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/2026-10-15.md.
Built-in capability skills for the Assistant: the Assistant gains two capabilities of its own — Create Agent and Create Task. You can pick one for the message you are writing, or let the Assistant enable one automatically from what the message says. When one is selected, the Assistant first produces an editable confirmation card; nothing is created until you confirm.
Create Agent / Create Task confirm in the conversation: the Assistant drafts the Agent or Task from your message, and you can fill in or change fields on the confirmation card before confirming. Cancelling creates nothing.
Skill selection is per message: only one of the two capabilities can be selected at a time. The selected capability shows as an inline chip in the composer and is removed with Backspace; manually picked skills are badged on your message with the skill name, while auto-inferred ones show no badge. Capability cards appear only at the start of a conversation.
Summon an Agent on a task: a new Tasks-node operation in automation workflows — Call Agent or Crew into a new chat — starts a conversation with a chosen Agent or Crew on the triggering task and gets it working from your prompt, without changing the task's assignee or executor, so you can chain Agent steps that run in order. The CLI gains sharkly task summon, so Agents can summon others too (HF-13955).
Stop a running Agent with a reaction: adding the ❌ reaction to the message that started a run stops the Agent in Slack and Discord, however the run was started, and anyone in the chat can do it. Stopping it in the app syncs back to the IM, taking down the working indicator. The stop emoji can be changed in the integration settings (HF-13399).
Agents can work in Docs: an Agent you hand work to can now create and edit collections, write documents, upload and replace files, and use My Docs — acting as you, under your permissions; the previous 403 on writes is fixed (HF-14000).
Retry a failed message: failed request blocks in conversations gain a Retry button that removes the block and resends the request, so you don't have to resend the whole turn (HF-12939).
Clearer new-conversation layout: starting a new conversation centers the composer and the capability cards as one group; after the first message it returns to the message list with the composer at the bottom.
CLI skill create/import targets a Space or Agent: new --space (create the skill in a chosen Space instead of the oldest one) and --agent (create and mount it as that Agent's skill; mutually exclusive with --space) (HF-14016).
Clearer import/sync errors (phase 1): binding validation and import errors for trackers such as Jira are more actionable — recoverable and unrecoverable failures are handled separately instead of retrying forever, and diagnostic records now reflect real outcomes (HF-13908).
Chat input polish: after you hit Stop, your own messages in that turn regain their copy entry; the ↑/↓ input-history recall now shows a visible hint instead of being guesswork (HF-13841).
Calmer task pages: decorative animations (thinking spinners, shimmers) only run while visible — paused off-screen or in the background — cutting idle repaints and CPU (HF-14059).
Remote computer settings match local: the CLI section in a remote computer's settings drawer lists everything local shows, greying out unsupported items with a reason (HF-13960).
Task drawers never overwrite each other: opening another task from a drawer task always opens a new drawer instead of replacing the current one, and Esc closes only the top drawer (HF-14006).
UI detail: pinned sidebar conversations get a clearer mark that no longer clashes with the selected state (HF-13952).
Frontend memory no longer creeps: fixes chat message lists holding timeline data alive through stale closures — UI memory used to climb for hours (into multiple GB) until a refresh (HF-14019).
Windows computers stop going offline mysteriously: fixes the daemon hanging while waiting for system dialogs such as Git authorization, losing heartbeats and being marked offline; Git now fails fast without prompts (HF-13976).
macOS auto-update retries quietly: a failed update check retries in the background and completes on its own once the network recovers, instead of jumping to the "cannot auto-upgrade" fallback (HF-13982).
Onboarding fixes: skipping "connect a computer" no longer seeds a "Connect a machine to run Agents" task in the default Space (HF-14027); finishing Agent creation now lands in the Space task board instead of Agent settings, and saving a new Space no longer surprises you with a "copy from another Space" option (HF-14005).
Storage migration messages: migrating the storage location over same-named files now explains itself clearly instead of a cryptic error (HF-13282).
Storage dialog: cleanup by time and by task status can be issued while usage is still being calculated — the buttons no longer stay disabled and the numbers show "Computing…" instead of 0; both usage figures gain hover explanations of how their scope differs from total storage use (HF-14015).
