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/2026-10-28.md.
Automation runs as full conversations: automation run records now render as complete conversations reusing the chat panel — continue the conversation, ask follow-ups, or add instructions inside a run, just like any other conversation (HF-14558).
Turn a standalone chat into a task conversation: the "⋯" menu of a standalone chat adds "Turn this conversation into a task" (the Agent creates a task from the conversation and binds it; messages queue while the Agent is busy) and "Bind this conversation to an existing task" (search and pick a task); the CLI gains sharkly chat bind-task (HF-14413).
Conversation fork across all runtimes: forking a new session from any historical message (AI reply turns or your own messages) is now available across every runtime — inherit the full context up to that message, original conversation untouched (HF-14400).
Finer-grained notification settings: personal notification settings now break events down per channel: task comments split by assignee / watcher and by Agent vs. human commenter, replies separated from comments, new task-status changes (assignee / watcher) and Agent replies in conversations added; the delivery channels add the "Sharkly inbox" with per-event and mark-important settings; each channel can restore defaults in one click (HF-14197).
Better auto-titles for task conversations: the automatic title generation for task conversations was refined (comment-body cleanup, fixed assignment phrasing), and existing conversation titles were backfilled with the new logic (HF-14562).
Relative times switch to absolute dates: relative timestamps such as "3 days ago" switch to absolute dates after a threshold, localized into 10 languages (HF-14697).
"Docs collections" renamed "Shared docs": Docs pages rename "Docs collections" to "Shared docs", unifying the docs wording across the CLI and the interface (HF-14439).
Agent card details: refined how Space and visibility display on Agent cards (HF-14685).
Built-in OMP Harness (beta): a platform-level harness tune-up for models with weaker long-instruction following (GLM-5, Gemini 3.5 Flash, …): tool-call syntax self-healing and anchored edits, automatic spilling and collapsing of huge outputs, no yielding before verifying edits, memory prefetch and background consolidation. Enable it under Settings → Beta features; applies to new sessions only, with zero chat-UI changes (HF-14572).
Boards failed to load: fixed the board-datasets endpoint returning concentrated 409 / 502 / 499 responses that made board pages unloadable (HF-14658).
Server stability: fixed the Agent queue and capacity slow queries (hundreds of millions of rows per scan) that maxed out database CPU and caused widespread 502s for daemon registration and message reporting (HF-14441); fixed the Jira import parent-child self-join slow query (134M rows scanned) (HF-14651); fixed workflow proxy streaming endpoints logging false 500 panics on client disconnects (HF-14653).
Conversation fork fixes: fixed a forked session's first message being invisible in the transcript while it queued (HF-14436); fixed Agent message delivery breaking after a fork — the run bound to the forked session while its context kept the pre-fork session id (403 Agent run resource mismatch) (HF-14584).
Task rename not reflected in the sidebar: fixed the "Task conversation" row in the chat sidebar keeping the old task title until refresh (a regression after server-directed list refreshes); front-end polling of session endpoints was also reduced (HF-14636, HF-14621).
Misaligned bubbles while reading history: fixed user message bubble overlays covering Agent messages when scrolling up through history during streaming output (HF-14639).
Computer online but not claiming proxy requests: fixed the machine access token's 30-minute rotation invalidating the control heartbeat's lease, leaving the Computer "online" but never claiming private-network proxy requests (self-hosted Jira imports) until a client restart (HF-14650).
"Assigned to me" listed Agent-owned tasks: fixed "My Tasks → Assigned to me" wrongly listing tasks whose assignee is an Agent (HF-14476).
No auto-recovery after offline: fixed opening the app offline and never recovering connectivity state after the Computer regained network access (HF-14543).
Redmine projects wrongly showed "Computer unavailable": fixed public-network Redmine OAuth connections still reporting "Computer unavailable" for projects after saving (HF-14575).
Run-steps spinner with multiple thinking blocks: fixed the spinner in run steps only stopping when the whole turn finished — it now stops as each thinking block completes (HF-14589).
Duplicate avatars no longer overwritten: fixed scheduled sync overwriting a duplicate Agent's avatar back to its parent's (HF-14635).
New task's conversation missing from the dropdown: fixed conversations of tasks created via the members page's "Create or assign task" not appearing in the home conversation dropdown until F5 (HF-14661).
Model selector display glitches in Agent settings: fixed the remembered model selection rendering abnormally and doubling up inside the input (HF-14677).
iOS details: iOS bottom-bar "My Tasks" and "More" icons align with the ClickUp style (HF-14372); inbox task rows tightened to match desktop information density (HF-14406).
Asana fixes (beta): fixed import job records showing the wrong source project name (HF-14578); two-way sync now writes comments back — the comment box adds a "Also send to Asana" checkbox, and checked comments post as Asana comments without duplicating on import (HF-14579); fixed the import-completion page showing "Unknown" instead of "Idle" for link status (HF-14580).
