On this page
On this page
Release notes
v2026.11.5
v2026.11.5
AI agents and tools can read these release notes as plain Markdown.
OpenClaw 2026.11.5 release notes document 138 bug-fix entries and 18 quality-of-life entries.
Quality of life
Chat and sessions
-
Adds a keyboard link to skip the chat transcript and reach the message composer. In archived chats it leads to the footer and Unarchive.
More details
#180896. Thanks to @goutamadwant, @liuwqgit, @obviyus and @carlopignatari-boop.
-
Simplifies paired-device folder selection in New Session. Admins can browse, other writers can enter an absolute path, and the selected folder name appears in the workspace chip. The menu also explains when the selected model cannot use device folders.
-
Adds global and per-agent limits for client-side compaction summaries. The default remains 16,000 characters; larger summaries use more context and cost more to replay.
More details
#196207. Thanks to @aniruddhaadak80, @garrytan, @RomneyDa and @vincentkoc.
-
Reduces JavaScript loaded when opening Chat by deferring the Template creator and workspace conflict recovery controls until needed.
More details
#187783, #187830. Thanks to @vincentkoc.
-
Clarifies the error for oversized compressed conversation archives, suggesting uncompressed JSONL as an alternative.
More details
commit 2b08b6f. Thanks to @RomneyDa.
-
Extends existing widgets to compatible private Telegram bot chats, including private topics. Telegram message widgets are opt-in and off by default; older Telegram clients cannot render them.
More details
#196991, #198022. Thanks to @obviyus for the widget transport and private-topic work, and @dankarization for private-topic routing.
Tools and providers
-
Adds a directory of eligible deferred MCP tools to Tool Search, so agents can discover listed tools without a separate search. Large catalogs may still require a search for omitted names.
More details
#196252. Thanks to @jalehman, @GrahamMcBain and @RomneyDa.
-
Adds compact input signatures to MCP Tool Search results, helping agents make straightforward calls without requesting every full tool description. Complex or missing schemas may still need the full description.
More details
#196266. Thanks to @jalehman, @GrahamMcBain and @RomneyDa.
-
Adds Gemini Decision 1 as a provider option for Boolean, choice and score evaluations of text, JSON or one supported image. It requires a Gemini Developer API key; evaluation sends evidence to Google and incurs API charges. This is not a chat or Vertex model.
More details
commit 01b98da. Thanks to @RomneyDa, @vincentkoc and @obviyus.
-
Adds a read-only GitHub publication readiness check so agents can inspect a session’s source and existing publication requests without starting a publication.
More details
#187152. Thanks to @fuller-stack-dev.
Maintenance
-
Extends retries for ClawHub reads during brief outages. Persistent failures may take up to 30 seconds longer to report.
-
Allows an idle cron worker to be reclaimed under memory pressure when automatic scheduling is disabled. Manual runs remain available.
More details
#196387. Thanks to @RomneyDa and @vincentkoc.
-
Allows idle skills discovery workers to stop under critical memory pressure without losing the skill list; refreshing can still pick up edits.
More details
commit 76acafe. Thanks to @RomneyDa and @vincentkoc.
-
Adds the maintenance owner kind to offline state database contention errors alongside the existing role and process ID.
More details
commit 016bcbe. Thanks to @RomneyDa.
-
Improves recovery cleanup refusal messages to identify which destination check failed.
More details
commit aea2145. Thanks to @RomneyDa.
-
Adds translations for the Mac app's send-failed retry and page-open error messages.
More details
commit f8cc0d9. Thanks to @RomneyDa.
-
Warns about detected model billing route changes after successful Gateway configuration writes. The warning is advisory and may be skipped.
More details
commit baba5d8. Thanks to @RomneyDa, @vincentkoc and @etzelm.
Automation notifications
-
Adds best-effort notices in the owning conversation when a manually started automation is skipped or fails before it starts.
More details
commit 6604030. Thanks to @RomneyDa.
Bug fixes
Automations
-
Fixes scheduled main-session automations failing before a model call or failing to send messages after the main session was used from a channel DM with the default DM scope.
More details
#180603. Thanks to @masatohoshino, @obviyus and @Suiah98.
-
Fixes quiet automation checks posting false failure messages and moving conversations to the top when their conditions do not fire.
-
Fixes interrupted
cron listandcron showcommands waiting for a timeout and printing a late failure. Already-sent add or edit requests still wait for their result.More details
#187726. Thanks to @SunnyShu0925 and @obviyus.
-
Fixes scheduled automation completion events omitting an available task run ID. A manual request run ID still takes precedence.
More details
commit b15ea1e. Thanks to @RomneyDa and @vincentkoc.
-
Fixes retained automation history failing to load when a scheduled run rotates its session during a read. Sharing access is checked again.
More details
commit be8e76f, commit 75812b5. Thanks to @RomneyDa and @vincentkoc.
-
Fixes scheduled plugin commands being rejected after their conversation is reset before the command starts.
More details
commit 94e1bfb. Thanks to @RomneyDa and @vincentkoc.
-
Fixes scheduled notices being able to remove their own automation when sent to the owner conversation.
More details
commit 28ebe54. Thanks to @RomneyDa and @vincentkoc.
-
Fixes manually forced runs of paused automations staying marked as running after they are disabled.
More details
commit c12366a. Thanks to @RomneyDa.
-
Fixes automation time limits not covering waits before a run starts.
More details
commit c12366a. Thanks to @RomneyDa.
-
Fixes manually started automations failing to run in the same conversation after the requesting turn ends.
More details
commit 6604030. Thanks to @RomneyDa.
-
Fixes queued channel messages being claimed after restart admission closes, and resumes delivery if the restart is rolled back.
More details
commit b3f5bdd. Thanks to @RomneyDa.
Gateway and recovery
-
Fixes Gateway shutdown or restart reporting a cleanup failure after an exec was refused before it started.
More details
#187137. Thanks to @fuller-stack-dev.
-
Fixes failed Gateway startups leaving the CLI waiting for usage work to finish.
-
Fixes the Gateway refusing to start after a successful configuration health check.
-
Fixes Doctor waiting indefinitely in selected startup checks and leaving startup work running when interrupted. Stalled checks now report which step timed out.
More details
#187936. Thanks to @steipete, @RomneyDa and @JordanNewell.
-
Fixes Gateway shutdown losing the chance to save sessions for resumption when a restart request stalls.
-
Fixes Gateway shutdown waiting on read-only worker-environment inspections when their providers do not respond to cancellation.
-
Fixes configuration updates hanging when they trigger a channel reload.
More details
#192701. Thanks to @RomneyDa and @vincentkoc.
-
Fixes Gateway startup and affected shared-state reads failing immediately during brief database schema contention.
-
Fixes draining Gateway restarts being deferred immediately by late final chat writes when time remains for them to finish. If writes remain pending at the deadline, the refusal now includes useful diagnostics.
More details
#196493. Thanks to @steipete, @RomneyDa and @vincentkoc.
-
Fixes accepted queued app messages disappearing during a graceful Gateway restart. Recovered messages retain their original limits and must still have current permission.
More details
#196923. Thanks to @joshavant, @RomneyDa and @vincentkoc.
-
Fixes queued sends after a graceful Gateway restart using broader permissions than they had when accepted. Sends whose original authorization cannot be verified fail and can be submitted again.
More details
#196540. Thanks to @joshavant, @RomneyDa and @vincentkoc.
-
Fixes Gateway requests such as subscriptions and artifact downloads failing when bulk session reads fill the worker queue.
-
Fixes startup recovery skipping agents whose databases are still preparing. Queued messages left by a crashed process are marked cancelled in chat history once their database is ready.
More details
commit 7f99793. Thanks to @RomneyDa and @vincentkoc.
-
Fixes Gateway start errors showing a generic health-check timeout when a more specific failure is known.
-
Fixes delayed cancellation of model sign-ins when the Gateway shuts down.
More details
commit 222d836. Thanks to @RomneyDa and @vincentkoc.
-
Fixes rescue commands losing their findings when they finish with a nonzero exit status.
More details
commit f6c8beb. Thanks to @RomneyDa and @vincentkoc.
-
Fixes eligible sessions interrupted during an upgrade from legacy restart recovery remaining idle. The bridge is best effort, and sessions stopped for more than six hours need a new message to continue.
-
Fixes Gateway status and Doctor’s macOS stale-updater checks using an unrelated CLI profile instead of the installed service’s settings.
More details
#187901. Thanks to @steipete and @shakkernerd.
-
Fixes Gateway CLI commands causing heavy disk activity and holding up other writes on affected large SQLite databases.
-
Fixes a Gateway crash when a progress update retries after its channel plugin has closed. Progress updates stop for that turn.
-
Fixes startup admission failing when a previous state owner exits during the check.
Security and privacy
-
Fixes browser actions allowing cross-origin requests after a rejected preflight.
More details
#185028. Thanks to @eleqtrizit.
-
Fixes Tlon citations expanding from unlisted channels when channel auto-discovery is off. Verified owner one-to-one DMs retain their broader access. In shared DM sessions, owner-cited content may be visible to other senders.
-
Removes PR attribution instructions from conversations when attribution is off or in Incognito, and keeps opt-in guidance specific to the requester and request. Start a new session to discard older retained guidance.
More details
#187805. Thanks to @lastguru-net and @beru-ant-king.
-
Redacts recognized credentials from quoted context in compaction summaries. Original transcripts are unchanged, and redaction does not cover every secret.
More details
commit 21bb4e7. Thanks to @RomneyDa, @vincentkoc and @obviyus.
-
Prevents browser tool tab and open results from exposing credentials contained in browser connection URLs.
More details
commit c78d58a. Thanks to @sallyom and @RomneyDa.
-
Fixes a Control UI reply after plugin reload going to the last external chat instead of remaining internal.
More details
commit 2b08b6f. Thanks to @RomneyDa.
Agents and tools
-
Fixes Goals created during a chat stopping after delegated work returns instead of continuing automatically.
More details
#183706. Thanks to @HermanZeng and @fuller-stack-dev.
-
Fixes agent turns stopping after ordinary tool errors when optional error details cannot be read.
-
Fixes output-limited agent turns skipping their selected fallback after the input was already saved.
-
Fixes matching MCP list and scalar results being sent to the model twice while retaining distinct text.
More details
#196243. Thanks to @jalehman, @GrahamMcBain and @RomneyDa.
-
Fixes turns waiting too long on provider rate limits after tools have already run. When the requested delay exceeds the configured limit, the turn ends with a rate-limit error so queued work can continue without replaying completed actions.
-
Fixes retained subagent resumes being rejected after a task metadata update when the owner has not changed.
More details
commit 4aa7856. Thanks to @RomneyDa.
-
Fixes one-shot Codex runs being forced to stop before their requested shutdown grace has elapsed.
More details
commit 2d03432. Thanks to @RomneyDa.
-
Fixes state lease cleanup giving up too early after a lease is renewed while offline maintenance delays its release.
More details
commit 2d03432. Thanks to @RomneyDa.
-
Fixes MCP command approval requests failing to reach the right ACP turn when multiple turns are pending.
More details
commit 12691f9. Thanks to @RomneyDa and @vincentkoc.
-
Fixes completed Workboard cards with stale running executions blocking new work for the same owner, including on other boards.
More details
#187880. Thanks to @atomicdjt.
-
Fixes tool progress labels freezing on commands with deeply nested paths followed by spaces.
More details
commit 5fe03ed. Thanks to @RomneyDa, @vincentkoc and @LeizhenZhang.
Updates and maintenance
-
Fixes Windows packaged updates rejecting a valid recovery copy when equivalent paths use different spellings.
-
Fixes successful updates leaving old recovery copies behind when the Gateway writes to its database during cleanup.
More details
#189628, #191275. Thanks to @RomneyDa, @steipete and @vincentkoc.
-
Fixes update progress checks blocking Doctor maintenance and leaving automatic database restoration unavailable.
-
Fixes update summaries omitting a failed post-plugin Doctor and its repair instructions.
-
Fixes updates and automatic database restoration checks failing when another connection opens or closes an unchanged state database.
More details
#193134. Thanks to @fuller-stack-dev.
-
Fixes restored global npm installations becoming inaccessible to non-root services after a failed update. Existing access is preserved; access lost in an earlier rollback is not restored.
More details
#193272. Thanks to @obviyus, @RomneyDa and @vincentkoc.
-
Retries briefly busy files during Windows backup capture and identifies files that remain unreadable. Incomplete backups are still refused, and the running updater must already contain this fix.
-
Fixes updates stopping when a successful candidate database check cannot remove its temporary copy.
More details
#193702. Thanks to @fuller-stack-dev and @RomneyDa.
-
Fixes Doctor and Gateway startup failing after successful SQLite snapshot checks when busy NFS cleanup cannot finish. Cleanup may be deferred; unrelated read and integrity failures remain errors.
-
Fixes eligible Gateway updates leaving the invoking CLI on a different installation when the service has only named environment overrides. Timed updates also retain the selected target when the service uses a different installation. The running updater must already contain this fix.
More details
#196371. Thanks to @RomneyDa and @vincentkoc.
-
Fixes hosted Doctor diagnostics getting stuck and copying large databases while the Gateway is running. Database, plugin, credential and workspace checks are deferred; stop the Gateway and run
openclaw doctor --linton its host for a full report. -
Fixes Doctor being unable to clear a persistent port-preparation warning on affected SQLite databases after upgrading from 2026.10.10.
More details
-
Fixes Git updates encountering SQLite catalog errors after activation when the running updater contains this fix. If an older source updater encounters this error, preserve recovery artifacts and follow the manual source update guidance before retrying.
More details
commit 8bedc56. Thanks to @RomneyDa and @vincentkoc.
-
Fixes managed updates of PostgreSQL-backed foreground Gateways failing before handoff is ready with a SQLite schema error.
More details
commit 7091456. Thanks to @RomneyDa and @vincentkoc.
-
Fixes Doctor refusing a SQLite migration because the newly activated Gateway wrote to its database during an update without a service restart.
More details
commit ee28307. Thanks to @RomneyDa.
-
Fixes held agent databases being opened during startup before they are explicitly restored. Stop the Gateway before restoring an older preserved database if it needs a schema migration.
More details
commit 5d93d5e. Thanks to @RomneyDa and @vincentkoc.
-
Fixes bundled dependencies failing to load after installation for packages built with the affected pnpm-linked layout.
More details
commit 44d9651, commit 8878714. Thanks to @alfredjbclaw, @RomneyDa and @vincentkoc.
-
Fixes MISSING_EXPORT errors reported with npm and Docker.
More details
commit a8a974d. Thanks to @RomneyDa and @vincentkoc.
-
Fixes a misleading timeout hiding the original PostgreSQL admission error when a schema retry expires.
More details
#193735. Thanks to @steipete, @RomneyDa and @vincentkoc.
-
Fixes update rehearsal failing to load plugins that use supported external data files through package links. Repair preserves existing private copies when restoring missing files. The running updater must already contain the fix.
Sessions and chat
-
Fixes overflow summaries losing the sender and conversation context of queued messages, including the context that identifies room activity.
More details
#187833. Thanks to @goslingmanagment, @dmsrg399 and @obviyus.
-
Fixes empty completed assistant attempts appearing as blank rows in chat history when no recovery is pending.
-
Fixes archived PostgreSQL conversations failing to resume after they have been idle.
-
Fixes existing folders on paired devices being unavailable for supported models using the default runtime.
-
Fixes new chats unintentionally pinning the default runtime after provider setup or return from a hosted environment.
-
Fixes interrupted turns being discarded after a Gateway restart while their managed workspace is still being prepared. Recovery remains available if another restart interrupts that wait.
More details
#196423. Thanks to @steipete, @RomneyDa and @vincentkoc.
-
Fixes sessions discarding compatible server compaction checkpoints when switching models within the same provider, API, endpoint and session. Providers can still reject retained items.
-
Fixes OpenAI Responses sessions discarding compaction checkpoints when the same API key is selected from a different credential source, avoiding unnecessary full-history replay.
-
Fixes eligible progress commentary disappearing from refreshed chat history when commentary fallbacks are enabled. It also keeps commentary from intentionally silent messages hidden.
More details
commit c4f6b26. Thanks to @RomneyDa.
-
Fixes completed Claude CLI answers missing from conversation history when a later answer follows.
More details
commit 2cfd37f. Thanks to @RomneyDa and @vincentkoc.
-
Fixes streamed ACP replies missing their final text when the reply finishes.
More details
commit e0430c8. Thanks to @RomneyDa and @vincentkoc.
-
Fixes recovered sessions with widget access losing the ability to create or update pinned widgets when no browser is connected.
More details
commit 860d766. Thanks to @RomneyDa and @vincentkoc.
-
Fixes actions on an already published GitHub PR becoming unavailable after a later publication attempt fails, and restores guidance for recovering a changed head.
-
Fixes earlier conversation history failing to load around large messages or hidden rows, keeps history navigation working in both directions, and avoids offering another page once available older rows are exhausted.
More details
commit 2b08b6f, #188612. Thanks to @RomneyDa as a co-author of the released carrier and @shakkernerd for the pagination correction.
-
Fixes compaction losing the active human request across repeated compactions.
More details
commit 2b08b6f. Thanks to @RomneyDa.
-
Fixes unnamed offline people appearing as opaque IDs in the chat mention picker when a verified GitHub handle is available.
More details
-
Fixes some saved Codex conversations failing to continue after a local transport error.
More details
#187375. Thanks to @wtfichigo and @obviyus.
-
Fixes stray blockquote markers appearing in sidebar and session message previews.
-
Fixes Node-based attachment validation crashing when a local preview is cancelled.
-
Fixes PostgreSQL session context reads failing on conversations containing certain unusual characters.
More details
commit 4ae9adf. Thanks to @RomneyDa and @vincentkoc.
Memory and search
-
Fixes memory indexing failing on wrapped embedding batch-limit errors when the provider’s limit is available in the retained error.
More details
#187797. Thanks to @fredtanxo and @obviyus.
-
Fixes interrupted memory indexing losing track of session updates still waiting to be processed.
More details
#191633, commit 3a7c113. Thanks to @RomneyDa and @steipete.
-
Fixes large memory-index rebuilds running out of memory and repeatedly restarting, leaving vector search unavailable.
-
Fixes attachment and fallback model searches missing matching names written with equivalent Unicode characters.
More details
#187674. Thanks to @huwei-xydt and @obviyus.
-
Fixes Mattermost channel searches missing matching names written with equivalent Unicode characters.
More details
#187199. Thanks to @wangmiao0668000666 and @obviyus.
-
Fixes joined Matrix rooms missing from search when their names use a different Unicode spelling from the search.
More details
#187881. Thanks to @xialonglee.
-
Fixes search dropping text after a null character without changing the saved conversation.
More details
commit 4ae9adf. Thanks to @RomneyDa and @vincentkoc.
Models and providers
-
Fixes Claude Code models disappearing from the model list or being rejected when Anthropic models use the API route.
More details
#191148. Thanks to @etzelm, @Takhoffman and @RomneyDa.
-
Fixes provider catalogs being skipped during concurrent host and plugin loading.
More details
#196448. Thanks to @1052326311 and @RomneyDa.
-
Fixes Anthropic Advisor reviews and other plugin session forks missing the parent conversation's prompt cache when thinking comes from agent or model defaults.
-
Fixes unnecessarily low default output limits for models without a known limit by sizing the default to their context. Fresh LM Studio setups also recalculate the inferred limit after context selection, and Doctor reports eligible older limits without changing them.
More details
commit 85be164. Thanks to @RomneyDa, @vincentkoc and @VACInc.
-
Fixes OpenAI conversations failing with an item conflict after compaction, so affected conversations can continue with their retained context. Adds recovery guidance to remaining item-conflict errors.
More details
commit 9e9bb27. Thanks to @RomneyDa and @vincentkoc.
-
Fixes Anthropic setup-token requests failing when a newer Claude Code version is required, by retrying the qualifying request once.
More details
commit df9d6ae. Thanks to @RomneyDa and @vincentkoc.
-
Fixes a configured OpenAI account blocking fallback to the Claude CLI.
More details
commit 25eac4b. Thanks to @RomneyDa and @vincentkoc.
-
Fixes Google tool calls failing when parameters use Python keywords.
More details
commit 2b08b6f, adapting #191561. Thanks to @steipete and @HighervGround for their work on #191561, and @RomneyDa as a co-author of the released carrier.
-
Fixes media-understanding provider requests failing on Node 26 when their remaining timeout is fractional.
Channels and apps
-
Fixes background Talk cutting off or failing to resume listening on a locked iPhone when background Talk is enabled. Background Talk remains best effort.
More details
#187864. Thanks to @brianmcguire and @ultrakorne.
-
Fixes ending Android Talk cancelling agent work that was already accepted. On supporting Gateways, the answer can still finish in chat after the call ends.
More details
commit 0118146. Thanks to @RomneyDa.
-
Fixes ACP replies appearing only in the transcript instead of their bound Telegram topic when progress streaming is enabled.
-
Fixes disabled Telegram buttons appearing active and accepting taps.
More details
#191647. Thanks to @aspalagin and @obviyus.
-
Fixes Telegram forum approvals losing their topic when the current turn omits it but the saved destination includes it.
-
Fixes automatic Tlon group replies targeting unsupported nested replies. Replies in a thread go to its root, while replies to top-level messages stay in the channel.
-
Fixes Matrix direct-message thread replies being rejected when session identity casing differs.
More details
commit d241291. Thanks to @RomneyDa and @vincentkoc.
-
Fixes Matrix direct-message thread IDs losing letter case, which could make separate threads appear identical.
More details
commit 1ecdad2. Thanks to @RomneyDa and @vincentkoc.
-
Fixes Discord progress updates continuing when a reply ends without a final message. Errors while clearing channel progress, typing indicators or acknowledgment reactions are also contained.
-
Fixes iMessage account status hiding observed Automation send failures after a healthy read check. A timeout does not confirm whether the message was delivered, so check before retrying.
More details
commit 3288a4e. Thanks to @RomneyDa, @vincentkoc and @anmol0b.
-
Fixes native Talk consultations getting stuck before reaching the agent during a handoff.
More details
commit 50994ca. Thanks to @bdjben, @brokemac79, @RomneyDa and @vincentkoc.
Browser and desktop
-
Fixes Lightpanda navigation timing out and preventing browser tasks such as reading pages and submitting forms.
More details
#196400. Thanks to @RomneyDa and @vincentkoc.
-
Fixes Stop leaving Mac browser tabs stuck loading when clicked as a page begins to load.
More details
commit 058dada. Thanks to @RomneyDa and @vincentkoc.
-
Fixes Host Desktop checks blocking connections to password-protected TigerVNC desktops after repeated readiness checks. Managed Linux desktops also check the configured password before reporting readiness.
More details
#193487. Thanks to @SebTardif and @obviyus for the change and @lakemike for reporting.
-
Fixes Codex plugin installation failing during macOS onboarding launched by the shell installer when an accessible terminal can be resolved.
More details
#200247. Thanks to @miguelbranco80.
-
Fixes interrupted Control UI builds leaving the interface unavailable after a source Gateway restarts on POSIX systems.
-
Fixes browser control failing to recover on its next use after a connection drops.
More details
commit 2b08b6f. Thanks to @RomneyDa.
-
Fixes stale Git results in Review and session diffs after the system clock changes.
-
Fixes arrow keys changing the suggested worktree branch while you are composing text with an input method.
Plugins and installation
-
Fixes Plugin SDK type-check errors caused by missing PostgreSQL declarations in package installations.
-
Fixes packaged plugins resolving runtime exports to source files that are missing from the package.
More details
#188639. Thanks to @galiniliev and @RomneyDa.
-
Fixes plugin state work repeatedly hitting retiring storage during Gateway shutdown. Work already accepted can finish, while new plugin state operations are refused once shutdown begins.
-
Fixes old native plugin data accumulating in Gateway memory when repeated plugin inspections replace earlier results.
-
Fixes registered legacy OpenShell sandboxes being unavailable with OpenShell v0.1.3.
More details
#190101. Thanks to @vincentkoc and @RomneyDa.
-
Fixes onboarding recommending LINE for Linear because their names overlap.
More details
commit bb674d1. Thanks to @RomneyDa and @vincentkoc.
-
Fixes errors after onboarding in fresh macOS installations when plugin work is still finishing as a command exits.
-
Fixes session export callbacks reporting stale or repeated messages when an export restarts after the source changes. Message-only consumers are also notified to reset.
More details
commit d072273. Thanks to @RomneyDa and @vincentkoc.
Command line
-
Fixes shell completion omitting available help flags. Regenerate cached completion scripts with
openclaw completion --write-stateto receive the change.More details
#187799. Thanks to @bajiaoshan and @obviyus.
-
Fixes
openclaw path resolveandsetrejecting dotted insertion paths in JSONC and YAML when the parent exists. -
Fixes Codex app-server errors hiding the final failure behind earlier warnings in long diagnostic output.
More details
#187873. Thanks to @shakkernerd.
-
Fixes Lobster workflows running in the wrong folder when the requested working directory starts or ends with whitespace.
More details
#187739. Thanks to @wangmiao0668000666 and @obviyus.
-
Fixes Lobster selecting the wrong workflow when a filename begins with whitespace. A trailing space after the extension is still rejected.
More details
#187785. Thanks to @wangmiao0668000666 and @obviyus.
-
Fixes cancelling a Claude authentication check leaving a refresh lock that can stall the next turn. Cancellation waits for a started check to finish within its existing limits.
More details
commit 29f996b. Thanks to @RomneyDa, @vincentkoc and @obviyus.