ELYTerminal/tasks/acceptance.md

Acceptance

PRD.md line by line, as of T23. Status: Pass when a test, a run or a review shows it; Partial when part of it was shown; Open when nothing has shown it yet. Reviews are in tasks/reviews.md; tests are named by module (cargo test -- MODULE).

What no check reached: no real KWin drew the blur; of Wayland desktops, only Hyprland, on the owner's Omarchy, has drawn it. On 2026-10-15 macOS gave shot the window's ground alone, for old and new builds alike, while the terminal it ran under lacked Screen Recording (CGPreflightScreenCaptureAccess false); that day's macOS checks read the app's log.

1. Product

PRD line	Task	Evidence	Status
One window: sessions left, terminal middle, Git or Files right	T11	T11 review; a real X11 window shows the terminal and the right column	Pass
Every setting has a control in the settings window	T16–T18, T30, T33	17 of the 24 fields have a row; show_sessions, show_panel and panel are the main window's toggles and switch, sessions_width and panel_width its columns' borders, and window_width and window_height its size, which §6 keeps in the file	Pass
Silent: no sound, bell, notification but a session's, update check, telemetry, tip or onboarding	T10, T12, T40	a bell is logged and marks a hidden row; the code holds no network call, and its one notification is a session's, through gpui's, which sets no sound on macOS and a silent toast on Windows; on Linux the notification daemon decides	Pass
Clean: no subtitle, tagline, logging option, status bar or badge without state	T11–T18	reviews of the layout and settings tasks	Pass
Built-in themes, each with a preview, Catppuccin among them; a theme colors terminal and chrome	T03, T17	themes::tests; T17 review	Pass
The background blurs on macOS, Windows, Wayland and X11	T19–T21	see §11	Partial

2. Foundation

PRD line	Task	Evidence	Status
One binary crate elyterminal, edition 2024, Rust 1.95	T01	Cargo.toml	Pass
Ely by pinned rev; gpui and gpui_platform from gpui-pre at Ely's version	T01	Cargo.toml: Ely a2cb531, gpui-pre =0.3.8	Pass
alacritty_terminal 0.26	T04	=0.26.0	Pass
Logs through log to stderr, RUST_LOG sets the level, no log setting	T01	src/main.rs; no such field in Settings	Pass
CI on Gitea Actions: macOS builds, tests and lints; Ubuntu and Windows build and test	T01, T23	.github/workflows/ci.yml; run 926 on 65bb4d3, see the CI notes	Pass
The repo, private until v0.1.1 and public since; no credential in any file or commit	T01, T26	T22's review scanned the tree and its 398 reachable objects: 0 hits; T23's scans for the Gitea token and the runners' SSH details: 0 hits; before the owner made it public, a scan of all 91 commits' contents and messages found no token, SSH password, gateway, port, SSH user or private key	Pass

3. Settings model and storage

PRD line	Task	Evidence	Status
One settings.json in dirs::config_dir()/ELYTerminal	T02	settings::store tests	Pass
Every field has a default; a missing file or field means its default	T02	settings::tests, settings::store tests	Pass
A bad file or value stops the start, naming file and field; never overwritten	T02	a_bad_file_is_an_error_naming_it, a_bad_value_is_an_error_naming_the_field	Pass
Temp file in the same folder, then rename; one writer	T02	T02 rounds 1–2	Pass

4. Themes

PRD line	Task	Evidence	Status
The 14 built-in themes	T03	assets/themes, byte for byte with upstream 21f281d; Ely's two	Pass
Background, foreground, cursor, cursor text, selection, 16 ANSI from upstream	T03	themes::tests	Pass
A theme maps onto Ely's Palette through Theme::set_palette	T03, T19	the_chrome_wears_the_theme	Pass
One theme, or a second for dark mode	T17	T17 review	Pass
On Omarchy 4, an Omarchy theme wears the current theme and follows each switch; the default there; another pick stops following	T27	omarchys_ghostty_conf_reads_without_its_glass, omarchys_file_is_worn_again_once_it_reads; on X11, as a user, with Omarchy v4.0.4's own renderer staging and swapping the folder as omarchy-theme-set does: with no theme saved the window wore Omarchy's, and through all 22 of its themes the terminal's ground matched each background within 1.5 s; Omarchy's glass lines changed nothing; a file without palette 7 was logged and the last theme stayed; a file swapped in unreadable was worn once made readable, its stamp unchanged; a folder unreadable for 2 s kept the theme; each failure was logged once while it lasted; a picked Catppuccin Mocha stayed through a switch, and with it worn a bad Omarchy file was logged and the start went on; with Omarchy's folder unreadable the start stopped, a theme saved or not, and with settings wearing it and no Omarchy too, each naming the file and the error; the first release build, on the owner's Omarchy 4.0.4 under Hyprland 0.56.2, wore the live theme, sepia-night, which is none of Omarchy's 22, and the release build from aae7fba, under Xvfb on Debian 13 with that theme's folder, painted its background	Pass

5. Terminal engine

PRD line	Task	Evidence	Status
PTY through alacritty's event loop; OSC 7 and OSC 133 scanned on the reader	T04, T34	term::pty, term::tap tests; round 3 failed at the cap, its fix stands on a red-green test	Pass
Shell integration for zsh, bash, fish, PowerShell; a shell without it reports nothing and the panels say so	T05	term::shell tests on macOS and Linux	Pass
Prompt marks, OSC 133, from the shell integration and any program; each session keeps its last command's status	T34	term::marks tests: each letter and option as Ghostty reads them; term::tap and term::pty tests: marks split across reads reach the session in order; term::shell tests: zsh, bash, fish and PowerShell send C, D;1, A and B around a failing command, and the user's prompt still shows the failure, a zsh theme's that sets PS1 in its own hook too, whose first prompt is marked before any command, and $pipestatus stays; bash's lone subshell ends with D;7, $_ stays, a command runs under extdebug after a failure, a key binding marks nothing, the user's DEBUG trap runs, bash 3.2 sends no C, and bash says at start why it marks nothing while promptvars is off; PowerShell reads 1 for a cmdlet's failure under -ErrorAction Ignore after a success, a native code twice, a native failure's code under $PSNativeCommandUseErrorActionPreference, on a line that mixes a program and a cmdlet the cmdlet's 1, and through a call by variable the program's code; on macOS fish 4.9 marks itself and bash 3.2 marks no C, in the Linux container fish 4.0 gets B and bash 5.2 marks C through PS0, CI's Ubuntu runs fish 3.7, which the script marks whole; by hand in a pty, PowerShell under strict mode, and a cmdlet's failure after a native one's reads 1	Pass
Jump to the prompt before or after the view's top, as Ghostty jumps	T35	term::prompts tests: a mark becomes its prompt's numbered link, read as the session reads marks, whole when split across reads at every byte, and other sequences, 133;AB among them, pass as they came; a prompt's cells carry its link through SGR 0 and a resize, its input none; a program's own link and the prompt's keep each other, a program's link inside a prompt carries the prompt's number, a prompt inside a link left open too, a target of 264 bytes in either order; controls after an OSC's closing ESC keep their line; a long escape of controls keeps the pty reading; a prompt over rows starts at its first, blank rows within it too; up and down land as Ghostty's scrollPrompt; term::links: a prompt's link is no target; term::pty: through the event loop the prompt's cell alone carries it; in the app on macOS, Cmd-Up once and twice put PROMPT-3 and PROMPT-2 at the top, and Cmd-Shift-Down and Cmd-Down came back to the screen's bottom; the review's fixtures of a prompt that starts with a link and of a prompt with a blank row jumped to PROMPT-3-FIRST, and its fixture of 133;AB marks found no prompt; the second review's fixtures of a prompt all in a link and of a linked first row jumped to PROMPT-3-FIRST. The keys off macOS are bound, not run	Pass
Program status, OSC 7501 revision 0.4: reports checked as the spec checks them, records per terminal, a new prompt and a full reset, the support query	T36	term::status tests: the spec's examples read as it means them; a malformed pair is skipped, the last of a key wins, unknown keys pass; a sequence over any limit, with bad base64, UTF-8 or a control character, or without a known state drops whole, a clear too, and an earlier pair of a key counts though the last wins; the sequence's length counts the controls its body skips; kind and progress count only with their states; records replace whole, clear with those under them, and past 256 the one updated longest ago goes; a new prompt keeps done and error alone. term::tap: a full reset found between bodies, and through DEL and high bytes in its escape. term::pty: a report, a query and a full reset reach the session in order through the event loop. In the app on macOS, a program that sent the query and then DA1 read the answer with every state and kind, then DA1's	Pass
Progress, OSC 9;4, as the root record until an OSC 7501 report; notifications, OSC 9, 777 and 99, read	T37	term::notify tests: OSC 9 reads as Ghostty reads it, ConEmu's commands no notifications, 9;4's progress clamped; OSC 777's notify; OSC 99 gathers chunks by id, base64 split before encoding or after and a character across chunks among them, keeps the newlines base64 carries, reads options, close, query and alive, and drops what breaks its rules with what it gathered; its answers name the asker's id. term::tap: an OSC 99 body keeps its metadata and a full payload. term::status: 9;4 maps to the root record until a report, and again after a full reset. term::pty: OSC 9, 777, chunked 99 and 9;4 reach the session in order, and a full encoded chunk with the lines base64 carries reaches it as one notification. In the app on macOS, a program read OSC 99's query and alive answers before DA1's, and four notifications reached the workspace's log	Pass
Inside herdr, the folder of herdr's focused pane, read from its socket, on macOS, Linux and Windows	T29, T32	term::herdr tests: the socket found as herdr 0.9 finds it, only the terminal client followed and a remote one refused, the snapshot read, a socket round trip, a process's path, arguments and variables; the_foreground_is_the_program_s_own_group; on X11 with herdr 0.9.5 run in the session: its start showed the focused pane's folder, a cd in the pane moved the panel, the focus moved to a pane in /etc and back moved it each time, and the client's end gave the panel back to bash's report, then exec herdr from that bash, with HERDR_SOCKET_PATH relative to its folder, was followed through the socket there; on macOS under shot, herdr 0.9.5 as the program and again by exec from sh with a relative socket, its pane's cd moved the panel; on Windows, term::herdr tests on the CI runner: the folder under APPDATA and USERPROFILE, the herdr.exe nearest under the shell, this test's name, arguments and variables read through the live process list and sysinfo, a round trip over a named pipe; no real herdr has run on Windows yet	Partial
Inside herdr, herdr's agent states are the session's records	T38	term::herdr tests: each agent in a state herdr knows is its pane's record, its app, title and message from herdr's fields, unknown and new states passed by, and a snapshot without agents lists none. term::status: herdr's agents replace their own records alone, a program's herdr/build kept through an empty snapshot, a full one and herdr's leaving, since a program's id ends at a colon; the same agents change nothing, and the next snapshot puts them back after a full reset	Pass
A row shows its session's status, hover each record; a hidden session that starts waiting marks its row	T39	term::status tests: blocked, then error, working, done and idle show, the record reported last in its state speaks; an app comes from the nearest ancestor that names one; typing sees done and error, herdr's aside; each change of the state a row shows is noted, a wait after work in one output among them. sessions tests: hover lists each record's state, app, title and message, ten at most, then a failed command's status; a program's text shows on one line without direction controls or invisible formatting, joiners kept, cut at 120 characters. In the app on macOS, shot showed each mark: the ring at 40%, the still loader, a shield over three records whose hover listed each, the root's app inherited, a check, a cross, a hollow circle, an alert after 133;D;1, a hidden session's question mark with its title strong, and a done record gone after a typed key; after review, a hidden session that waited, worked and waited again in one output set its title strong, an open hover followed a record replaced under it and work from 20% to 80%, ten long records scrolled to the last within the window, and a selected row at 0% in Ely Dark showed its loader; after round 2, herdr's records replaced in the same state showed at once in an open hover, and a hidden session that waited and worked again in one output set its title strong	Pass
A row's mark moves: work ripples from the centre, progress fills the outer ring clockwise from the top, waiting stays lit and dips, a settled state draws its shape and holds; the window draws at the marks' steps only while one moves	T42	sessions::orb tests: the ripple peaks at the centre, then every edge, then every corner, once every 15 steps; at 50% the first four ring dots are lit, the next pulses and the rest stay dim, and at 100% the centre pulses; waiting dips on the 1st, 2nd, 5th and 6th of 20 steps, every dot alike; a done check draws from its left end and holds after 5 steps, idle is a ring; work and waiting never stop. sessions test: the mark takes the most pressing look, a failed command as an error. In the app on macOS, at 70% glass in Ely Dark, the recorder's stills showed each look in one column: the ripple, 40% filled, waiting lit, a check, two crosses and a ring; two stills 0.4 s apart showed the ripple at two steps	Pass
System notifications on macOS, Windows and Linux: a session that starts waiting, finishes or fails, and OSC 9, 777 and 99, while the window is inactive or the session hidden, one a session every 3 s; a press shows the window and session	T40	term::reports tests: a session posts once in 3 s, its own under one tag; a program's notifications go by their id and live until pressed, closed or expired, an expiry taking only its own post and none a later post holds; a session keeps its last 64; the text shows as text, and as markup's text on Linux; a session posts unless it is the one shown in an active window; waiting, done and error post, work and rest do not; a tag names the session that posted it, by itself, an id or a count. term::titles: OSC 0 and 2 set the title as vte splits it, 15 parts past the number at most, and as alacritty joins it; CSI 22 t and 23 t save it and bring it back; a full reset empties both. term::tap: the scanner finds CSI 22 t and 23 t as vte reads them; a title's body holds 65536 bytes, another's 4096. term::pty: titles, notifications and full resets reach the session in the output's order, one write and a write a sequence alike, a title of 5000 bytes and one of 17 parts among them. term::notify: the query, on Linux without close and w, alive, untracked and activation answers. In the app on macOS, outside a bundle, where gpui posts nothing and the log names each post: an OSC 9 posted and an OSC 777 within 3 s was dropped; an OSC 99 with c=1 and w=800 posted, its program heard p=close;untracked, alive listed it, and once it expired alive listed none; a closed one left alive empty; the query's answer came back; a command's end without its C posted nothing, and a failed one after its C posted; a notice after a title set, or after a full reset, in the same write named the session as split writes do, as did one after a full reset and 2 KB to 32 KB more of output, and a title of 17 parts or of 4100 bytes named the notice as it named the row, one write or two; a notice, alive and DA1 in one write came back as alive naming it, then DA1's answer; a notice set to expire whose session ended expired after it; a hidden session that waited and worked again in one output posted its wait; and a notice sent while the window was active and the session shown posted nothing, the next, once the window lost focus, posted. No system showed one yet: on macOS gpui posts only from the app bundle, which asks the owner's leave first, and Windows and Linux did not run it; a press needs the system's notification center, so the owner posts and presses one on each	Partial
Colors, styles, five underlines, wide characters, combining marks	T06	term::frame, term::ink, term::paint tests	Pass
Box drawing, blocks and powerline shapes as geometry	T07	term::boxes tests	Pass
JetBrains Mono, any installed font, Nerd Font icons	T06, T17	tests/fonts.rs	Pass
Cursor shapes, blink by setting and DECSCUSR, hollow unfocused	T06	T06 review	Pass
xterm keys, cursor mode, Option as Alt, kitty protocol, keypad; a shifted symbol as its own key with Shift on macOS and Windows; win32-input-mode on Windows	T08, T44	term::keys tests, kitty_reports_the_key_shift_turned_into_a_symbol red at CSI 123;13u before the fix; on macOS herdr 0.9.5, its [keys] previous_workspace = "cmd+shift+ctrl+[" and next_workspace = "cmd+shift+ctrl+]", run by shot in a fresh HOME: before, ⌘⇧⌃] sent CSI 125;13u and herdr stayed on its workspace; after, CSI 93:125;14u moved it to the next and ⌘⇧⌃[, CSI 91:123;14u, back; term::win32 and term::tap tests on Windows	Pass
IME: marked text at the cursor, commits to the program	T08	T08 review	Pass
Selection by character, word, line; Shift extends; mouse reporting; wheel	T09	term::mouse, term::tap tests	Pass
Copy, paste, bracketed paste, copy on select, OSC 52	T09	T09 review	Pass
OSC 8 and plain links on Cmd-press, Ctrl-press off macOS; inside herdr, herdr opens them whole	T09, T45	term::links tests; on macOS with real herdr 0.9.5 and a stand-in open on its PATH, Cmd-press on a URL herdr wraps, on its wrapped tail, on an OSC 8 link and on a URL over three rows opened the whole URL and the OSC 8 target, where before the app opened .../ELYTermin and nothing; Cmd-hover underlined the long URL across its rows; outside herdr the app opened the URL itself; an OSC 8 link whose label holds file:///etc/hosts and whose target is a web page opened the page through herdr, which keeps the target from the terminal while it takes the mouse; Shift-Cmd-press on a plain file:///etc/hosts inside herdr the app opened itself, caught by an NSWorkspace stand-in	Pass
Focus reporting, title, scrollback limit, find	T10	term::find tests; round 3 failed at the cap, its fix stands on a red-green test	Pass
A bell makes no sound; the row marks it	T10, T12	T12 review	Pass

6. Main window and layout

PRD line	Task	Evidence	Status
ColumnShell with three columns; blur when the setting is on	T11, T19	shot names a BlurredView under the window	Pass
Columns toggle and keep their state	T11, T23	a click on each title bar button hid its column and saved it, and a restart opened the window with both hidden (147 columns against 87); Cmd-B with the app active; Ctrl-Shift-B on a real X11 window	Pass
The window opens at the size it was left at; maximized or full screen, the size before stays; a kept size past the screen shrinks to it	T30	a_value_out_of_range_names_its_field; on X11 under xfwm4: the first start opened at 1200 by 760 and wrote nothing; a resize to 900 by 600 was kept with one write; maximized by xfwm4's Alt+F10 the window spanned the 1400 by 900 screen while the file kept 900 by 600, and restored it was 900 by 600 again; maximized, minimized and restored, the file kept 900 by 600; resized to 950 by 650 and minimized at once, the file kept 900 by 600 while minimized and 950 by 650 once restored; the next start opened at 950 by 650 and wrote nothing; WM_NORMAL_HINTS names a 100 by 100 minimum, and a resize to 900 by 50 held at 900 by 100, kept so; a kept 3000 by 2000 opened at 1400 by 900, logged; on macOS, through gpui's own window: a kept 900 by 600 opened at 900 by 600, where a centered edge on a half point had grown it to 601, and a move wrote nothing; a resize to 923 by 617, covered by an opaque window within 150 ms and uncovered, was kept once shown, and the next start opened at 923 by 617; zoomed, minimized, restored, full screen and back, the file kept 923 by 617; a 100 by 100 window was kept; a kept 5000 by 4000 opened at the visible 1920 by 1113, logged; on Windows the owner dragged v0.1.2's window to a new size, closed it, and it opened again at that size; a mouse drag on macOS not yet shown	Pass
The sessions and right columns resize by a drag on their border toward the terminal, 160 to 960 pixels, kept in settings	T33	Ely 5c2822c and 005faef: shell::tests::border: a border toward the filling column drags its width, before it by its right and after it by its left; a drag stops at each limit; a drag in a header moves no window in Windows', macOS' and Linux's style (the test window panics on a move); a drag sets off from the width drawn, which holds Windows' caption buttons inside the window; a short shell squeezes both columns, and a drag moves each border the pointer's way from the width kept, a drag with no sideways move changing none (Ely 0b4848b): on macOS 900 and 900 kept in a 1200 window drew 572 each, and a drag of 20 from the left border saved 920 and moved it right, where review round 1 saw it jump to 435; on X11 900 and 900 in a 1380 window, a drag of 40 saved 940 and moved both borders right, and a drag with no sideways move kept them; a drag out and back to where it set off gives the width it had (Ely da3fdca): on macOS each border dragged 50 out and back kept 240 and 240 and came back to its pixel, where review round 2 saw 290 kept, and on X11 both settled at 240 with nothing written; a resizing column with no border toward a filling one fails; a_value_out_of_range_names_its_field (panel_width 1000); on macOS under shot, the sessions border dragged in the header and the right border in the body moved 50 points each, saved 290 and 290 once, and the next start opened at them; on X11 under xfwm4: a drag in the body set 300, one on the right border in the header set 300 and left the window at 0,0, drags stopped at 960 and 160, the next start opened at 160, and both borders paint the theme's border color, #37384A, as before; the line had painted half a pixel on macOS and none on X11 until Ely 005faef; no real drag on Windows yet	Partial
The title shows the session's title, nothing else	T11, T40	log: "the title is", per session shown; since T40 the title comes from the tap in the output's order, and the shell tests read each shell's title from it	Pass
The pointer over the terminal is the arrow	T26	the terminal sets no pointer, so gpui's arrow shows; on X11 with the pointer over the grid, gpui asked for no I-beam ("text"), which it had asked for, and failed to load under Xvfb, before	Pass
The title bar is transparent	T11, T19	T19 review	Pass
Every title bar control takes its press, Ely's window buttons on Windows and Linux too	T26	Ely cdc5ce8: window_controls_occlude_what_lies_under (each style; before, the box under Windows' caption buttons took the press), a_wobbling_press_on_a_slot_moves_no_window (before, the window moved), a_wobbling_press_on_a_column_title_moves_the_window; on the owner's PC, window messages asked the app for Windows' hit test along the title bar: v0.1.0 answered caption from end to end, T26's build answers client over the toggles and +, caption over the titles, min, max and close over Ely's buttons, and a press on close closed the app; the owner, given v0.1.1's build and these controls to click, reported it works on Windows	Pass
Ely peeks over the window's top edge from behind it, or stands inside at its bottom left where no room lies above it; her face follows the sessions; a press pets her and wakes her from a doze; nothing but her shows	T46, T48, T49, T51, T52, T53	pet::look tests: the most pressing mark sets her mood; she moves in every mood, weeks into one too; she blinks on her own, twice at times, her eyes opening as an act begins; calm glances, then dozes with zs; a finish cheers her with a hop; work runs her eyes along the lines and taps her fingers; a wait widens her eyes and hops her twice a round; an error shakes her and runs a drop down; a press winks with a rising heart; she rises from behind the edge; a change of act eases her from her last pose, her brows easing and the marks she leaves fading; pet tests: a mood that changes inside a wink, an old finish cleared mid-glance and a second press mid-hop each ease her from where she was, where round 1 of T49's review found her drop 12 points; pet::rig tests: every layer of the icon's paths parses and stays in its box, and every mark has its layer; on macOS shot's snaps, 40 in 10 s while calm, caught her breathing, blinking and glancing, and two a second apart in each mood caught her reading along, rocking her mark, fretting with her drop, cheering, and easing back to calm; pet::x11 test under Xvfb: her window stands on the pane, just above the window manager's frame, and unmaps and maps with the main window; on macOS shot found her the main window's child, 112 by 124 points on its top edge 140 in, a capture of the screen showed her over the edge, and OSC 7501 reports gave her surprise, worry, eyes down and a cheer; a press on her window left the main window key and every key typed before and after it reached the program; switched off and on during an error she came back worried; on a 100-point window she stood at its left edge, and moved, resized and minimized she followed it; on Windows the pet::windows test, red before T48, finds the framed popup Windows makes, 1898 by 1481, bare in her own box, 168 by 185 at 150%, on the pane's top edge, and no size until her window takes the place posted to it; pet::windows tests, red before T48's second round: with the main window at 1, 1.5 or 2 and hers at 1, 1.25 or 1.5, her height is her own and her bottom the pane's top, where it fell 62 pixels into the terminal, and at a new DPI gpui takes the rect Windows suggests, then her box comes back on her corner, or Windows' rect stands where that box would lie on a display of another DPI; two displays side by side at 144 and 96 DPI, simulated, settle her after one change, where a corner kept always, red before T48's third round, changed her DPI back and forth; and two places piled up before either ran drain on the last corner in her DPI's size, where a keep whose DPI had passed, red before T48's fourth change, set it back and forth, and where a place sized when it was posted left her 112 by 123 at 144 DPI, red before its fifth; on a Windows 11 VM at 150%, with the main window 49 pixels from the screen's top, 0.1.4 opened her 2326 by 1654 with a caption, drawn at its bottom over the terminal, and a press inside the terminal went to her; T48's build opened her 168 by 185 without one, on the edge, the press went to the terminal, one beside her box to the program behind, and one on her to her; with the window mid-screen she showed whole on its edge; T51: pet tests: she stands on the top edge while her box fits above the pane, 123.4 points, else the bottom; on macOS 0.1.4 opened her, over a window under the menu bar, at 1081 to 1205, inside its top, as the owner saw, and T51 at its bottom edge, 92, 140 in, and over a centered window still on its top edge, 1028; the pet::windows tests on the owner's PC put her on the top edge 400 pixels down the work area and on the bottom edge at its top, and there the release build stood her on the top edge of a centered window, on the bottom edge of one moved to the work area's top and of a maximized one, 210 pixels in at 150%, and on the top edge again once it came back; the pet::x11 test under Xvfb puts her on the bottom edge once _NET_WORKAREA leaves no room; on WSLg's X11 she stood over a window with room and inside one as tall as the screen; on WSLg's Wayland, Weston, her popup stood on the pane's top edge and, once the window was maximized, opened again on its bottom edge, at its bottom left; under Xorg with two monitors she moved to the bottom edge of a window at the lower one's top, 112x123+240+1277, and of one held mostly by a staggered monitor at its top, 112x123+1240+877; under headless Sway her anchor went (140, 0) on the top edge, then (140, 999), (140, 799) and (140, 999) on the bottom edge as the window went full screen and the output 800 and 1000 high, and full screen at the same size moved her too; T52: the pet::look test wakes her from a doze 400 s into calm with a press, which winks her, then she stays awake, calm, her eyes open, until 5 minutes after the press, when she dozes again, her eyes falling shut, and a finish long past wakes alike; the pet test presses her 400 s into calm, as her mouse handler does, and finds her awake 2 s later and asleep again 301 s after; T53: a user's 0.1.4 on macOS drew a hairline along the top of her box, over the clear room above her head, where only the titled frame of gpui's popup panel draws; on macOS 27.2 the log reads pet: her panel drops its titled frame, style mask 0x8081 to 0x80, her window's own picture holds no pixel on any edge, and the screen around her box, the app active and inactive, matched the titled build's but for her moving figure, since this macOS draws no edge for it; a real click on her face winked her, pet: Ely is pressed, the front app stayed in front, and abc typed after it reached cat; over a window under the menu bar she still stood inside on its bottom edge; on the owner's PC, WSLg's X11 and Weston her box showed no frame, border or shadow	Pass

7. Sessions panel

PRD line	Task	Evidence	Status
Rows: title, then the folder, home as ~	T12	sessions::tests	Pass
New (+, Cmd-T), close (×, Cmd-W), select (press, Cmd-1…9, Ctrl-Tab)	T12	2026-10-15: + opened session 2, a press showed session 1 again	Pass
A program's end closes its row; the last row closes the window	T12	T12 review	Pass
A bell in a hidden session marks its row	T12	T12 review	Pass

8. Right panel: Files

PRD line	Task	Evidence	Status
A switch at the top, centered below the title bar: Git, Files	T13, T26	T13 review; a capture on X11 shows it centered below the title bar, the right toggle and Ely's window buttons in the bar above	Pass
The session's folder as a tree, folders first, Git colors	T13	files::tests	Pass
Double press opens a folder, or a file in the default app	T13	T13 review, with a logging open stub	Pass
Reloads on a new folder and at each prompt	T13	log: "the tree shows", per folder reported	Pass

9. Right panel: Git

PRD line	Task	Evidence	Status
Through the git command line	T14	git::tests on temp repos	Pass
Branch, staged and unstaged, commit field and Commit, recent commits	T15	git::panel tests; T15 review	Pass
Stage and unstage; a press opens the diff in DiffViewer	T15	T15 review	Pass
Discard asks first	T15	T15 review, in temp repos	Pass
Outside a repository it says so; a failed command shows its error	T15	"Not a Git repository." on a real X11 window	Pass
Reloads at each prompt and on a folder change	T15	log: "git panel: shows", per folder	Pass

10. Settings window

PRD line	Task	Evidence	Status
Cmd-, opens it; categories left, rows right, each scrolling on its own; changes apply at once and save	T16, T26	T16 review; on X11, eight wheel notches over the Terminal page carried it down to its Edit and View shortcuts while the category column and both headers stayed	Pass
On Windows and Linux, a gear in the title bar opens it too	T26	on X11 under Xvfb with xfwm4, xdotool's click on the gear, just inside the right toggle, opened the 900×640 Settings window ("settings: the window opens"), after a click on + opened a session; sent through App::dispatch_action, the same click had opened nothing; the owner, given v0.1.1's build and the gear to click, reported it works on Windows	Pass
Terminal: Shell, Start in	T16, T26	settings::tests; the General page is gone, and a capture on X11 shows its two rows leading Terminal	Pass
Appearance: theme grids, dark theme, font, size, line height, ligatures, cursor and preview, opacity, blur, padding	T17	T17 review	Pass
On Omarchy, the Omarchy theme first in the grid	T27	on X11 with Omarchy's theme folder, the Omarchy card leads the grid, chosen, over a preview in its colors	Pass
Terminal: scrollback, Option as Alt (macOS), copy on select	T18	T18 review	Pass
Terminal: Ely, a switch	T46	on macOS the switch closed her window and saved "ely": false, a start with it showed no Ely, and the switch brought her back	Pass
Terminal: every shortcut, with a search	T26	read from the keymap, the bindings that apply in the main window's terminal; on X11 the list showed this platform's keys (Ctrl+Shift+T, Alt+1 and on), and typing "paste" left Edit's Paste, Ctrl+Shift+V	Pass

11. Background blur

PRD line	Task	Evidence	Status
macOS: Blurred (NSVisualEffectView)	T19, T23	shot reads BlurredView, blur 30 off the view's backdrop layer; the owner saw the desktop blur	Pass
Windows: Blurred (acrylic)	T19	the same gpui call; on 2026-10-15 the owner saw the window blur on Windows	Pass
Wayland: ext_background_effect_v1 beside gpui's org_kde_kwin_blur	T20, T28	blur::wayland tests on a wayland-server compositor; on the owner's Omarchy 4.0.4, Hyprland 0.56.2 took ext_background_effect_v1 from a T27 release build: the log named it and the pane's region, and the wallpaper behind the window blurred while Hyprland's blur was on and showed sharp while it was off; Plasma and GNOME not run	Partial
X11: _KDE_NET_WM_BLUR_BEHIND_REGION, removed when blur turns off	T21	blur::x11 test on Xvfb; the property on a real xfwm4 window; KWin not run	Partial
Neither protocol: transparency without blur; the log names the path	T20, T21	the log lines in src/blur	Pass

12. Menus and shortcuts

PRD line	Task	Evidence	Status
The macOS menu bar	T22	shot lists every item with its key	Pass
The same shortcuts off macOS, Ctrl-Shift where they collide	T22	xdotool on a real X11 window: select all, size, toggle, find, quit	Pass
On Windows Ctrl-V pastes too, as in Windows Terminal	T26	the binding is Windows' alone; on a real X11 window, with a command on the clipboard through xclip, Ctrl-Shift-V pasted its 33 characters and the shell ran it, and Ctrl-V pasted nothing, as Linux keeps it for programs; the owner, given v0.1.1's build and Ctrl-V to press, reported it works on Windows	Pass

13. Acceptance

PRD line	Evidence	Status
Each task ticked with a passing review	each task is ticked only after its review, as tasks/reviews.md records; T04 and T10 reached the three-round cap without a passing round, their last fixes standing on red-green tests	Partial
scripts/check.sh passes	macOS: 135 tests, clippy, fmt, 500 lines; Linux through scripts/linux.sh: 138 tests, clippy, on aarch64 in Colima and on x86_64 in WSL's Docker	Pass
CI green on macOS, Ubuntu and Windows	run 955 on 43ea5b1, v0.1.1's commit: all three jobs pass; see the CI notes below	Pass

14. Icon and the macOS app

PRD line	Task	Evidence	Status
The icon: Ely, a girl with long pink hair, holding a terminal	T24	assets/icons, drawn at 16 to 1024 px and legible at 32; the owner saw it	Pass
macOS: the rounded square; Windows and Linux: the figure alone, flat	T24	elyterminal-macos.svg on Apple's grid; elyterminal.svg, which the Linux tarball carries and the Windows .ico renders	Pass
scripts/mac-app.sh builds the app, signed ad hoc, or for a release with Developer ID and notarized, and its disk image	T24, T26, T31, T41, T43, T47, T50, T54	its run: codesign --verify --strict passes on the app and on its copy in the image; the .icns holds 16 to 1024 px; the image verifies and mounts with the app beside Applications; LaunchServices opens the app as com.zacharyzhang.elyterminal; for v0.1.1 the Release workflow (run 959) ran it on macos-signing: notarization Accepted, the ticket stapled, and on this Mac the downloaded image and its app pass codesign --verify --strict under the Developer ID and Gatekeeper's accepted, source=Notarized Developer ID; the owner opened it; for v0.1.2 the Release workflow (run 1001) ran it again: the downloaded image's stapled ticket validates and Gatekeeper accepts it as Notarized Developer ID, and its app, version 0.1.2, passes codesign --verify --strict --deep under the Developer ID with the hardened runtime and is accepted the same way; for v0.1.3 the Release workflow (run 1157) ran it again, and its image and app, version 0.1.3, check the same way, as does the image rebuilt for T43 (run 1181); for v0.1.4 the Release workflow (run 1216) ran it again, and its image and app, version 0.1.4, for macOS 11.0 or later, check the same way, as does the image rebuilt for T50 (run 1256); without the key the script stops before building; v0.1.5's disk image, from the Release workflow (run 1312) at 0119776: its stapled ticket validates, Gatekeeper accepts the image and the app inside as notarized Developer ID, the app passes codesign --verify --strict --deep and reads 0.1.5, macOS 11.0, arm64, beside a link to Applications	Pass

15. Release

PRD line	Task	Evidence	Status
A Gitea release, vVERSION, with the macOS disk image, a Windows zip and a Linux tarball	T25, T26, T27, T31, T41, T43, T47, T50, T54	v0.1.5 at 0119776: the zip and the tarball from their scripts on the owner's PC and its WSL, the disk image from the Release workflow (run 1312), signed and notarized; published once CI run 1311 on that commit passed on macOS, Ubuntu and Windows; each file downloaded back anonymously matches the SHA-256 in the release notes; each binary holds T51's Ely moves to the pane's and the failed glue's line, the disk image T53's her panel drops its titled frame and the tarball the monitors' line; the downloaded app, run with a HOME of its own, logged her panel going borderless, 0x8081 to 0x80, and Ely on the pane's top edge; the tag names 0119776. v0.1.4 rebuilt for T50, at a54f4ff: the zip and the tarball from their scripts on the owner's PC and its WSL, the disk image from the Release workflow (run 1256), signed and notarized, each in place of the first build of its name, once CI run 1253 passed on macOS, Ubuntu and Windows on a9be8ab, which a54f4ff follows with TASKS.md and tasks/reviews.md alone; the tag moved from 2d9180e to a54f4ff, and the release's target with it; the notes give the new SHA-256, add T48's fix and T49's motion, and say the builds were replaced; each file downloaded back anonymously matches its SHA-256, each binary holds the rig's log line and none the old faces' string, and the tag's source archive holds src/pet/rig.rs and no src/pet/faces.rs. The first v0.1.4 builds, at 2d9180e: the zip from its script on the owner's Windows VM, the tarball from its script on the owner's PC under Omarchy, the disk image from the Release workflow (run 1216), signed and notarized; published once CI run 1215 on that commit passed on macOS, Ubuntu and Windows; each file downloaded back anonymously matches the SHA-256 in the release notes, each binary holds Ely's log line, and the tag named 2d9180e. v0.1.3 rebuilt for T43, at eb96109: the zip and the tarball from their scripts on the owner's PC and its WSL, the disk image from the Release workflow (run 1181), signed and notarized, each in place of the first build of its name, once CI run 1180 on that commit passed on macOS, Ubuntu and Windows; the tag moved from 7569f00 to eb96109, and the release's target with it; the notes give the new SHA-256 and say the builds were replaced; each file downloaded back anonymously matches its SHA-256, each binary holds the moving marks' log line and none the old ring's element id, and the tag's source archive holds src/sessions/orb.rs. The first v0.1.3 builds, at 7569f00: the zip and the tarball from their scripts on the owner's PC and its WSL, the disk image from the Release workflow (run 1157), signed and notarized; published once CI run 1156 on that commit passed on macOS, Ubuntu and Windows; each file downloaded back anonymously matches the SHA-256 in the release notes, and the tag named 7569f00. v0.1.2 at 4364802: the zip and the tarball from their scripts on the owner's PC and its WSL, the disk image from the Release workflow (run 1001), signed and notarized; published once CI run 1000 on that commit passed on macOS, Ubuntu and Windows; each file downloaded back anonymously matches the SHA-256 in the release notes, and the tag names 4364802. v0.1.1 at 43ea5b1: the zip from its script on the owner's PC, the disk image from the Release workflow, signed and notarized; the Linux tarball, rebuilt for T27 from aae7fba by its script on the owner's PC under Omarchy, took the first one's place, and the notes say so; each downloaded back anonymously, once the repository was public, matches the SHA-256 in the release notes. v0.1.0 at 0075667, its three files built from that commit by their scripts and checked again: the disk image verifies and its app passes codesign --verify --strict; the zip's executable has the icon and no C runtime DLL; the tarball opens its window on Ubuntu 22.04; each file downloaded back matches the SHA-256 in the release notes	Pass
The Windows executable carries the icon	T25, T31, T41, T47, T50, T54	its resources: icon group 1 with the seven images of elyterminal.ico, 16 to 256 px, and gpui's manifest (PerMonitorV2); a GUI program, importing no C runtime DLL; on a Windows 11 VM the first 0.1.4 executable, running, showed it in the taskbar; v0.1.2's, v0.1.3's and v0.1.4's executables the same, T50's rebuild with icon and manifest byte for byte the first's; v0.1.5's executable a GUI program importing no C runtime DLL, its seven icon images, icon group 1 and manifest byte for byte 0.1.4's	Pass
The Linux tarball carries the icon and a desktop entry; the window names itself com.zacharyzhang.elyterminal	T25, T27, T41, T47, T50, T54	on Ubuntu 22.04: desktop-file-validate reports nothing, every library resolves, glibc 2.35; under Xvfb WM_CLASS reads com.zacharyzhang.elyterminal twice, and under headless sway the window's app_id reads the same; v0.1.2's passes the same checks on Ubuntu 22.04, run in the owner's WSL, WM_CLASS too; T27's rebuild passes the same checks on Ubuntu 22.04, whose llvmpipe (LLVM 15) leaves the window black in a capture, the first tarball's too, while Debian 13's paints it; v0.1.3's holds the same three files, its desktop entry names com.zacharyzhang.elyterminal, and its binary asks for glibc 2.35 at most; v0.1.4's the same, T50's rebuild too; v0.1.5's holds the same three files, its desktop entry and icon byte for byte 0.1.4's, and its binary asks for glibc 2.35 at most	Pass
Each build comes from its script	T25, T27, T31, T41, T43, T47, T50, T54	scripts/mac-app.sh on macOS, scripts/windows-zip.ps1 on the owner's Windows PC, scripts/linux-tar.sh in the owner's WSL, and for T27 on the owner's PC under Omarchy; v0.1.2's and v0.1.3's three the same way, and v0.1.3's again for T43, the disk image's on macos-signing; v0.1.4's zip on the owner's Windows VM, its tarball on the owner's PC under Omarchy, its disk image on macos-signing; T50's zip on the owner's PC, its tarball in the PC's WSL, its disk image on macos-signing; v0.1.5's three the same way: scripts/windows-zip.ps1 on the owner's PC, scripts/linux-tar.sh in its WSL, scripts/mac-app.sh in the Release workflow on macos-signing	Pass

CI notes

On 2026-10-15 every job first failed on its runner, not on the code. With the owner's leave, each runner was put right:

• macOS: fish and PowerShell were missing, so the two shell tests failed, as they should without their shells. Homebrew installed both.
• Windows: rustup's .cargo\bin held rustup.exe alone, without the cargo.exe and rustc.exe proxies, so the toolchain step stopped at rustc: command not found. rustup's installer, run again, put them back.
• Ubuntu: no runner carried ubuntu-24.04, so 71 runs waited; they were cancelled, and a runner for this repository alone runs the job in Docker, in a VM of its own. Its first build was killed linking gpui's test binaries with debuginfo in 8 GB, so the Linux and Windows jobs build without debuginfo, as scripts/linux.sh does.

A clippy run for x86_64-pc-windows-msvc on macOS, with stand-ins for the MSVC assembler and llvm-rc, found testlog unused on Windows; it now builds only beside the Unix tests that use it.

Run 907 then failed the PowerShell test on macOS: it typed its probe right after the first folder report, which pwsh 7.6.6 sends before PSReadLine reads its first key, and the line was echoed, never run. The shell tests now type a line only once the line editor reads raw keys.

Run 926, on 65bb4d3, passed all three jobs: macOS ran scripts/check.sh (135 tests, the fonts check, clippy, fmt, the 500-line rule) and the build; Ubuntu ran 138 tests and the build; Windows ran 113 tests and the build, without the Unix-only pty and shell tests.

After run 928 cargo failed on the Windows PC: Swatinem/rust-cache's post step had deleted rustup from the host's ~/.cargo/bin, on the Mac mini too, leaving the proxies pointing at nothing. The cache step now runs only in the Linux container. On the hosts, rustup toolchain install --no-self-update reads rust-toolchain.toml in place of dtolnay's action, which ran rustup default, so CI leaves their default toolchain alone. Run 934, on 0075667, passed all three jobs, each on 1.95.0 "overridden by rust-toolchain.toml".

The owner's checks

On 2026-10-15 the owner ran the app on macOS and found three faults, now fixed:

• The desktop showed through sharp: on macOS 27 gpui's Selection material builds no blur; the window's view now takes underWindowBackground, radius 30.
• The debug build lagged; the release build does not.
• No button hid the side columns; each now has one at its end of the title bar, its state saved.

On 2026-10-16 the owner ran it on Omarchy 4.0.4 under Hyprland:

• No window blurred: Omarchy turns Hyprland's own blur off, and Hyprland then blurs nothing; T28 says how to turn it on.
• The v0.1.1 Linux tarball built from aae7fba, installed by hand, wore the Omarchy theme.

On 2026-10-16 the owner ran v0.1.2 on Windows: dragged to a new size, closed and opened again, the window opened at that size.