Command line

jevtest has two commands: run runs tests, and inspect shows what's on the app's screen as steps name it, for writing them.

jevtest run PATH... --lock MODE --out DIR [--test NAME]... [--platform android|ios]... [--prune-lock] [-v]

Argument	Required	Meaning
PATH...	yes	Test files and/or folders. A folder runs every test file in it and its subfolders (rules).
--lock MODE	yes	How the lockfile is used.
--out DIR	yes	Where results go. Each run adds a timestamped folder inside it.
--test NAME	no	Only run this test. Repeatable; names can come from any of the files.
--platform android\|ios	no	Only run each file's tests on this platform: its other builds needn't exist, its other devices aren't looked for, and the ${NAME}s only they use needn't be set, so a Linux CI job can run the Android part of files that also run on iOS. A file without the platform doesn't run, and says so. Repeatable.
--prune-lock	no	After a run where every test passed, remove saved steps and recorded answers the run didn't use.
-v, --verbose	no	Also print every Jev question with its top answers and probabilities.
--version	no	Print the version.

Lock modes

--lock	A do: with saved steps	A do: without	Other Jev questions (expect:, …)	Writes the lockfile
record	repeats them; if one no longer fits, Jev works the rest out again	Jev works it out; the steps are saved	recorded answer, else Jev (recorded)	yes
frozen	repeats them; if one no longer fits, the step fails	the step fails	recorded answer, else the step fails	no
refresh	Jev works it out again	Jev works it out	Jev is asked again	yes: new answers replace recorded ones; --prune-lock drops the rest
off	Jev works it out	Jev works it out	Jev is asked	no

Before anything runs

jevtest checks everything it can before it installs or starts anything, and exits with code 2 and a message saying what to fix:

• every test file and library loads (a file's problems are listed together), and every ${NAME} is set;
• every step a file's tests take can run on each platform the file runs on: no network: or Android-only key: in a file that runs on iOS, and every grant: names its permission for each platform;
• every app build exists;
• every device a file names is running (a missing one is listed with the devices that are), and no other jevtest run is testing it: a second run on the same device would restart the app under the first. The claim ends with the run, however it ends;
• with --lock refresh or --lock off, TYPESAFE_API_KEY is set if a test that runs has a do: or an expect:, since those always ask Jev then;
• --prune-lock has every test on every platform and a lockfile to prune (below).

--prune-lock

The lockfile only grows: a do: or a screen that no longer exists keeps its saved steps or recorded answers. --prune-lock removes them, and the answers Jev gave while working out a do: whose steps are now saved (later runs repeat the steps without asking). It only acts after a run of every test in which every test passed, because a failing test stops early and skips screens that are still real. So it refuses to combine with --test, --platform or --lock off, and after a failure it says not pruned, because a test failed.

Prune with the same devices you record on: decisions recorded on one device's screens are unused on another's.

Exit codes

Code	Meaning
0	every test passed
1	a test failed
2	setup error: bad test file, missing value, no such device, bad arguments
130	interrupted (Ctrl-C)
143	stopped by SIGTERM (a cancelled CI job)
129	stopped by SIGHUP (the terminal closed)

A stopped run puts every device back (orientation, appearance, network, autofill, location) and says so: stopped (SIGINT): devices put back; this run wrote no report.

inspect

jevtest inspect APP --device NAME --out DIR [--once]

Argument	Required	Meaning
APP	yes	The build: .apk/.aab for Android, .app/.zip/.ipa for iOS.
--device NAME	yes	The running device, named as a test file names it.
--out DIR	yes	Where to save. Each inspect adds a timestamped folder inside it.
--once	no	Save the screen once and stop, instead of waiting for Enter: for scripts and agents.

It installs and starts the app as a run does, keeping the app's data (no fresh:), waits until the screen stopped moving as a screenshot: step does, and saves it as 001.png with the notes beside it: the elements and every name a step can find each by, as a list (.txt), as JSON for an agent (.json), and as a page that boxes each element on the screenshot (.html). Then it waits: use the app on the device, press Enter to save the screen again (002.png, …), or type q to stop. The list is printed too:

text           'Welcome back,'
text           'View All' (2 on screen)
button         'Sign in' | 'login_button'

The exit codes are those of run: 0 when done, 2 when something needs fixing first, 128 + the signal when stopped.

Environment

Variable	Meaning
TYPESAFE_API_KEY	Your TypeSafe API key, for Jev. From the environment or the .env next to the test file. Not needed when every decision comes from the lockfile.
JEVTEST_CACHE	Where the built on-device agents are cached (default ~/.cache/jevtest).