[smarzban]

Why

#297 did not fix the board in T3 Code. T3 Code rebuilds a hidden terminal pane by replaying a capped log of raw output (5,000 lines or 8 MB). Once the log is trimmed, tsk's startup escape codes are gone: alternate screen, mouse, focus and paste reporting. The board replays as scattered cells and ignores clicks. Focus-in never fires, and ctrl+l redrew the board without turning the mouse back on.

What changed

• ctrl+l and focus-in send every terminal mode again (alternate screen, mouse, focus, paste, keyboard flags), then repaint in full (repaint_on_request).
• Every two minutes the board sends mouse, focus and paste reporting again, then paints a keyframe that rewrites every non-blank cell without clearing (paint_with_keyframe). A live terminal sees no change. A trimmed replay now always ends with a complete frame and its modes. The keyframe never touches the alternate screen or the keyboard flags: re-entering the screen can clear it, and keyboard flags stack.
• Docs: the keys.md row and note, the Help label, and the existing Unreleased CHANGELOG line, edited rather than duplicated.

Scope

• Covers: restoring the board after a host replays a trimmed log, automatically within two minutes or at once on ctrl+l.
• Leaves out: T3 Code's replay itself, which is the root cause. A bug report to T3 Code is the owner's call.

Done when

#	Result	Proved by	Proof	Step
1	ctrl+l and focus-in send every mode again, then repaint every cell	agent	repaint_request_resends_every_mode_and_repaints_in_full; smoke 1	5
2	Every 2 min, mouse, focus and paste are sent again and every cell is rewritten, with no clear, no alternate screen and no keyboard flags	agent	keyframe_restores_modes_and_a_dropped_screen_without_clearing, keyframe_is_due_every_interval_and_reschedules; smoke 1	5
3	A keyframe never leaves stale text	agent	keyframe_on_a_paint_that_blanks_cells_leaves_no_stale_text	4
4	Docs, Help and CHANGELOG match	agent	diff	4
5	In T3 Code, a reopened board comes back whole and clickable within 2 min, or at once on ctrl+l	owner	Your steps in the report

Verification

Commit: 82f15a2. The smoke ran on 9ac9c97, and the two later commits change only tests for Windows. There, crossterm refuses keyboard flags, and supports_keyboard_enhancement is always false, so the board never asks for them. Crossterm also turns mouse capture on through the console API rather than by writing a code, so the tests don't look for one.

1. Replay smoke, isolated TSK_STATE_DIR, TSK_CONFIG_DIR and HOME under /tmp. The board runs in a 100×30 pty. The script cuts the output log after the first paint, the way T3 Code trims it, moves the cursor, and replays only the tail into a fresh pyte screen. The "wait" run uses a build with the keyframe interval patched to 3 s (not committed):

   main 93d085e8, wait:     modes in tail: mouse=False focus=False altscreen=False | replayed: PARTIAL (15 words vs 116)
   main 93d085e8, ctrl+l:   modes in tail: mouse=False focus=False altscreen=False | replayed: FULL
   branch (3 s), wait:      modes in tail: mouse=True  focus=True  altscreen=False | replayed: FULL (matches live)
   branch, ctrl+l:          modes in tail: mouse=True  focus=True  altscreen=True  | replayed: FULL (matches live)
   

   On main, ctrl+l redraws the board but leaves the mouse off, which is the reported symptom.
2. cargo fmt --check && cargo clippy --all-targets -- -D warnings && cargo test: 2433 passed, 0 failed on 82f15a2. cargo clippy --all-targets --target x86_64-pc-windows-gnu -- -D warnings and cd site && npm test (63 passed) are clean too.
3. Each new test was watched failing with its fix reverted by hand: the old keyframe order, a full re-send on the keyframe, and the re-send removed from repaint_on_request.

Could break

Every two minutes the board writes one extra frame, about 3 to 4 KB. Turning a mode on when it is already on does nothing in any terminal. On Windows, crossterm treats focus reporting as a no-op.

For the owner to decide

Whether to report the trimmed replay to T3 Code. The real fix there is to restore terminal modes, or save the screen itself, when it replays.

Review

Pi review panel on e942a9b: 8 of 8 seats voted (sol and opus; holistic, tests, specification-conformance, subtle-correctness). It found 7 mediums. Kept F-1/F-7 (a keyframe on a paint that blanks cells left stale text), F-4/F-5 (no test that the keyframe sends only the light set) and F-2/F-6 (no test reached the ctrl+l/focus wiring). All were fixed in 9ac9c97, and the sol seat's verify passed with no regressions.
Set aside: F-3, advisory. It asked for a test that both paint paths schedule the keyframe, and both now go through one shared helper.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated
herdr-tsk	Ignored	Preview	Nov 14, 2026 8:12am UTC

[smarzban]

Replaced by #300: ctrl+l alone restores the modes and redraws; the 2-minute refresh and the focus redraw are dropped.