feat(agents): support model priority sorting

[xuyaqist]

1. 支持多选模型时调整模型顺序

2. 在对话时，特别渲染agent名称 3. 令markdown不渲染删除线

[Copilot]

🔵 Needs a closer look

Four unresolved findings remain, including three moderate issues and one accessibility nit.

Pull request overview

Adds ordered model selection, agent-name subtitles in new-chat conversations, and plain-text Markdown strikethrough handling.

Changes:

• Added drag-and-drop model priority controls and translations.
• Displayed agent names in conversation headers.
• Updated Markdown rendering and added regression tests.

File summaries

File	Reviewed changes	Findings
frontend/tests/newchatThreadAgentBinding.test.ts	Agent subtitle regression coverage	—
frontend/tests/newchatMarkdownRendering.test.ts	Markdown rendering regression coverage	—
frontend/tests/modelPriority.test.ts	Model priority tests	—
frontend/public/locales/zh/common.json	Chinese translations	—
frontend/public/locales/en/common.json	English translations	—
frontend/package.json	Drag-and-drop dependencies	—
frontend/lib/agent/modelPriority.ts	Reordering and selection helpers	—
frontend/app/[locale]/newchat/ui/markdown-text.tsx	Plain-text strikethrough rendering	moderate (1 vote): Apply the same del override to the streaming reasoning renderer.
frontend/app/[locale]/newchat/assistant-ui/thread.tsx	Agent-name subtitle	—
frontend/app/[locale]/agents/components/agent-prompt.tsx	Model priority drag-and-drop UI	moderate (1 vote): Guard no-op drops before persisting. moderate (1 vote): Include form-lock/read-only state in sortable controls and handlers. nit (1 vote): Give drag handles model-specific accessible names.
docs/plans/2026-09-11-newchat-agent-name-subtitle-plan.md	Implementation plan	—
docs/plans/2026-09-11-newchat-agent-name-subtitle-design.md	Agent subtitle design notes	—
docs/plans/2026-09-11-disable-newchat-strikethrough-design.md	Markdown design notes	—

Review details

Suppressed comments (4)

frontend/app/[locale]/agents/components/agent-prompt.tsx:221

• onDragEnd also fires when the item is dropped back onto itself. In that case reorderModelIds returns the same array, but this still calls updateModelSelection and enqueues a full agent update. Guard the no-op case before persisting so a click/no-op drag does not issue an unnecessary API request.

      updateModelSelection(modelIds);

frontend/app/[locale]/agents/components/agent-prompt.tsx:191

• The parent agent-config form is disabled while useAgentReadOnly() is true, and that state includes the NL2 form lock, but this new flag only checks role/speed. During a locked NL2 flow for an otherwise authorized user, the priority button and sortable items remain active, so handleModelPriorityChange can call updateAgent and persist a reorder through the form lock. Include the form-lock/read-only state in the sortable control and handler guard.

  const isModelSelectionDisabled = !canManage && !isSpeedMode;

frontend/app/[locale]/agents/components/agent-prompt.tsx:83

• All drag handles use the same accessible name, so a screen-reader or keyboard user cannot distinguish which model will be moved when the popover contains several items. Include the model's display name in the handle's accessible label (or otherwise expose it as part of the handle name).

        className="flex cursor-grab touch-none text-muted-foreground disabled:cursor-default"
        aria-label={reorderLabel}

frontend/app/[locale]/newchat/ui/markdown-text.tsx:696

• This override is only used by defaultComponents; the newchat reasoning path still renders streaming Markdown through its separate ReactMarkdown component with remarkGfm and no del override (ui/reasoning.tsx:200-267). While reasoning is streaming, ~~text~~ will therefore still render as a strikethrough and only change after streaming completes. Apply the same del mapping to that renderer as well.

  del: ({ children, ...props }) => <span {...props}>{children}</span>,

• Files reviewed: 10/10 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.