Getting past character-select: the host does it, not the launcher

From the aowlspt repository — docs/AUTOENTER.md

Published from aoughwl/aowlspt with only its links repaired. It was written for someone with the checkout open, so it is terse and points at source files by path — browse the tree as you read. A few of the documents it cites were held back from publication and are not here.

The manual · The aowlspt section

Who answers Tarkov's character/mode screen when the game starts — the host, natively, rather than the launcher driving the UI.

The default since 2026-09-02: the launcher writes NOTHING to the live inspector at boot. The host answers the character/mode screen natively. Pass --auto-enter to aowlspt-launch.exe if you want the launcher to drive that screen instead.

The two paths

The host path (default). host/Aowlspt.Host.Il2Cpp/modeskip.nim, gated by uxSkipModeScreen in aowlspt-host.json, calls Submit for the bound profile from inside the client. It runs on the Unity main thread, needs no channel, no polling and no file, and it lands at ~26s of a normal boot. Its line in aowlspt-host.log is:

skip mode screen: Submit called for profile <id>

The launcher path (--auto-enter). tools/aowllaunch.nim waits for the CharacterSelectionScreen to be up and presses the "PvE" slot through the live-inspector file channel, using tools/aowlui.nim. Keep it for debugging the selector itself, or for a build where uxSkipModeScreen is off.

Why the default flipped

Measured on a normal boot, auto-enter wrote nine batches into aowlspt-inspect.txt in the first 22 seconds: eight roots polls and then a 58-command batch carrying allow write and 29 call name:get_activeInHierarchy. That is a call into game code that nobody asked for, on a channel that has exactly ONE command file and no arbitration beyond a lock, at the same moment the host was already doing the same job. It also left that 58-command probe batch on disk, where the next boot's host reads it first.

Doing a job twice is only cheap when the second attempt is free. This one cost the channel: any tool driving the inspector during those 22 seconds had its batch delayed or dropped.

What --auto-enter does now, when you do pass it

Three things changed at the same time as the default, because the traffic was worth fixing whether or not it is on by default:

• The readiness poll is READ-ONLY. It is roots + children with no allow write and no call into game code, and it is what runs for the whole cold-boot minute while the answer is "the screen is not there yet".
• The allow write batch only fires once the selector node exists.activeInHierarchy is reachable only through call name:get_activeInHierarchy, and the host gates every call behind allow write -- so confirming the screen is UP genuinely needs a write batch. It is now a separate batch from the readiness poll and from the press, and it is not issued when there is nothing to confirm.
• The command file is blanked when the launcher is done, on every exit path -- pressed, already past it, timed out, no channel. If another writer holds the channel lock at that moment the file is left alone and the launcher says so, because clobbering someone else's in-flight batch is the collision the lock exists to prevent.
• Every batch names its writer. Sentinels are aowl-batch-launch<pid>-<ms>-<n>, the convention plugins/aowlsptcode/mcp/channel.py uses. The sentinel is the only part of a batch that reaches the host log (the host echoes > echo aowl-batch-... and skips #-prefixed serial lines), so unattributed traffic previously had to be traced to its writer by the SHAPE of the sentinel. It is a grep now.

Flags

flag	effect
(none)	the host answers character-select; the launcher writes nothing to the channel
--auto-enter	the launcher drives character-select over the inspector channel
--no-auto-enter	accepted and ignored -- it asks for what already happens. Kept so existing scripts and shortcuts keep working

The launcher says which it chose, once, on stdout and in aowlspt/aowlspt-autoenter.log:

auto-enter verdict: OFF (default) -- the host answers character-select
natively (uxSkipModeScreen, modeskip.nim); pass --auto-enter to drive it from
the launcher instead

aowlspt-launch --dry-run prints the same decision as a would enter line in its plan, from the same option the real run branches on.

The test, and what it does not cover

python tools/test_autoenter.py drives the real launcher binary against a scratch install: the default plan enters nothing, --auto-enter flips it (the positive control), --no-auto-enter is still accepted, and a real default run leaves no aowlspt-inspect.txt -- with channel.py writing into the same directory as the control that proves the detector is looking in the right place.

Not covered: no client is booted, so the --auto-enter side of "and then it writes" is NOT exercised. The auto-enter block only runs after a successful process start and a successful host injection into a real IL2CPP client. Proving the write side needs a live boot with --auto-enter and a grep of aowlspt-inspect.txt for an aowl-batch-launch<pid>- sentinel.