After a server restart the byte recording a tab switch replays (`?tail=`) holds only the tmux client's attach repaint, so an inline Claude session came back with an empty local buffer while tmux still held its scrollback. The hollow-buffer fallback then sent PageUp/PageDown to Claude, which moved the composer cursor instead of the transcript, until a page reload. Pre-existing on an unmodified build; every restart of the server showed it.
The paging fallback now stands down when the last capture reported tmux scrollback (`paneHistoryLines` > 0), dropping the gesture debt and any page keys already queued, so the gesture falls through to the local path, whose scroll-to-top pull fetches `full=1` and the gesture scrolls on from there. Zero and unknown counts keep paging (fullscreen Claude, codex, opencode, remote and Docker panes). The pull no longer inherits a downgrade refusal's 60 s latch for such a buffer, drops a completion that outlived a tab switch back to the same session (select generation, not only the session id), records no count from an error envelope (a JSON 5xx would otherwise forget it and hand the next gesture back to paging), and re-positions the local-echo overlay after its rewrite. The refresh path records its count only once it still owns the terminal, for the same reason. The `[scroll]` line prints `paneHistoryLines` so an unloaded buffer reads apart from a hollow one, and the invariant docs record the exception.
Measured with phone emulation after a restart: before, the second visit landed at baseY 0 and every swipe sent PageUp; after, the first swipe fetched the history once with nothing sent to Claude and the gesture scrolled 21 rows, with an unsent draft intact on the overlay.
After a session's first load per page, a tab switch fetched a 1 MiB tail of
the server's byte stream. For a fullscreen claude pane (alternate screen,
tmux `history_size` 0) that tail is old repaints of one frame, and an idle
tab parsed it twice (the cached copy, then the fresh one).
Every terminal response now records `paneHistoryLines` per session, and
`selectSession` takes `full=1` whenever the last report was 0. For such a
pane that capture is the visible frame, a few KB, which is also what a page
load already shows. A response without the field forgets the entry, so
unknown never counts as empty; a pane that starts keeping history goes back
to the tail on the next switch. Shells and the refresh path are unchanged.
Measured with real prod payloads through the real client in chromium:
median switch 1.26-1.66 s -> 0.10-0.19 s, parse ~1.1-1.5 s -> 2-3 ms.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
JD