I have been digging into Claude Code's settings lately and found several genuinely interesting things that I (and I assume others) did not know. I will share them as a series. Let us start with the setting that fixes a bug that annoyed me endlessly: the terminal flicker.
This is part 1 of a series on hidden Claude Code settings. Part 2: the skills listing budget. Part 3: the session history cleanup. Part 4: remote control at startup. Part 5: disabling forced worktrees.
What is the problem?
Every time an answer streams in, the terminal shakes. Code, tables, everything jumps in front of your eyes. If you have lived with that flicker for a year, this is the fix.
How do you turn it on?
Activation is simple: inside an existing session, run /tui fullscreen and it reattaches without losing the conversation. If you prefer an environment variable:
CLAUDE_CODE_NO_FLICKER=1 claudeTo turn it off: /tui default.
What do you gain?
- Zero flicker. Output flows smoothly.
- Flat memory. A 6 hour session consumes the same RAM as the first minute, because off-screen messages do not stay in the render tree.
- A working mouse. Click to place the cursor, open tool output, scroll, and click links.
What is the one price to know about?
Your terminal's regular search (Cmd+F) no longer finds the conversation, because it lives in a separate buffer, not the normal scrollback. Search inside the conversation works differently now: press Ctrl+o to open transcript mode, then / to search, exactly like less.
And if mouse capture bothers you (vim, tmux, SSH), add CLAUDE_CODE_DISABLE_MOUSE=1 and you get the smooth rendering without the mouse being captured.
What is happening under the hood?
For those who like the technical side: Claude Code switches to an alternate screen buffer, exactly like vim or htop. Instead of redrawing the entire output on every token, it draws only what is visible. That is where the flat memory comes from, and also the change in search behavior.
The cost of trying it is one line. Did not like it? /tui default and you are back.
It is still a research preview, but most of the internal team at Anthropic already prefers it over the old renderer. Squeezing real productivity out of these tools is a big part of what I teach in hands-on AI workshops; small settings like this one add up fast when you live in the terminal all day.
FAQ
How do I turn on the Claude Code fullscreen renderer?
Inside an existing session run /tui fullscreen and it reattaches without losing the conversation. If you prefer an environment variable, launch with CLAUDE_CODE_NO_FLICKER=1. To go back, run /tui default. It is a research preview, but most of the internal team at Anthropic already prefers it.
What is the catch with the alternate screen buffer?
Your terminal's own search no longer sees the conversation, because it lives outside the normal scrollback. Search now works like less: press Ctrl+o for transcript mode, then / to search. If mouse capture bothers you in vim, tmux, or SSH, set CLAUDE_CODE_DISABLE_MOUSE=1 and keep the smooth rendering.