Agents can now pause mid-conversation to ask you structured questions before continuing, so discovery flows collect the right information upfront. The skills and methods experience has been completely redesigned with dedicated routes, display controls, and a template library. A new sync engine lays the groundwork for two-way Linear integration, and agents can now spawn child threads to work on sub-tasks in parallel.
Agents can now present sequential, structured questions during a conversation. Instead of asking for everything in a single prompt, agents break discovery into focused steps -- each question appears as a set of selectable options you can click through. Keyboard navigation is supported, and you can type a custom answer when none of the options fit. This powers the getting-started onboarding and will extend to brief and blueprint discovery flows.
Skills and methods now live on separate, deep-linkable routes with their own navigation. A new library view lets you browse template collections, filter by type, and preview templates in a full-size modal before activating them. Both list and grid layouts are available, with display preferences persisted across sessions. Empty states have been replaced with welcome modals that explain what each section does and offer a quick path to get started.
A new sync engine module provides the foundation for two-way integration with Linear. The initial scaffold includes a Linear adapter with field mapping, webhook ingestion with HMAC verification, conflict resolution, change detection, and rate limiting. Controllers for managing connections, entity mappings, sync logs, and conflict resolution are in place. Queue consumers handle inbound sync, outbound sync, document polling, reconciliation, and retention cleanup.
Agents can now spawn child threads to handle sub-tasks independently. When an agent encounters a multi-part request -- such as analyzing several uploaded documents during onboarding -- it creates dedicated threads for each piece of work and synthesizes the results. A new spawn indicator in the UI shows when child threads are in progress, and a post-spawn summary is generated once they complete.
Web research flows now use Perplexity Pro for deeper, citation-backed results. Research triggers have been expanded, and the flow uses thread spawning to run multiple research queries in parallel. Results are deduplicated using text extraction before being persisted.
Briefs now get an automatic title as soon as the agent writes the first content. The title updates in real time via the realtime hook and applies consistently across all accept paths, so the sidebar and navigation always reflect the latest title.
The blueprints and methods sidebars now show a live document tree that updates as agents create or modify documents. Recent threads also appear in real time with a highlight animation when new items arrive.