Driver-specific configuration, passed through untouched.
OptionalAllowWhether a failed replay may fall back to the agent and re-derive the goal.
Defaults to true. Set false to pin a test to deterministic execution,
wherever a silent re-derivation could paper over the regression the test
exists to catch.
A green fallback records to PendingReplayScript, not
ReplayScript — replacing a promoted script is always a reviewed act.
This governs the whole goal; to stop a single STEP being repaired during
replay, use the driver's replayHeal.
OptionalPendingA newly recorded script awaiting review, written when a run re-derives a
test that already had a ReplayScript. Replay always uses the
promoted script, never this one, so a UI change never takes effect until
someone has seen the diff and promoted it (mj test scripts).
A test's first script skips this and lands in ReplayScript directly — there is no baseline to diff it against, and the run that produced it already passed the judge and every gating oracle.
OptionalReplayThe recorded, replayable trajectory for this test — written by a passing
agent-driven run, replayed by later runs with no model calls. Absent until
the first run records one. Held structurally identical to ComputerUseTrace
by compile-time assertions in @memberjunction/computer-use-engine.
Shape of the
Configurationcolumn onMJ: Tests.The column is shared by every test type, and each driver parses its own shape out of it — so the named properties here are the ones the framework understands and the index signature carries the rest through untouched. Adding a framework-level option is an edit to this interface plus
mj sync push, never a migration.See
plans/regression-testing/dom-selection-and-replay-design.md