fix(web): stop auto-sending Ctrl+L from session selection paths

[TeigenZhang]

Symptom

Quickly refreshing the page (or an SSE reconnect firing in quick succession) would sometimes silently wipe the current conversation — the active Claude session got cleared with no warning or confirmation.

Root cause

Claude Code 2.x treats Ctrl+L (\x0c) as a two-step "clear conversation" command (first press shows a confirmation prompt, second press clears). The frontend fired \x0c from three places to force Ink to redraw stale CUP-positioned frames in the tailed buffer. When a fast page refresh / reconnect ran the same path twice within Claude's confirmation window, the first \x0c opened the clear prompt and the second \x0c confirmed it — destroying the conversation.

Fix

Removed the \x0c sends from:

• selectSession() — the main offender, runs on every tab switch & page reload
• restoreTerminalSize() — manual "restore size" button
• sendPendingCtrlL() — dead code path (pendingCtrlL was never populated)

Trade-off: occasional stale Ink frame immediately after refresh; the user's first keypress causes Ink to redraw and the artifact vanishes. Losing the conversation silently is far worse than a brief cosmetic glitch.

Test plan

• esbuild parses app.js / terminal-ui.js cleanly
• Verified: tab switches / page reloads no longer risk clearing the conversation; transient redraw artifact self-heals on first keypress

🤖 Generated with Claude Code

[Ark0N]

Thank you very much — this is an excellent catch and a really well-reasoned fix. 🙏

Silently wiping the conversation on a fast refresh / SSE reconnect is about the worst failure mode we could have, so trading it for a transient stale Ink frame that self-heals on the next keypress is clearly the right call.

I verified the whole change before merging:

• pendingCtrlL really is dead — it's only ever read inside sendPendingCtrlL, never assigned anywhere, so the if (!this.pendingCtrlL) guard always returned early. The function was already a no-op in practice and has zero callers, so the explicit stub changes nothing.
• The user-initiated Ctrl+L keymap in input-cjk.js is correctly left intact — only the automatic/programmatic sends are gone.
• No correctness regression in the anti-flicker pipeline: redraw correctness there comes from resize/SIGWINCH + the buffering layers; the selectSession \x0c was purely a cosmetic force-redraw.
• The backend CTRL_L_PATTERN filter only strips form-feed from PTY output, so it's unrelated to these input sends.

Love the "Do NOT re-introduce Ctrl+L here" guard comments too — that'll save a future regression. CI is green. Merging now, thanks again! 🎉