Start a sealed session: a display of its own (X11 under Xvfb, or Wayland under a headless sway), its own D-Bus and accessibility bus, and a scratch home. Nothing in it reaches the desktop of the person you're working for, and nothing of theirs reaches it. The answer gives its name, display and size, the programs spawn_program can start in it, and anything it couldn't seal. It lasts until stop_session.

Stop a session start_session made, when you're done with it: every program in it is killed, unsaved work and all, and its scratch directory is deleted, with any clip record_video saved there (one saved at a path is kept). In a session started with usage, the answer says its programs' peak memory and CPU time. You needn't stop sessions before the server ends: they all stop with it.

Start one of the programs the person allowed this server to run, by its name, in a session; nothing else can run. The answer gives its pid. Then wait_for its window, by the program's name, before looking at it. If it doesn't come up, get_program_output says why.

Read the last lines (40, or lines) a spawned program wrote to stdout and stderr, after a line saying whether it's still running or exited, with its status. It works after the program has exited, until the session stops; for a program spawned more than once, it reads the latest. Use it when a window never comes up or a program seems to have crashed; to wait for a line in a log file, use wait_for line.

Read what a screen reader would find: every visible node, a line each, indented, with its role, name, description and the states a screen reader says (focusable, focused, checked...). The cheapest way to see what's on screen and what can be clicked by name. A node with an empty name ("") is one a screen reader can say nothing for. Use take_screenshot for how it looks.

Take a picture of the screen, or of one window, RETURNED AS AN IMAGE for you to look at: the only way to see colors, layout and what's drawn. Pictures cost far more than get_accessible_tree; take the window rather than the screen, and crop when you need a part. vision checks nothing is told apart by color alone; burst, as one image, is for motion, an animation or a transition. The answer gives its size, and if it was shrunk, the scale.

Record a video of the screen, or of one window, and save it as an .mp4: in the session's scratch directory, gone when the session stops, or at path. The answer is the saved path, how long the clip is, and up to 8 frames from it RETURNED AS ONE IMAGE, with the times they show: you can't play the clip, but a person or another tool can. A clip stops at 100 MB. sound also says what it heard, as record_audio does. until_line ends it at an event instead; the answer gives the line. For a few moments of motion, take_screenshot's burst is cheaper.

Click an element by its accessible name, or a point by x and y, on the screen or inside a window. By name is the one to use: it waits up to timeout seconds for the node, raises its window and clicks its middle, and the answer gives the node and its bounds. When two nodes match it lists them, with where they are; when none does, the nearest names. A click gives a window the keyboard, which nothing else does on X11. hold and state, by x and y, hold the button as press_keys holds keys.

Turn the mouse wheel over a node by its name, found as click_element finds it (waiting for it, its window raised), or at a point by x and y, leaving the pointer there. Use it for what's past a list's, a page's or a window's edge, then wait_for stable and look again; to move the keyboard focus, use press_keys.

Press keys in turn, each a chord: Tab, Return, Escape, ctrl+s, alt+shift+Tab, F6. They go to whatever has the keyboard; click a window first with click_element if nothing does. hold presses them all together and keeps them down that long (walking while running); state down leaves them down, across calls, until a call with state up (move or click while a key is held). The answer says what it pressed.

Work a virtual gamepad, in a session started with gamepads: press buttons together, tapped, held for hold seconds, or with state down left down until a call with state up; push a stick to x, y; pull a trigger to value; or neutral, letting go of everything. A stick or trigger stays where it's put, unless hold says for how long. Programs in the session see a real Xbox 360 controller, and none of the machine's. For keys and the mouse, press_keys and click_element.

Type text into whatever has the keyboard focus, a key at a time. ASCII only: a character with no key fails the call, once what's before it is typed. For keys and shortcuts (Return, ctrl+s), use press_keys.

Wait, instead of sleeping, for: a window (value: its title, exactly or a regex, or app=RE, or the name of a program spawn_program started, for a window of that program's, whatever its title; giving up, it lists the windows there are), stable (the screen, or the window, unchanged for value seconds, 0.5, up to 60: after a click, for the repaint), an announcement (value: a regex in what was announced to a screen reader; each announcement is waited for once), focus (value: the name of the node with the keyboard focus, as click_element takes one), match (value: a PNG, by its path from the allow file's directory, that the screen, or the window, cropped to crop, should look like), change (from how it looked when the wait started), or line (value: a regex found in a line of file, written after the wait starts, or with from_start, anywhere in it: for a program that logs its events, rather than a sleep; a file not there yet is waited for, and one rotated or rewritten is read from the top). match and change say how long they took; line answers with the line. Fails after timeout seconds, saying so.

Lint the accessibility tree for nodes a screen reader has no name to say for: controls, images (not inside a named control) and anything else that takes the focus. The answer: a tree line each, with bounds, then what's wrong; or "No problems.". Use it to check labels after a change; for colors, use check_contrast.

Check that text, buttons, links and images stand out from their background: WCAG's 4.5:1 for text (3 for large text) and 3:1 for images unless text_ratio and graphics say otherwise, and with edges, a button's or field's edge too. It raises the window and measures a picture of the screen in each node's bounds. The answer: each node that fails, a tree line with bounds, its ratio and colors, or "No problems."; then what it couldn't measure, which doesn't fail (too small, or GTK 4's, whose positions are wrong), and why. A window where nothing can be measured, as GTK 4's, is an error, not a pass. Switches, sliders, check boxes, disabled nodes and what's scrolled out of view are skipped. For missing names, use check_accessible_names.

Check the keyboard focus order in one call: press Tab, or key, until the focus comes back round, and list where it went, a tree line each, after a "began in" line. Then the controls the key never reached, and how it ended: round again (fine), stuck (a trap, or a text view nothing leaves), not round after most presses, lost (no node has the focus), stopped (a text view closed as it was left), or nothing focused to begin with (click the window with click_element first). With visible, "focus not shown on N" counts stops that don't show it. A text view that takes Tab is left as a person would, with ctrl+Tab, shift+Tab, Escape or F6. The focus stays where it ended. To press a few keys yourself, use press_keys.

List what programs in the session have announced to a screen reader since it started, oldest first, a line each with its politeness: "polite: Saved", "assertive: Error". That's GTK's announcements and a browser's live regions. Reading doesn't clear them; "Nothing announced yet." means none so far. Use it to check what a screen reader user heard after an action; to wait for one, use wait_for announcement.

Record what a session started with audio plays, from a call with action start to one with action stop, across the calls between that make the sound. stop's answer: the sound's length, peak and RMS level (dBFS), samples clipped, and each gap of exact silence between the first sound and the last, as a program that falls behind leaves; then what's wrong, if anything (gaps, clipping, silence). Sounds meant to be apart, two beeps say, need a gap longer than the silence between them. For a video of the screen with its sound, up to a minute, use record_video with sound.