For AI agents: the complete documentation index is available at /llms.txt, the full documentation bundle is available at /llms-full.txt, and this page is available as Markdown at /changelog/2026-10-21.md.
Sprint owner: Manual-mode Sprints gained an optional Owner field, set when creating or editing a Sprint (the creator by default) and shown in the Sprint list and on the Sprint page; the Sprint list gained an Owned by me filter; and Sprints gained a Delete entry — deleting moves the Sprint's tasks back to unscheduled (the tasks themselves are not deleted) while the Sprint's statistics, including burndown data, are removed (HF-14248).
Session status for standalone chats: standalone chats with Agents and Crews now carry a conversation state, just like task conversations: the sidebar chat list and the chat page show the current state, and clicking it opens actions such as mark as completed, retry, or stop; the CLI can set the state on direct conversations too (HF-14215).
Development environments on the Computer page: a Computer's detail page now shows the development environments its local service detects — tools installed on it such as Node.js, Python, uv, and bun, plus the built-in environments the local service maintains itself — so you can confirm the Computer has the toolchain a task needs (HF-13335).
Desktop and IM notifications open the conversation directly: clicking a conversation notification on the desktop opens the conversation drawer in place (maximized by default) and returns you to where you were when you close it; Agent replies in standalone chats now raise a system notification (to the conversation's owner); a transferred conversation notifies its recipient through the Inbox, a desktop banner, and an IM notification; an @ mention inside a conversation shows a red Mentioned you in the sidebar chat list; and conversation notifications in Slack now link straight to the conversation instead of going through the Inbox (other notifications still open the Inbox and are marked read) (HF-14153).
Inbox remembers your last segment: the Inbox's All / Primary / Other segments remember your last pick — if it was Primary, the next visit opens straight on Primary (instead of loading All first and then switching); All and Other both reopen on All (HF-14282).
Inbox rows are plain text now: Inbox list rows no longer render Markdown; markers such as headings are stripped for a cleaner read (HF-14284).
Long conversations open and switch faster: the conversation page renders only what is visible on first paint, and charts, syntax highlighting, and long tables initialize as they scroll into view; opening and switching long conversations no longer stalls for a long time, and returning from a minimized window is faster too (HF-14221).
Board counts now match the cards: column counts and the cards inside a column now come from the same read model, ending issues such as "the header says 4 but the column shows no tasks", counts disagreeing with reality, and pull-to-refresh spinning with nothing new (HF-14277, HF-12407, HF-12935, HF-13063, HF-13436).
Tidier Agent comments: Agents writing task comments avoid H1 / H2 headings in favor of bold text and lists, so task details no longer get stretched by oversized type (HF-14259).
CLI system-prompt upgrades: a built-in prompt-patch table per Runtime version and model now injects the correct argument-passing rules at run time, cutting mis-passed-argument errors; the prompt states that ending a session also ends its child processes, so Agents stop assuming background processes are still running; and Agents now know Sharkly's own core features — tasks, Sprints, Projects, Views, Agents, and more (HF-14149, HF-14077, HF-13284).
Steadier task pickup on idle Macs: the local service's available-memory admission was redesigned with a corrected macOS view of free memory (compressed and purgeable memory now count), so small-memory or heavily used Macs keep claiming new tasks after a run instead of sitting queued forever (HF-14220).
Server performance work: tuned high-frequency slow SQL such as the inbox projection under pg18, and Agent-run broadcasts no longer invalidate the inbox cache — large Spaces respond faster (HF-14117, HF-14108).
Settings no longer show others' private Computers: the private section of the Settings → Computers list shows only your own private Computers (HF-14250).
Collapsed chat groups show unread and state counts: the headers of collapsed and pinned conversation groups in the sidebar now show the unread count and per-state counts (HF-14278).
Task drawer maximization syncs the address and selection: maximizing the task drawer, or opening a task conversation, now updates the browser address and the selected sidebar item, so refreshing and sharing links no longer land in the wrong place (HF-14128).
Model picker in Call Agent → Assign: the Agent picker in the task's Call Agent → Assign entry gained model selection, matching the other Agent pickers (HF-14230).
Hover to read long activity text: when an activity entry runs long, hovering shows the full text in a tooltip (HF-14241).
Mobile tab in the website download dialog: the download dialog on the website gained a Mobile tab (coming soon) (HF-13969).
Transfer impact failed to load after an account switch: fixed Windows takeovers by another account always failing with "transfer impact failed to load" when clicking View transfer impact, which blocked the transfer from being submitted (HF-14287).
Windows incremental update failures: fixed some Windows desktop packages (Main asar) failing to update incrementally, which left the shell on an old version until a manual reinstall; update permissions and diagnostics were strengthened (HF-14210).
Two ClickUp import problems: fixed the board's to-do column reporting "cannot load more tasks" after an import finished (a 400 on archived-status columns), and the status sticking to "syncing" after clicking replay (HF-14281, HF-14280).
Linear import stuck on comment conflicts: when an import hits a row conflict where a comment's external ID is already mapped to another local task, it now skips the row and records row-level diagnostics, so the import continues and resuming no longer hangs (HF-14279).
Assistant click opened the wrong chat: fixed clicking the Assistant in the sidebar not opening the new-chat page and instead jumping to an existing Assistant conversation (HF-14265).
Interface details: a batch of small fixes — trailing whitespace in the composer model pill's short label; the narrower, taller empty-state composer for new standalone chats now applies to ordinary Agents and Crews; re-tapping a duration in Prevent sleep no longer dismisses the computer-list popover; the narrow-window sign-in layout aligns; the Antigravity default-model area in Computer settings no longer breaks its layout; all avatars render as circles (including custom uploads); and the working Agent's loading animation was refreshed (HF-14292, HF-14226, HF-14228, HF-14247, HF-14237, HF-14291, HF-14263).
