[FEATURE]

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

I'm using Claude Code for Windows (the desktop GUI app, not the CLI), and on long conversations the UI becomes very heavy. Hundreds of rendered messages, tool calls, code diffs and sub-agent transcripts stay in the window, and the app starts to lag — typing delay, scroll stutter, slow re-render on every new token. On a lower-spec machine this makes long sessions effectively unusable.

The core problem: there is no way to clear the visual chat without also wiping the agent's memory of the conversation. /clear drops the context entirely. /compact is lossy and designed for token economy, not UI relief. Starting a new chat forces me to rename it, drag it back into the right group, and re-attach project context every single time — which I do not want to do multiple times a day for chats I've intentionally kept long-lived per project / per role.

This is especially painful when working with agents and sub-agents (Agent Teams), which is the main reason I'm filing this. Every sub-agent invocation returns a full transcript block — often hundreds of lines of tool calls, intermediate reasoning, file reads. A single productive orchestrator session can easily produce 5–15 sub-agent calls (a research team, a board meeting with multiple C-level agents, a multi-platform audit). All of those transcripts stay rendered in the parent chat on top of my own messages.

Result: agent teams generate a lot of display weight relative to decision weight. The desktop UI has no way to shed the display weight without also shedding the decisions, so the only options today are "live with the lag" or "throw away everything the agent knows."

Proposed Solution

Any one of the following would solve it. Listed in order of how clean they'd be:

1. Virtualized message list in the desktop app. Render only the last N messages in the DOM, lazy-load older ones on scroll. This fixes the root cause without any new user-facing command and would also help #4896 / #31643 / #26302.

1. /clear --display (or --visual, --ui). A flag on the existing command that clears the rendered message list in the window but keeps the full conversation context in memory, so the agent still remembers everything.

1. Keyboard shortcut (e.g. Ctrl+L) that does the same as #2 — one-key "wipe the screen, keep the brain."

1. Collapse-all sub-agent transcripts. A toggle in the UI that folds every Agent(...) result block into a single summary line ("✓ satur-seo — 12 tool calls, 4.2k tokens"), without dropping it from context. Click to expand again. This alone would dramatically lighten chats heavy with Agent Teams.

Ideal: ship #2 or #3 as a quick fix, then do #1 properly. #4 is independently valuable for any agent-team workflow.

Alternative Solutions

What I've tried / currently work around with:

• /clear — wipes the context. Unacceptable: the whole point of a long-running named chat is that the agent remembers the project.
• /compact — lossy summary, designed for token budget, not UI weight. Loses nuance from earlier in the thread that I still need.
• Starting a new chat — forces rename + drag into group + re-attach project context every time. I maintain named chats per project/role on purpose; constantly re-creating them defeats that.
• Closing and reopening the same chat — sometimes the desktop app re-renders only the visible portion and gets lighter temporarily, but it's inconsistent and the lag returns as soon as I scroll.
• Disabling skills / MCP servers — helps a little, but I actually use them; this is treating the symptom.

Other tools (e.g. terminal-based chats, VS Code's chat panels) solve this either by virtualizing the scrollback or by allowing a "clear display, keep history" operation. Terminal CLI users already have a related ask in #37656 — this issue is the desktop-GUI counterpart.

Priority

High - Significant impact on productivity

Feature Category

Other

Use Case Example

Concrete scenario from yesterday:

1. I run a multi-project AI-agent system. I have one long-lived chat in Claude Code for Windows for each role (architect, SEO, designer, analyst, CMO, ads, co-founder, board). Each chat carries weeks of context I deliberately keep alive.

1. In the architect chat, I run a board meeting — a 6-phase protocol that spawns ~10 sub-agent calls (CEO, CFO, CTO, CMO, CPO, COO, etc., each via the Agent tool with a different subagent_type). Each sub-agent returns a several-hundred-line transcript.

1. After two such meetings in one chat, the desktop app on Windows starts to lag noticeably: ~1s input delay, stutter when scrolling, sluggish re-render as new tokens stream in.

1. I want to keep working in this same chat — all the upstream context (company baseline, prior decisions, role definitions) is here and the agent should keep referencing it. I do NOT want to /clear and lose that.

1. With the proposed feature, I would either hit Ctrl+L or run /clear --display. The window would empty out, the agent would still remember everything we decided, and I'd continue the conversation on a responsive UI.

1. Time saved: ~5–10 min per session that I currently spend on the "rename new chat → drag to group → re-paste project context" dance, multiplied by every long-running chat I maintain. More importantly, it removes a daily friction that's pushing me toward shorter, less context-rich sessions — which is exactly the opposite of what a long-context agent product should encourage.

Additional Context

• Platform clarification (important): this is about Claude Code Desktop on Windows, not the CLI. There's an existing related issue #37656 for the terminal/CLI display — this request is the desktop-GUI counterpart, where the problem manifests differently (React-rendered message list in a window, not a terminal buffer that can be cleared with Ctrl+L).

• Related performance issues already filed against the desktop app that this feature would partially mitigate from the user side:
• #4896 — severe UI performance regression, React render loop in skills system
• #31643 — 18+ second input lag from React render loop
• #26302 — Windows 1.1.3189 UI lag and mouse stutter

• Why this matters more now than 6 months ago: Agent Teams / sub-agents are a newer feature, and they shift the ratio of rendered content per chat dramatically upward. The UI design assumptions from the pre-agent era no longer hold — a single user message can now produce 10x the rendered output it used to.

• Environment:
• OS: Windows 10 / 11
• App: Claude Code Desktop (Windows build)
• Usage: long-running named chats per project/role, heavy use of sub-agents (Agent Teams), 100+ skills / multiple MCP servers loaded