Pattern Automation
← All releases

v2026.1005.0 Breaking

Breaking Changes

  • Operator UI snippet injection removed — NEURO_CLOUD_UI_SNIPPET and NEURO_CLOUD_UI_SNIPPET_B64 no longer inject HTML into pages. Move integrations to a plugin UI contribution before upgrading.
  • Keyboard shortcuts always on — Instance and profile toggles for shortcuts are gone, along with /api/auth/preferences shortcut fields. Every signed-in user has shortcuts enabled.
  • @-mention no longer starts a run — Mentions in task comments stay as context for the assignee; only assignment and review requests start work.

Highlights

  • Grok Build on the native runner — Grok Build joins the Neuro OS Runner through ACP for local and Daytona execution, with subscription and API-key credential paths. Shared harnesses move to Codex 0.156.0, Claude Agent SDK 0.3.280, and OpenCode 1.18.32.
  • Agents keep their own files — An agent’s directory (AGENTS.md and supporting files) persists across tasks and sessions and syncs with the Instructions Editor.
  • Skills synced from GitHub — Skills gains a Sources tab: pick a repository through an existing GitHub connection, choose skill folders, and refresh on demand. Agents use installed snapshots — never fetch GitHub mid-run.
  • Browser Use Cloud, live in the task — A Browser Use Cloud connector gives agents governed browser automation; tasks show interactive browser tabs where people can follow along and take over.
  • Connections set up in one click — Connector setup collapses to one screen. Zapier, Arcade, Composio Connect, and Executor MCP aggregators are on by default. New catalog entries include Fireflies and experimental memory providers.
  • Slack goes both ways — Board replies reach the linked Slack thread; assigned agents get Slack tools during ordinary tasks and scheduled routines.
  • Agent Chat and a smarter composer — Chat gets its own navigation and durable task handoffs. The task composer picks model and effort per message, and tasks can be created from a prompt alone.

Improvements

  • Governed API tools on by default — Native runner search_api and call_api tools are enabled without an environment variable; large API responses stream to disk up to 1 GiB.
  • Operator controls — NEURO_HIDDEN_SETTINGS accepts wildcards with named exceptions; plugins can supply an organization switcher; native agents can hire teammates through a governed hire_agent action.

Fixes

  • Native runner and sessions — Tool outcomes survive shutdown; failed warm sessions retire cleanly; quota exhaustion is classified for proper backoff; provider auth failures offer an inline connection card.
  • Scheduler and recovery — A queued run whose claim is permanently rejected is cancelled instead of crash-looping the server; exhausted retries escalate to a visible blocked state.
  • Board UI resilience — The board recovers during server restarts with retries; mobile pickers stay in view above the keyboard; “Recovery needed” is distinguished from “Recovery in progress”.

Upgrade Guide

  • Migrations 0284–0293 run on upgrade (keyboard shortcuts, assets, agent instruction heads, chat handoffs, Browser Use Cloud, skill sources).
  • Remove reliance on UI snippet env vars and shortcut preference APIs.
  • Use assignment or review request instead of @-mention to pull another agent onto a task.
  • Optional: NEURO_RUNNER_API_COMPANY_CAPTURE_MAX_BYTES, NEURO_WORKSPACE_GIT_SNAPSHOT_TIMEOUT_MS, NEURO_WORKSPACE_MANIFEST_MIN_FREE_BYTES. Refresh provider CLIs to match harness pins.