T59: round 4 of its review

tasks/reviews.md

@@ -294,3 +294,4 @@ Each task gets at most three codex rounds. Each round: the commit, the verdict,
 - Round 1, 82cfe91: FAIL. fish 4.9 moves a code point per Left, so a counted click put `X` inside `é`; fish's redraw dropped the wrap flag of a long line, so a second click on its next row moved nothing; a press let go in another cell still moved the cursor; a program that turned on mouse tracking while the button was down still got arrows. Fixed in 0c55d41: OSC 133 A's `click_events`, which fish sends on its own, makes the click go to the shell as an SGR press, as kitty and Ghostty send it; the release must land in the press's cell, and the program must track no mouse at the release.
 - Round 2, 0c55d41: FAIL. The four findings of round 1 replayed fixed, and zsh, bash, the prompt jump and the shell tests held; but a click on fish's `$` went to fish as a press, and fish moved its cursor to the start, where T59 asks for clicks past the prompt. Fixed in e74ea74: the press goes only past the prompt's last cell before the cursor, and never on a prompt cell.
 - Round 3, e74ea74 and 9fa68b2: a first run found that fish's empty left prompt beside a right prompt carrying its own marks panicked in `click::press`, then stopped without a verdict when Codex's content filter cut it off; 9fa68b2 fixed the panic. A second run, at 9fa68b2: FAIL. Every earlier finding replayed fixed and the panic gone, but a prompt that prints nothing, fish's empty left prompt with a plain or no right prompt, left no cells to find the input by, so clicks on the typed line did nothing. The owner chose a fix and a fourth round over leaving it documented. Fixed in 84364ce: OSC 133 B stamps the line typed after a prompt with a link of its own, numbered by the commands ended, which every prompt drawn again on that line shares, so the input starts at its first typed cell where no prompt cell of its line comes before it.
+- Round 4, 84364ce: FAIL. Round 3's finding replayed fixed, every earlier one held, and the rewrite passed 5,291 split reads, 15 pty replays, Cmd-press in prompt, input and output, bash 3.2, find, select all and copy; but the newest typed line was taken by its place, lowest on the screen, so after `printf '\e[H'` fish's prompt at the top over older lines took no click. Fixed in 28d7615: the newest line is the one of the highest number, and a press goes no lower than its last row or the cursor's, where fish moved its cursor for a click on older text below. Round 4 was the last the owner allowed.