Reliability

public boolean run(MyCoolConfig config) {
    reliability()
            .named("My Cool Script")                         // the name the player sees if it stops
            .trackProgress(() -> Inventory.count("Logs", true))
            .onRecover(ctx -> {
                Checkpoint cp = ctx.getCheckpoint();
                return cp != null && cp.getLocation() != null && Walker.walkTo(cp.getLocation());
            });

    mainScheduledFuture = scheduledExecutorService.scheduleWithFixedDelay(() -> {
        try {
            if (!super.run() || !ProjectX.isLoggedIn() || !reliability().tick()) return;

            if (Inventory.isFull()) {
                ActionResult open = reliability().step("Open bank")
                        .action(Bank::walkToBankAndUseBank)
                        .verify(Bank::isOpen, 5_000)
                        .run();
                if (!open.isSuccess()) return;           // already retried, recovered or stopped

                ActionResult deposit = reliability().step("Deposit logs")
                        .require(Preconditions.bankOpen())
                        .action(() -> Bank.depositAll("Logs", true))
                        .verify(() -> !Inventory.hasItem("Logs", true), 3_000)
                        .run();
                if (deposit.isSuccess()) reliability().checkpointAtPlayer("Banked");
            }
        } catch (Exception e) {
            log.error("MyCool loop failed", e);
        }
    }, 0, 600, TimeUnit.MILLISECONDS);
    return true;
}

The classes are in net.runelite.client.plugins.projectx.util.reliability. Every Script has its own service from reliability(); nothing changes for a script that never calls it. Script.shutdown() resets its failures and checkpoint, and keeps the recovery handler and progress signal, so set those once in run().

A step describes one action before it runs: what must be true first, what to do, how to tell it worked, and how hard to try. Nothing happens until run().

An ActionResult tells you what happened: isSuccess(), getOutcome() (SUCCESS, RECOVERABLE or UNRECOVERABLE), getAttempts(), getReason(), getError() and isScriptStopping(). When a step did not succeed, return from the loop: the service has already retried, recovered or stopped the script, and the next pass starts fresh.

Ready-made checks in Preconditions, all safe to call from the script thread:

An unmet precondition fails the step without running the action, as recoverable. Add .fatal() to one (for example Preconditions.hasItem("Hammer").fatal()) to stop the script instead. The description appears in the log, and in the message to the player if the script stops.

1. A failed attempt is retried after the back-off wait, up to the player's “Retries per action”.
2. A step that still fails is counted and handed to your recovery handler, with a RecoveryContext: getTrigger() (STEP_FAILED or NO_PROGRESS), getStep(), getReason(), getConsecutiveFailures(), getStepFailures() and getCheckpoint().
3. When “Failures before stopping” is reached, either that many failed steps in a row or the same step failing that many times without once succeeding, the script is stopped safely.

The recovery handler runs on the script thread, so it may walk, bank and use guarded steps (those never start a second, nested recovery). Return true when the game is back in a known state. To give up, throw UnrecoverableActionException: from an action, a precondition, a verify check or the handler, it stops the script at once with its message as the reason.

A safe stop logs the reason, shows the player a message (“My Cool Script stopped itself so it would not carry on in a broken state. Reason: ... Last good step: ...”) and calls shutdown(). Stop the script yourself the same way with reliability().stop("Out of runes.").

Call reliability().tick() once per pass, after super.run(), and return when it is false. It counts time without progress. When the player's “No-progress timeout” passes, the recovery handler is called once; if a second timeout passes with still no progress, the script is stopped (straight away if there is no handler). Time logged out, paused, on a break, or with the loop not ticking for over 30 seconds does not count.

Players tune this under Project X settings > Script reliability. A running script picks up a change on its next step.

With reliability switched off, steps still run once and still return their result, but nothing is retried, recovered or stopped automatically. stop(reason) works either way.