This is a nightly release of Vibeshine and contains several bug fixes and may contain new features. Check the commit log for more information.
Windows installer · Linux (Arch x86_64)
Nightly Windows installers are unsigned and include self-signed gamepad components. They may cause false-positive antivirus alerts.
Source: fa218c69d881
Since the preceding successfully published nightly Commit log
2.0.2-daily.d20261015.1: compare exact sources.
fix(playnite): authenticate the SYSTEM host without elevation
Normal-user Playnite cannot inspect the service host process image.
Adopt the reviewed connection-token fallback only for host access denial,
retain PID, role, executable and ACL checks, and restrict host pipe tokens
to identification. Publish connector 0.5.2 so existing copies can update.
Addresses Nonary/Vibepollo#608
Addresses Nonary/Vibepollo#618
Addresses Nonary/vibeshine#311
(cherry picked from commit 3ed0834514b8e7cdc55e1e814dbb4578cdd875e3)
Generated with [Codex](https://openai.com/codex/)
Model: GPT-6.1-Sol X-High
fix(installer): preserve large environments during interop builds
Windows PowerShell CodeDOM rejects otherwise valid large environments.
Invoke the Framework compiler directly for both driver interop wrappers,
while preserving the environment, verification policy and visible errors.
Exercise the original limit and production wrappers without installing drivers.
Addresses Nonary/Vibepollo#610
Generated with [Codex](https://openai.com/codex/)
Model: GPT-6.1-Sol X-High
fix(ui): keep historical stats within the mobile viewport
The fixed 680-pixel drawer pushed session details off the left edge of
phone screens. Cap its width at the viewport while retaining the desktop
width so metadata, charts, and actions remain reachable.
Refs: Nonary/Vibepollo#612
Generated with [Codex](https://openai.com/codex/)
Model: GPT 6.1-Sol X-High
(cherry picked from commit 37cde2d3e87941feb47f4c15fe7777ee8927ccb5)
fix(steam): discover Windows libraries from service sessions
LocalSystem could not see the desktop user's SteamPath, leaving non-default
installs undiscovered. Resolve the selected user hive and both machine registry
views, retain folder fallbacks, and scope library/artwork reads to that user
while preserving protected service cache writes and nested failure handling.
Addresses Nonary/vibeshine#314
Generated with [Codex](https://openai.com/codex/)
Model: GPT 6.1-Sol X-High
(cherry picked from commit 451aee7735c5939f497bc14aa34d33bc3c2bb0cd)
fix(tray): cancel startup waits during Windows shutdown
Stopping before Explorer appeared left the tray waiting for a shell while
the shutdown watchdog forced a crash. Cancel startup waits, serialize owned
thread joins, and cover early main returns before process resources disappear.
Exercise missing/delayed desktops, initialization races and repeated teardown.
Addresses Nonary/Vibepollo#614 and Nonary/Vibepollo#615
Generated with [Codex](https://openai.com/codex/)
Model: GPT 6.1-Sol X-High
(cherry picked from commit fc37d8b9ccd771dbe3c3606c9d7cbb8f2be6ab22)
fix(kms): refresh display maps before capture initialization
Encoder probes could consume empty or stale connector maps before an
output query. Refresh maps for the requested device before capture starts
and serialize enumeration with initialization so concurrent probes cannot
replace shared routing state. Cover factory routing and transaction locking
without opening GPUs or changing the live display.
Addresses Nonary/Vibepollo#611
Generated with [Codex](https://openai.com/codex/)
Model: GPT-6.1-Sol X-High
fix(capture): Retry transient output observation failures
Failed DXGI observations immediately latched a capture restart even when the
output recovered unchanged. Retry null observations in a bounded window while
preserving confirmed display changes and capture backend errors. Keep retries
active under event subscription and report recovery or exhausted observations.
Replay the production worker for transient, repeated, initial and persistent
faults, including the populated capability-change boundary. This fixes the
conditional capture regression; #301's reported periodic cause remains unproven.
Generated with [Codex](https://openai.com/codex/)
Model: GPT-6.1 Sol X-High
(cherry picked from commit 145d5e27ea3556d9b9fb96f1ae059d25b75f04be)
fix(display): preserve owned inactive displays during recovery
Recovery replaced responsive control transports and treated missing Windows
mode data as unsupported, unnecessarily cycling securely owned displays.
Preserve healthy transport provenance and consult the live exact preferred
descriptor when enumeration is unavailable. Keep ownership, adapter and
explicit mode checks strict; cover both recovery paths with portable tests.
Addresses recovery defects in Nonary/Vibepollo#617; the initiating fullscreen
deactivation remains unproven.
Generated with [Codex](https://openai.com/codex/)
Model: GPT-6.1 Sol X-High
(cherry picked from commit 8dd058b005d171b608b83c4594681d745ac54350)
fix(input): retain touch for automatic PlayStation profiles
Automatic VHF can still select DualSense for a PlayStation client with
both selection heuristics off. Keep controller-touch negotiation enabled
for that reachable profile while preserving fixed touchless profiles and
composite overrides. Cover the selection and capability contract in tests.
Generated with [Codex](https://openai.com/codex/)
Model: GPT-6.1 Sol X-High
(cherry picked from commit a3434b0a708bf5e7e5f78dae8247b5053f8c6049)
test(installer): isolate the CodeDOM environment size control
The Windows regression combined oversized environment and Unicode scratch
paths, and failed with a path error before proving the expected size limit.
Establish a working baseline and vary environment size first; then require
both production compilers to handle the full Unicode fixture without
altering the environment or hiding compiler failures.
Related: Nonary/Vibepollo#610
Generated with [Codex](https://openai.com/codex/)
Model: GPT-6.1-Sol X-High
fix(amf): retain fresh captures when older output wins submission
An older completion can race the output drain and discard the latest
unaccepted capture. Deliver that output, then retry the retained capture
once with its original pixels, frame ID and timestamps. Preserve bounded
backpressure and recovery, with extracted production-code replay coverage.
Ported from Nonary/Vibepollo commit
886ef840d82003e3225242ed241c6ca64b5e5832. Preserve Vibeshine's two-timestamp
API and register only the new AMF regression; this addresses the demonstrated
submission race, not the full reported loss symptom.
Related: Nonary/Vibepollo#613
(cherry picked from commit 886ef840d82003e3225242ed241c6ca64b5e5832)
Generated with [Codex](https://openai.com/codex/)
Model: GPT-6.1-Sol X-High
test(installer): load package constants after helper declarations
The inline interop helper now precedes the package constants. The nightly
MSI fixture stopped at that helper and compared a valid manifest with an
empty expected list. Load all direct top-level assignments while keeping
installer entry points inert and all package/security checks intact.
Generated with [Codex](https://openai.com/codex/)
Model: GPT 6.1-Sol High
test(linux): make wedged app polling independent of scheduler delays
A one-second wall-clock budget can expire before the first mocked stop,
incorrectly reporting a repeated request. Control time only in this fixture
and require one stop, three polling passes and deadline rejection. Preserve
the separate real command-timeout checks and propagate subshell failures.
Generated with [Codex](https://openai.com/codex/)
Model: GPT 6.1-Sol High