Tanium Cloud Atlas Updates & Release Notes

Features

• Endpoint Must Gather from chat: You can now ask Atlas to collect an Endpoint Must Gather from an endpoint. Atlas picks the gather steps that fit the problem you describe, starts the collection, and adds a status panel to the page that shows progress and offers the ZIP for download when it's done. The collection keeps going in the background, so you don't need to keep the page open while it runs. Requires the Direct Connect must-gather privileges.

Improvements

• Atlas no longer opens every reply by restating your request. It now opens with what it's about to do, and skips that line when the request is simple or has only one reasonable reading.
• Reordering panels within a row now keeps each panel's width, so the widths move with the panels.
• When the Atlas Agent adds several panels of the same width, whether they share a row no longer depends on the order they were added.
• Closing a panel no longer ends the Direct Connect session for other panels connected to the same endpoint.
• Empty KPI tiles in Risk Summary, Compliance Summary and the other Endpoint Details summaries now say why there's no number, instead of showing N/A or a count that was really an error.
• When a sensor returns the same error from every endpoint, Atlas is now more likely to treat it as a problem with its query rather than reporting the sensor as broken across your fleet.
• Atlas is less likely to report that a result was cut off when it actually returned every matching row.
• Skills Workspace has a clearer Skill actions menu, and enabling or disabling a skill now lives in that menu.
• Typing a prompt on a New page no longer causes a brief hitch every few words.
• Command palette endpoint rows no longer repeat the same details on a second line when you hover over them.

Resolved Issues

• Fixed an issue where dragging or arranging a panel reset the widths of every row on the page, not just the row that changed.
• Fixed an issue where a dropped Direct Connect session could report that you didn't have permission to start a session and remove the Connect button.
• Fixed an issue where conversations could fail with "something went wrong" on every message after the page scope changed to a persona with different privileges.
• Fixed an issue where Risk Summary and Compliance Summary tiles cut off numbers and labels on a narrow panel.
• Fixed an issue where prompts sent with Queue were skipped when pressing Up to recall earlier prompts.
• Fixed an issue where Core Identity on Endpoint Details showed a Deploy button that couldn't be used yet.
• Fixed an issue where the info icon on the Deleted folder had a button-style hover background.

Improvements

• Markdown panels that the Atlas Agent adds are now sized to fit their text, up to a limit, the first time someone who can edit the page views them.
• A panel that the Atlas Agent adds to a row now matches the height of the panels already in it, instead of leaving a ragged edge.
• Result tables now take their name from the panel title, so screen readers can tell several tables on one page apart.
• Panel titles now sit above the headings inside the panel in the page's heading structure, so navigating by heading reflects how the page is laid out.
• Users with the Atlas Audit Reader role can now open the Audit page in the Atlas console, with user names shown instead of IDs.
• When the Atlas Agent adds or changes panels while you are looking at another part of the page, those panels now highlight and the page scrolls to the topmost one, once per reply.
• Endpoint Details panels no longer repeat the same heading as a label, a header, and a table title.

Resolved Issues

• Fixed an issue where an endpoint with no applicable patches showed one applicable patch in Core Identity and in Patch and Software Summary.
• Fixed an issue where exporting a page to PDF could print some panels as empty frames, each on its own page.
• Fixed an issue where Export transcript failed with a generic error in a browser tab that had been open while Atlas was updated.
• Fixed an issue where clicking the Activity icon while already on Activity returned you to the previous screen.
• Fixed an issue where a background agent run that handed off a long task could be stopped partway through as if it had stalled.

Features

• Import and export skills as ZIP files: In Browse Skills, select one or more skills and choose Export selected, or use Export ZIP from a skill's menu. To import, choose Import, drop a ZIP onto the upload card or click it to pick a file, and choose where the skills should go. Atlas previews every skill in the archive before anything is created, showing any problems, flagging names that are already taken, and letting you rename a skill or leave it out. The selected skills are imported together, so an import either fully succeeds or creates nothing.

Improvements

• When you ask which users hold a role, such as Administrator, Atlas is now more likely to account for user groups that hold the role, not just people assigned it directly.
• The persona drop-downs for page scope, new pages and your default page scope, and the persona switcher in Settings, now list only personas that can use Atlas.
• Panels on a page created from a template are once again based on the access of the persona you signed in to Atlas with, rather than the page's scope, so the panels you see stay consistent if the page scope differs from your own or changes later.
• Background agent spend emails now include links to the AI Credit Usage or My Usage page.
• Deleting a published skill now warns that other users will lose access and that workflows relying on it may break, and requires acknowledgment before deleting.
• On Endpoint Details pages, Install and Deploy buttons now sit below the table, and Core Identity treats missing risk counts the same way as the other panels on the page.

Resolved Issues

• Fixed an issue where the Activity panel on Endpoint Details could show no Deploy or Patch history and no scheduled activity, and where Patch rows showed a number instead of the deployment name.
• Fixed an issue where a background agent could not pause itself after completing its work, and kept running on its schedule.
• Fixed an issue where peeking an endpoint from a results table did not offer View full device profile or Add to new page on the Core Identity card.
• Fixed an issue where new Endpoint Details pages included an Installed applications button in Core Identity.
• Fixed an issue where recovering a page from Deleted left the address bar on the previously open page, so refreshing or sharing the link opened the wrong page.
• Fixed an issue where clicking a page in the Pages sidebar did not highlight it until the page finished loading.

Features

• Export a conversation: You can now save an Atlas conversation as a single HTML file with Export transcript in the chat header. The file opens in any browser and includes the full transcript, including the request and response of each tool call.

Improvements

• Deleting several pages at once is much faster, and any background agents on those pages are now handled in a single dialog instead of one dialog per page.
• Clicking a page in a long Pages sidebar list responds faster.
• A page recovered from Deleted now appears in its proper place in the Pages sidebar instead of dropping in at the bottom and then jumping.
• Panel headers and footers have a new design. Long titles wrap to two lines and then truncate instead of getting cut off, header actions appear on hover or focus, and the main footer action comes first.
• The endpoint details drawer has a lighter shadow, on its left edge only.
• The Background Agents panel has a date picker next to the Owner toggle. Choose Current month, Last 7 days, Last 14 days or a custom range, and the Credits column and the list of agents follow it. The panel now opens on the current month instead of lifetime totals.
• Endpoint search in the command palette now shows each endpoint's IP address, platform, user and status, along with the computer name.
• A new Atlas Audit Reader role gives read-only access through the API to all users' Atlas audit data, including conversations, transcripts and usage, without full administrator access.
• Atlas is less likely to treat an endpoint's last check-in time as the age of its cached data, or to report old inventory as the endpoint's current state.
• When filtering on a numeric sensor, Atlas is more likely to leave out endpoints that returned errors or no results, which could otherwise be counted as matches.
• If a Python script Atlas runs hits its memory limit, Atlas now knows the script failed and can see how far it got.

Resolved Issues

• Fixed an issue where selecting several pages with Cmd, Ctrl or Shift could leave the page area stuck on a loading spinner.
• Fixed an issue where going Back to a page that had since been deleted did not show the "Couldn't find that page" message.
• Fixed an issue where panels whose content fits showed a scrollbar and scrolled by a few pixels, and where the scrollbar in Suggested Prompts covered the prompt text.
• Fixed an issue where Atlas could finish a response without answering, sometimes leaving only the suggested follow-up prompts.
• Fixed an issue where a response could hang after Atlas had finished, or fail when the connection to the AI model dropped partway through.
• Fixed an issue where chat messages could briefly draw on top of each other, or the chat could jump to its scroll position, when a page opened.
• Fixed an issue where a panel's refresh button could get stuck loading.
• Fixed an issue where the Web Searches setting appeared in environments where web search is not available.
• Fixed an issue where donut charts in Tanium dashboard panels were invisible when the panel was short.

Features

• Background agent spend notifications: Atlas now lets you know when background agents are using a large share of your AI credits, with no setup needed. Administrators are notified when background agents across the environment reach 25% and then 75% of the monthly credit allocation. Owners are notified when their background agents together reach 25% of their personal credit cap, or 15% of the environment's allocation if they don't have a cap. Each notification arrives in Activity Center and by email, and opening it shows which background agents are spending the most. You can change or turn off these thresholds in Configure threshold warnings, on the AI Credit Usage page for administrators or the My Usage page for owners.

Improvements

• Fleet queries that will never finish now give up sooner instead of holding your conversation for up to ten minutes.
• A fleet query that Tanium rejects now comes back right away, so Atlas can fix it and try again.
• The home screen's prompt bar and starter prompts have a refreshed design.
• The command palette footer now has a / Switch Mode hint, so slash shortcuts like /e for endpoint search are easier to find.
• When you duplicate a page, the original page's chat now notes that a copy was made from it. Anyone who opens the original page later sees the note too.
• When you delete a page, background agents that are already paused or disabled now default to Delete in the confirmation dialog, rather than Disable.
• Artifacts in chat can no longer be renamed.

Resolved Issues

• Fixed an issue where sensor searches could come back empty, even for an exact sensor name, causing Atlas to report that a sensor did not exist.
• Fixed an issue where asking for the same sensor twice with different parameters in one query silently dropped one of the answers.
• Fixed an issue where Shift-clicking pages in the sidebar to select a range opened the clicked page in a new browser tab.
• Fixed an issue where panels could stay on "Atlas is still starting up" until the page was refreshed, if Atlas was opened while the service was starting.
• Fixed an issue where resizing the browser window froze the chat for several seconds and made the sidebar stutter.
• Fixed an issue where some panels showed their heading twice, which made navigating by heading with a screen reader confusing.

Note: Atlas memories have been temporarily turned off for performance evaluation and improvements.

Improvements

• Background agents with no recent activity are no longer included in spend notifications, though a paused agent that spent during the current period still appears.
• Clicking an agent's name in the Background Agents panel now opens its details in place, instead of expanding a table row.
• The command palette now lets you know when your persona is limiting which templates and skills appear in search results.
• Template suggestions when starting a new conversation are more relevant.
• Audit exports can now include the custom instructions that were active for a user during a conversation.

Resolved Issues

• Fixed an issue where exporting an Atlas page to PDF could overlap panels or let table content spill outside its panel.
• Fixed an issue where an Atlas Tip above the chat box caused the chat input to jump.
• Fixed an issue where pressing Backspace in the command palette could unexpectedly exit a category you'd clicked into.
• Fixed the move and Arrange controls appearing on panels that can't be rearranged.
• Fixed an issue where a single narrow panel could disable dragging, resizing, and Arrange for an entire page.

Features

• Template favorites and ownership filters: Favorite templates from a card's menu and find them under the new Favorites filter in the template gallery. A Created by me / Shared with me toggle in the All Templates and Custom Templates views narrows the list to your own templates or to ones other people made, and works alongside search.
• Duplicate a page: Duplicate a page from the page header or the sidebar menu. The copy has the same panels and page scope, starts with a fresh, empty chat, and gets a new name. Background agents stay with the original page, and panels you aren't entitled to, or that are no longer supported, aren't copied.

Improvements

• Atlas can call Tanium APIs from the Python scripts it runs, so it can fetch data and work with it in one step. For now, a call that would need your approval is declined when it comes from a script.
• The Add to existing page and Add to new page actions on template gallery cards have moved from the hover buttons into the card menu.
• Atlas is more careful when recommending a change to an endpoint, not just when making one. It's more likely to explain the consequences in plain terms, including effects on other services and on security, and for a consequential repair, why it's justified, how to recover, and how to confirm it worked. It's also less likely to call an endpoint healthy when a check returned no data or couldn't finish.
• The Page Scope selectors are hidden for users who have no personas assigned.

Resolved Issues

• Fixed an issue where Atlas could say you had interrupted a response that actually stopped because of an error or a timeout.
• Fixed a rare issue where moving between pages could leave the previous page's conversation in the chat panel until a hard refresh.
• Fixed lag when swapping the chat to the other side, expanding it, or restoring a popped-out chat.
• Fixed an issue where an endpoint's name shifted down while its details loaded after opening them from a Computer Name cell.
• Fixed chat response buttons (Copy, Good answer, Bad answer) and Pages sidebar links being smaller than the 24 by 24 pixel minimum target size.

Features

• Save to Gallery: Any page you have built can now be saved to the template gallery, so the next person starts from your work instead of assembling it again. Save to Gallery is on the page menu, the panel kebab and the sidebar row; you give the template a name and a description and choose whether it is visible only to you or to everyone in Tanium. Your own templates appear in a Custom Templates section of the gallery with your name on them, and you can rename one, edit its description, change its visibility or delete it at any time. Changing a template to Only me leaves it in place for pages already created from it. Saving requires page-write access.
• Endpoint details from a results table: Computer names, computer IDs and IP addresses in a results table are clickable. Clicking one opens that endpoint's core identity in the drawer along the right edge of the page, without leaving what you were doing or searching for it again by name.
• Keyboard shortcuts panel: Atlas lists every keyboard shortcut it supports, grouped by category, so the shortcuts are discoverable without trial and error.
• Charts over time: Atlas can chart a series of values over time, as a line or an area.

Improvements

• The chat welcome screen has a Browse Templates button that opens the gallery directly, with the starter prompts collapsed behind a Suggested Prompts button beside it.
• Hovering a template card in the gallery shows a small hint, Add to new page or Add to page, instead of a large popover that repeated the card's own title and description and sat on top of the card's action buttons.
• In the gallery Edit view, the note explaining what a visibility setting means appears only when you change the setting, instead of describing the state you were already in.
• The endpoint detail view in the command palette has the same full-screen toggle the scratchpad, skills and gallery modes already had, so the identity card is no longer confined to a short scrolling area.
• Reordering pages in the sidebar with the keyboard is now discoverable, rather than something that worked only if you already knew it was there.
• Atlas is more careful about sensors that do unbounded work. A new sensor whose script scans a caller-supplied path, or walks the filesystem without a bound, is no longer rated low risk on the approval card just because it only reads.
• When asked for a VBScript sensor, Atlas writes it to Tanium's content authoring standards and tells you VBScript is deprecated. It still prefers PowerShell and shell by default.
• Atlas is less likely to lose a turn to an invalid call when it hands a question to its fleet query helper.
• The chat progress indicator says Tempering where it used to say Forging.

Resolved Issues

• Fixed an issue where a model provider failing part-way through a response ended in a red error box. When the provider returns a retryable server error and no output has reached the user, Atlas now retries once and carries on.
• Fixed an issue where a background agent whose trigger pointed at a route removed from the API catalog reported an internal message about a path that does not resolve. The error now states that the route is no longer in the catalog and that the trigger needs editing.
• Fixed an issue where duplicating the same page twice produced two pages with identical names.
• Fixed an issue where re-expanding a panel that had been collapsed for a long time quietly re-ran its fleet question. The panel now shows the expired state and the refresh control.

Features

• Guided skill authoring: Atlas can walk you through writing a skill. Ask for /skill-creator and it interviews you one question at a time — what the skill is for, when it should trigger, what it needs from you — without repeating anything you have already answered or it can infer. When a draft is ready to look at, it opens the Skill Workspace panel so you can read, adjust and save it yourself. A skill built this way can carry supporting files and Python scripts for the parts that should be deterministic rather than left to the model.
• Create Skill in the command palette: Browse Skills has a Create Skill control with two paths. Create New opens the editor on an empty skill, and Create with Atlas opens a chat with a skill-creation prompt already in the input.

Improvements

• A panel can now open a detail view in a drawer along the right edge of the page, rather than stacking it inside the panel or putting it in a dialog. Drilling again from inside the drawer goes a level deeper, with Back and a breadcrumb trail, and any level can be promoted onto the page as a panel of its own. Each panel opts into this separately, so it appears in some places rather than everywhere — a row on the Threat Response alerts panel offers View alert details.
• Atlas reads the skill you have open in the Skill Workspace before answering questions or proposing edits, so it works from your unsaved draft rather than reporting a brand-new skill as unavailable.
• Administrators can set warning thresholds for background agent spend specifically, alongside the existing organization thresholds on the AI Credit Usage page.
• The remaining controls that start an AI run stop offering themselves once you have reached your credit cap: the starter prompt chips on the welcome screen, Run on current page and Run on new page in the skills palette, and send-to-chat from a drawer or a deprecation banner. Controls that only fill the input stay available, since Send is where the cap applies, and text you have typed stays editable so a draft survives the cap being raised.
• The category and subcategory list in the template gallery starts collapsed, which makes the gallery easier to read.
• Scrollbars are consistent across Atlas. Some surfaces had shown the browser's own scrollbar, always visible and with a white track in dark mode, next to others using the subtle one that fades out when you stop scrolling.
• The page header reflows instead of clipping when the window is very narrow.
• When an activity cannot be loaded, the message now raises page scope as a likely reason and suggests trying another one, rather than only saying it cannot be found.
• The actions column in the Skills catalog has a header.

Resolved Issues

• Fixed an issue where a page created from a template was missing panels. Panels were dropped at creation if the person instantiating the template could not reach the data behind them, leaving an incomplete page with no indication anything was missing. Access is now determined when the page is read, so a panel whose data you cannot see shows the standard permissions placeholder and begins working once the access is in place, or once the page's scope changes to a persona that has it.
• Fixed an issue where Atlas pages created through the Atlas MCP Server appeared under Deleted. The pages were created as shared pages, so another user deleting one from their own view archived it for its owner as well. Pages created this way are now ordinary private pages, and anything already archived can be restored from the Deleted view.
• Fixed an issue where page and folder actions in the sidebar were offered without page-write access. Rename, archive, delete, restore, and creating, renaming, deleting or reordering a folder all stayed available to a read-only user and failed at the API. Move to folder is unchanged, since it only affects your own sidebar.
• Fixed an issue where the model selector applied across every conversation. Changing the model in one chat changed it for all other chats, including conversations pinned to a different model.
• Fixed an issue where text typed into the chat box on the landing screen was lost when switching browser tabs and returning. This occurred on tenants with no Atlas pages yet.
• Fixed an issue where an activity notification could print the same relative time twice, such as "Last activity 23 hours ago · Sent 23 hours ago".
• Fixed an issue where the enabled toggle in the Skills catalog disappeared and reappeared when clicked instead of animating.

Features

• Web searches: Atlas can search the web and read what it finds, so a question that needs an outside source, such as a CVE write-up, a vendor advisory, or a patch note, no longer stops at the edge of your environment. Answers built this way cite what they drew on. Web searches are off until an administrator turns them on: the Web Searches setting in Atlas Settings carries a disclosure you must accept, and a user needs the Atlas Web Searcher role before Atlas will search on their behalf. Which sites Atlas can reach is controlled through the Internet Access Policy panel, an administrator-maintained allow list or block list. When a search reaches that boundary, Atlas names the domain it could not reach and notes that the policy can be changed to permit it. Atlas reads the HTML a site returns, so a page that assembles its content in the browser can come back thin or empty.

Improvements

• Atlas checks its full tool surface before reporting that a capability does not exist, rather than concluding a capability is missing after searching only part of it.
• Atlas is more likely to find an existing endpoint group before building a target from scratch. A zero-hit result on a common cohort is treated as a reason to search more broadly rather than to create a new group.
• Atlas routes a deploy request by what you asked for. An exploratory or incomplete request stages a pre-filled deploy panel for you to review and submit; a complete directive, with package, target, and parameters stated, goes to the approval gate as an inline approval card. A person reviews before anything runs in either case.
• When Atlas delegates a fleet query, the result reports what was asked, how many rows returned, and what could not be determined, leaving severity, threat labels, and recommended actions to the main agent.
• Atlas explores unfamiliar modules directly instead of delegating to a separate explorer, reducing the time spent on discovery.
• Skill listings show each skill's description rather than repeating its name.
• Page rows in the Pages sidebar are links, so right-click, Cmd/Ctrl+click, and middle-click behave as they do elsewhere in the browser, including Open in new tab and Copy link. The kebab menu keeps all of its items.
• The Activity Center filter menu groups options under Activity type and Status and adds Agents and Approvals filters. A filter with nothing currently matching it stays selectable and shows an empty state.
• Reaching a notification by deep link, refresh, or browser Back and Forward keeps the Activity rail, the sidebar, and the originating card selected.
• The chat rail header reflows at narrow widths. The AI credits text moves to its own line, and the swap, pop-out, and collapse buttons collapse into a single kebab menu when space runs out.
• Atlas displays a session reconnect notice while session heartbeats are failing, and clears it on the next successful heartbeat. The popped-out chat window does not display the notice.
• Panels display the component type above the title only when the panel type opts in.
• Escape closes the command palette in one press whether or not it is full screen. The v key still toggles full screen.
• Clicking anywhere in a sidebar folder header row expands or collapses the folder.
• The page actions menu labels the action "Move to folder," matching the sidebar's terminology.
• Page-creation controls are not offered to a persona without page-write access.
• The out-of-compliance notice on the AI Credit Usage page is suppressed while allocation enforcement is enabled.
• The Background Agents table includes a Credits column showing each agent's lifetime total.
• The welcome heading, logo, and starter prompt chips are centered on the new page screen.
• The scope-change message in a conversation is localized.
• Panels in the grid follow the same order in the DOM as they do on screen, so keyboard navigation moves through them in the order shown.
• Pages and Endpoints entries in the command palette are reachable by keyboard.

Resolved Issues

• Fixed an issue where switching between pages briefly displayed text from the previous conversation, or text that the incoming page's pinned prompt should have covered.
• Fixed an issue where typing "/" on a new page listed the skills available to the logged-in session instead of the persona staged in the page scope selector.
• Fixed an issue where the Activity notification view resolved panel types as the logged-in user rather than the Activity scope, causing a panel the selected persona could use to report that the activity could not be displayed, and a panel only the logged-in user could use to load indefinitely.
• Fixed an issue where a page whose bound persona lost page-write access was never detected or re-scoped.
• Fixed an issue where clicking an Activity card from Settings dismissed the Activity Center sidebar and reset the nav rail to Pages.
• Fixed an issue where two notifications with identical definitions opened inconsistently, with one navigating to a page and the other remaining in the Activity notification view.
• Fixed an issue where the last card in a long Activity Center list was clipped and could not be scrolled fully into view.
• Fixed an issue where a long panel title overflowed the panel header instead of truncating.
• Fixed an issue where adding or removing a panel immediately after a page loaded could permanently mis-size panels on that page or render a chat panel as a standard panel.
• Fixed an issue where creating or cloning a skill with a name already in use returned a generic error instead of identifying the duplicate name and the scope it conflicts with.
• Fixed an issue where syntax highlighting was missing in the Skills workspace editor for .json, .md, .py, .yaml, and .yml files.
• Fixed an issue where a live query results table displayed its loading overlay on every poll even when the results were unchanged.
• Fixed an issue where the collection percentage on a multi-sensor live query understated progress by reporting the slowest sensor's value.
• Fixed an issue where scrolling the command palette allowed popovers to render outside its bounds.
• Fixed an issue where keyboard focus outlines on page section rows were clipped, a truncated title's tooltip covered the action menu, and the expansion chevron had insufficient padding.

Features

• Write your own skills: You can now write your own skills, not just run the ones Tanium ships. A skill is a set of Markdown instructions Atlas follows when you invoke it. Ask Atlas to write one and it saves a skill to your private collection; write it yourself in the new Skill Workspace panel, which gives you a file tree and an editor and holds as many nested files and folders as the skill needs; or work on it together, with Atlas proposing changes into the panel as an unsaved draft for you to review, adjust and save. A skill can include .csv, .json, .yaml, .toml, .py and .txt files alongside its SKILL.md.
• Sharing skills: Skills start private. Publish one to a content set and everyone with access can use it straight away. Edits to a published skill apply immediately, so you are asked to confirm before saving one. You can clone any skill you can see, including Tanium's own, into a private copy and change it however you like. Enable, disable, run, edit, clone, publish and delete are all available from the Skills catalog in the command palette, and typing "/" in chat picks a skill to run. Authoring follows the Atlas Page roles, which carry the new skill privileges by default.
• Background agents in the chat rail: The chat rail header has a Chat / Agents toggle. Switching to Agents replaces the transcript with the background agents on the page and a usage rollup, plus the runs for whichever agent and time window you pick. Toggling back restores the conversation with your draft intact.
• Environment AI credit enforcement: Administrators can enforce the environment's AI credit allocation as a hard ceiling, from a single switch on the AI Credit Usage settings page. It is off by default. While the environment is at its allocation, sending is blocked with a message identifying the limit as environment-wide rather than personal; your draft is preserved and Stop remains available.

Improvements

• Atlas now looks to the REST API first when exploring an unfamiliar module, rather than the GraphQL gateway, which covers far fewer operations.
• The Pages sidebar marks a page unread when the main chat agent has posted something you have not seen, the way it already did for background agent activity. It clears when you open the page.
• Typing "/" into an empty command palette opens the list of modes, navigable with the arrow keys. The slash shortcuts previously worked only if you already knew the alias.

Resolved Issues

• Fixed an issue where clicking into Atlas from the Tanium Console showed a black screen for several seconds with no loading indicator.
• Fixed an issue where the popped-out chat on a page with a scope read the page as the logged-in user rather than the page's scope, so it could show a different set of panels than the main window, or a panel Atlas could not identify.
• Fixed an issue where the panel types available on a scoped page were read as the logged-in user, so a panel type only the scope could use rendered as an error panel, and one only the login could use appeared available until the write was refused.
• Fixed an issue where setting a page's scope, or the New Page Persona preference, to your own user context while signed in as an alternate persona left the page unusable, failing every message and reload with a permission error until the scope was changed back.
• Fixed an issue where panels on a page whose panel types could not be loaded showed a startup notice and promised a reload that never happened, instead of reporting the failure.
• Fixed an issue where right-clicking a page row or folder header in the sidebar could open two identical action menus at once.
• Fixed an issue where an ampersand in a report panel appeared as its raw escape sequence.
• Fixed an issue where renaming a page in the sidebar lost the typed text if the page stopped matching the active filter mid-rename.
• Fixed an issue where a new folder disappeared when created while a folder move was still in progress, and where a second drag started during that window reported a successful drop before snapping back.
• Fixed an issue where the Steps tab of a background agent's run detail printed stray markup where a conversation summary appears in the run.
• Fixed an issue where the Landing Page preference seeded itself to Classic and wrote that value on any save of the General preferences tab, recording Classic as an explicit choice.
• Fixed an issue where an approved write to certain older API routes was sent without the request body the operator reviewed, so the write failed reporting a missing required field.