Docs · Thanks @steipete, @azuretek.
• Models offered by a ChatGPT account can run through its sign-in on the default Codex runtime even before built-in route lists catch up. #168033 Thanks @obviyus.
• Refreshing LM Studio discovers newly installed models while preserving saved names and context limits. #168039 Docs · Thanks @steipete, @jb-lopez.
• LM Studio gives slow loads more time and stops with clear guidance when a required larger instance cannot load. #168044 Docs · Thanks @steipete.
• ClawRouter makes fewer repeated model-list requests, and Refresh fetches current models and grants. #168050 Docs · Thanks @steipete.
• Long Codex sessions avoid another unnecessary OpenClaw compaction after native-only compaction leaves the transcript unchanged. #168051 Docs · Thanks @ruel225, @steipete, @xydigitLybnnnn, @robertoronderosjr.
• ChatGPT can recover a specific rejected continuation while retaining completed tool results instead of repeating the work. #168055 Docs · Thanks @steipete, @nikosml.
• Thinking selections now reach compatible custom reasoning models through OpenAI-compatible chat requests. #168056 Docs · Thanks @steipete, @j-nolan.
• Disabling streaming now works for compatible chat servers while retaining reasoning, usage and tool calls in completed replies. #168057 Docs · Thanks @steipete.
• Rerunning llama.cpp setup offers the selected model for download instead of substituting a hardware recommendation. #168064 Docs · Thanks @steipete.
• Earlier Claude CLI sign-ins can repair model routing through update or doctor --fix instead of falling through to a missing API key. #168074 Docs · Docs · Thanks @obviyus.
• Gemini keeps partial replies when a generation limit is reached and avoids running incomplete tool calls. #168103 Docs · Thanks @steipete.
• Memory-saving runs keep the original conversation when provider compaction is enabled; this release retains that fix without enabling Anthropic compaction by default. #168164 Docs · Docs · Docs · Thanks @obviyus.
• llama.cpp setup honors the selected model through aliases and stops visibly if it cannot save a required server credential. #168232 Docs · Thanks @steipete.
• Model requests that never return response headers reach OpenClaw's timeout and recovery path instead of waiting for longer transport defaults. #168282 Docs · Thanks @steipete, @Yigtwxx.
• llama-server router refresh supplies verified model limits and tool support, and agent requests use those facts instead of guessed budgets. #168309 Docs · Thanks @steipete, @guarismo.
• Claude CLI models stay in the picker after another plugin is enabled or agent settings change, without needing a manual refresh. #168315 Thanks @obviyus.
• Cleans up unused Codex processes after model discovery times out while ongoing work finishes. #168365 Docs · Thanks @steipete, @609NFT.
• Recovers missing endings in Codex replies when the completed answer extends the streamed text. #168400 Thanks @RomneyDa.
• Allows affected scheduled jobs and chats to start when the model catalog changes during startup. #168410 Thanks @RomneyDa.
• Avoids false sign-in errors and provider penalties caused by numbers in malformed tool-call diagnostics. #168500 Thanks @TYY-2005, @steipete, @r3n3x.
• Sends runtime context in a supported message role to compatible and local Responses servers. #168546 Docs · Thanks @steipete, @flanker-stack.
• Honors thinking levels advertised by the admitted provider catalog across chats and scheduled runs. #168566 Docs · Thanks @obviyus.
• Discovers provider models after an API key is saved using the actual saved credential. #168617 Thanks @obviyus.
• Keeps permitted native and MCP tools available after Claude CLI background commands finish. #168670 Docs · Thanks @steipete, @Oldrich333.
• Keeps chat text consistent between first send and later replay when surrounding whitespace is present. #168725 Thanks @steipete.
• Refreshes OAuth model catalogs despite historical or empty agent databases without deleting agents or signing in again. #168957 Thanks @obviyus.
• Lets model requests try working IPv6 routes when IPv4 fails. #168973 Docs · Thanks @steipete, @an2in, @georgenaz, @motookoyama, @orthogonalistic.
• Cross-session messages keep stable model-visible text from first send through replay, preserving prompt-cache eligibility and sender context. #169238 Docs · Thanks @RomneyDa, @TourAround.
• Supported ChatGPT-authenticated Codex conversations retain explicitly selected Ultrafast on resumed and follow-up turns. #169265 Docs · Thanks @SanderHelgesen.
• OpenAI models can be selected without the Codex plugin when the existing OpenClaw runtime fallback can run them. #169412 Thanks @obviyus.
• Compatible Codex follow-ups resume after failed thread preparation while native background commands continue. #169431 Thanks @jjjhenriksen, @RomneyDa.
• Explicit cache-TTL pruning works for OpenRouter Grok and supported Gemini 2.5/3 models. #169548 Docs · Thanks @RomneyDa, @gregb100.
• The next response waits for saved history to become readable after native compaction instead of failing immediately. #169583 Docs · Thanks @obviyus.
• DeepSeek-format replies hide internal tool markup and recover valid managed tool calls instead of displaying payloads. #169728 Docs · Thanks @steipete, @TOMUIV.
• Personal-account quota checks and OAuth refresh stop aborting replies over harmless differences in unset stored fields. #170065 Thanks @steipete, @modlethub, @Yigtwxx.
• Supporting Meta model controls retain Muse Wonder and max reasoning for eligible Wonder and Spark models already available in 2026.10.3. #170232 Docs · Thanks @RomneyDa.
• Repeated Talk questions after tool use avoid the identified Anthropic message-ordering failure while retaining changed instructions. #171267 Thanks @LiuwqGit, @steipete, @marleyman5000-boop.
• New Sonnet5.5 usage estimates use the corrected default cached-input price. #171723 Thanks @vincentkoc.
• Gemini4 Argon can be selected through Google and Vertex, with supported thinking controls and continuing tool conversations. #171852 Docs · Thanks @steipete.
• Google adds Gemini 3.8 Flash and updates Gemini aliases; rerunning API-key setup selects Gemini 4 Argon as the existing primary model. #172339 Docs · Thanks @steipete.
• CLI-backed agents avoid a connection-reuse race that could drop tool calls or message sends before they reached the Gateway. #172980 Thanks @steipete, @Oldrich333.
• Claude CLI conversations can deliver their first reply when startup hooks emit many events instead of stalling initialization. #173011 Thanks @ericcaiwx-star, @steipete, @sopa-gpters.
• OpenRouter setup stops rejecting valid keys or saved model choices because discovery and activation disagree about the model name. #173292 Thanks @chandraveshchaudhari, @obviyus.
• Tasks can recover from the OpenAI SDK's malformed server-sent-event JSON error without requiring resubmission, while reusing recorded tool results. #173404 Docs · Thanks @steipete.
• Codex one-shot commands keep completed answers and payloads if later cleanup fails, helping automation distinguish finished work from a failed task. #173516 Docs · Thanks @RomneyDa, @Compassboot.
• Pressing Stop during an early Codex session handoff can preserve the healthy replacement process for the next turn and other sessions sharing it. #173580 Thanks @LLagoon3, @jalehman, @RomneyDa.
• Google Interactions preserves multiline streaming replies and reports malformed streams instead of silently dropping content and appearing successful. #173612 Docs · Thanks @steipete.
• Claude CLI readiness checks recognize login credentials explicitly preserved for live CLI execution, allowing the affected child-agent admission path to proceed. #175114 Thanks @RomneyDa.

• Tlon preserves complete link and image URLs containing balanced parentheses. 193f7e6.
• Incoming and replayed channel messages wait for their agent to be ready after startup instead of failing prematurely. 8e3c6be Docs
• Discord's editable preview includes the last character of a complete batch without an unnecessary wait. 9df0cd0.
• Feishu group subjects beginning with # keep their group metadata. #121463 Docs · Thanks @zw-xysk.
• Matrix recovery rejects corrupted saved messages instead of sending altered text. #133338 Thanks @RileyJJY, @obviyus.
• Groups receive recovery guidance when saved conversation history cannot load instead of typing followed by silence. #135253 Docs · Docs · Docs · Thanks @izshaw, @steipete.
• Outbound audit history retains completed-send records for affected constructor-prefixed destinations. #137763 Thanks @teddytennant, @obviyus.
• Overlapping Voice Call questions receive a clear busy response instead of the previous request's answer. #145776 Docs · Thanks @tianhaotian, @Leon-SK668, @steipete, @rmyers64.
• Telegram polling starts correctly from source checkouts and captured-plugin installations. #154286 Docs · Thanks @xydt-juyaohui, @steipete, @markfietje.
• Short Slack channel names such as general resolve to their actual rooms so allowed messages can receive replies. #155882 Thanks @MertBasar0, @obviyus, @jared-rebel.
• Normal Slack logs show a sanitized reason when an unresolved mention is skipped. #157953 Thanks @Olli0103, @shannon0430, @obviyus.
• Malformed LINE question-button replies no longer start unintended agent turns. #158694 Docs · Thanks @Alix-007, @obviyus.
• Teams and Zoom meeting commands reject explicitly empty timeout values before contacting the Gateway. #158695 Thanks @Alix-007, @obviyus.
• Newer Discord corrections can reach reply processing while an earlier follow-up waits. #159265 Docs · Docs · Thanks @steipete, @aatreya.
• Channel tools remain usable after the agent or operator task that started the account ends. #159396 Thanks @ceckert, @RomneyDa, @slinkoff, @fuller-stack-dev.
• Long IRC replies with CJK text or emoji leave room for server forwarding so chunks are not lost. #159730 Thanks @rodricksz4h5, @obviyus.
• Telegram node-command results and error follow-ups stay in the conversation and topic that started them. #160675 Docs · Thanks @brokemac79.
• Agent deletion guidance explains retained bot accounts and when queued messages may reach a new owner. #161427 Docs · Docs · Thanks @charlesBerg-ux, @obviyus.
• Matrix warns when invite allowlist entries cannot match a room. #161607 Docs · Docs · Thanks @xydigitLybnnnn, @captain-marlow, @obviyus.
• WhatsApp can deliver a later decrypted message after an initial empty encrypted placeholder instead of discarding it as a duplicate. #162078 Thanks @stanleys12, @obviyus, @hundredhandsio, @jmejia-appropia, @gabrielrojasc.
• Broken bundled channel media support reports its original initialization error instead of silently losing attachment-directory declarations. #162650 Thanks @wangmiao0668000666, @obviyus.
• Slack places subsequent streamed progress and answers below a follow-up human message while preserving earlier output and Stop behavior. #163511 Thanks @Olli0103, @Patrick-Erichsen, @leovoloshin, @Alix-007.
• Long mostly-ASCII replies require less CPU work to split for channel delivery. #163573 Thanks @boeingchoco, @obviyus.
• Slack no longer floods Gateway logs with two irrelevant ping/pong diagnostic warnings. #164305 Thanks @jayzhou2309, @Johannes-Berggren.
• Talk agent consultations can start when finalized speech updates shared history during startup. #164372 Docs · Thanks @sunnyshu0925, @steipete, @samiralibabic.
• Incoming channel pairing checks move database work off the Gateway thread while keeping sender permissions unchanged. #164476 Docs · Thanks @steipete.
• An outdated or closed channel setup wizard no longer starts unwanted WhatsApp QR pairing after cleanup. #164577 Thanks @steipete.
• Telegram keeps a visible streamed preview until its replacement is accepted rather than letting it disappear while waiting. #164649 Docs · Thanks @jayzhou2309, @obviyus, @jmar5439.
• Internal relative attachment instructions stay hidden in chat replies that use carriage-return line breaks. #164657 Thanks @ly85206559, @obviyus.
• Clean UTF-16 text and JSON attachments retain their declared type instead of being guessed as CSV because they contain commas. #164687 Thanks @yetval, @obviyus.
• Public IPv6 HTTPS media references pass attachment parsing while private destinations stay blocked. #164689 Docs · Thanks @yetval.
• Voice preference preparation uses the existing worker while preserving selected paths and fresh preference contents. #164694 Thanks @steipete.
• Eligible background-command results return to the requesting chat with its conversation history. #164719 Docs · Docs · Thanks @obviyus, @SparcleAI, @arduano, @igs-rogenlo, @wpu911, @jalehman.
• Chained conversation-owned command results no longer wait for the next periodic heartbeat interval. #164756 Docs · Thanks @obviyus, @ghostllm, @hexsprite.
• Matrix drops a pre-March raw-sync cache reader while retaining supported normalized saved state from July onward. #164816 Thanks @steipete.
• Tlon messages turn bare URLs into links and preserve Markdown images embedded in formatted text. #164828 Thanks @jayzhou2309, @yetval, @obviyus.
• Loaded history retains useful command status and spawned-session information; channel listeners continue using their own Gateway after the starting client disconnects. #164848 Docs · Thanks @sjf-oa, @sjf.
• The agent is told when an inter-session reply was recorded only in a transcript and was not delivered to an external channel. #164860 Docs · Thanks @jayzhou2309, @obviyus, @Jackten.
• Completion checks can keep the original detailed answer when it already explains an approval request or other concrete pause. #164896 Docs · Thanks @obviyus.
• A completed message command keeps its result when a plugin shutdown hook times out, reporting deferred cleanup instead of an exit13 failure. #164915 Docs · Thanks @steipete.
• Voice Call continuation waits use elapsed time, avoiding early or overlong timeouts when the calendar clock changes. #164951 Thanks @SunnyShu0925.
• WhatsApp media uploads on Bun retain proxy routing and work through redirects. #165017 Docs · Thanks @steipete.
• IRC rejects oversized incoming records and spaces out repeated reconnect attempts. #165018 Thanks @eleqtrizit.
• Ordinary Gateway logs now explain more attachment staging failures. #165066 Thanks @akshatshahh, @someone378, @obviyus.
• Telegram streaming replies keep the original message after temporary server errors instead of leaving duplicate answers. #165072 Thanks @steipete.
• Meeting-library page reads leave the Gateway thread and respect access revoked while a page loads. #165111 Thanks @steipete.
• Discord streaming previews preserve paragraph spacing and continuous code blocks across chunks. #165120 Docs · Thanks @steipete.
• Telegram message sends and inline buttons work after chat handoffs without an invalid inferred reply target. #165137 Thanks @mbelinky.
• Approved X accounts can mention configured bots and receive text reply chains in the public conversation. #165138 Docs · Docs · Thanks @steipete.
• Matrix accepts and retains valid recovery keys even when verification does not newly enable a backup. #165236 Thanks @steipete.
• Removing stored X permission stops mentions and replies still being authorized from reaching dispatch or posting a public response. #165244 Docs · Thanks @steipete.
• Slack membership caching now uses a fixed 60-second expiry, removing the environment override previously used to disable or adjust it. #165330 Thanks @steipete.
• X mention streaming uses the correct subscription token and accepts X's documented event envelope, with clearer reasons when it falls back to polling. #165337 Docs · Thanks @steipete.
• WhatsApp QR startup deadlines return their expected failure without an unhandled rejection, and failed worker reconnect preparation clears its timer. #165340 Docs · Thanks @steipete.
• X accounts gain configurable API spending budgets, visible cost estimates and pauses that preserve mentions for processing after reset. #165351 Docs · Thanks @steipete.
• Optional X guest mode answers public repository questions in isolated sessions with restricted read-only tools and daily admission limits. #165354 Docs · Thanks @steipete.
• X streaming periodically checks for missed mentions without replying twice to posts already received, and reports repeated unreadable events. #165376 Docs · Thanks @steipete.
• Matrix attachments without captions no longer pass their filenames off as typed messages. #165419 Thanks @ooiuuii.
• Images split across Markdown lines can arrive as attachments with captions and ordering preserved. #165439 Docs · Thanks @Milbaxter.
• X guests can use hidden helpers for repository questions with the same restricted access. #165543 Docs · Thanks @steipete.
• Talk finds open calls without scanning retained closed-call payloads and preserves calls resumed during stale recovery. #165546 Docs · Thanks @steipete.
• Delegated realtime calls can end their current call, delay outbound greetings until pickup speech, and optionally hang up after speech inactivity. #165715 Docs · Thanks @mbelinky.
• Give a phone call a task brief, follow and steer the conversation, and receive an outcome report; optional callbacks and Twilio voicemail handling support follow-up. #165716 Docs · Docs · Docs · Thanks @mbelinky.
• IRC messages and NickServ passwords keep literal backslashes, including paths and code snippets. #165719 Docs · Thanks @Sedrak-Hovhannisyan, @obviyus.
• IRC login accepts server passwords containing spaces or beginning with a colon. #165721 Thanks @Sedrak-Hovhannisyan, @obviyus.
• Messages waiting on startup database checks stop before dispatch so their channel can retry, and session-sharing reads can recover once from concurrent changes. #165738 Docs · Thanks @RomneyDa.
• Replies explaining pipe operators keep the paragraphs between them instead of silently losing text. #165753 Thanks @Sedrak-Hovhannisyan, @obviyus.
• Photos sent as WhatsApp documents reach vision-capable models after their downloaded bytes are recognized as images. #165774 Thanks @yetval, @Takhoffman.
• Photos sent as LINE files reach vision-capable models while actual documents remain documents. #165778 Thanks @yetval, @Takhoffman.
• Default logs explain why a channel message did not join the running turn and whether it entered the follow-up queue. #165876 Docs · Thanks @steipete, @PilonQV, @VACInc.
• Plugin channels accept queue overrides; X guest questions must wait rather than alter an active reply. #165965 Docs · Docs · Thanks @steipete.
• Telegram progress cards stay visible while delegated work continues and the parent resumes. #166101 Docs · Docs · Docs · Thanks @Kimiyu-186, @steipete, @obviyus.
• Answered voice calls retain their remaining talk time after a restart; later transcript lines can resume without replaying an uncertain pre-crash batch. #166431 Thanks @steipete.
• Long formatted chat replies spend less time being split before delivery, including Telegram previews. #166493 Thanks @boeingchoco, @obviyus.
• Channel status returns recorded health promptly, while explicit live probes share a bounded deadline and preserve sibling results. #166538 Docs · Docs · Thanks @steipete.
• The mock voice-call provider preserves valid caller details so developers can simulate inbound call lifecycles. #166552 Thanks @ly85206559, @vincentkoc.
• Skips outdated pending Feishu streaming previews so completed replies do not wait behind every intermediate update. #166610 Thanks @rudycelekli, @feie22, @cas1per, @L-Trump, @xiaobao-k8s, @csxjihk.
• Discord account health inspection avoids loading activity and voice code that could delay Gateway connections and approvals. #166736 Thanks @RomneyDa.
• Eligible long replies using native reply instructions stream faster by avoiding repeated parsing of the growing answer. #166782 Thanks @boeingchoco, @obviyus.
• Telegram and Discord show useful temporary progress for ongoing and delegated work, with Discord progress enabled by default. #166861 Docs · Docs · Docs · Thanks @obviyus, @samiralibabic, @dankarization.
• Successful emoji reactions stop adding redundant work rows to Slack progress cards while failed or unresolved reactions remain visible. #166880 Thanks @obviyus, @notmnp.
• Long Slack and IRC replies keep emoji and accented characters together when split into messages. #167025 Thanks @steipete.
• Doctor preserves Discord owner DM targets and can restore lost heartbeat routing from matching backups; the later typed-owner fix covers discord:user:<id>. #167059 Thanks @steipete, @jaiden0.
• CLI sends that expire while queued explicitly report confirmed non-delivery and advise retrying with a longer timeout. #167106 Docs · Thanks @steipete, @A1fred-AI.
• Messages sent into a busy session wait until its current reply and delivery finish, preserving queued order. #167126 Thanks @holgergruenhagen, @steipete, @kseczhouyb.
• Tlon links and images keep their full destination when an address contains balanced parentheses. #167191 Thanks @ly85206559, @obviyus.
• Pairing administration moves database writes away from message handling and correctly removes older approved-sender IDs padded with spaces. #167239 Thanks @steipete.
• Mattermost stops retrying permanent refusals even when error details mention temporary failures. #167244 Thanks @Leon-SK668, @obviyus.
• Feishu quotes, fetched thread context and forwarded text preserve readable mention names instead of opaque placeholders. #167264 Thanks @Leon-SK668, @Raven-wy, @Takhoffman.
• Channel progress from Claude and Gemini CLI runs uses familiar tool labels and preserves authored execution titles. #167274 Docs · Thanks @VACInc.
• Chained background-command results stay in the Control UI chat that started them instead of falling back to the heartbeat destination. #167284 Docs · Thanks @Marvinthebored, @Peetiegonzalez, @steipete, @oywino.
• Telegram and Discord progress cards remain visible during background media generation and retain a failed status if results fail to arrive. #167362 Thanks @obviyus.
• Telegram detailed progress rows regain recognizable tool emoji, including retained background progress cards. #167392 Thanks @obviyus.
• Messages and progress-card updates avoid hanging when saving a card overlaps a conversation-history search. #167396 Thanks @fuller-stack-dev.
• Completed Talk consultations keep their completion status and duration without repeated terminal-persistence warnings. #167455 Thanks @steipete, @Bartok9, @samiralibabic.
• Telegram replies rejected before sending receive an accurate failure notice without misleading delivery-recovery warnings. #167470 Docs · Thanks @obviyus, @mgr-nick.
• Readiness checks report terminal channel logouts and blocked accounts as unavailable so operators can recover them. #167478 Docs · Thanks @steipete.
• Telegram actions honor withdrawn authorization after chat lookup, and closed transports stop queued retries. #167635 Thanks @steipete.
• Mattermost and Discord preserve HTTP refusals when error details cannot be read; Signal returns results without waiting for optional cleanup. #167637 Thanks @steipete.
• Telegram streamed answers respect an explicit reply target even when automatic reply threading is off. #167895 Docs · Thanks @obviyus.
• Queue guidance explains why collect mode keeps Discord and Telegram messages as separate follow-up turns. #167940 Docs · Thanks @obviyus.
• Plain-text replies retain command placeholders and generic type examples instead of silently deleting them. #167953 Thanks @obviyus.
• Long formatted code replies avoid unnecessary tiny messages between larger chunks while retaining their text and spacing. #167958 Thanks @obviyus.
• Telegram settings and skill commands wait behind a pending question instead of discarding it. #168004 Thanks @obviyus, @vicary.
• Telegram GIF animations reach the agent as video attachments, including captionless messages, without duplicate downloads. #168011 Thanks @obviyus, @ThisisPeggy, @steipete, @manus-pi.
• Telegram delivers requested tool summaries with /verbose on and completed output with /verbose full in default progress mode. #168016 Docs · Thanks @obviyus.
• Enabled Discord tool-progress logs use tool-specific icons to make activity easier to scan. #168032 Docs · Docs · Thanks @obviyus.
• Persistent Thinking messages arrive before streamed answers on Telegram and Discord without repeating reasoning across answer items. #168062 Docs · Thanks @obviyus.
• Opt-in X maintainer trust uses eligible GitHub collaborators' declared X accounts, with hourly revocation checks that can retain last-known trust during outages. #168071 Docs · Docs · Thanks @steipete.
• Background replies stay in their requesting WebChat conversation, transcript mirrors preserve future reply destinations, and Reef accepts ordinary GitHub revision links without hash-only false alarms. #168248 Docs · Docs · Docs · Thanks @steipete.
• Keeps Telegram rich-reply tables visible and correctly placed before quotes and lists. #168383 Docs · Thanks @dhruv1220, @steipete, @barry-ssb.
• Recognizes silent reasoning placeholders even when the final silence token has punctuation. #168571 Thanks @liwenjie200543, @steipete, @cipp-ashe, @eonicman, @he-yufeng.
• Prevents queue recovery from interrupting messages still being sent. #169010 Thanks @RomneyDa.
• Telegram messages stay queued when a retry finds the original timed-out attempt still pending. #169202 Thanks @diregoblin, @obviyus, @Grynn, @travellingsoldier85, @PaGrom, @cablackmon.
• Feishu DMs continue receiving replies after overlapping settings reload and old monitor cleanup. #169226 Thanks @ruel225, @obviyus, @junniar, @PratikRai0101.
• Scripts receive clean JSON from voicecall steer without startup chatter mixed into stdout. #169430 Thanks @steipete.
• Questions that pause an agent can reach the messaging conversation where the request began. #170039 Docs · Thanks @steipete.
• The first old Telegram spool import skips entries older than 24 hours, including unprocessed entries; review any old instructions you still need. #170068 Docs · Thanks @steipete, @angeliti999.
• Slow pasted Teams images get time to download and reach the agent, with a separate window for message context. #170437 Thanks @RomneyDa, @EbulenConsult.
• A Buzz relay disconnect during authentication no longer shuts down the Gateway and interrupts other channels. #170591 Thanks @masatohoshino, @obviyus, @jarvis-modern-mail.
• Stopping an ACP turn no longer cancels a separate queued follow-up's reply. #170883 Docs · Thanks @steipete, @fanmeng-felix.
• Discord owners configured as discord:user:<id> regain voice controls, owner recognition, and inferred command-approval eligibility. #172385 Docs · Thanks @yetval, @RomneyDa.
• Slack owners configured as slack:user:<id> can be inferred as command approvers when automatic approvals use the owner fallback. #172424 Docs · Docs · Thanks @yetval, @RomneyDa.
• IRC detects silent disconnects and uses the existing reconnect flow without a manual Gateway restart; missed messages are not restored. #172961 Thanks @rodricksz4h5, @steipete.

• Supporting integration refreshes skill eligibility for new sessions, retains selected skill sources, recovers parent turns, and preserves supported scheduled reasoning settings. 3f9636b
• Web-search availability checks reuse prepared authentication instead of rediscovering it. 4a63135
• Windows plugins resolve managed and lazy imports from their captured source with correct nested alias priority. 56d714d
• Integrations canceling transcript operations get consistent errors that retain the original cancellation details. ea7cc10
• Prepared cron and Doctor checks keep their selected web providers without registering the same plugins again. f8f80ea Docs
• Linux node commands can find tools from their configured service PATH after login-shell startup. #136677 Docs · Thanks @harjothkhara, @steipete, @abacha.
• Blank camera-facing options fail before contacting a device. #145538 Thanks @hugenshen, @altaywtf.
• Plugin authors can read the documented after-tool observation fields and limits. #149377 Docs · Thanks @arpanghoshal, @obviyus.
• Custom transcriber guidance explains empty optional arguments and how to omit language flags for autodetection. #151941 Docs · Thanks @fede-kamel, @obviyus.
• Firecrawl searches retain news and image results, including mixed results beside an empty web array. #155885 Docs · Thanks @qingminglong, @obviyus.
• Meme-maker SVGs label GIF and WebP images correctly, and CLI failures retain non-Error diagnostics. #158059 Thanks @NoelDavies, @obviyus.
• Windows node commands preserve quoted prompts, URLs and paths with spaces. #158160 Thanks @Dimitri-lcht, @igs-rogenlo, @steipete, @kassol.
• Code Mode module refusals point agents toward enabled tools and catalog search. #158716 Thanks @sxh313, @pennymolt, @VACInc.
• Tools retain already-valid values such as the string ID 00123 while other invalid fields can still be converted. #158974 Docs · Thanks @Shenrui-Ma, @steipete.
• Video results warn when an unrecognized requested size was ignored. #159037 Thanks @YangManBOBO, @obviyus.
• Retired plugin registries can release obsolete resources after successful cleanup, reducing retained Gateway and catalog-worker memory. #159448 Docs · Thanks @steipete, @JFLeonM, @Cyb3rb1ade, @highfly-hi.
• Agents API operators can connect and reuse executors on their own infrastructure; files, tools, and credentials remain operator-managed, and a shared workspace is not isolation. #161609 Docs · Docs · Thanks @sjf-oa, @sjf.
• Plugin approval waits can use their full response window without ending the waiting turn early. #161821 Docs · Thanks @fransqaas, @obviyus.
• The plugin session-lifecycle documentation link reaches the relevant hook catalog. #161907 Docs · Thanks @yigtwxx.
• Stopped-Tailscale startup attempts no longer suppress channel autostart after Tailscale reconnects and the Gateway restarts. #162266 Docs · Thanks @RomneyDa, @odrobnik.
• Applications importing OpenClaw as a library skip an unnecessary CLI version helper. #162313 Thanks @sunlit-deng, @obviyus, @love-ai-open-source.
• Crabbox worker setup keeps its intended timeout when the host clock changes. #162813 Thanks @SunnyShu0925.
• Gateway clients can browse and search ClawHub plugins and skills, filter official listings and show selected-agent installation status. #163459 Docs · Thanks @jacobtomlinson.
• The bundled GitHub skill explains asynchronous merge submission and status tracking without resubmitting an uncertain request. #163579 Thanks @vincentkoc.
• Administrators can connect installed OAuth MCP plugins from eligible HTTPS Gateway dashboards without duplicating server configuration. #163875 Docs · Docs · Thanks @Patrick-Erichsen.
• Nextcloud Talk consistently uses the documented nested permission for private-network access. #164258 Thanks @steipete.
• Separately installed ACPX restores already-enabled plugin and OpenClaw tools by finding the installed host bridges. #164272 Docs · Thanks @jayzhou2309, @ajmtrz, @obviyus.
• Paired-node command results stop warning about PATH changes that came only from the Gateway launcher. #164506 Docs · Thanks @holgergruenhagen, @obviyus.
• Opted-in plugin tools can deliver finished reports or confirmations without another model response. #164558 Thanks @mbelinky.
• Commands executed on updated CLI nodes retain the originating chat and subagent routing context. #164584 Docs · Thanks @steipete, @idanzhavi09, @MikyWang, @thearabchad, @goslingmanagment, @kassol, @vgranciero.
• Cloud-worker bootstrap failures identify the Gateway address and suggest a reachability check when appropriate. #164627 Docs · Thanks @steipete.
• Node-hosted sessions preserve the selected agent instructions, persona, memory guidance and conversation context. #164647 Thanks @steipete.
• Source-checkout plugins using SDK worker helpers and offline memory status load again after worker package extraction. #164662 Thanks @shakkernerd.
• Mixed-case plugin IDs work across enablement, activation, uninstall and enforcement of configured settings. #164683 Thanks @steipete.
• Tool Search handles unusually large Unicode descriptions without a regular-expression stack crash. #164737 Docs · Thanks @steipete.
• Turn-completion and reply checks move selected database work off the Gateway thread; harnesses can await transcript preparation. #164747 Docs · Thanks @steipete.
• Skill reads explain that a skill is unavailable to the agent and guide it toward available skills without exposing hidden entries. #164884 Thanks @obviyus.
• Timed-out plugin reloads retry automatically after the old instance finishes its work, applying queued configuration edits together. #164894 Docs · Thanks @obviyus.
• Plugin session state survives normal Gateway stops and restarts, and settled cleanup-hook errors warn without alone failing shutdown. #164975 Docs · Thanks @steipete.
• More credential-source checks run through the existing background readers, and retained media tools notice credentials saved after an earlier unsuccessful check. #164992 Docs · Thanks @steipete.
• Code Mode identifies invalid nested tool arguments as input errors so the agent can correct the call. #165008 Docs · Thanks @steipete.
• Tool preparation avoids blocking other Gateway work while preserving approval checks. #165012 Docs · Thanks @steipete.
• Plugin configuration types now reject retired settings already rejected at runtime. #165014 Docs · Thanks @steipete.
• Deployment cleanup guidance now identifies plugin installation directories that must be preserved. #165023 Docs · Thanks @steipete, @morikplay.
• Background plugin cleanup reads installation records without blocking the Gateway thread. #165029 Thanks @steipete.
• After replacing a plugin, Doctor uses the new version’s repair code. #165049 Docs · Thanks @steipete, @ADLBAS26, @Torsie.
• Local and linked plugins can save their own state and use durable incoming-message queues. #165063 Docs · Docs · Thanks @steipete.
• Concurrent GitHub previews share requests, and publication-account discovery returns a retryable error after five seconds. #165094 Docs · Docs · Thanks @steipete.
• Large skill-library browsing and new-session defaults avoid repeated database work and unnecessary avatar copying. #165106 Docs · Thanks @steipete.
• Headless node commands report signal termination so partial output is not mistaken for completion. #165116 Docs · Thanks @steipete.
• Plugin authors can await complete saved-session context with source checks; the synchronous reader remains temporarily available with a deprecation warning. #165134 Docs · Thanks @steipete.
• New MCP discovery hides disconnected server actions during recovery and restores them after reconnection while healthy servers remain available. #165158 Docs · Thanks @steipete, @BorisZhangYYY.
• Windows Bun plugins load affected asynchronous imports and release retired module-cache records on reload or disable. #165169 Docs · Thanks @steipete.
• Concurrent agents searching identical installed skill instructions can share indexes and use less memory. #165173 Docs · Thanks @steipete.
• Windows plugins resolving SDK imports receive native Windows filenames, including require.resolve under Bun. #165282 Docs · Thanks @steipete.
• Plugins containing native prebuilt binaries avoid repeated full-file scans that could cause long loading and Gateway startup delays. #165301 Thanks @goutamadwant, @steipete, @exGeni.
• Large watched skill collections reuse unchanged directory discovery and avoid repeating escaped-path warnings during unrelated refreshes. #165304 Docs · Thanks @steipete.
• Team Reports leaves startup quiet and performs catch-up at the first eligible scheduled collection; manual generation remains immediately available. #165356 Docs · Thanks @steipete.
• Codex plugin migration honors a configured custom app-server command when installing selected plugins. #165365 Thanks @steipete.
• GitHub visitor invitations and revocation by login use configured Gateway credentials and give clearer quota and account errors. #165368 Docs · Docs · Thanks @steipete.
• GitHub publication lookup again reports an ownership error when a session's recorded live checkout belongs to another session. #165407 Thanks @steipete.
• Zalo and Synology attachments stay correct across runtime replacement, and Copilot cancellation can prevent a request from starting. #165431 Thanks @steipete.
• Incoming messages move activity-time database lookups off the main message-handling thread; plugin authors gain async preparation APIs. #165436 Docs · Thanks @steipete.
• Repeated Tool Search and Code Mode lookups reuse prepared catalogs while checking current tool visibility. #165491 Docs · Thanks @steipete.
• Managed private portals and hostname discovery work on larger Tailscale networks, with clearer discovery-failure warnings. #165507 Docs · Thanks @obviyus, @redasadki.
• New sessions recognize newly installed skill requirements without restarting the Gateway when the binary is already on its PATH. #165529 Docs · Thanks @steipete.
• Skills retain complete descriptions and metadata when structured YAML values span several lines. #165531 Thanks @SashaMIT.
• On Windows with Bun, active plugins keep using their loaded source until reloaded, including later TypeScript imports. #165533 Docs · Thanks @steipete.
• Resuming external ACP agents avoids a scan of every retained session. #165550 Docs · Thanks @steipete.
• Remote-exec worker preparation includes the runtime files required by installed Codex and other affected external plugins. #165669 Thanks @LiuwqGit, @JakeBotFleet.
• Plugin help and errors explain connection and readiness checks in more familiar language. #165741 Thanks @RomneyDa.
• Pictures returned by tools found through Tool Search reach image-capable models as images instead of encoded text. #165762 Thanks @obviyus, @jlapenna.
• Plugin authors can await strict transcript-message preparation while eligible harness appends use existing background workers. #165793 Docs · Thanks @steipete.
• Plugin authors can await upstream session-link writes before adopting, binding or cleaning up a session. #165796 Docs · Thanks @steipete.
• Valid plugin skills, including Discord and WhatsApp skills, stay available while source-checkout updates retain the running tree. #165882 Docs · Thanks @RomneyDa.
• Plugin authors can prepare ACP session reads from an existing private binding; related Incognito operations prepare for a later ownership change. #165895 Docs · Docs · Thanks @steipete.
• Plugin snapshots use filesystem clones where supported, reducing repeated disk writes with direct evidence on macOS APFS. #166189 Docs · Thanks @steipete, @matthematics1137.
• Agents using a different model can consult the built-in helper and call it again after a benign plugin reload. #166235 Thanks @PollyBot13, @Colton-Harris, @tyler194653, @obviyus.
• Finished local ComfyUI workflows without media report missing output promptly instead of waiting for a misleading timeout. #166249 Thanks @chelsealong, @catnipluss, @obviyus.
• Repeated native plugin SDK imports reuse validated lookup facts instead of repeating path checks. #166278 Docs · Thanks @steipete.
• Background title requests avoid unrelated plugin initialization. #166310 Docs · Thanks @jjjhenriksen, @obviyus, @cmrstudio56.
• Live monitoring shows current running and queued work, while shared conversation reads wait for refreshed publication facts. #166387 Docs · Docs · Thanks @steipete.
• Native Codex children can use OpenClaw hooks when the app-server and Gateway run on separate hosts. #166400 Docs · Thanks @Kimiyu-186.
• Linux plugin capture can hash cloned files without bad-descriptor failures. #166457 Thanks @steipete, @MertBasar0.
• Shows Feishu error explanations and request log IDs when document or Drive requests are rejected. #166599 Thanks @rudycelekli, @wenbiyou.
• Lets forced plugin replacements activate after active work finishes when the only config notification repeats their own unchanged settings. #166618 Docs · Thanks @steipete.
• Restores message_sent failure callbacks after a narrow partial-delivery fallback acknowledgment. #166659 Thanks @steipete.
• Plugin replacement releases old diagnostic tracking and runs the retiring plugin's cleanup before registering its replacement. #166683 Docs · Docs · Thanks @steipete.
• Lets slow but advancing native plugin verification finish and reuses its result during model catalog refreshes. #166687 Thanks @steipete, @martinderm, @swampgorilla.
• Repeated plugin reloads retain less retired plugin state while active calls and streams keep their selected plugins until completion. #166751 Docs · Thanks @steipete.
• Code Mode keeps core shell, process and file arguments visible in crowded tool catalogs, reducing avoidable failed calls and retries. #166785 Docs · Thanks @steipete.
• Claw-created agents can be removed through the CLI while the Gateway runs, preserving unrelated scheduled jobs. #166889 Thanks @steipete.
• External ACP coding agents start under the correct OpenClaw owner, configured aliases use the correct reply account, and authorized existing conversations can resume. #166908 Docs · Docs · Thanks @saariuslystoned, @steipete, @josephbergvinson, @lachiefreedom, @llagoon3.
• GitHub authorization recovery moves configuration reads off the Gateway thread and defers stale results when settings change during recovery. #166936 Thanks @steipete, @Patrick-Erichsen.
• External plugins with many small dependency files can prepare sooner for Gateway startup and Doctor while retaining complete resource snapshots. #166955 Thanks @jjjhenriksen, @RomneyDa.
• Unchanged plugin source warnings stop repeating after configuration reloads; inspection diagnostics stay available. #167023 Thanks @steipete.
• A replacement Crabbox can be created after stopping one even when the provider repeats a tool-call ID. #167058 Thanks @obviyus, @yetval.
• Coding-agent session checks and resets avoid repeated metadata work on the Gateway main thread. #167086 Thanks @steipete.
• Plugin completion hooks receive the originating scheduled job ID for successful, failed and aborted agent runs. #167124 Docs · Thanks @xyrolle, @obviyus.
• Sessions requiring native workers can finish workspace setup and reach their first model request instead of failing with HTTP 500. #167162 Thanks @RomneyDa.
• Code Mode guidance explains how to read declaration text from the file record, helping agents avoid an initial failed discovery call. #167286 Thanks @obviyus, @rclarke0.
• Context-engine plugins clear abandoned unaccepted turns and warn when cleanup or restart recovery skips them. #167333 Thanks @steipete, @setieroth.
• Plugin builds and validation reject malformed metadata text before silently rewriting author-owned files. #167391 Thanks @Alix-007, @obviyus.
• SSH sandboxes can initialize workspaces on affected Linux volumes, with quicker guarded plugin captures and more useful temporary-directory errors. #167476 Docs · Thanks @steipete, @valkyriweb, @joequant.
• Official ClawHub plugin installs first try a build matching the running OpenClaw version, reducing avoidable compatibility failures. #167485 Docs · Thanks @steipete, @charlesBerg-ux.
• Meeting transcript exports remain consistent during concurrent edits and protect previously saved artifacts if exporter ownership is revoked. #167532 Thanks @steipete.
• Scheduled model-catalog updates reuse unchanged external plugins, avoiding an unnecessary post-startup reload that could stall the Gateway. #167544 Docs · Thanks @obviyus, @cmrstudio56.
• Windows plugins installed beneath directory junctions, including version-manager or Scoop paths, can pass package validation and load. #167588 Docs · Thanks @steipete, @JIRBOY.
• GitHub issue and pull-request hovercards overlap avatar loading with the final visibility check to reduce waiting. #167597 Docs · Thanks @steipete.
• Repeated Prometheus histogram scrapes use less formatting work. #167618 Thanks @steipete.
• Plugin-started turns correctly save finished status and timing without duplicate answers. #167621 Thanks @steipete.
• Known MCP tools stay visible while their server reconnects, with clearer explanations of genuine tool-definition changes. #167642 Docs · Thanks @steipete.
• Watching large skill trees creates less temporary background allocation. #167644 Thanks @steipete.
• Selected bundled plugins can reload despite an unused registry record, avoiding an unnecessary Gateway restart. #167647 Docs · Thanks @steipete, @mbelinky, @jayzhou2309, @Alix-007, @giodl73-repo.
• Worker workspace synchronization rejects timed-out file listings instead of accepting incomplete results. #167673 Thanks @steipete.
• Plugin administration and initial Board dashboard writes occupy the Gateway's main thread with less database work. #167702 Thanks @steipete.
• Plugin authors can prepare transcript messages asynchronously before committing them under the original session's authority. #167715 Docs · Docs · Thanks @steipete.
• Conversation and plugin-state consumers receive committed updates and deletions without late results restoring superseded values. #167731 Thanks @steipete.
• GitHub reader descriptions and comments render recognized emoji shortcodes, including status marks in tables. #167830 Docs · Thanks @jacobtomlinson.
• Large Session Share catalogs refresh with less repeated work on the source gateway. #167909 Docs · Docs · Thanks @steipete.
• Prometheus duration percentiles can distinguish long-running work through one hour instead of flattening at ten minutes. #167930 Docs · Thanks @obviyus, @etzelm.
• Plugins gain asynchronous session lookups, and Active Memory keeps incognito recall debugging transcripts private. #167952 Docs · Docs · Docs · Thanks @steipete.
• Token and cost metrics include successful background work such as session titles and Activity recaps. #167986 Thanks @ericcurtin, @ruel225, @etzelm, @obviyus.
• OpenAI-compatible APIs identify gateway-cancelled runs as cancellations instead of blaming a provider timeout. #167995 Docs · Docs · Thanks @steipete.
• Agents API conversations continue after native tool use while respecting web-search settings and session file permissions. #168067 Docs · Thanks @steipete.
• Supported Agents API conversations saved by v2026.9.9 can continue after upgrading while keeping their remote conversation identity. #168150 Docs · Thanks @steipete.
• Workboard cards keep running while fallback models are still trying and settle from the task's overall outcome. #168215 Docs · Thanks @obviyus, @skyzhao1223, @asikorskiy, @ericcurtin.
• Applies saved per-session MCP server choices to the next Claude CLI turn. #168393 Docs · Thanks @steipete, @dprev.
• Refuses oc-path edits that would corrupt unsupported file bytes and preserves the original file. #168491 Thanks @wangmiao0668000666, @yetval.
• Keeps plugin summaries and new model requests working after an unrelated plugin reload. #168916 Docs · Thanks @jalehman.
• Keeps incompatible-plugin errors and repair guidance inspectable without blocking healthy sibling inspection. #168977 Docs · Thanks @steipete.
• Catalog refreshes reuse unchanged installed native plugins instead of retaining another loaded copy each time. #169272 Docs · Thanks @obviyus, @mcooks-cmd, @svdinu-jpg, @eduherranz.
• A failed native provider catalog refresh no longer leaves later requests using rejected configuration. #169321 Thanks @RomneyDa.
• Node commands and requests keep working when unrelated pairing activity updates only part of the device cache. #169486 Thanks @steipete.
• Node commands and events continue while other devices' pairing information is being written. #169575 Thanks @steipete, @gisk0.
• The deprecated session-header helper imports again without restoring return buttons; plugin authors should move to custom accessories before January 12, 2027. #169746 Docs · Thanks @RomneyDa.
• MCP sign-in can finish when a provider such as Dropbox registers the same local callback without its port. #170876 Docs · Thanks @steipete, @9fw2pq8sgb-art.
• Agents can explicitly end their active turn, reload a plugin after cleanup, and continue the same conversation with the result. #171257 Thanks @jacobtomlinson, @steipete.
• Explicitly enabled standalone plugin files now register their Gateway hooks, including local tool-blocking policies. #171330 Docs · Docs · Thanks @holgergruenhagen, @obviyus.
• After removing a local MCP server whose executable is missing, affected conversations can continue without a Gateway restart. #171397 Thanks @joelagnel.
• When Tailscale Serve is disabled by tailnet policy, the Gateway can keep authenticated access available on the same computer. #171509 Thanks @RomneyDa.
• Eligible skills using the macos platform label appear in new agent sessions on compatible local or remote Macs. #172214 Docs · Docs · Thanks @fhgffy, @steipete, @jessecq1995.
• Temporary SDK exports restore 29 affected plugin imports, including the reported Zulip approval import; migrate before January 15, 2027 and use upsertAuthProfileWithLock for the removed auth-profile helper. #173777 Docs · Thanks @RomneyDa, @FtlC-ian.

• Keep replies visible and the reading position steady while live output is handed over to saved history. 1f9022d.
• The app preserves its Gateway process's existing startup readiness allowance 3d181e1.
• Supporting browser integration keeps agents' tabs separate; compact iOS chat gives messages more room and preserves accessible speaker names. 3f76fe0
• Pending chats retain their normal frame and saved layout with sending disabled until the new session is confirmed. 5a9a6d3
• Retired iOS Gateway connections close their private network transport. b3a37c2
• Session-list refreshes share work when startup and background refresh overlap. d60b095
• The Logs page shows current logs after a Gateway restart and keeps visible rows when refresh fails. ddd39ec
• Saved iOS gateways coordinate Cloudflare Access sign-in and cleanup across profiles. #147238 Thanks @vincentkoc, @Solvely-Colin.
• Connect native iOS chat and media to gateways protected by Cloudflare Access through browser sign-in. #147244 Docs · Thanks @vincentkoc, @Solvely-Colin, @TheAngryPit, @joshavant.
• Saved assistant commentary appears once while a Control UI run continues, preserving surrounding output order. #158459 Thanks @RomneyDa.
• macOS can wait up to five minutes for the same launchd Gateway to finish starting after reboot. #158954 Thanks @MertBasar0, @cjn119-ui.
• Seven rare Lobsterdex visitors bring the collection to 49 characters. #159815 Docs · Thanks @hannesrudolph.
• Removing a Settings list or map entry keeps keyboard focus in the same collection. #160131 Docs · Thanks @moerai, @steipete, @patrick-erichsen.
• Change model and reasoning preferences during a reply so waiting messages use the latest saved choices. #160440 Docs · Docs · Thanks @hxy91819, @steipete.
• Future timestamps keep future wording when calendar-date formatting fails. #164269 Thanks @steipete.
• Chat file-link tooltips let users copy the full path without opening a preview, with keyboard access and copy feedback. #164431 Docs · Thanks @jjjhenriksen, @vyctorbrzezowski.
• Windows nodes reduce reply delays from workspace controls and can run fresh commands after successful runtime refresh cleanup. #164523 Docs · Thanks @steipete.
• Review no longer attributes unchanged pre-session edits to the session in the covered Uncommitted view case. #164671 Thanks @yetval, @obviyus.
• Device-only text loads when Devices opens, reducing the initial Control UI download. #164735 Thanks @shakkernerd, @steipete.
• Waiting Control UI browsers show declined or expired access requests and offer an explicit retry. #164793 Docs · Docs · Docs · Thanks @steipete.
• Sending a chat message keeps the model picker steady, with loading feedback reserved for the actual model catalog. #164963 Docs · Thanks @vyctorbrzezowski.
• Activity recaps refresh in subscribed clients after conversation compaction. #165024 Thanks @steipete.
• New session tabs no longer show Disconnected during normal initial connection. #165051 Thanks @obviyus, @vyctorbrzezowski.
• Native chat names Tool Search activity clearly and aligns tool icon meanings across desktop, mobile, and web. #165095 Docs · Docs · Thanks @steipete.
• Settings preserves composing text and lets the focused control handle Escape without accidental saves or dismissal. #165100 Docs · Thanks @steipete.
• Review uses available header space for long branch names and shows the full comparison on hover. #165126 Thanks @steipete.
• Control UI metadata refreshes send less duplicate model data and avoid unnecessary projection work. #165190 Docs · Thanks @steipete.
• Choose an agent avatar for the browser tab icon to distinguish agents while keeping activity and attention indicators. #165201 Docs · Thanks @vyctorbrzezowski, @jacobtomlinson.
• Concurrent authorized connections can share successful Gateway reads, reducing repeated work and queues for status-like requests. #165203 Docs · Docs · Thanks @steipete.
• Talk settings text downloads when Settings loads, slightly reducing the initial Control UI download. #165228 Thanks @shakkernerd, @steipete.
• macOS session actions and sleep/wake retries stay bound to their original Gateway; logging verbosity changes apply immediately to existing loggers. #165307 Docs · Thanks @steipete.
• macOS respects the native Claude catalog opt-out; microphone diagnostics share local wake-word capture, yield to Talk/push-to-talk and release capture when closed. #165318 Docs · Docs · Thanks @steipete.
• Unusual model-provider IDs display readable labels instead of selecting inherited object properties. #165339 Thanks @steipete.
• Control UI dialogs open and close without animation when the operating system requests reduced motion. #165353 Docs · Thanks @steipete.
• The macOS no-speech message gains translations in 21 non-English locales. #165369
• The compact System busyness widget keeps CPU and memory in the same order as the expanded view, before Gateway ping. #165393 Docs · Thanks @steipete.
• Agent names and avatars remain visible while failed identity refreshes wait before retrying during Gateway restarts. #165443 Thanks @steipete.
• iOS keeps spoken replies visible, preserves local-answer order, and lets large continuation notices expand on demand. #165473 Thanks @Marvinthebored, @Peetiegonzalez, @Solvely-Colin.
• Android releases decoded attachment image memory after JPEG encoding succeeds or fails. #165562 Thanks @steipete.
• The macOS CLI rejects oversized numeric ports and handles very large discovery timeouts without overflow crashes. #165630 Thanks @steipete.
• Android’s accent colors stay easy to tap on narrow screens. #165639 Thanks @IWhatsskill.
• Unrecognized model, Mac, plugin and update identifiers now receive the intended labels, icons or help. #165647 Thanks @steipete.
• Native iOS chat does less repeated parsing and menu work as replies stream. #165927 Thanks @Marvinthebored, @Peetiegonzalez, @Solvely-Colin.
• Native app translations use the updated degraded-check wording. #165959
• Expanded agent updates show one source link instead of duplicate attribution rows. #166006 Thanks @RomneyDa.
• Apple patch previews and macOS failed update notifications avoid crashes caused by extreme numeric values. #166036 Thanks @steipete.
• Optional lobster-pet interactions handle a browser refusal to resume sound without an unhandled rejection. #166043 Thanks @steipete.
• Compact iOS chat gives replies more width, preserves accessible speaker names, and adds tails to completed standalone bubbles. #166067 Docs · Thanks @Marvinthebored, @Peetiegonzalez, @Solvely-Colin.
• Dictate a follow-up while the assistant responds, with normal Send feedback and no duplicated mobile transcript insertion. #166075 Thanks @jacobtomlinson.
• Normal shell exit closes its terminal tab and an otherwise empty bottom or right dock. #166087 Docs · Thanks @jacobtomlinson.
• History and session lists can load while unrelated catalog or subscription preparation waits. #166122 Docs · Thanks @steipete.
• Attention warnings clear while messages actively wait for settings or execute commands. #166124 Thanks @vincentkoc.
• Open supported running-work counts or child names to inspect delegated work beside chat; narrow panes show the work panel full width. #166142 Docs · Thanks @shakkernerd.
• Terminal prompts display bundled Nerd Font symbols, and a browser-local font setting updates open terminals without restarting shells. #166180 Docs · Thanks @jacobtomlinson.
• Link previews consume only a bounded page head, reducing Gateway parsing work. #166195 Docs · Thanks @steipete.
• Moving a terminal between bottom and right docks keeps its tabs, selection and output beside the selected chat without replacing the shell. #166213 Thanks @jacobtomlinson.
• Parent chats identify failed or blocked children, show their full diagnostic, and link directly to the affected session. #166224 Docs · Thanks @vincentkoc.
• Open chats reuse unchanged command menus and model choices instead of repeatedly fetching the same metadata. #166250 Docs · Docs · Thanks @steipete.
• Chats following the latest message settle at the true end after late layout measurements. #166261 Thanks @RomneyDa.
• Long chat history loads in smaller pages while complete referenced tool output remains available for inspection, copying and download. #166263 Docs · Thanks @steipete.
• Background PR counts avoid long missing-object scans in partial clones and show unavailable counts until required contents are fetched. #166264 Docs · Thanks @steipete.
• Child diagnostic notices align with the chat composer on desktop and mobile. #166312 Thanks @vincentkoc.
• Native iOS and macOS chat shows one copy of a streamed reply already recorded in the current turn. #166313 Thanks @Marvinthebored, @Peetiegonzalez.
• Native task lists display saved progress markup as readable bars and cleaner collapsed summaries. #166321 Docs · Thanks @Marvinthebored, @Peetiegonzalez, @Solvely-Colin.
• Unlocked Lobsterdex artwork can identify personal Gateway browser tabs. #166325 Docs · Thanks @vyctorbrzezowski, @jacobtomlinson.
• Empty child-session notices stop leaving blank space above the chat composer. #166347 Thanks @vincentkoc.
• Twenty locales receive refreshed approval-discard prompts and worker status wording. #166353
• Native chat keeps tool titles readable and shows missing outcomes separately as No result. #166359 Thanks @Marvinthebored, @Peetiegonzalez, @Solvely-Colin.
• A session can hide the branch and idle Publish PR row above its composer. #166366 Thanks @RomneyDa.
• Twenty existing locales receive translated Claude Code sign-in guidance. #166390
• Shortened widget repair requests no longer split a boundary emoji into malformed text. #166392 Thanks @ly85206559, @obviyus.
• Android web chat can select STL and other files previously disabled by its File picker. #166402 Thanks @DonnieFi.
• Board and widget-document reads incur less overhead under host load. #166403 Docs · Thanks @steipete.
• Repeated multi-agent views reuse prepared avatar thumbnails instead of reprocessing unchanged images. #166429 Docs · Thanks @steipete.
• Subagents stay in Running while their descendants are active and move to Finished once that work settles. #166442 Docs · Thanks @steipete.
• Control UI keeps the actual reply when a ClawHub search finds no recommendations. #166452 Thanks @a6xa3, @obviyus.
• Background PR comparisons avoid network fetches and stop dependent checks after timeout while keeping the branch visible. #166461 Thanks @steipete.
• Link previews include queue time in their deadline, reuse shared image downloads, and withhold stale results after access changes. #166474 Docs · Thanks @steipete.
• Valid Codex configuration no longer hides paired-Mac sessions, and catalog browsing does not require supervision tools. #166510 Docs · Thanks @steipete.
• Android notification events can start processing when recurring heartbeats are disabled with every: 0m. #166522 Docs · Thanks @jayzhou2309, @BensonChen1204.
• Canvas widget error notices and repair requests avoid broken emoji at truncation boundaries, including reports from saved widgets. #166563 Thanks @ly85206559, @obviyus.
• Restores selecting and copying Workboard notes and gives longer notes a larger resizable editor. #166625 Thanks @rogeriochaves, @llmsommelier, @obviyus.
• Opens long conversations with nested tool activity faster by avoiding unrelated history reads. #166631 Thanks @steipete.
• Confirms successful terminal text copies with a brief notice while keeping keyboard focus in the terminal. #166669 Thanks @jacobtomlinson.
• Makes Linux Quick Chat respect Escape and the latest show or hide request during startup. #166673 Thanks @steipete.
• Restores native chat Mermaid diagrams when the bundled library uses hashed asset filenames. #166680 Thanks @steipete.
• Handles unusual Talk microphone errors with the normal generic failure result. #166682 Thanks @steipete.
• Large compressed conversation histories can load faster with less temporary allocation while preserving selected messages and corruption checks. #166778 Thanks @steipete.
• Gateway shutdown finishes accepted voice persistence, and post-call summaries avoid duplicate sends after marker failures and respect custom conversation stores. #166795 Docs · Thanks @steipete.
• Activity shows compact app summaries and watched conversations before expandable grouped connection diagnostics. #166814 Docs · Thanks @steipete.
• The mobile model selector has a slimmer highlight and more label spacing while retaining its touch target. #166819 Thanks @steipete.
• Keyboard, goal and command submissions clear the submitted collaborator draft preview before sending. #166824 Thanks @steipete.
• Managed chat images have clear previews with Copy and Download in an adjacent image actions menu. #166829 Thanks @steipete.
• The shared native-app runtime improves worker termination at memory limits and idle-worker cleanup. #166838 Thanks @steipete.
• Holding the microphone to begin dictation then releasing it keeps recording until a separate Stop tap. #166843 Thanks @alfredjbclaw, @obviyus.
• Starting a chat keeps the header, submitted prompt and disabled message box visible without disrupting navigation or saved splits. #166865 Thanks @steipete.
• iOS Access settings and expired setup-code guidance gain translations in 21 non-English locales. #166873
• Native chat keeps later backtick code examples literal in an affected combination of inline formulas and code spans. #166951 Thanks @steipete.
• PR automation checkboxes respond immediately and remain independently usable while saves finish, with less popup movement. #166976 Docs · Thanks @steipete.
• Recent projects and folders can load without waiting for session display details to refresh. #166987 Docs · Thanks @steipete.
• Stopped replies keep their interruption marker in the conversation without a duplicate badge above the editor. #166994 Docs · Thanks @Patrick-Erichsen.
• Workboard downloads its full page when opened and explains how to recover if that download fails. #167006 Docs · Thanks @vincentkoc.
• Open chats pause automatic read requests and identity refreshes during Gateway restarts and pace temporary retries. #167020 Thanks @steipete.
• Returning to a tab preserves instructions to reload, fix credentials or request pairing after a rejected connection. #167031 Thanks @steipete.
• Workboard's column color selector displays a readable Color label instead of a translation key. #167082 Thanks @vincentkoc.
• Repeated Publish PR option requests avoid rediscovering the same managed worktree's Git target. #167189 Thanks @steipete.
• Opening a chat no longer briefly flashes timeout cards for expired subagent links. #167232 Thanks @jalehman.
• Browser tab icon settings gain translated labels across 20 Control UI languages. #167289
• Browser-tab icon settings and their browser-local collection explanations are translated into 20 languages. #167307
• Recent and discovered projects load faster when substantial session history is stored. #167322 Thanks @steipete.
• iOS readiness-check instructions and completion messages gain translations in 21 non-English locales. #167329
• Renaming or recoloring a Sessions board preserves columns saved elsewhere, and mistyped Workboard status filters report valid choices. #167348 Thanks @steipete.
• Control UI reconnects avoid a Gateway-wide stall caused by repeatedly loading large external plugins for a cold tool list. #167374 Thanks @jalehman.
• Failed GitHub account discovery in chat can be retried or dismissed without leaving an intrusive publication error above the composer. #167375 Thanks @jalehman.
• Plugin catalog status dots align with the plugin icon and name, including long-name cards. #167400 Thanks @Patrick-Erichsen.
• Control UI initial loading waits through Gateway restarts and resumes without repeatedly retrying the same unavailable connection. #167407 Thanks @steipete.
• Native chat labels for background tasks and injected context gain translations in 21 non-English locales. #167408
• Workboard dispatch shows each failed card's reason and offers the approved admin-request route when the card and caller qualify. #167427 Docs · Thanks @Leon-SK668, @mightyx, @obviyus.
• iPhone and iPad Safari show one Attach action, with camera, photos and files offered by the native picker. #167590 Thanks @steipete.
• Status checks for isolated macOS installations promptly report their absent Gateway instead of waiting on an unrelated host service. #167677 Thanks @RomneyDa.
• Inbox mentions stay dismissible during recap, automation-badge, and turn-status updates. #167689 Thanks @steipete.
• Workspace icons load after Gateway startup, and disconnected HTTP requests stop cleanly. #167697 Docs · Thanks @steipete.
• Talk saves ordinary call records and transcript bookkeeping in a background database worker while preserving accepted writes through cleanup. #167720 Thanks @steipete.
• Firefox keeps selected agent-avatar or Lobsterdex tab artwork after appearance updates. #167759 Docs · Thanks @jacobtomlinson.
• Agent-avatar browser tab icons can use square, rounded, or circular shapes from Appearance settings. #167809 Docs · Thanks @jacobtomlinson.
• Opening chat avoids fetching unopened conversations and shares overlapping app-discovery requests. #167906 Docs · Docs · Thanks @steipete.
• Gateways reuse shared update payloads when sending agent and presence events to multiple viewers. #167961 Thanks @steipete.
• The Control UI sends fewer idle identity requests and pauses background reads during announced gateway restarts. #167968 Docs · Thanks @steipete.
• Repeat Control UI visits reuse unchanged uploaded avatars and retained build assets with fewer downloads. #167993 Docs · Thanks @steipete.
• Conversation navigation appears only when the loaded chat needs scrolling, keeping short chats clear. #168010 Thanks @steipete.
• Reloading the Control UI reuses unchanged plugin styles and scripts, including after a scoped cookie renews. #168023 Docs · Thanks @steipete.
• iPhone and Mac apps recover unresponsive Gateway connections sooner after network changes and resume retries when connectivity returns. #168024 Thanks @steipete.
• Themes can customize dashboard names, icons and working indicators; terminal themes use a quieter neutral presentation. #168027 Docs · Docs · Docs · Thanks @steipete.
• Optional question cards use less space by leading with the question and removing repeated visible instructions. #168040 Thanks @steipete.
• Take photo opens device capture directly in recognized Android browsers in Chat and New Session. #168049 Docs · Thanks @steipete.
• Reopened dashboards show saved titles and panel positions consistently while keeping one loading placeholder until widgets are ready. #168073 Docs · Thanks @steipete.
• The Mac app reconnects automatically after a trusted Gateway certificate renewal, including recovery after network loss or a restart. #168151 Docs · Thanks @steipete, @fuller-stack-dev.
• Refreshes Activity message previews when archived conversation history is restored. #168397 Thanks @steipete.
• Lets chats recover after a server update and reopen the same conversation link. #168941 Thanks @steipete.
• File attachments finish with one usable card instead of a false error or duplicate. #169093 Thanks @steipete.
• Automatic recovery reloads reopen the same chat with its draft instead of a Not found page. #169194 Docs · Thanks @steipete.
• Mac local setup saves missing Gateway authentication and reduces runtime preparation and connection delays. #169403 Docs · Thanks @steipete.
• Control UI chats continue through unrelated pairing activity and ordinary tab reconnects while explicit revocation still stops affected work. #169457 Thanks @steipete, @Oldrich333.
• The Mac app recognizes its protected Bun Gateway command during update recovery instead of remaining on the help screen for this parser error. #170006 Thanks @steipete, @jokerman1302-eng, @davidmackrory-byte, @josephsilvagnoli.
• Spoken Talk requests can continue after voice setup finishes instead of cancelling immediately while still connected. #170083 Thanks @brokemac79.
• Queued Swarm session updates retain their publication work after task completion; the later shutdown fix waits for accepted updates before closing clients. #170231 Thanks @steipete.
• Verified external macOS terminals can restart the Gateway despite an unreadable process ancestor. #170354 Thanks @RomneyDa.
• Gateway shutdown waits for accepted session updates to publish before closing connected clients. #170555 Thanks @steipete.
• macOS Gateway repair recognizes overlapping service loading after native inspection confirms the service is loaded. #170740 Thanks @RomneyDa.
• Mac app recovery keeps a newer managed Gateway and can restore its missing Node service rather than attempting a downgrade. #170766 Thanks @RomneyDa.
• Open dashboard tabs stay connected when a busy Gateway delays application heartbeats, preserving pending requests. #171253 Thanks @steipete.
• Gateway request failures can report an error while keeping a healthy connection open for recovery requests. #171303 Thanks @steipete.
• Mac node commands consistently find tools in preferred folders after launching the app from Finder, Dock or Spotlight. #171460 Thanks @brianmcguire, @steipete, @jamiezigelbaum.
• Gemini3.8 Talk saves spoken questions alongside replies so both sides remain after Stop and reload. #171506 Thanks @masatohoshino, @steipete.
• Mac app Retry repairs plugin links, checks the bundled Gateway and clears the incomplete-update warning after recovery succeeds; failures show the cause. #171508 Thanks @RomneyDa, @shanselman.
• Mac Companion setup can finish checking a saved AI sign-in after wizard replies instead of failing with expired caller authority. #171667 Thanks @steipete, @enter501.
• Mac service restarts preserve multiline credentials, quotes, and whitespace; restore any already-truncated value and run openclaw gateway install --force. #171740 Thanks @steipete, @yetval.
• Settings saves involving secret references can acknowledge changes already applied by the Gateway instead of timing out. #171922 Thanks @steipete.
• Workspace editor saves over 2 MiB of UTF-8 text are rejected before replacing the existing document, matching the agent's read limit. #171938 Thanks @steipete.
• Connection timeouts show their explanation before automatic retry discards it; this improves diagnosis rather than repairing the underlying outage. #172626 Thanks @RomneyDa, @najef1979-code, @steipete.
• A prompt sent immediately after Stop during preparation keeps its place in the queue and receives a reply after the stopped turn settles. #172685 Thanks @shakkernerd.
• Returning to a tab preserves an opening connection, and Logs resumes its saved position on the same Gateway; in-place log rewrites can still leave stale cached text. #172864 Thanks @steipete, @najef1979-code.
• Following a large log avoids rescanning old history on each poll; this reduces Logs-page work without repairing the underlying connection outage. #172885 Thanks @steipete, @najef1979-code.
• Fresh Mac profiles initialize their Gateway correctly, and automatic provider setup waits through supervised startup retries. #173246 Thanks @steipete.
• Downloads from the native macOS dashboard and conversation window open Save while keeping the page available, instead of replacing it with archive bytes. #173654 Thanks @Patrick-Erichsen, @RomneyDa.

• Chrome automation can recover on a fresh worker after a temporary harmless-cleanup storage failure without losing pairing. #151037 Thanks @ethangui.
• Finished MCP agent runs release desktop control, and rebuilding tools preserves the same run's computer session. #151127 Docs · Thanks @ylcn91, @steipete, @vovarbv, @ghbalf, @aleps001.
• The browser panel keeps refreshing during sustained typing or scrolling when live streaming is unavailable. #153580 Thanks @alexsimeonoww, @obviyus.
• Chrome extension uploads totaling strictly less than 47 MiB work without local-file read permission; larger uploads still need a readable path and permission. #160013 Docs · Docs · Thanks @LiuwqGit, @obviyus, @elandesberg, @holny, @Hensarj.
• Chrome extension pages and discarded tabs no longer block the listing and use of ordinary tabs. #162110 Docs · Thanks @openclaw-agent-man, @dlmccoy, @obviyus.
• Browser tool guidance tells agents how to use the configured browser node rather than encouraging a host override. #164575 Thanks @miguelbranco80, @obviyus.
• Linux browser startup finds Playwright Chromium in relative, npm-configured and package-local caches. #164659 Thanks @ly85206559, @obviyus.
• The Chrome extension relay avoids ports already supplied in configured browser URLs. #164678 Thanks @yetval, @obviyus.
• Fetched pages retain paragraph and list boundaries in plain text. #164692 Thanks @yetval, @obviyus.
• DuckDuckGo searches avoid the browser-identifying header that triggered observed bot challenges. #164726 Docs · Thanks @takyyon, @obviyus.
• Web fetch text extraction retains literal syntax inside recognized fenced code examples while converting surrounding prose. #164936 Thanks @ly85206559, @obviyus.
• Local sandbox workspace operations avoid loading large recovery records just to check existence. #165093 Thanks @steipete.
• Eligible first sandbox sessions reuse dependency preparation started after Gateway readiness, reducing setup waits. #165125 Docs · Thanks @steipete.
• Quiet commands receive fewer false stall warnings, unchanged polling is detected, and stuck sandbox-control calls get bounded cancellation. #165167 Docs · Thanks @steipete.
• New managed sandbox workspaces avoid a repeated source scan, and setup timing logs and metrics help identify slow preparation phases. #165411 Docs · Docs · Thanks @steipete.
• Sandbox setup, synchronization and retained-workspace cleanup move database persistence off the Gateway thread. #165534 Thanks @steipete.
• Sandbox creation and cleanup reduce blocking Gateway database work and wait for accepted removal to finish during shutdown. #165558 Thanks @steipete.
• Agents on macOS and Linux can read Windows-style UTF-16 text files that include a byte-order mark, including sandbox reads. #165777 Thanks @yetval, @Patrick-Erichsen.
• Empty desktop and portal streams stop waiting sooner after their target closes, while streams with final data allow it to drain before shutdown. #165806 Thanks @SebTardif, @obviyus.
• Agents stop trying managed web search before a provider is configured and can point users to setup. #166199 Docs · Thanks @jacobtomlinson.
• Discovers computer tools without starting an unused desktop helper; actual use still prepares the desktop. #166602 Thanks @steipete.
• Enables newer Peekaboo clients to use window-specific desktop actions through the macOS app's embedded host. #166615 Thanks @steipete.
• Closes untracked or older managed-browser tabs before recently used session tabs when enforcing the existing tab limit. #166666 Thanks @ruel225, @obviyus, @vokaplok.
• Browser response and navigation URL filters avoid demonstrated long stalls on troublesome wildcard patterns. #166727 Thanks @SebTardif, @obviyus.
• Missing or unauthorized portal WebSocket connections return clear errors and close cleanly instead of hanging. #167065 Thanks @hugenshen, @obviyus.
• Browser management commands honor an explicitly requested timeout and reject invalid values before connecting. #167252 Docs · Thanks @Leon-SK668, @obviyus.
• Browser form and batch commands reject malformed text input before changing page values instead of silently substituting characters. #167412 Docs · Thanks @Alix-007, @obviyus.
• Cleaning up one agent's session preserves another agent's browser tabs even when both sessions share the same name. #167484 Docs · Thanks @steipete.
• Failed portal connections close cleanly when their worker is unavailable instead of accumulating open sockets. #167645 Docs · Thanks @steipete.
• Authorized commands sent to paired POSIX nodes work when the command ends in a newline, space, or tab. #168120 Thanks @yetval.
• Headless browser pages can dismiss invisible passkey prompts to offer another sign-in method; passkeys need a visible profile, and an already-waiting attempt needs cancellation and reload. #170857 Docs · Thanks @trukhinyuri, @steipete.
• Chrome relay cache settings, extra headers, and service-worker bypass survive allowed navigation without incorrectly reporting that access was revoked. #173486 Thanks @LiuwqGit, @steipete, @ValantisV.

• Exit-triggered automations preserve escaped spaces and newer edits; reapply commands whose saved escaping was already trimmed. #161321 Thanks @LinzeShi, @obviyus.
• Cron guidance explains why a trailing NO_REPLY suppresses a text announcement, including its preceding report. #161964 Docs · Docs · Thanks @heranliu, @jackson2w, @obviyus.
• Viewing a scheduled run transcript transfers less unnecessary history data internally. #163885 Thanks @vincentkoc.
• Automations model suggestions follow a changed selected agent without losing the draft or All agents scope. #164357 Thanks @TUARAN, @obviyus.
• Automations keeps same-named models from different providers distinct and saves the chosen provider. #164574 Thanks @ooiuuii, @obviyus.
• Cron permission updates publish in order, expiring standing grants are rechecked, and accepted persistence settles during shutdown. #164805 Docs · Thanks @steipete.
• Workboard refuses to redispatch a retained checkout with a different explicit base ref, preserving work and explaining which refs differ. #164829 Docs · Thanks @ArcueidMP, @mightyx, @obviyus.
• Workboard avoids repeatedly downloading unchanged cards while keeping live edits current. #165058 Thanks @steipete.
• Doctor-migrated heartbeat tasks for secondary agents can have their prompts edited again. #165064 Thanks @steipete.
• Sessions boards now hide automation, system, and Home sessions by default, including existing boards. #165076 Docs · Thanks @steipete.
• Scheduled messages recheck permission at transport handoff and retain accepted parts when later sends fail. #165129 Docs · Thanks @steipete.
• Workboard Sessions boards avoid unnecessary reloads and repeated work for equivalent viewers while real changes still update. #165141 Docs · Thanks @steipete.
• Sessions boards show live activity, retain known PRs with outage warnings, offer PR links and display readable plain-text message previews. #165189 Docs · Thanks @steipete.
• Full reminder queues preserve accepted reminders and explicitly fail new overflow instead of silently discarding older work. #165305 Docs · Docs · Thanks @steipete.
• Scheduled turns targeting busy conversations wait for active work and final metadata updates, preventing them from invalidating in-progress history summaries. #165310 Docs · Thanks @Linux2010, @steipete, @automaton1379, @Oldrich333.
• User messages queued behind heartbeats keep normal tools, context and reply delivery when execution is deferred or retried. #165313 Thanks @steipete.
• Scheduled runs prepare and save session state through database workers, reducing work on the Gateway main thread while accepted writes still settle during shutdown. #165332 Docs · Thanks @steipete.
• Scheduled and webhook-driven Codex agents honor configured max thinking when the model supports it. #165444 Thanks @Oldrich333.
• Repeated automation lists and scheduler status reuse unchanged results to reduce server work. #165629 Docs · Thanks @steipete.
• Team Reports retains the missed day before startup when its first scheduled catch-up crosses UTC midnight. #165780 Docs · Thanks @yetval.
• Automation cleanup waits for agent databases to finish startup preparation instead of reporting a premature maintenance failure. #165814 Docs · Thanks @steipete.
• Automation guides explain that manual CLI runs bypass conditions by default and show how to test eligible conditions with due mode. #165828 Docs · Docs · Thanks @obviyus, @lidbai.
• Doctor stops repeatedly warning about tool authority on command cron jobs that do not use agent tools. #166384 Docs · Thanks @steipete, @jayzhou2309, @davidchyi-beep.
• Webhooks reject malformed supplied session keys and honor transform requests to wake now. #166406 Docs · Thanks @steipete.
• Scheduled runs keep reading their captured history when a later queued message arrives. #166487 Docs · Thanks @steipete, @AXEG0, @hippoley.
• Repeated timezone-aware scheduling calculations reuse date formatters. #166507 Thanks @obviyus, @kzclaw.
• Scheduled command announcements save the destination conversation route without a misleading policy warning when no source session is configured. #166836 Thanks @obviyus, @jlahullier.
• Agent webhooks can wait through overlapping heartbeat database registration instead of failing before the agent starts. #166860 Thanks @steipete.
• Independent heartbeat work can resume after an explicitly stopped agent command finishes cleanup. #166968 Thanks @shakkernerd.
• Paused child tasks retain ownership needed to continue after conversation storage changes, and expired deliveries show their original sanitized error for inspection. #166990 Docs · Thanks @steipete.
• The first scheduled condition check after restart reuses loaded Gateway plugins, avoiding duplicate-loading timeouts and stalls. #166991 Docs · Thanks @67-luck, @steipete, @lucianosmori.
• Conversations can receive delegated results and answer later messages after a queued follow-up yields to child tasks. #167007 Docs · Thanks @MoerAI, @steipete, @hannesrudolph, @hippoley.
• Immediate background results and notifications enter their conversation's normal queue independently of periodic heartbeat settings. #167113 Docs · Docs · Thanks @steipete.
• Cron script and private-note imports reject malformed UTF-8 instead of silently altering supplied text. #167120 Docs · Thanks @Alix-007, @obviyus.
• Scheduled-session cleanup sends less ordinary-conversation metadata through the Gateway. #167248 Thanks @steipete.
• Isolated WebChat automations return results to their creating conversation; resets or deletion can leave a destination binding that needs repair. #167363 Thanks @steipete.
• Subagent restart recovery avoids repeated environment setup while retaining checks against the original database. #167372 Thanks @steipete.
• Removing an active scheduled job preserves its canceled status in saved session history after cleanup. #167524 Thanks @steipete.
• Automation prompts collapse into expandable chat rows, with a link to the originating run in History. #167600 Docs · Thanks @steipete.
• Fresh scheduled sessions stop raising misleading prompt-cache drop warnings. #167608 Thanks @steipete.
• Webhook retries avoid a false HTTP 502 when their first session database appears before the read completes. #167629 Thanks @steipete.
• Older announcements and background tasks can proceed while newer interactive requests keep arriving. #167683 Docs · Thanks @steipete, @SebTardif.
• Large sub-agent task registries use less repeated CPU and temporary allocation for status and maintenance. #167684 Thanks @steipete.
• Session cleanup continues for healthy agents while another agent is waiting for deletion cleanup. #167907 Docs · Thanks @steipete.
• Temporary provider outages get their scheduled recovery retries before automations request repair or send failure alerts. #167957 Docs · Docs · Thanks @obviyus.
• Manually started automations retain their permitted tools after the chat request that launched them ends. #168037 Docs · Thanks @obviyus.
• Creating many sessions leaves room for Home reminders and notifications by combining new-session notices into one pending summary. #168239 Docs · Docs · Thanks @steipete.
• Cron releases its current database write when host approval stalls and retries later, reducing interference with updates and other writes. #168291 Docs · Thanks @steipete, @jlapenna.
• Lets scheduled jobs reach their configured backup model after a primary-provider timeout without duplicating the prompt. #168605 Docs · Thanks @steipete, @wongwilliam425-ui.
• Stops unchanged one-shot automations from causing redundant Doctor recovery backups. #168938 Thanks @harshitgupta31415, @wg-mojo.
• Agents get clearer guidance for returning after a wait, and eligible failed one-time automations request repair in their original conversation. #169102 Docs · Docs · Thanks @steipete, @erlightbourn-source, @lonexreb.
• Eligible dashboard automation interrupted by restart draining can resume after startup instead of losing its recovery claim. #170162 Thanks @steipete.
• Agent webhooks reject unavailable execution capacity so callers can retry or back off; HTTP 200 confirms admission rather than completion, and session conflicts can still return 409. #175111 Docs · Thanks @RomneyDa.

• Portable worker archives include the helper needed to open their state during startup. 22cd7d6
• Older agent databases can finish Doctor migration without failing a schema check before legacy migrations complete. 343e56d
• Database-reader cleanup can finish despite an unreadable discovery path instead of leaving old checks retained. 357466e
• Supporting restart integration retains interrupted chat state and original CLI failures and avoids temporary password files. 383cb4b
• Node workers retain their configured memory limit even when the Gateway starts with a larger limit. 436bc7c Thanks @xydigitLybnnnn.
• Doctor memory-wiki checks avoid loading the pairing database until a pairing read is needed. 4bac149
• The updater recognizes stable 2026.10.2 as a source for the first upgrade step. 5290333
• Doctor maintenance refusals show the underlying database or filesystem cause. 54cfa2f
• Doctor refuses damaged older databases instead of masking missing structures during upgrade. 6785728 Docs
• OpenClaw refuses to start work if its shared database has invalid saved ownership information, even when no writes are pending. 7cbba45.
• Updates recognize retained runtime dependencies through Windows short paths and case differences. 836fd79
• Fix an unnecessary cache setup error that stopped some anonymous diagnostic requests before they could run. 8d91b9c8
• Doctor checks tool schemas using the selected plugin generation, avoiding duplicate registrations. 965ac0f
• Background database verification keeps its writer available until scanning and the saved verification result finish. 978478f Docs
• Explain disk-full plugin capture failures and direct users to free space and rerun Doctor. a6496a2.
• New session database work can resume after configuration changes. b798788
• Docker builds pass workspace lockfile validation while retaining optional-plugin selection. b7e788b
• Automatic update repair can read configuration and proceed during an update. b8602f4
• Valid database writes can proceed after an earlier write fails. c190df1
• Completed updates avoid unnecessary failure triage or a second failure report. d357abf
• Database work already in progress can finish against its original storage after configuration changes. d81f112
• Normal SQLite worker shutdown preserves queued diagnostic file logs for troubleshooting. f8f1c8a Docs
• Supporting diagnostics retain useful paths and original errors; the compact System widget matches the expanded CPU, memory, and ping order. faf63c5 Docs · Docs
• Diagnostic queries reject empty filters and limits instead of silently using defaults. #145532 Docs · Thanks @hugenshen, @altaywtf, @nocodet666.
• Blank saved-bundle selectors fail instead of choosing unintended diagnostic evidence. #145537 Thanks @hugenshen, @altaywtf.
• Blocked settings updates explain the refusal and next step while leaving current configuration unchanged. #155427 Docs · Thanks @RomneyDa.
• Gateway readiness waits keep their elapsed-time limits when the system clock changes. #155799 Thanks @SunnyShu0925.
• Source builds use less peak memory during runtime bundling by scheduling bundles one at a time. #157739 Thanks @emes, @obviyus, @steipete.
• Linux Gateways remain responsive when log rollover encounters a temporary directory removed by system cleanup. #160433 Thanks @masatohoshino, @WadeOwenWatts, @Patrick-Erichsen, @obviyus.
• Model-catalog worker failures now show a reason in logs and remain counted after worker replacement. #161740 Docs · Docs · Thanks @matthematics1137, @obviyus.
• Worker diagnostics preserve both the operation failure and the cleanup failure when both occur. #162124 Thanks @michaelswissa, @obviyus.
• Doctor can complete deferred external-plugin upgrades through their public setup entries. #162275 Docs · Thanks @romneyda.
• Doctor preserves older ACP conversation bindings during upgrades and protects closed sessions from being revived by old metadata. #162920 Docs · Thanks @steipete.
• Independent terminals on affected macOS Monterey hosts can stop or restart the Gateway despite denied launchctl inspection. #163317 Thanks @NovaUnboundAi, @xXG0DLessXx, @RomneyDa.
• Doctor moves default legacy Clawdbot state and configuration to canonical OpenClaw paths; ordinary runtime no longer discovers old paths. #163376 Docs · Thanks @steipete.
• Deleting a workspace-free session also removes its leftover local GitHub publication records. #163833 Thanks @steipete.
• Exec-approval documentation directs operators with saved legacy policy fields to Doctor repair. #164234 Docs · Thanks @wakqasahmed, @cristianoc, @obviyus.
• Multiple-Gateway guidance includes sandbox listener ports to help avoid conflicts. #164275 Docs · Thanks @Satoshi-Kusumoto, @obviyus.
• Doctor backups no longer fail falsely when Linux cannot report stable file creation times. #164363 Thanks @jayzhou2309, @steipete, @StevenKrebs, @kirylh, @stunt4711.
• Fish completions preserve backslashes in descriptions and avoid broken suggestions. #164455 Thanks @LinzeShi, @obviyus.
• Reusable Gateway resources release completed request data, and Doctor refuses repairs when damaged indexes hide active or conflicting Gateway ownership. #164465 Thanks @steipete.
• Gateway startup errors retain the original cause even when cleanup also fails. #164500 Thanks @sunlit-deng, @waynegault, @obviyus.
• Package updates can continue after full verification reaches resource limits, with a visible reduced-verification warning. #164504 Docs · Thanks @steipete.
• Persistent database workers on supporting Bun runtimes can reclaim released JavaScript heap while idle. #164567 Thanks @steipete.
• Non-root package updates and rollback tolerate launcher ownership that the updater cannot reproduce. #164636 Docs · Docs · Thanks @steipete, @hrp6vv9mwr-glitch, @cmillwat.
• Valid workspace avatar paths no longer fail the avatar check in disposable update validation. #164638 Thanks @nitinjwadhawan, @vincentkoc, @BottaniCals.
• Doctor repairs stale string subagent targets without crashing on malformed allowlist entries; retired sandbox registries now require an intermediate upgrade. #164669 Docs · Thanks @steipete.
• Database recovery begins earlier after unclean restarts, and recovered agents remain available in chat metadata and model listings. #164670 Thanks @steipete.
• SQLite inspection subprocesses inherit explicit supported Node compiler settings, and Google Meet connection failures use consistent errors. #164675 Thanks @RomneyDa.
• Affected recovery and package-install paths keep installation scripts on the selected compatible Node runtime. #164677 Docs · Thanks @steipete, @jarvisnextgeneration20-lang.
• Windows plugin preparation avoids redundant file opens, reducing a measured source of Gateway and model-status delay. #164680 Thanks @steipete, @absol89.
• Doctor shows the actual database refusal or recovery information without temporary snapshot cleanup masking it. #164697 Thanks @steipete.
• Doctor can re-enable and start an owned disabled Windows Gateway task and explains its disabled state. #164728 Docs · Thanks @steipete, @taoxia255-del.
• The updater announces each validation check and command before it starts. #164746 Docs · Thanks @steipete.
• Doctor preserves and can migrate legacy audit logs when native no-replace filesystem moves are unsupported. #164748 Docs · Thanks @steipete, @bbs2i58.
• Device setup links move join-code storage work off the Gateway thread while retaining one-use exchange. #164759 Thanks @steipete.
• Updated npm updaters can recover a Docker image-layer package rename failure using a verified rollback copy. #164769 Docs · Docs · Thanks @steipete, @potterdigital.
• Slow-response logs retain recent request timings when the bounded timing cache fills. #164772 Thanks @etzelm, @obviyus.
• Update history and Settings can retain the actual refusal and its repair guidance. #164782 Docs · Thanks @steipete, @coygeek.
• Installing completion preserves significant escaped trailing spaces in existing shell-profile commands. #164785 Thanks @yetval, @obviyus.
• Settled agent databases can close cleanly while unrelated shutdown cleanup remains pending. #164788 Thanks @steipete.
• POSIX npm updates admit writable prefix parents owned by another user and reject unsafe package ownership earlier with useful details. #164797 Docs · Thanks @steipete, @potterdigital.
• Interrupted-update recovery accepts identical files despite timestamp-only drift. #164823 Docs · Thanks @steipete.
• Eligible Linux Gateway workers can reopen agent sessions sooner after a verified process crash while full integrity checks continue in the background. #164846 Docs · Thanks @steipete.
• Update repair can settle obsolete recovery ownership after its database was replaced, allowing subsequent updates while retaining evidence. #164849 Docs · Thanks @steipete, @szef-svg, @izrl613.
• Automatic session cleanup moves cold subagent-protection reads off the Gateway thread while preserving active and pending-delivery sessions. #164851 Docs · Thanks @steipete, @todddickerson.
• Doctor drains retained workers before CLI exit, and update history preserves completed-check diagnostics even if a later crash fails the candidate. #164853 Docs · Thanks @steipete, @cunzai007-hub.
• Refused package updates identify the recovery object and unsafe metadata without exposing absolute installation paths or contents. #164858 Docs · Thanks @steipete, @xydigitLybnnnn, @creanet-64, @jrpasolini, @jb168, @burchk007-jpg.
• Doctor imports sessions blocked by older migration receipts while preserving archives and existing SQLite metadata when history is uncertain. #164867 Docs · Thanks @steipete, @hshs38033-cmd.
• Update repair can finish its own completed Doctor work while a separate Doctor command remains active. #164868 Docs · Thanks @steipete.
• Unused agent databases no longer block Gateway startup and readiness; their bytes remain intact and a warning points to Doctor. #164929 Thanks @steipete, @Oldrich333.
• Session maintenance prepares remote-worker preservation facts outside the Gateway main thread and rejects stale facts before cleanup commits. #164943 Docs · Thanks @steipete.
• Reconnecting clients share update-status preparation instead of repeating the same notification reads, while later outcomes continue to refresh. #164984 Docs · Docs · Thanks @steipete.
• Doctor and updates can publish and roll back configuration in a writable state folder owned by another user, including Kubernetes fsGroup setups. #164990 Docs · Thanks @steipete, @saftall.
• Update recovery retains the service identity needed after systemd unloads a stopped unit, and the repaired CLI can retire a verified preparation that never began installation. #164997 Docs · Thanks @steipete, @jorden2895.
• Prepared child commands keep using the release they were prepared with when an update changes a launcher symlink before they start. #164998 Docs · Thanks @steipete, @Grynn.
• Source-checkout Gateways prepare their first worker environment faster. #165015 Thanks @steipete.
• Scheduled automations remain available after Doctor repair without an extra restart. #165032 Thanks @steipete.
• Memory warnings reflect actual runtime capacity, avoiding unnecessary retirement of warm workers. #165042 Docs · Thanks @steipete.
• macOS Gateway recovery preserves a newer runtime or service choice made by the operator. #165044 Thanks @steipete.
• Busy Gateways avoid memory buildup behind slow requests. #165052 Thanks @steipete.
• Source updates to locally cached pinned commits avoid unnecessary fetching and accept full SHA-256 IDs. #165060 Docs · Thanks @steipete.
• Agent databases can close cleanly during busy restarts even if unrelated plugin cleanup stalls. #165069 Thanks @steipete.
• Eligible Linux crash restarts restore usable sessions sooner, and large session cleanup avoids repeated work. #165070 Thanks @steipete.
• Cleanup of large diagnostic histories does less database work. #165074 Docs · Thanks @steipete.
• Cleanup preserves temporary plugin and model-catalog files still in use or whose use cannot be checked. #165079 Docs · Thanks @steipete, @Dhotei.
• Session cleanup planning reads less unrelated saved data in large installations. #165082 Thanks @steipete.
• Routine maintenance helps delivery-history lookups choose a more selective database index. #165085 Docs · Thanks @steipete.
• Rosetta installations avoid unsafe native calls during worktree creation and process checks. #165088 Thanks @steipete.
• Large transcript stores spend less work on status checks and stale search-index cleanup. #165090 Thanks @steipete.
• Smaller agents can become available after restart while larger agents are still preparing, with recovery progress in logs. #165107 Thanks @steipete.
• Nested documentation examples retain their authored indentation when displayed and copied, including Slack setup manifests. #165127 Docs · Thanks @x-n2o, @obviyus.
• Updates finish their selected run's deferred model repairs even when plugins are unchanged. #165148 Docs · Thanks @steipete.
• Eligible Windows Gateway tasks can start before login, and managed startup allows longer cold starts. #165163 Docs · Docs · Thanks @steipete, @desksk, @runvouch.
• Restart recovery settles orphaned spawned and archived sessions left marked running while preserving resumable main sessions and active work. #165174 Docs · Thanks @steipete.
• Delegated work resumes after deferred database readiness, and one failed requester group no longer blocks unrelated recovery. #165184 Docs · Thanks @steipete.
• Runtime database opens avoid repeated completed integrity checks, and slow-hold warnings identify the operation and time spent waiting or committing. #165186 Thanks @steipete.
• Agents with small conversation-history rebuilds can become available sooner after a multi-agent Gateway restart. #165207 Thanks @steipete.
• Fresh installs without Matrix or legacy Matrix data avoid an unnecessary Doctor migration failure while the Gateway is running. #165209 Docs · Thanks @steipete.
• Non-interactive onboarding returns requested JSON error details when affected provider setup fails, helping installers explain the failure. #165211 Docs · Thanks @steipete.
• Large Gateways can serve requests before optional startup maintenance finishes, with clearer phase timings for diagnosis. #165218 Docs · Thanks @steipete.
• In-place updates avoid a misleading missing-module cleanup error after success, and failed commands retain their original error. #165233 Docs · Thanks @steipete.
• Update repair can recover the covered blockage when reboot or temporary-file cleanup removes the handoff database. #165238 Docs · Thanks @steipete, @Hensarj, @dawid-sitko.
• Updates to an already-current package can still repair supported legacy configuration and complete plugin maintenance without --channel. #165242 Thanks @shakkernerd.
• Retained workers exchange task messages directly, spending less CPU and time moving inputs and results. #165249 Thanks @steipete.
• Old plugin-session artifact cleanup skips recent snapshots and reads historical windows in pages, reducing planning work and long read holds. #165254 Thanks @steipete.
• Updates can copy file bytes when a container blocks file cloning, avoiding the covered updater-runtime-retention failure. #165257 Docs · Thanks @steipete, @MrJanvar, @rybing7, @Mohl.
• Windows service reinstalls replace the old running process so runtime paths, arguments and environment changes actually take effect; status verifies the registered command. #165261 Docs · Thanks @steipete.
• Full source builds compile overlapping public declarations together instead of repeating SDK declaration work. #165271 Thanks @huntharo, @obviyus.
• Doctor recognizes native Claude CLI installations outside the service PATH and avoids misleading reinstall advice. #165273 Docs · Thanks @steipete, @mightyx.
• Cleaning old diagnostic run history causes shorter interruptions to other writes in the same agent database. #165274 Docs · Thanks @steipete.
• A retained background-worker initialization failure stops repeated replacement starts driven by queued tasks; healthy siblings keep working. #165277 Thanks @steipete.
• Busy Gateways avoid repeated process work when preparing PR information and protected HTTPS commands to the same hostname. #165290 Docs · Thanks @steipete.
• Session worker reads reuse unchanged ownership metadata and existing read transactions to reduce repeated database work while keeping related facts current. #165293 Docs · Thanks @steipete.
• Initial saved-session reads for replies and fast commands move off the Gateway thread while preserving session and save safeguards. #165295 Thanks @steipete.
• Updates preserve existing Feishu callback endpoints before activation even when plugin installation is deferred; shared fallback covers Telegram, Teams and Nextcloud Talk. #165303 Docs · Docs · Thanks @shakkernerd.
• Update status avoids inflated ahead/behind counts in shallow Git histories, returning unknown when exact comparison cannot be justified. #165316 Thanks @ooiuuii.
• Existing shared Bun consumers receive VM rendering, asynchronous import-context and worker heap/stack-limit engine fixes. #165317 Thanks @steipete.
• Generated macOS service definitions request a longer shutdown window, with Doctor and update finalization migrating recognized older generated definitions while preserving custom settings. #165320 Docs · Thanks @steipete, @yuzh086-create.
• An agent repairing its startup indexes promptly asks callers to retry while other ready agents, Control UI and health checks remain usable. #165323 Docs · Docs · Thanks @steipete.
• Updates of prepared source builds and package installations skip an unnecessary plugin-state lock during runtime completion, avoiding unrelated lock admission failures at that step. #165326 Thanks @steipete.
• Pending session notices are restored after restart through background database workers, keeping their recovery SQL off the Gateway main thread. #165331 Docs · Thanks @steipete, @vincentkoc.
• Workspace recovery spends less time checking large numbers of files that are already deleted. #165341 Thanks @steipete.
• Dev-channel updates can prepare a source checkout before its dependencies are installed instead of failing with a missing-package error. #165347 Thanks @steipete.
• Doctor can wait for an exited container's fresh ownership lock to expire instead of immediately refusing maintenance, while protecting active Gateways. #165355 Docs · Thanks @steipete.
• Queued health checks and inference requests keep waiting until their actual deadline when the clock moves backward or the deadline exceeds the native timer range. #165359 Thanks @Chessing234, @steipete.
• Configuration checks show the actual SQLite helper launch error and affected file instead of a misleading trim TypeError. #165373 Docs · Thanks @steipete, @nnljl2003.
• File copying falls back safely when containers block cloning, and the filesystem dependency also corrects fresh file creation on gVisor/9p. #165374 Docs · Thanks @steipete, @MrJanvar, @Mohl, @rybing7.
• Two retired configuration formats require an intermediate upgrade through 2026.9.5 and openclaw doctor --fix before installing a newer release. #165375 Docs · Thanks @steipete.
• Six retired JSON state families require 2026.9.7 Doctor on the original host before upgrading; current Doctor preserves the files and explains that step. #165402 Docs · Docs · Docs · Thanks @steipete.
• Managed npm installs honor uppercase and mixed-case retry and timeout settings, including zero retries. #165409 Thanks @Chessing234.
• Agent diagnostic recorder setup uses background session readers, reducing synchronous Gateway-thread database work without changing captured diagnostics. #165410 Docs · Thanks @steipete.
• Authorized updates no longer fail because Doctor's permission check mistakes its own maintenance session for another owner's lock. #165417 Docs · Thanks @steipete.
• Unused old login files no longer block Doctor, upgrades or container startup. #165420 Thanks @obviyus.
• An obsolete update-notification cache no longer blocks Doctor or upgrade admission and remains preserved. #165461 Docs · Thanks @steipete.
• Sessions accept the next instruction after restart recovery settles the interrupted reply. #165469 Docs · Thanks @steipete.
• Opening messages and usage reports moves their database reads off the main request thread. #165470 Docs · Thanks @steipete, @shakkernerd.
• Large worker-session inventories cause less repeated work during cleanup checks, and changed placements invalidate old cleanup decisions. #165475 Docs · Thanks @steipete.
• Restart inspection no longer produces false secret-failure warnings or unnecessary reload advice. #165485 Docs · Thanks @steipete.
• Gateway startup skips unnecessary repair-rule loading, and recovery inspections read a consistent state view. #165498 Docs · Thanks @steipete.
• Routine session cleanup does less repeated work on large watcher registries while preserving retention safeguards. #165511 Docs · Thanks @steipete.
• Usage reports reject dates skipped by the Gateway timezone instead of returning backwards ranges. #165523 Thanks @SashaMIT, @obviyus.
• An external Gateway restart can hand over after required data closes even when plugin shutdown stalls. #165524 Docs · Thanks @steipete.
• Doctor explains retired runtime files without letting advisory inspection failures block an otherwise valid startup. #165528 Docs · Thanks @steipete.
• Windows Gateway status preserves a saved Node or Bun choice after installation and gives Task Scheduler repair guidance. #165538 Docs · Thanks @steipete.
• Optional configuration fallbacks during updates appear as warnings instead of misleading errors. #165539 Thanks @Patrick-Erichsen.
• Linux Gateways automatically warm bounded recent-history data in the background to reduce later disk reads. #165545 Docs · Thanks @steipete.
• Prometheus separates worker queue waiting from request duration so operators can diagnose contention. #165591 Docs · Thanks @steipete.
• Update repair can unblock a stranded publication after external package changes once modified inventoried files are restored. #165607 Docs · Thanks @steipete, @NovaeonStudio.
• After restart, a long recovered turn no longer delays unrelated interrupted sessions beyond normal main-lane limits. #165612 Docs · Thanks @steipete.
• Gateway startup reuses configuration checks and plugin information to reduce repeated setup work. #165667 Thanks @steipete.
• Gateway shutdown logs identify slow cleanup steps and show what is still pending when a deadline expires. #165670 Docs · Thanks @steipete.
• Old diagnostic run history can be cleaned up while new events continue arriving. #165678 Thanks @steipete.
• First-use agent database checks move off the Gateway thread, and background observer writes stop blocking turn cleanup through an inherited write context. #165680 Thanks @steipete.
• Large subagent histories cause fewer Gateway pauses during recovery and cold reads, and restored state is published together. #165684 Docs · Thanks @steipete.
• The earlier removal of Voice Call conversion is superseded by the restored Doctor migration below; current supported older settings need not all be converted manually. #165707 Docs · Docs · Thanks @steipete.
• Session activity follows live runs and queues so interrupted restarts do not leave stale Working indicators; local destructive cleanup now requires exclusive ownership. #165733 Docs · Docs · Docs · Thanks @steipete.
• Setup and troubleshooting guides describe connection checks in clearer everyday language. #165736 Thanks @RomneyDa.
• Connection buttons, results and native diagnostics use clearer check wording. #165737 Thanks @RomneyDa.
• CLI setup, status and repair messages describe connection and authentication checks in clearer language. #165740 Thanks @RomneyDa.
• Runtime errors and health messages describe existing checks more clearly. #165742 Thanks @RomneyDa.
• Linux updates respect systemd maintenance holds and show how to recover when a hold prevents restart. #165747 Docs · Docs · Thanks @steipete, @gregb100.
• Database schema histories identify the releases that introduced stored-data changes, helping operators plan upgrades and recovery. #165757 Docs · Docs · Thanks @steipete, @gregb100.
• Affected Linux hosts no longer reject healthy database snapshots solely because metadata timestamps changed. #165775 Docs · Thanks @steipete, @donquiche61, @naturecurly, @dreamkeeper, @wangmiao0668000666, @clockthunder, @honghongxinxi-zhb, @jayzhou2309.
• Interrupted Gateway restarts can finish database closure when scheduled continuations are still active. #165782 Thanks @steipete.
• Worktree cleanup remembers invalid Git metadata instead of repeatedly probing it, and the X plugin regains its Plugins-page icon. #165795 Thanks @steipete.
• Slow preparation of individual agent databases no longer holds up Gateway readiness; available agents and the Control UI can be used while the remaining agents finish. #165803 Docs · Thanks @steipete.
• Prometheus metrics identify the worker tasks and database operations behind queue delays. Dashboard queries using older task labels need updating. #165816 Docs · Thanks @steipete.
• Doctor and update backups preserve vector memory data on CPUs that cannot load the bundled vector-search extension, protecting recovery before repairs. #165820 Docs · Thanks @steipete, @NovaUnboundAi.
• The experimental fleet command is retired; retained containers and data need Docker or Podman management for deployment, updates, and backups. Related transcript completions keep their recorded owner. #165825 Docs · Docs · Docs · Thanks @Patrick-Erichsen, @RomneyDa.
• Managed CLI updates respect explicitly selected temporary directories when planning snapshots, while retaining capacity-based fallbacks. #165830 Docs · Thanks @steipete, @DonnieFi.
• Doctor can again repair supported older Voice Call settings so they pass validation after an upgrade, while preserving current-format values. #165831 Docs · Docs · Thanks @steipete.
• Doctor archive repairs keep progressing through empty historical session IDs and respond to interruption, preserving completed batches for retry and reporting the affected archive on failure. #165837 Docs · Thanks @steipete, @A1fred-AI, @goncalonc, @mpquemel.
• An unfinished channel-health check no longer causes an otherwise verified update to fail solely for timing out; the update keeps a warning directing users to check health. #165852 Docs · Thanks @steipete, @vasaeru, @jlapenna, @rogerallen1, @TimofeyK.
• Doctor and updates can resume interrupted archive publication, retain missing-copy recovery information, and restore a managed Gateway Doctor stopped when a repair fails. #165854 Docs · Thanks @RomneyDa, @Olli0103, @h6y3, @waynegault, @rauf543, @mpquemel, @goncalonc, @A1fred-AI, @steipete.
• Updates from 2026.10.1-beta.1 can finish cleanup after package replacement without losing required modules. #165855 Thanks @shakkernerd.
• Doctor can clear unfinished-plugin warnings for eligible empty configurations or a completed official replacement, while retaining plugin data and genuine migration work. #165861 Docs · Thanks @steipete, @Franklinyung, @ekinnee.
• Catchable interruptions during local update finalization give recovery time to restore a consistent package and restart the Gateway before reporting the signal and phase. #165863 Docs · Thanks @steipete, @chrisreed14, @dankarization.
• Doctor and update-time migration can import older saved pending replies without repeatedly failing, preserving exact conversation IDs, transcripts and archived originals. #165866 Docs · Thanks @SunnyShu0925, @steipete, @609NFT.
• Gateway suspension can finish after already-disconnected requests or failed final dispatches instead of waiting on stale work, and reports the conversations still awaiting writes. #165871 Docs · Thanks @steipete.
• The Control UI's generated assets are readable by arbitrary container users, addressing the permission failure that prevented it loading in OpenShift deployments. #165880 Thanks @RomneyDa, @Conan-Scott.
• Secret-file inputs reject unwritten named pipes, plugin logs stay out of machine output, Doctor preserves replaced links, and custom Mistral costs survive configuration cleanup. #165900 Docs · Thanks @steipete.
• Prometheus metrics identify shared-state read families, and node pending-work drains avoid a redundant pairing lookup while retaining fresh authorization. #165910 Docs · Thanks @steipete.
• Background worktree cleanup releases sessions sooner and backs off repeated Git timeouts. #165913 Docs · Thanks @steipete.
• Git and worktree failures retain their real diagnostic when a generated path contains dns. #165921 Thanks @steipete.
• Large worker fleets do less repeated background reconciliation work. #165922 Thanks @steipete.
• Idle Gateways avoid unnecessary scheduler wakeups and temporary allocations. #165931 Thanks @steipete.
• Repeated node and environment discovery does less work when paired devices and node state are unchanged. #165941 Thanks @steipete.
• Gateways restore large retained subagent registries faster before becoming ready. #165942 Thanks @steipete.
• Worktree cleanup reports complete totals for eligible removals, deferrals and failures even beyond its detail limit. #165969 Docs · Thanks @steipete.
• Managed worktree Git indexes can share unchanged file data on APFS and other clone-capable filesystems. #165981 Docs · Thanks @steipete, @matthematics1137.
• Updates accept unchanged plugin files after harmless hard-link metadata changes while still checking their contents. #165982 Docs · Docs · Thanks @steipete, @wg-mojo.
• Managed-worktree maintenance recovers when a stale Git pack index refers to replaced pack files. #165995 Docs · Thanks @steipete.
• Global npm installs avoid retaining native binaries for unrelated platforms, reducing installation size. #165997 Thanks @steipete, @tommyfive, @kenmege.
• Gateway stops can finish accepted storage writes and close databases cleanly after active-work draining times out. #166018 Docs · Thanks @steipete.
• Routine chat history checks and session saves perform fewer database operations. #166041 Thanks @steipete.
• Doctor can migrate legacy state on Linux systems such as QNAP where no-replace rename is unsupported. #166048 Docs · Docs · Thanks @steipete, @StarGazer1995, @bbs2i58, @MertBasar0.
• Chat preparation and worker recovery avoid repeated local database reads. #166053 Thanks @steipete.
• Ongoing chats and history reads reuse unchanged session information instead of repeating database queries. #166054 Thanks @steipete.
• Windows installs and upgrades accept equivalent short and long session-database paths instead of refusing startup. #166062 Thanks @RomneyDa.
• Session-history reader workers retain less memory by loading fewer unrelated dependencies. #166069 Thanks @steipete.
• Credential-free Hermes migration plans can proceed despite an unreadable destination credential store. #166070 Thanks @steipete.
• Managed-worktree cleanup avoids parsing entire saved conversations to check whether checkouts are still in use. #166072 Docs · Thanks @steipete.
• Chat turns and session history reads perform less database bookkeeping while preserving consistent snapshots. #166074 Thanks @steipete.
• Failed transcript archive writes clean up their own temporary files, including storage-full failures. #166085 Thanks @steipete.
• Agent operations avoid rechecking unchanged database catalog information. #166102 Thanks @steipete.
• Busy Bun Gateways avoid expensive detailed heap checks between database and history operations. #166104 Thanks @steipete.
• Gateway startup waits through temporary database schema ownership instead of exiting before channels come online. #166111 Docs · Thanks @NianJiuZst, @steipete, @ericcaiwx-star, @mashaohua1999-stack.
• Completed background full database checks can be reused after clean shutdown, avoiding a repeated scan on the next eligible restart. #166112 Docs · Thanks @steipete.
• Future updater runs avoid redundant database snapshots during readiness checks. #166126 Thanks @vincentkoc.
• Account refresh and personal model discovery move database work away from the Gateway request thread and reduce repeated reads. #166183 Thanks @steipete.
• Git-operation diagnostics release data retained after completed operations. #166214 Thanks @steipete.
• New exec and stdio MCP launches can find an available stable Node path after Homebrew removes the old executable. #166222 Docs · Thanks @jayzhou2309, @steipete, @mightyx.
• Linux updates handle older stopped-service handoffs when systemd has unloaded the unit and no manager ID was recorded. #166232 Thanks @RomneyDa.
• Setup and repair helpers give failure-specific recovery advice instead of recommending onboarding for temporary runtime failures. #166234 Thanks @PollyBot13, @obviyus, @Colton-Harris, @tyler194653, @stemkat100, @MaikiMolto, @explainanalyze, @edith-vitaetpax, @spiral-cmd, @FtlC-ian, @jj5638521, @iamade, @dh-js, @JoPaMu, @skorfmann, @ryanography.
• Crash-loop safe mode pauses automatic session replay, preserves interrupted work, and schedules resumption after the stable window clears. #166238 Docs · Thanks @steipete, @masatohoshino, @KoalaShen.
• Restart readiness inspects shared-state databases outside the Gateway request thread while retaining refusal checks. #166252 Thanks @steipete.
• Ongoing chats avoid repeated database preparation when validated structure remains unchanged. #166253 Thanks @steipete.
• Conversation discovery and session associations batch database work through existing workers. #166259 Docs · Thanks @steipete.
• Active SQLite paths reduce redundant retention, append and initialization work. #166265 Thanks @steipete.
• Older node hosts fail worker admission early with an update-required explanation instead of rejecting new prompt context during launch. #166268 Thanks @vincentkoc.
• Profile and recent-project requests prepare data in workers and avoid repeated lookups. #166283 Thanks @steipete.
• History and file reads move database work off the Gateway thread. #166293 Thanks @steipete, @vincentkoc.
• Updates prepare private validation state faster for large plugin installations while retaining existing checks. #166317 Thanks @hannesrudolph.
• Doctor and migration maintenance avoid redundant blocking database reads. #166331 Thanks @sasan1200, @steipete.
• Unrelated system processes with unreadable working directories stop blocking cleanup of unused temporary runtimes. #166335 Thanks @vincentkoc.
• Doctor tool inspection reuses plugin registrations and owns provider discovery cleanup consistently. #166364 Thanks @steipete, @hippoley.
• Enabled journal diagnostics explain where slow Git reads spend time without logging repository content. #166381 Docs · Thanks @steipete.
• Eligible completed Linux update receipts can reconcile changed device numbers so a later update can proceed. #166394 Docs · Thanks @steipete, @neilofneils404.
• Doctor can reconcile leftover session files without replacing current sessions or stopping unrelated repairs. #166397 Docs · Thanks @steipete, @coolcoolwcr.
• Failed update checks explain the requirement and identify the selected installation and runtime. #166410 Docs · Thanks @steipete, @TOM-WANG-GUANG.
• Background integrity checks retain successful verification through idle worker retirement for reuse after a clean restart. #166412 Thanks @steipete.
• Pre-repair captures can publish through an exclusive fallback when native no-replace rename is unsupported. #166425 Docs · Thanks @steipete, @MertBasar0.
• Exec and stdio MCP relays recover after Homebrew removes the running Node, with restart guidance for exact-runtime workers. #166430 Docs · Docs · Thanks @steipete, @mightyx.
• Session moves and approval preparation do more database work in the background, and deletion can finish when another cleanup has already retired the same placement. #166444 Thanks @steipete.
• Opening session file lists, downloading files, and viewing members avoids repeated database work while preserving current access checks. #166447 Thanks @steipete.
• Gateway error logs identify the session and reason when a requested wake fails. #166458 Thanks @steipete, @hawkit.
• macOS worker failures clean up background commands and release worker capacity without a stalled process lookup. #166478 Thanks @steipete, @hippoley.
• Interrupted Doctor recovery captures remain available and no longer hide nearby healthy captures. #166495 Docs · Thanks @steipete, @MertBasar0, @hippoley.
• Repeated session history and search work avoids unnecessary database reopening and validation. #166503 Thanks @steipete, @hippoley.
• Memory reindex locks no longer cause false held-database warnings in Doctor. #166517 Docs · Thanks @steipete, @noelff.
• Chat turns avoid rereading session details already known after saving. #166519 Thanks @steipete.
• Concurrent Gateway CLI calls load credentials with less overhead and stop clearly if required device identity cannot be read. #166523 Docs · Thanks @steipete.
• Rescue approvals and OpenRouter catalog storage move their database operations off the Gateway thread. #166528 Thanks @steipete.
• Windows update candidates can pass valid live handoffs despite changed launcher ancestry. #166530 Thanks @steipete, @wqj11a, @H0Nm4M1suZu.
• Cold device identity preparation uses the shared worker, and idle plugin cleanup skips unnecessary database work. #166539 Thanks @steipete.
• Failed managed updates retain actionable handoff reasons and late diagnostic details. #166540 Docs · Thanks @steipete, @dondonjang, @dahuangForToby.
• Creating sessions in large catalogs avoids repeated full scans when worker readers reopen. #166541 Thanks @steipete.
• An already-current update can safely restart the Gateway after refused Doctor maintenance stopped it. #166548 Docs · Thanks @steipete, @Shirleyyyhk.
• Multiline terminal table cells no longer cause unnecessary wrapping in neighboring columns. #166549 Thanks @ly85206559, @obviyus.
• Git update previews explain unavailable local target data instead of misleading remote-access advice. #166577 Thanks @Patrick-Erichsen.
• Allows plugin repair through pending migrations and official registry lag while preserving activation safeguards. #166588 Thanks @steipete, @CodeForcer.
• Reduces redundant copying when session updates are published to clients. #166591 Thanks @steipete.
• Restores Memory and Skills worker startup for source installations whose workspace is outside the checkout. #166606 Thanks @RomneyDa.
• Lets the Gateway serve requests before saving rebuilt plugin metadata when the registry is missing or stale. #166617 Thanks @steipete.
• Reduces startup work for one-shot Gateway calls and preserves credential resolution and port-cleanup errors. #166657 Thanks @steipete.
• Lets Doctor settle completed historical Codex tasks whose results cannot be delivered to their original requester, allowing otherwise-complete migrations to finish. #166668 Docs · Thanks @steipete, @dwonshin, @Torsie, @Hensarj, @HMKR2025, @flawedhumanoid-sudo, @BsnizND, @gisk0, @cortezhub, @KoennaOpenclaw, @juanbamorelli-source.
• Prevents device-identity inspection from changing SQLite files and invalidating Doctor migration plans. #166671 Thanks @steipete.
• Reduces chat-turn database bookkeeping and keeps failure cleanup tied to the accepted session. #166674 Thanks @steipete.
• Checks Windows Gateway task access before staging updates and bounds Task Scheduler waits with actionable errors. #166695 Thanks @steipete, @xiaozishan, @oxen-horse.
• Avoids repeating database validation after installing session tracking's known temporary objects. #166698 Thanks @steipete.
• Reduces CPU work when masking secrets in large contexts and logs while preserving masking behavior. #166701 Thanks @steipete.
• Reduces repeated session database work during replies and keeps accepted reads valid through legitimate compaction. #166703 Thanks @steipete, @vincentkoc.
• One-shot Gateway calls spend less time loading unused dependencies before making their request. #166722 Thanks @steipete.
• Large update inventories can use parallel hashing when host resources allow, while constrained hosts avoid the worker cost. #166741 Thanks @hannesrudolph, @vincentkoc, @steipete.
• Configuration reads correctly identify the optional root $schema key as valid but unset. #166745 Docs · Docs · Thanks @obviyus, @sxh313.
• Windows alternate spellings of the same state database share one connection and lifecycle identity, addressing an alias-related access conflict. #166755 Thanks @steipete, @taihartman.
• Chat turns avoid repeated conversation-history preparation and delivery checks while preserving fresh saved-history validation. #166804 Thanks @steipete.
• New-session initialization avoids waiting behind unrelated archival work for its final session count. #166822 Thanks @steipete.
• Configuration reads explain that missing version and migration metadata is managed automatically instead of suggesting writes that fail. #166830 Docs · Thanks @obviyus, @sxh313.
• Preserved old external-agent records without current session bindings no longer prevent Gateway startup after an update. #166839 Docs · Thanks @VACInc.
• Delegated update repair turns avoid spending their preparation allowance on repeated expensive permission snapshots. #166847 Docs · Thanks @steipete.
• Doctor JSON and lint inspections leave existing SQLite databases and support files unchanged while reporting pending upgrades. #166857 Docs · Thanks @steipete.
• Source-checkout Gateway status works when prepared input contents still match despite Git marking them modified. #166864 Docs · Thanks @steipete.
• The first reply after Gateway restart can start when healthy database connections open concurrently. #166867 Docs · Thanks @steipete.
• Eligible new chats use fewer storage commits to start their first turn, reducing work during bursts of session creation. #166891 Docs · Thanks @steipete.
• Older builds give compatible-build or pre-update-backup guidance when newer database state also has an unreadable catalog. #166906 Docs · Thanks @steipete, @shakkernerd.
• Administrators can capture CPU profiles, allocation profiles and heap snapshots on OpenClaw's supported Bun runtime. #166933 Docs · Docs · Thanks @steipete.
• Gateway start waits for an already-running Gateway to become ready and reports meaningful recovery guidance without restarting it. #166942 Docs · Docs · Thanks @RomneyDa, @odrobnik.
• Gateway diagnostics keep recognized startup failures visible after cleanup messages and direct unavailable services to useful checks. #166946 Docs · Thanks @RomneyDa, @odrobnik.
• Doctor warns early when configured local Tailscale Serve or Funnel cannot start because its backend needs attention. #166947 Docs · Thanks @RomneyDa, @odrobnik.
• Interactive Doctor asks to pause the managed Gateway during separately approved repairs and restores its prior running or stopped state. #166948 Docs · Docs · Thanks @RomneyDa, @odrobnik.
• Doctor previews the actual Gateway service repair before approval and identifies warnings that still require separate action. #166950 Docs · Thanks @RomneyDa.
• Inspecting newer database state preserves its files and gives compatible-build or matching-backup guidance across more diagnostic commands. #166963 Docs · Thanks @steipete.
• Chat and maintenance perform fewer repeated database checks while retaining existing storage and permission rules. #166966 Docs · Docs · Thanks @steipete.
• Successful background database checks survive the startup timing race and clean shutdown, avoiding repeated full scans on the next open. #166972 Docs · Thanks @steipete.
• Normal Gateway shutdown avoids repeated closed-session-observer warnings while chat handling and final session writes finish. #166978 Docs · Thanks @steipete.
• Doctor preserves the original agent for ownerless scheduled jobs when upgrading older multi-agent configurations. #166996 Docs · Thanks @steipete.
• Fresh database workers and maintenance checks avoid rebuilding expected database layouts already prepared by the Gateway. #167001 Docs · Thanks @steipete.
• Large saved subagent histories use less CPU and allocate less memory during maintenance and first session-list reads. #167010 Docs · Thanks @steipete.
• Background worktree cleanup causes fewer long pauses, and Git housekeeping no longer downloads missing objects automatically. #167030 Thanks @steipete.
• Linux Gateway stops allow more time for service-manager inspection, retry one inspection timeout and identify the stalled check. #167040 Thanks @steipete, @jlapenna.
• Overlapping update and Doctor commands can wait through brief coordination locks instead of failing immediately. #167051 Thanks @steipete.
• Manual session cleanup and transcript trimming keep more database work off the Gateway main thread. #167070 Thanks @steipete.
• Connecting with a target-file path preserves boundary spaces and consumes the intended file instead of its similarly named neighbor. #167116 Thanks @Alix-007, @obviyus.
• Doctor inspection can let Node finish closing native resources instead of forcing process exit. #167125 Thanks @RomneyDa.
• Cooperative Gateway suspension status can wait briefly for protected writes to finish before reporting deployment blockers. #167184 Thanks @steipete.
• Plugin property updates and session discovery avoid redundant allocation and repeated path resolution. #167235 Thanks @steipete.
• Gateway housekeeping skips redundant work when pairing notifications, queued recovery and worker-maintenance state are empty or unchanged. #167268 Thanks @steipete.
• Larger Gateway hosts keep fewer general-purpose reader workers, reducing potential retained worker memory. #167295 Thanks @steipete.
• Verified completed package updates can be cleared with update repair without discarding the retained package or recovery evidence. #167302 Thanks @shanselman, @dankarization, @fermentolactico, @chrisreed14, @catnipluss, @steipete.
• Repeated chat and scheduled-session cleanup releases unused agent lookup state that previously accumulated in long-running Gateways. #167308 Docs · Thanks @steipete.
• Idle checkout cleanup preserves affected work and pauses repeated failures when a shared Git repository has missing or corrupt objects. #167323 Thanks @steipete.
• Completed package recovery no longer permanently blocks later updates after verified cleanup failures, filesystem changes or replaced temporary ownership records. #167330 Thanks @RomneyDa, @shanselman, @dankarization, @neilofneils404, @fermentolactico, @catnipluss, @chrisreed14, @GitHoubi.
• Diagnostic logs distinguish long-held database reservations from writes waiting behind them and preserve session timing details. #167332 Thanks @vincentkoc.
• Deep Gateway status correctly identifies an unreadable state database, exits with failure and directs operators to verified-backup recovery. #167352 Thanks @steipete.
• Validated completed recovery history and obsolete-package cleanup stop blocking the next update while unfinished restoration remains protected. #167358 Thanks @RomneyDa.
• Doctor can recover scheduled agent work blocked by orphaned session windows while preserving the original database in a backup. #167366 Docs · Thanks @steipete, @changjunfeng8-hue.
• Background managed-checkout cleanup continues through unrelated sidebar, logging and model settings changes. #167367 Thanks @steipete.
• Backup verification rejects corrupted manifest text instead of reporting success with silently altered metadata. #167389 Thanks @Alix-007, @obviyus.
• Connect preserves its one-shot target file when malformed text encoding prevents the file from being read. #167399 Docs · Thanks @Alix-007, @obviyus.
• PowerShell completes quoted option values with valid choices instead of local-file suggestions. #167421 Thanks @LinzeShi.
• New managed checkouts use the fetched or cached remote default instead of silently substituting older local code when fetching fails. #167426 Thanks @steipete.
• Node identity inspection returns a parseable JSON failure and recovery hint when no identity exists. #167437 Docs · Thanks @Leon-SK668, @obviyus.
• Agent Git commands avoid automatic maintenance loops that could stall work in partially cloned repositories. #167443 Thanks @steipete.
• Overlapping storage measurements for independent session stores can finish sooner on hosts with spare CPU capacity. #167449 Thanks @steipete.
• Update checks avoid redundant temporary database copies and give clearer guidance when snapshot space runs out. #167451 Thanks @steipete, @iFadi, @dankarization, @azuretek, @davidcittadini.
• Managed worktree maintenance consolidates accumulated small Git packs, with cautious cleanup of abandoned temporary packs on Linux. #167453 Thanks @steipete.
• Doctor, updates and message delivery avoid premature interruption when clock differences create unusually long expiry delays. #167467 Thanks @steipete.
• Managed projects and worktrees can recover fetches blocked by missing Git commit objects while preserving local work. #167473 Docs · Thanks @steipete.
• Detailed update summaries, saved reports and human status output retain complete recorded warnings and recovery paths. #167509 Docs · Thanks @steipete, @GitHoubi.
• Interrupted Gateway starts clean up their own Tailscale claim so it does not block the next start. #167514 Docs · Thanks @steipete, @pandysp.
• Doctor and Gateway startup reduce repeated directory checks while capturing plugin files. #167516 Thanks @steipete, @palewire.
• Triage can open an interactive repair conversation in an installed, signed-in Google Antigravity CLI using --agent agy. #167530 Docs · Thanks @RomneyDa, @guarismo.
• Doctor reuses verified, unchanged plugin source files across repair phases, reducing repeated copying during maintenance. #167535 Docs · Thanks @steipete, @palewire.
• An old import record for another database no longer blocks Doctor from recovering available conversation files or allowing the normal update restart path. #167538 Docs · Thanks @steipete, @bipolar84.
• Removing agents or Claws shifts database work off the Gateway main thread while preserving shared history. #167547 Docs · Docs · Thanks @steipete.
• Skill learning, sandbox operations, checkout cleanup and discussions repeat less database work while checking current permissions before acting. #167571 Docs · Thanks @steipete.
• Gateway restarts avoid reusing an already-consumed request, and accepted update-failure reports can finish before shutdown. #167572 Docs · Thanks @steipete.
• An agent store awaiting preparation no longer prevents the Gateway from accepting connections. #167627 Thanks @steipete.
• Unrelated paused Git operations no longer block an eligible local default branch from advancing during managed-worktree creation. #167631 Docs · Thanks @steipete.
• Renaming, archiving, and changing session models spend less time waiting behind stale restart recovery. #167643 Docs · Thanks @steipete.
• Update repair can clear a pending recovery record when the intended candidate is already installed. #167648 Docs · Thanks @steipete, @cbhawthorne84.
• Failed updates avoid the unnecessary drain wait before restoring an installation when the Gateway is confirmed stopped. #167649 Thanks @steipete, @hannnnn-l, @alkor2000.
• Authentication-profile migration warnings name the cause and direct operators to Doctor repair. #167650 Docs · Thanks @steipete, @samson1357924, @vegisearchteam.
• Doctor archives removed accounts' old sender lists and continues later repairs without importing their permissions elsewhere. #167651 Docs · Thanks @steipete, @oshunter.
• Queued heartbeats no longer hold idle Gateway shutdown open, and deadline logs name pending owners. #167666 Docs · Thanks @steipete.
• Healthy Bun processes avoid false shared-state ownership expiry when a renewal worker starts. #167669 Docs · Thanks @steipete.
• Session and model reads just after restart wait briefly for the selected agent's database instead of immediately failing. #167674 Docs · Thanks @steipete.
• Simultaneous managed worktree creation avoids branch-registration conflicts and shares remote discovery and fetching. #167691 Docs · Thanks @steipete.
• Managed worktree removal avoids repeating shared repository maintenance for each checkout. #167695 Docs · Thanks @steipete.
• Doctor clears obsolete inspection warnings for plugins with no migration contract; a declared-empty contract with missing modules can still be misclassified. #167699 Thanks @steipete, @ruicozibon.
• Doctor verifies a restart only after the Gateway answers health checks, and gives diagnostic steps when a running service remains unresponsive. #167713 Docs · Thanks @steipete, @CamilleValley.
• Deferred update repairs are saved as pending and show the service-owner stop-and-rerun action needed to finish maintenance. #167714 Docs · Thanks @steipete, @ruicozibon.
• Android and Termux accept their existing state database after ordinary SQLite writes instead of mistaking it for a replaced file. #167719 Thanks @yxshee, @steipete, @grabowskidaniel359-design.
• Gateway status and Doctor clear temporary startup-preparation warnings once agents recover, avoiding unnecessary repair advice. #167723 Thanks @steipete.
• Normal lease retries stop producing misleading SQLite lock-failure warnings during healthy Gateway starts. #167754 Thanks @steipete.
• Updates with --no-restart report blocked state checks as unverified readiness and explain the follow-up restart and verification steps. #167760 Docs · Thanks @steipete, @ruicozibon.
• Triage detects pending package recovery even when Doctor and update history look healthy, and directs operators to update repair. #167782 Docs · Thanks @steipete, @waynegault.
• Update refusal messages identify the changed coordination database and point to profile-aware repair. #167794 Thanks @steipete, @waynegault, @masterdevill.
• Update reports explain which updater ran safety checks and clarify the permanent consequences of retiring recovery originals. #167802 Docs · Thanks @Patrick-Erichsen.
• Post-install Doctor can run its commands when an older updater omits the run ID, avoiding misleading admission failures. #167833 Docs · Thanks @steipete, @GoDiao.
• Agent-management requests report success after the saved agent changes are usable by the running gateway. #167894 Docs · Thanks @steipete.
• Expired worktree recovery references are removed together, keeping them available for retry when cleanup fails. #167899 Docs · Thanks @steipete.
• New pnpm gateway launchers follow the active package after replacement instead of pointing at a deleted version. #167905 Docs · Thanks @steipete, @SMF12138.
• Saving chats avoids repeated database validation work while rejecting malformed session records before writing. #167932 Docs · Docs · Thanks @steipete.
• Structured logs use less temporary memory while applying the same secret-masking rules. #167943 Thanks @steipete.
• Parallel managed worktrees from partial clones avoid repeated scans and prepare missing files before advancing local main. #167944 Docs · Thanks @steipete.
• Eligible large agent databases can open sooner while full integrity checks continue in the background. #168008 Docs · Thanks @steipete.
• Update repair can clear eligible abandoned updater records that kept later updates blocked after a manual installation change. #168043 Docs · Thanks @steipete, @farisoftfry.
• Declining Doctor maintenance still shows read-only findings in the same command without applying repairs. #168065 Docs · Docs · Thanks @RomneyDa.
• Doctor and update repair avoid repeated integrity scans when an existing repair record needs no change. #168069 Docs · Thanks @steipete, @Tosko4, @obviyus.
• Updates preserve a local plugin's settings and enablement when an unsafe installation path temporarily blocks discovery. #168111 Docs · Thanks @obviyus.
• Slow local database checks no longer cause false service-inspection timeouts during Linux Doctor and update repair. #168161 Docs · Thanks @obviyus, @Tosko4.
• Update repair can finish beside an active updater without disturbing that updater's ownership. #168197 Docs · Thanks @steipete, @farisoftfry.
• A busy state database no longer makes optional update progress logging abort the update. #168210 Docs · Thanks @steipete, @jlapenna.
• Gateway stops that exit the process avoid waiting for each native skills watcher to close individually, addressing a macOS cleanup stall. #168273 Thanks @obviyus.
• Windows plugin preparation uses the existing bounded copy path for small ordinary files, addressing long Gateway startup stalls. #168313 Thanks @RomneyDa.
• Lets users unblock older updaters by safely archiving an already-completed update receipt with an independent helper. #168388 Docs · Docs · Thanks @steipete, @k94118, @kimjinsuhh, @GoldenrodScript, @neilofneils404.
• Prevents temporary progress-output congestion from aborting update metadata inspection. #168493 Thanks @vincentkoc, @creanet-64.
• Lets commands and Gateway shutdown finish owned work and cleanup while restoring Node compiler defaults. #168587 Docs · Thanks @steipete.
• Offers a ready-to-copy device enrollment command for a background session host and correct hosting steps for paired devices. #168623 Docs · Docs · Thanks @obviyus.
• Supplies compatibility modules needed for cleanup when updating from stable 2026.10.1. #168740 Thanks @RomneyDa, @steipete.
• Preserves pending cleanup before shutdown during upgrades from published stable 2026.10.1. #168883 Docs · Thanks @RomneyDa.
• Preserves configured heap controls in captured service and installation-child update environments. #168889 Thanks @steipete, @bluet.
• Points failed updates after data migration to the existing repair command with clearer recovery instructions. #168905 Docs · Thanks @steipete, @ezimerman.
• Allows clean Gateway restarts when stored cloud workers are disconnected and defers their cleanup. #168907 Thanks @steipete.
• Avoids copying unrelated dependencies and failures caused by those files changing during updates. #168927 Docs · Thanks @steipete.
• Detects known incompatible Codex plugins early and gives useful plugin-update recovery instructions. #168930 Docs · Docs · Thanks @steipete, @ezimerman, @adlbas26.
• Reduces repeated file reads during package updates. #168958 Docs · Thanks @steipete.
• Avoids false Windows update rollbacks when the same recovery folder has a short-name or case-variant path. #168968 Thanks @RomneyDa.
• Reports accepted background Gateway updates as pending rather than failed. #168993 Docs · Thanks @steipete, @sayitfast.
• Clears stale completed-plugin repair warnings and keeps package-verification deadlines steady across clock changes. #169004 Docs · Thanks @steipete, @adenzhou1350, @WG-Mojo.
• Settles certain blocked 2026.9.8 recovery records and reports failure honestly while older recovery remains pending. #169027 Docs · Thanks @steipete, @chedong.
• Windows upgrades from 2026.10.1 can complete the plugin migration request expected by the older updater. #169031 Thanks @RomneyDa.
• Updates preserve existing webhook callback settings when an official plugin migration is deferred. #169035 Docs · Thanks @steipete, @sirius1024.
• Verified successful updates retain success when result publication fails, and repair reports stopped channels with recovery guidance. #169037 Docs · Thanks @steipete, @kvnloo, @RomneyDa.
• Eligible stale update recovery records can settle automatically once the live installation is verified. #169038 Docs · Thanks @steipete, @mkdelta221, @waynegault, @clawdgawd, @yoyo837, @lishewen, @nukk-pain, @jullianatolentino, @joshpetras, @creanet-64.
• Confirmed pacman-managed installations receive package-manager update guidance instead of self-update. #169055 Docs · Thanks @steipete, @sahilsatralkar, @Pegorim.
• Clock corrections no longer distort the recovery timeout after an interrupted update. #169168 Docs · Thanks @SunnyShu0925.
• Incomplete update recovery captures identify affected databases and relevant backup warnings. #169282 Docs · Thanks @DonnieFi.
• Update-history approval reduces blocking Gateway writes; the later host-approval handoff fix also releases its writer while permission is pending. #169283 Thanks @DonnieFi.
• Update warnings explain why an optional candidate configuration read failed while preserving the existing service definition. #169284 Docs · Thanks @DonnieFi, @RomneyDa.
• Updates recognize alternate installation path spellings and keep workers on their original runtime, with abandoned-runtime cleanup restored. #169292 Thanks @RomneyDa.
• Doctor distinguishes unknown plugin migration backup coverage from bundled migrations already covered by core, with useful next steps. #169295 Docs · Docs · Thanks @RomneyDa.
• Windows SQLite recovery no longer rejects its own protected temporary snapshot directory. #169397 Thanks @steipete.
• A separately installed helper can clear a narrowly eligible unfinished publication after verified filesystem device drift while preserving package and evidence. #169410 Docs · Docs · Thanks @patrickschiller, @RomneyDa.
• New recovery copies preserve legacy Canvas documents and Memory Core files before migrations change them. #169436 Thanks @RomneyDa.
• Git updates clean owned temporary worktrees even when progress reporting fails and retain the original failure for diagnosis and recovery. #169454 Docs · Thanks @hannesrudolph.
• Equivalent Windows task account names no longer prevent the Gateway restarting after an update. #169481 Docs · Thanks @steipete.
• Update and Doctor checks accept explicitly empty agent rosters and reject unsafe Canvas folder links before reading them. #169516 Thanks @RomneyDa.
• Npm-global updates avoid scanning unrelated global tools while retaining dependencies OpenClaw actually needs. #169551 Thanks @RomneyDa, @goutamadwant, @holgergruenhagen.
• Doctor can continue an update after a later process-inspection failure once its owning updater was already verified. #169563 Thanks @RomneyDa.
• Updates can restart a revalidated preserved Gateway service after refresh safely refuses to rewrite its definition. #169658 Docs · Thanks @steipete.
• Update-time Doctor can verify its updater without inspecting higher supervisors and retains command identity when binding fails. #169665 Docs · Docs · Thanks @steipete.
• Doctor imports deferred ACP metadata from verified archived legacy sessions once its plugin becomes available. #169759 Thanks @RomneyDa.
• Doctor can attempt Gateway restoration after transient resource-close errors settle successfully on retry. #169783 Thanks @RomneyDa.
• Mac updates avoid self-lock timeouts when temporary directories have equivalent path spellings. #169860 Thanks @steipete.
• Older Node updaters retain valid ownership across the heartbeat clock transition instead of unnecessarily rolling back. #169915 Thanks @vincentkoc.
• New or repaired pnpm Gateway service commands survive package replacement, and nested pnpm installs are recognized for updates. #170003 Docs · Thanks @steipete, @xilopaint, @k-sekula.
• Background database checkpoints pause during retained offline maintenance to keep Doctor and update recovery captures stable. #170022 Docs · Thanks @steipete.
• Configured agents can start despite bogus legacy sentinel registrations, and standalone Doctor can clean invalid registrations. #170043 Docs · Thanks @steipete, @ourcoms2018.
• Update repair can reconcile an interrupted installation after filesystem device numbers change and unblock later updates. #170046 Docs · Thanks @steipete, @patrickschiller, @SergeiAug.
• Doctor rejects unsupported legacy runtime files before changing the database or stopping the Gateway, preserving the intermediate-upgrade path. #170051 Docs · Thanks @steipete, @bautrey.
• Updates requested through chat can finish Doctor cleanup without a false maintenance-permission failure. #170070 Thanks @hannesrudolph, @ezimerman, @steipete, @yoyo837, @aia007731-lab, @justinforsound.
• Updates avoid reopening migrated data during configuration saves and can report a recorded schema migration as advisory; same-version wrong-build concerns remain unresolved. #170130 Thanks @steipete, @angeliti999.
• Native pnpm 12 updates keep the openclaw command pointing at the installed package after temporary files are removed. #170131 Docs · Thanks @steipete, @goutamadwant, @k-sekula.
• Doctor can find abandoned npm updater runtimes that leave later updates blocked by an extra recovery-database link. #170170 Thanks @steipete, @sunnyshu0925, @warhangel.
• Managed updates use per-user temporary storage and can recover eligible receipts that point into another user's directory. #170174 Docs · Thanks @steipete, @holgergruenhagen.
• Older updaters stop rejecting a validated candidate solely because Doctor changed version or wizard bookkeeping. #170195 Thanks @steipete, @maestrokling, @frxiaobei.
• Failed-update reports keep Doctor recovery instructions when a healthy Gateway does not prove Doctor writers have safely finished. #170207 Thanks @steipete.
• Update repair loads the target installation's settings before choosing a model instead of stopping with a configuration-not-prepared error. #170276 Thanks @steipete.
• Gateways with deferred database preparation can finish startup instead of leaving worker recovery and client readiness waiting on each other. #170348 Docs · Thanks @steipete.
• Update snapshot capture stops falsely treating an unchanged agent database as changed by its own read. #170351 Docs · Thanks @DonnieFi, @RomneyDa.
• Large configurations can reach the updater intact instead of causing candidate-read failures during Gateway service refresh. #170373 Docs · Thanks @RomneyDa, @DonnieFi, @steipete.
• Slow update snapshots stop consuming the next command's inspection time, and genuinely deferred checks get follow-up guidance. #170374 Thanks @RomneyDa, @DonnieFi, @steipete.
• Final update summaries and saved history stop calling completed plugin upgrades unfinished and retain actionable warnings for pending ones. #170413 Docs · Thanks @RomneyDa, @DonnieFi.
• Stalled transcript indexing gets a five-second grace then cancellation so it does not indefinitely block Gateway shutdown. #170493 Docs · Thanks @steipete, @Oldrich333, @O96a.
• Update and Doctor recovery can continue when optional dreaming schedule checks or settled temporary-snapshot cleanup fail. #170754 Thanks @RomneyDa, @aniruddhaadak80, @anhnt3.
• Standalone Doctor can repair matching account state when an app-owned Gateway is verified offline. #170768 Docs · Thanks @RomneyDa, @shanselman.
• Expected exclusive Doctor maintenance no longer appears as a database checkpoint failure in logs or health. #170783 Thanks @steipete.
• Update state inspection continues when an advisory progress message temporarily cannot be written. #170955 Thanks @steipete.
• Early checks for 2026.9.2 upgrades keep database changes private so a refused update can preserve compatibility with restored code. #171015 Docs · Thanks @steipete, @bamorem.
• Doctor recognizes equivalent imported attachments instead of falsely rejecting healthy legacy session history. #171388 Docs · Thanks @he-yufeng, @steipete, @brookpw, @junnie-stack.
• Terminal update progress waits through temporary Doctor maintenance contention and resumes when it clears. #171454 Docs · Thanks @RomneyDa, @DonnieFi.
• Update failures during snapshot preparation identify the failed snapshot step and give matching retry guidance. #171457 Thanks @RomneyDa, @DonnieFi.
• Linux update and Doctor cleanup can reclaim abandoned updater folders despite inaccessible working directories of narrowly identified user-session processes. #171488 Docs · Thanks @RomneyDa, @DonnieFi.
• Gateway worker setup avoids a silent successful exit when Node diagnostic flag checks meet an explicit heap limit. #171490 Thanks @RomneyDa.
• A busy updater releases its database write lock while waiting for permission to save status, giving scheduled jobs and cleanup a chance to proceed. #171493 Docs · Thanks @RomneyDa, @DonnieFi.
• Doctor and updates allow terminated child processes longer to disappear before reporting a cleanup failure. #171497 Thanks @RomneyDa, @steipete.
• Updates that change core and plugins still remind users who choose --no-restart to run the appropriate Gateway restart command. #171687 Thanks @steipete.
• Managed updates wait through brief maintenance overlap and preserve recovery permission so an otherwise completed update is less likely to leave the Gateway stopped. #171699 Thanks @steipete, @fdovarela.
• State reads can resume after a database-resource cleanup failure, avoiding a Gateway restart solely to restore state-dependent work. #171709 Thanks @RomneyDa, @residentb2b, @homeworld520, @sunny251008-debug.
• Doctor avoids silent empty reports when a child closes its input, while completing admitted repairs before exiting if Doctor's own output closes. #171780 Thanks @steipete, @dtump, @joshpetras, @cbhawthorne84.
• Updates of standard pnpm installations find the installation's own package manager and bin folder despite the Mac Gateway service's restricted PATH. #171936 Thanks @steipete, @AntaresGG, @MoerAI.
• The setup assistant diagnoses problems without applying repairs or migrations to local configuration and state. #171960 Thanks @vincentkoc.
• Later updates can settle an unfinished completion record when the already-installed package verifies successfully. #171993 Thanks @steipete, @teetee2tone.
• Updates avoid failing after a short Doctor command has already completed successfully. #172132 Thanks @steipete.
• If the new version's Doctor fails before database writes, updates can restore the previous package and databases without mistaking preliminary reads for changes. #172151 Thanks @RomneyDa.
• Updates can inspect busy agent databases without refusing solely because normal writes reuse old log space. #172196 Thanks @RomneyDa.
• A verified update stays installed when Doctor cannot inspect its launching process, with a warning to run openclaw update repair. #172247 Thanks @steipete, @homeworld520.
• Read-only-root Kubernetes deployments need XDG_CACHE_HOME=/home/node/.openclaw/.cache or an equivalent writable cache alongside the image update. #172377 Docs · Thanks @saftall, @steipete.
• Non-interactive setup preserves its actual result and credential-activation instructions when temporary credential files remain locked. #172379 Thanks @skyzhao1223, @steipete, @giannis88.
• A verified, ready Gateway update can finish with named channel warnings instead of reporting failure because a messaging channel needs attention. #172494 Thanks @steipete, @sbc2016.
• A timed-out optional shell-completion refresh can warn without failing an otherwise successful update after cleanup is confirmed. #172502 Thanks @steipete.
• Mac updates can proceed from an external login shell when native checks prove it is outside the Gateway service despite unreadable parent ancestry. #172640 Thanks @steipete, @giulianoaccorsi.
• Linux upgrades can regenerate eligible owned legacy service files with private permissions without requiring manual chmod repair. #172790 Thanks @obviyus, @wei887-os.
• Doctor avoids repeatedly rescanning migration manifests when stale import records refer to archived transcripts, reducing this cause of stalled repair and update checks. #172890 Thanks @steipete, @DongwonGim, @skyzhao1223.
• Updates and Gateway status launched through a remote node avoid mistaking the node service for the Gateway. #173304 Thanks @steipete, @Leon-SK668, @davidcittadini.
• Verified updates can continue with a repair warning when follow-up plugin inspection cannot finish, instead of rolling back solely for incomplete checks. #173341 Docs · Thanks @steipete, @yoyo837.
• Updates can move past an unhealthy Gateway in the same installation with a warning, leaving time to start and fully verify its replacement. #173358 Docs · Docs · Thanks @steipete.
• Gateway startup can follow a replacement model preparation instead of failing when an older preparation is cancelled. #173366 Thanks @steipete, @Jackten.
• --no-restart skips the global package updater's restart step, while a running Gateway's package watcher can still restart it after files change. #173409 Docs · Thanks @steipete, @mwbroxson.
• Updates wait briefly for a newly started Gateway to publish ownership before reporting restart preparation failure. #173588 Docs · Thanks @obviyus.
• Package updates verify the running build as well as its version; --no-restart still permits a package-watcher restart. #174649 Docs · Thanks @RomneyDa.
• Keep scheduled workspace cleanup and worker-bundle maintenance running on paired nodes. #168780 Docs · Thanks @steipete.

• Supporting session integration rechecks access during compaction, membership reads, and downloads and explains unreadable device identity storage. 2770dae
• Suggestions recheck current sharing access and session identity before returning results. 6a42f13
• The first profile read after GitHub sign-in can complete while initial identity verification finishes. 90ad159
• Gateway logs identify later rejected proxy sources while keeping repeated warnings bounded. #134404 Thanks @lzhan011, @abacha, @obviyus.
• PluralKit lookup failure diagnostics mask reflected credentials while retaining useful error details. #141317 Thanks @hugenshen, @obviyus.
• Plugin hooks must obtain fresh approval before changing trusted inline-approved tool arguments. #142700 Docs · Thanks @giodl73-repo, @giodl73.
• Switching approval-policy targets asks before discarding unsaved edits, and Cancel keeps the draft. #149181 Docs · Thanks @tommyjoseph, @steipete, @obviyus.
• Truncated or errored replies keep recognized unfinished reasoning hidden and explain empty output-token-limit results. #156939 Docs · Thanks @quinszouls, @steipete, @ingenuity2k.
• Configured plugin tool guards run in standalone agent exec and can block commands before execution. #159333 Docs · Thanks @joseairosa, @obviyus.
• Built-in credential masking covers secrets across long tool output while avoiding several redaction stalls and crashes. #161480 Thanks @eleqtrizit.
• Remote-node commands no longer fail an approval identity check when the prepared plan omitted session metadata. #161980 Thanks @Program-123, @steipete, @gaojian125, @enderwillsaveyou, @rozenpl, @daddydubzzz.
• Exec-approval guidance now explains that approved command values replace later command-field edits while changed environment overrides remain rejected. #164216 Docs · Thanks @ericcurtin, @cristianoc, @obviyus.
• Agents can ask explicitly approved peers for help and receive the reply without gaining access to their conversation histories. #164444 Docs · Thanks @steipete, @prasad-yashdeep, @cipp-ashe.
• Anchored conversation history rejects transcripts whose recorded owner does not match the requested conversation. #164512 Thanks @steipete.
• Delayed chat startup stays bound to the current caller and conversation while initial session lookups run through workers. #164682 Thanks @steipete, @galiniliev.
• Oversized policy-watch intervals no longer trigger near-continuous evaluations and overflow warnings. #164850 Docs · Thanks @KrasimirKralev, @obviyus.
• Scheduled commands retain valid approval through process startup and account for each approved use once, refusing uncertain or stale launches. #164864 Docs · Docs · Thanks @steipete.
• Covered scheduled messages and retries recheck permission at delivery handoff; revoked work cannot accept protected read results or commit an internal reply. #164906 Docs · Docs · Thanks @steipete.
• A2A peers with missing environment-backed credentials can no longer authenticate or send reference text as a token. #165021 Thanks @eleqtrizit.
• Code Mode cells cannot replace real host tool replies with forged failure results. #165071 Docs · Docs · Thanks @steipete.
• Incognito reset safely removes its managed checkout and branch, preserving unsaved-work snapshots or reporting why a checkout was kept. #165185 Docs · Docs · Thanks @jayzhou2309, @steipete, @obviyus.
• Localized configuration guidance clarifies outbound agent messaging permissions without implying access to other sessions. #165213
• Restricted senders can delegate only to hidden helpers in the same agent and workspace. #165418 Thanks @steipete.
• Answering or cancelling an agent question checks the current session before applying the response, refusing stale or replaced callers. #165552 Thanks @steipete.
• Queued chat messages retain current sharing checks while repeated permission reads move off the Gateway thread. #165600 Thanks @steipete.
• Replies with MCP app or connection buttons retain their original session-writer delivery checks. #165760 Thanks @steipete.
• Tlon group-DM author claims can no longer grant owner commands or elevated permissions. Previously pending DM requests need the sender to resend after approval. #165801 Docs · Thanks @eleqtrizit.
• HTTPS through the secret egress proxy works with strict TLS clients, including modern Python on macOS; restart the Gateway and start new commands after updating. #165818 Docs · Thanks @steipete, @vortexopenclaw, @RyanSandoval, @A1fred-AI, @gisk0, @jfk-hue.
• The source-map-js dependency is updated to its patched release to address pathological source-map processing. #165902 Thanks @steipete.
• Patched MCP dependencies preserve saved credentials when background refresh encounters a different login issuer and require explicit login to switch. #166217 Thanks @Patrick-Erichsen, @vincentkoc, @RomneyDa.
• Previously verified Tailscale GitHub users retain profile and session access through retryable GitHub outages. #166254 Docs · Thanks @obviyus.
• Approve and deny actions acknowledge saved decisions before slow notification delivery finishes. #166396 Docs · Thanks @steipete.
• Manual history trimming refuses requests whose access or selected session changed while older records were being restored. #166506 Thanks @steipete.
• Stops canceled or no-longer-authorized worker requests before further device preparation. #166608 Thanks @romneyda.
• Lets operators require a named worker profile and reject local fallback before inference. #166611 Thanks @RomneyDa.
• Applies required-worker policy to explicit destinations, moves and recovery while preserving accepted workspace results and Stop. #166613 Thanks @RomneyDa.
• Recovers watched child runs under their recorded agent after restart and rejects coordinate actions with stale screenshots even when target references are blank. #166691 Thanks @steipete.
• Node commands approved before the deadline can still run when forwarding crosses that deadline within the valid handoff window. #166971 Docs · Thanks @steipete.
• Native Codex follow-ups are checked against the sender's permitted conversation tree, and restricted instructions match available tools. #167075 Docs · Thanks @vincentkoc.
• Execution-policy edits and cleanup move database writes off the Gateway main thread while current permissions remain enforced. #167115 Thanks @steipete.
• Malformed text encoding in configuration files or piped patches is rejected before saved values can change. #167118 Thanks @Alix-007, @obviyus.
• LINE and Tlon mask recognizable credentials echoed in failed-request diagnostics. #167234 Thanks @xialonglee, @obviyus.
• Failed exec-policy commands requested with JSON return one structured error document and a nonzero exit. #167279 Thanks @leon-sk668, @obviyus, @masatohoshino.
• Device commands return structured JSON refusal reasons while preserving nonzero exits and human recovery guidance. #167294 Docs · Docs · Thanks @Leon-SK668, @obviyus, @masatohoshino.
• Authorized agent commands avoid spurious launch failures when existing shared-database features initialize again. #167486 Thanks @steipete.
• Secrets guidance identifies unsupported command consumers before asking users to enter a protected credential. #167976 Docs · Docs · Thanks @romneyda, @markma84.
• Reply permission checks refuse to authorize a prepared response while its database execution remains unfinished. #168029 Thanks @steipete.
• Unfinished hidden reasoning stays out of Chat Completions replies when a native tool call is followed by a token limit or content filter. #168294 Thanks @steipete.
• Network tool results keep their untrusted-source label through Tool Search and Code Mode and into later replies. #169560 Docs · Thanks @RomneyDa, @farisoftfry.
• Gateway exec blocks recognized variable-bearing SQLite file URLs from opening live OpenClaw state. #170253 Docs · Thanks @eleqtrizit.
• Doctor preserves saved OAuth profiles when an older auth file is corrupt or unreadable and warns where to inspect it. #170256 Docs · Thanks @steipete, @nosmile99.
• Graphical pairing keeps browser credentials out of launcher arguments by using private files; access to those files remains sensitive and cleanup is best effort. #170289 Thanks @eleqtrizit.