Skip to main content

达摩院AgentOS v10.0.195

Release Date: 2026-10-29

v10.0.195 更新概览

DAMO Academy AgentOS v10.0.195​

New Features​

  • Unified archive page: Added a “Data & Privacy → Archive” entry in Settings for viewing and restoring archived teams and archived conversations, with capabilities such as undoing disbandment and conflict backfill.
  • Mute regular conversation messages: You can mute regular messages by recipient, agent, and conversation. Approval and Q&A reminders remain active.
  • Explicit scope selection on the Decisions page: Added a personal/team scope dropdown on the Decisions page. The selection is persisted and no longer changes when you enter other task boards.
  • Personal task board discovery and queue reasons: Agents can discover and write to personal task boards in regular conversations. When dispatch is queued, the UI explains which conversation it is waiting for and why.
  • Unlimited delegation between agents: Removed the level limit for delegation between agents. Subtasks can only create sibling subtasks, making delegation chains more flexible.
  • Backend thinking feedback configuration: Supports per-model management of whether thinking content is sent back. Turning it off does not affect tool continuation turns.

Fixes​

  • Fixed an issue where the agreement consent page was incorrectly shown when local user profile reading failed, and ensured that the same version of an already consented record is not overwritten repeatedly.
  • Fixed issues where the cold draft recovery entry was missing, attachments were uploaded repeatedly after recovery, and multi-select in the history timeline collapsed.
  • Fixed an issue where large tool outputs could interrupt tasks or cause incomplete conversation recovery. Streaming save and precise recovery are now supported.
  • Fixed an issue where old task boards could freeze for tens of seconds on each write. The backend now automatically backfills historical indexes.
  • Fixed vague browser tool error messages. It now returns actionable modification instructions.
  • Fixed an issue where skill discovery incorrectly registered nested skills within a skill directory as independent skills.
  • Fixed misdetection of commands such as python3 in skill dependency detection, and completed the built-in runtime commands.
  • Fixed an issue where approval policy files with a BOM or parse failure could clear policies or fail to write.
  • Fixed error handling when preference file paths contain special characters such as colons.
  • Fixed an issue where a new Smart conversation missing a frozen routing decision would fail only after side effects were persisted. It is now rejected before the request.
  • Fixed an issue where a run bound to a task could not read or modify newly created tasks on the same board.
  • Fixed an issue where the delegation overview incorrectly displayed agent questions as “Waiting for your approval”; it now correctly displays “Waiting for your answer”.
  • Fixed issues with inconsistent message time display, missed cross-board task refresh events, and race conditions when updating recovery status.
  • Fixed an issue where message style preview was inconsistent with actual chat.
  • Fixed an issue where a canceled exit confirmation could still authorize other exit operations.
  • Fixed an I/O exception for run indexes linked to a data root on Windows.
  • Fixed an issue where identity conflicts in notification reply reminders across different agents or conversations caused notifications to be lost.
  • Fixed an issue where application publishing manifest validation failure returned only an error code. It now includes specific field errors.
  • Fixed an issue where route normalization rewrote and auto-committed team member configuration merely because key ordering differed.
  • Fixed multiple underlying security vulnerabilities.

Improvements​

  • Optimized pagination performance for cross-board task queries, making pages such as delegation overview load faster.
  • Reduced duplicate model echo for new file Write operations, lowering context usage.
  • Optimized tool usage guidance to reduce model invocation errors in scenarios such as delegation continuation, task cancellation, and plan lookup.
  • Upgraded several security dependencies.