Repository navigation
chore: release v0.8.18 - #2007
Merged
Merged
Conversation
 [Low impact] This release preparation appears safe to merge. SummaryPrepares release
Reviews (19) · Last reviewed commit: "chore(release): write whole-repo changel..." · Reviewed by Greptile |
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
davidbudnick
added
dependencies
aprilnea
Bot
force-pushed
the
release-plz/2026-10-23T11-41-03Z
branch
16 times, most recently
from
October 25, 2026 16:21
729f546 to
aa2fceb
Compare
aprilnea
Bot
force-pushed
the
release-plz/2026-10-23T11-41-03Z
branch
from
October 25, 2026 20:09
36cbc84 to
8b2a0f9
Compare
JayHome137
added a commit
to JayHome137/OpenLogi
that referenced
this pull request
Oct 26, 2026
* ci(release): drop the Codex release-notes job and use GitHub's generated notes (AprilNEA#1999) Codex never wrote the notes of a release; every run fell back to GitHub's generated list. Publish now asks GitHub for that list directly with `gh api .../releases/generate-notes` and `github.token`, passing the `previous_tag` the version job prints. A failed call leaves a compare link and a warning, so the notes still never block a release. Remove the Node tool under .github/scripts/release-notes, its pnpm install in .agents/setup, the ast-grep rule that only matched its TypeScript, and the docs that described it. Co-authored-by: Manus <dev-agent@manus.ai> * fix(release): name Sparkle deltas so Sparkle applies them (AprilNEA#2002) Sparkle treats a download as a delta only when its file name ends in `.delta`. The `.sparkle-delta` deltas of 0.8.16 and 0.8.17 failed with "No valid unarchiver", and every client then downloaded the full DMG. gpui-updater-release 0.4.4 names them `<dmg>.from-<version>.sparkle.delta` and refuses a delta URL Sparkle would not apply. Match the new names in the build artifact upload and the R2 metadata exclude, and let `release appcast` use the upstream `is_delta_file`. Co-authored-by: Manus <dev-agent@manus.ai> * feat(agent): drive the mock agent from a stdin control console (AprilNEA#2003) feat(agent): drive the mock agent from a stdin control console openlogi-agent-mock now reads one command per stdin line while the GUI and overlay stay connected. A developer can set a battery reading, take a device offline, make its reads and writes fail, show a lock-key or backlight indicator, move frozen test time, and drop every client connection, all without hardware. Battery readings and indicators go through the agent's own state machines. LowBatteryWatch::take_reading and LockKeyIndicator::announce become the only public entry points and apply the user's settings, so the orchestrator and the mock cannot skip them. The mock's observe_backlight is now served by BacklightOverlayManager, and its demo Caps Lock follows the user's switch and language. The console reads stdin on a std thread, because a blocking read on Tokio's blocking pool cannot be cancelled. Each connection also ends on a shared disconnect signal, so the overlay's agent-lost path can be tried without restarting the mock. fix(agent): refuse a mock time jump past the clock's range MockClock::advance added the requested time unchecked, so a large enough `advance` panicked the console task and left the mock without a console. It now refuses the jump and leaves the clock as it is. The lock-key platform test assumed every platform but Windows hides Num Lock. Only macOS does, so Linux CI failed. The test now expects what `LockKey::is_indicated_here()` gives on the platform it runs on. * feat(xtask,ci): compose update channel files from per-release records (AprilNEA#2000) * feat(xtask): compose update channel files from per-release records Each release now uploads an immutable feed record to releases/<tag>/feed/: its own latest.json and one single-item appcast per architecture. The new `cargo xtask release channels` composes every channel file from every record on R2, so a beta publish moves the beta channel only, and a stable hotfix published after a newer beta leaves that beta on the beta channel. - latest-json and appcast write the record into --stage; the manifest's channel is the tag's, and an item's sparkle:version is the build version. - A beta item carries <sparkle:channel>beta</sparkle:channel> and no deltas; sparkle-deltas skips a beta tag. - Channels offer their newest release in Sparkle's order of build versions. Items are copied byte for byte, so the signatures stay release appcast's. - Until a stable release has a record, the current stable channel files stand in for it. - release.yml uploads the record, composes the channel files, and uploads them last; the transitional stable-to-beta copy is gone. The version job still refuses a beta tag. - New guard xtask-release-layout-owner: support::update_feed owns the releases/<tag>/ layout and the GitHub Release page URL. * fix(xtask): keep a stopped record upload from blocking later releases A record's manifest now uploads after its appcasts and marks the record as complete. `release channels` skips a record directory without its manifest, so an upload that stopped part way no longer fails every later publish. A record with its manifest but a file missing is still an error. The old stable channel files are read only while no stable release has a record. The channel files are written to a scratch directory and renamed into place, so a failed write leaves no partial output. The docs now say that the built-in engine compares release versions, not build versions. * test(xtask): give the channel tests a fixed pubDate Without one, each item takes the time it renders at, so two renders a second apart differ and a byte comparison fails now and then. * fix(xtask): warn on every beta numbered past the beta channel, accept `dir/.` The built-in engine and Sparkle also disagree between two betas when the release PR's version goes down: an app on 0.9.0-beta.3 never gets 0.8.19-beta.10 from the engine. `release channels` now warns whenever a beta is numbered past the release the beta channel offers, stable or beta, and prints the warning as a workflow annotation in Actions. The atomic writer now resolves the output directory before it stages, so `dir/.` works again. Replacing the current directory is refused. * test(xtask): let the channels tests pass on Windows A removed record file fails with the OS's own words, which differ on Windows, so match fs_err's instead. Win32 drops `missing\..` before it reaches the disk, so there `missing/..` is the current directory: still refused, with that message. * feat(release): build a beta at its own version (AprilNEA#2004) * feat(release): build a beta at its own version A beta tags master without a version bump, so its tree carries the stable release it follows. Each build leg now runs `cargo xtask release stamp-version` before it compiles, which writes the beta's version into the workspace manifests and Cargo.lock without committing. The update engine, the About page, diagnostics, the bundles and the Linux packages then report the beta's version. `release version` prints the version to stamp (empty for a stable tag), and the version job no longer refuses a beta. * ci(build): let a dispatch build at a beta version A dry run of a beta build before its tag exists: the Build workflow takes `stamp_version` on dispatch too. A stamped dispatch only reads sccache, because its objects serve no other build. * fix(release): stamp by path, keep Arch betas' version, refuse signed dispatch stamps Review of AprilNEA#2004: - `release stamp-version` decided a dependency was a workspace crate's by its name, so a registry crate of the same name had its requirement rewritten. It now goes by `path`, also stamps `[patch.*]` entries that point at a workspace crate, and checks the stamped tree with a full `cargo metadata --locked`. If the stamp or the check fails, every file goes back to its bytes. - nfpm 2.46.3 leaves a semver pre-release out of the Arch `pkgver`, so a beta's Arch package said 0.8.18. `linux package` now spells the Arch version `0.8.18beta.1` and reads `.PKGINFO` back to check; the Linux leg installs `zstd` for that. - The Flatpak leg's stamp step ran with the workflow's sccache wrapper and no toolchain step. It now sets up the toolchain and switches the wrapper off. - A signed stamped dispatch has no build version, so Sparkle would read its `CFBundleVersion` `0.8.18-beta.1` as `0.8.18`. Such a run now fails before anything builds. - DEVELOPMENT.md no longer says the release workflow refuses beta tags. * docs(release): say which legs stamp and which beta the channel offers Review of AprilNEA#2004: the MSI leg compiles nothing and has no stamp step, the stamp itself compiles xtask at the old version, the beta channel offers the newest beta above the newest stable release rather than every beta, and signed non-release dispatches do embed Sparkle. * fix(xtask): edit each manifest once and restore every file a stamp wrote `stamp` kept its edits by the path it was given, so a root manifest that is also a member, spelled two ways, was loaded twice; the later write undid the earlier one, and a compatible requirement still passed `cargo metadata --locked`. Edits and backups are now keyed by canonical path, so each file is edited and written once. Restoring after a failed stamp stopped at the first file it could not write. It now tries every file and names each one it could not restore. The docs and workflow comments now say what the code does: restoring is best effort, the cache comment gives the real reason a stamped dispatch only reads, the Sparkle refusal names the update range it protects, and the beta channel's selection, the beta's notes base and the bundle version fallback are stated as the code implements them. * fix(device): bound Bolt receiver identity reads (AprilNEA#2009) Check receiver liveness before optional UID metadata. Reuse the existing receiver timeouts so missing UID replies preserve readable paired devices and per-slot cache fallback. Both regressions failed before the fix and passed afterward. Hardware attribution for AprilNEA#454 and the separate Bluetooth symptoms remain unverified. Co-authored-by: Manus <dev-agent@manus.ai> * feat(release): cut betas from a workflow and tell the tap about them (AprilNEA#2005) * feat(release): cut betas from a workflow and tell the tap about them Add `.github/workflows/prerelease.yml`, dispatched from `master`. It finds the open release-plz PR whose head contains the commit, runs `cargo xtask release next-beta --release-ref <its head>`, and creates the tag it names with the GitHub App's token, so the push runs release.yml. `release next-beta` names `v<release>-beta.<N>`: the release PR's workspace version, numbered by the commits since the stable release before it. It refuses a release not above that stable release, a bumped but untagged `master`, and a build version past the MSI limits, as `release version` does on the pushed tag. It also refuses a commit that is the stable release or already has a beta, an existing tag, and a tag not above the newest beta or with the newest beta off the commit's history, so betas only rise. `ReleaseTag::beta` spells the tag. release.yml's publish job exports what the beta channel serves. The tap gets `update-openlogi` for `openlogi@latest` on a stable release, as before, and a new `update-openlogi-beta` for `openlogi@beta` when a publish changes the beta channel. A tap without that event ignores it. * fix(xtask): plan a beta from the HEAD commit, refuse shallow history `release next-beta` read the workspace version from the working tree and counted commits in whatever history the clone had. An uncommitted edit could name a tag that `release version` then rejects on the tagged commit, and a shallow merge could undercount the beta number. It now resolves HEAD once and reads the committed manifest, the tags and the count from that SHA, and refuses a shallow repository. Git-backed tests cover both cases; disabling either fix fails them. The docs now say that the workflow tags the commit it was dispatched on, that the version checks do not promise a build, how pending runs queue, what the command needs to run locally, and that the event filter, not the cask, keeps betas out of `openlogi@latest`. * docs(agents): beta tags come only from prerelease.yml (AprilNEA#2001) * docs(agents): beta tags come only from prerelease.yml AGENTS.md said a single v{version} tag, created only by release-plz. Beta tags v{version}-beta.{N} are release tags too, and prerelease.yml will be the only place that creates them. Name both owners, and keep "never hand-create either tag". release-plz.toml's header says the same, and that release.yml builds and publishes both kinds of tag. * docs(release-plz): say the release-plz tag is the stable one * docs(release-plz): a beta needs an open release PR * fix(deps): update Tokio for AsyncFd soundness advisory (AprilNEA#2019) Update Tokio 1.53.1 to 1.53.3 for RUSTSEC-2026-0386. Keep existing package versions otherwise; Cargo corrects Windows-only dependency edges while resolving the update. Co-authored-by: Manus <dev-agent@manus.ai> * feat(hidpp): add capability-bound force sensing buttons (AprilNEA#2016) Add the 0x19c0 protocol foundation for adjustable press force. Bind each button to its own endpoint and reported writable range, preserve unknown flags, and separate acknowledgement from read-back. Co-authored-by: Manus <dev-agent@manus.ai> * feat(core,agent,gui): add a press force setting for the MX Master 4 Haptic Sense Panel (AprilNEA#2021) * feat(hidpp): add capability-bound force sensing buttons Add the 0x19c0 protocol foundation for adjustable press force. Bind each button to its own endpoint and reported writable range, preserve unknown flags, and separate acknowledgement from read-back. Co-authored-by: Manus <dev-agent@manus.ai> * feat(device): read and write the haptic panel press force Builds on the 0x19c0 ForceSensingButton protocol feature. Adds a ForceSensingInfo type and per-device force_sensing config, a force_sensing capability set by the probe, and get/set_force_sensing_on, which take the panel's ForceButton, refuse an unchangeable one, clamp to its reported range, and confirm the write by reading it back. * feat(agent): serve and re-apply the haptic panel press force Adds read_force_sensing and set_force_sensing to the agent IPC (protocol v40), re-applies a saved force on reconnect, wake and config reload when the device supports it, and covers the setting in the mock agent, device fixtures and fixture capture. * fix(device): keep a 0x19c0 probe failure from hiding the haptic panel A device-answered feature error now reads as an unsupported press force, only Busy or a lost reply leaves the probe incomplete, and the ReprogControls walk always runs. * feat(desktop): add a Light to Firm press force slider for the haptic panel Shows a Press force slider under the haptics row when the device's force sensing button is changeable. It spans the device-reported range, starts at the current force, writes on release and saves the force the device reads back. * fix(desktop): save only the newest confirmed press force Numbers each press force write the way DPI does, so an answer to a superseded write neither saves nor moves the slider, and only a read-back matching the request is saved. The agent skips reload writes to disabled devices, and the row stays visible with a reading line while a read is retried. * feat(desktop): move the press force slider to the haptic panel inspector and add a reset The slider shows in the mouse page's inspector when the Haptic Sense Panel is selected, outside app profiles. A reset button writes the force the mouse reports as its default, read back and saved like a slider release. * refactor(desktop): read the press force reading from one owner and keep its strings with the pointer One AppState helper answers whether the active device has a changeable press-force reading; the slider, the commit and the reset all read it. The press-force strings move to the pointer table with the rest of the mouse inspector's, the reset button gains an accessibility label, and the configuration docs name the slider's new place. * refactor(desktop): share the unsupported-error check and the force sensing test helpers * fix(device): treat a failed press force read-back as an error --------- Co-authored-by: AprilNEA <github@sku.moe> Co-authored-by: Manus <dev-agent@manus.ai> * feat(core,agent,gui): add guarded disable-keys controls (AprilNEA#993) * feat(gui): add guarded disable-keys controls * fix(keyboard): reconcile unconfirmed disable-keys writes * fix(gui): refresh config projections after disable-keys save retry * docs(gui): plan disable-keys feedback follow-up * fix(gui): localize disable-keys feedback and scope control ids * docs(gui): plan shared retry disabled-state correction * fix(gui): preserve shared retry disabled styling Let Button own semantic foreground colors and remove the unused palette argument from retry_line and its callers. Cover painted light/dark disabled colors, Enter/Space activation and disabled input, plus real-panel Save Retry recovery after a config revision conflict. Add gallery examples and verification evidence. * fix(deps): update tokio for RUSTSEC-2026-0386 Lock Tokio to 1.53.3 to address the AsyncFd I/O-safety advisory. Preserve all other locked packages and dependency edges. Validation is delegated to GitHub CI because external PR code must not run on this computer. Co-authored-by: Manus <dev-agent@manus.ai> * fix(device): gate hidden disable keys descriptors Co-authored-by: Manus <dev-agent@manus.ai> * fix(core): bump config schema for disabled keys Co-authored-by: Manus <dev-agent@manus.ai> * fix(agent): retain disable keys write ownership Co-authored-by: Manus <dev-agent@manus.ai> * fix(agent): isolate disable keys recovery Co-authored-by: Manus <dev-agent@manus.ai> * fix(agent): preserve queued disable keys response budget Start the requester deadline after the owner admits device I/O. Keep the structural guard aligned with the specialized write owner. Co-authored-by: Manus <dev-agent@manus.ai> * docs(core): describe disabled keys schema protection Co-authored-by: Manus <dev-agent@manus.ai> * feat(ipc): report saved config revision verdict Co-authored-by: Manus <dev-agent@manus.ai> * fix(desktop): reconcile stale settings revisions Co-authored-by: Manus <dev-agent@manus.ai> * fix(agent): fence write admission after receiver lease Check the current intent only after waiting for receiver access. Remove stale imports and use named replay construction parts without relaxing lints. Co-authored-by: Manus <dev-agent@manus.ai> * fix(device): use feature set descriptor type Co-authored-by: Manus <dev-agent@manus.ai> * fix(desktop): invalidate keys after external config reload Invalidate active and retained snapshots for both general and Disable Keys save acknowledgements. Close writes until the fresh query completes. Co-authored-by: Manus <dev-agent@manus.ai> * fix(agent): preserve disable keys transaction ownership Retain native I/O ownership across discovery and readback, confirm manual intent only on requester receipt, and queue reconciliation reads behind admitted policies. Preserve generic setter replies and distinguish physical replacement from identity enrichment. Add deterministic regressions and owner guards. * fix(gui): reconcile exact disabled keys config revisions Correlate coalesced reload receipts with each saved revision and keep editing blocked until the refreshed revision is accepted and reread. Preserve retry state across refresh failures and disconnects. Add stale-receipt, repeated external-edit, and recovery regressions plus an owner guard. * docs(config): explain disabled keys reconciliation guarantees * docs: drop disable-keys planning artifacts The plan and spec files under docs/superpowers/ are working notes from the implementation. The shipped behavior is described in docs/CONFIGURATION.md and the user documentation, matching how 75c7e2c retired the AprilNEA#344 artifacts. * fix(agent-core): guard the write-through keyboard setters agent-keyboard-write-owner matched set_disable_keys and set_*_on but not set_disable_keys_write_through_on, the entry point the Disable Keys owner actually calls. Another agent caller could reach the device setter without SharedHandles ordering or native completion ownership. * fix(desktop): name disable-key switches by their action Each switch is on when its key is disabled, but its accessibility name was only the key name, so a screen reader announced "Caps Lock, switch, on". Name the switch "Disable <key>" through a localized catalog entry. * fix(agent-core): preserve requests across identity enrichment * fix(agent-core): wake cancelled disable-keys queue waits * fix(desktop): fence external config adoption until revision confirmation * docs(config): describe the disable keys fixes Cover the revision fence for Save Retry and other saves, cancelled queued requests, direct keyboard identity enrichment, and the switch accessible name. * test(desktop): end statement closures with semicolons * fix(desktop): adopt the shared retry line in the press force row * fix(agent): announce Disable Keys recovery to clients A recovery restores the keyboard without a 0x4521 report, so the device-reported settings revision stays put and a client that read the abandoned mask keeps showing it. Count each started recovery per route and add it to the published revision. The recovery write already holds its turn, so the client read that follows waits for it. * docs(core): note that Disable Keys recovery refreshes the app * refactor(agent): move Disable Keys recovery dispatch out of the inventory refresh Keeps refresh_inventory_inner under the too_many_lines limit. --------- Co-authored-by: AprilNEA <github@sku.moe> Co-authored-by: Manus <dev-agent@manus.ai> * fix(i18n): translate the disable-keys strings in 22 locales (AprilNEA#2024) PR AprilNEA#993 added 25 Disable keys strings and translated them only in Russian. The other 22 catalogs showed English text. Key names follow each catalog's existing key-name convention. * fix(agent): re-apply volatile settings after replacing a dead channel (AprilNEA#1997) * fix(agent): re-apply volatile settings after replacing a dead channel When the enumerator retires a channel for failing probes and later reopens the node, the device comes back without the published inventory ever showing it offline: the ledger replays it as online throughout. `reapply_targets` therefore sees no offline->online transition, and any `0x1d4b` reconnect broadcast arrives while nothing listens on the new node. The firmware has lost its volatile settings by then, so DPI and wheel inversion stay at their defaults until something else re-applies them. Track the nodes retired for failing probes and report when a pass opens a replacement for one. The inventory watcher sends `DeviceWake` before that pass's snapshot, so the snapshot that publishes the new channel re-applies volatile settings through the existing reconnect path. * test(agent): cover DeviceWake ordering after channel replacement Drive the inventory worker through a direct mouse whose handle dies while its node stays listed: two failed probes retire the channel, and the pass that reopens it must send exactly one DeviceWake immediately before the snapshot that publishes the replacement. * fix(device): report a replacement only while its node replayed online A node the OS keeps listing but that never answers again keeps being retired and reopened: the ledger asks for a fresh channel on every second failed probe. Each reopen set `replaced_channels_last_tick`, so the watcher sent `DeviceWake` once per cycle, and the forced re-apply rewrote every online device's volatile settings and bumped the global capture re-arm generation about once per recovery period. The signal is meant for a reconnect no snapshot observed. Ask the ledger, which owns that fact: report a reopen only if the node's last-good inventory was still replayed when its channel reopened and is still published after the pass settles. Once the replay grace expires the agent has seen the node go offline, and its return re-applies through the ordinary offline-to-online transition. * fix(device): keep the replacement wake for a partial live result The previous commit dropped the replacement signal whenever the ledger no longer held a last-good inventory after the pass settled. That also dropped it when the reopening probe failed past the replay grace but still returned a partial inventory, such as a Bolt receiver with one unreadable slot. That snapshot keeps publishing the readable devices online, so no offline-to-online transition would restore their settings. Decide on what the pass publishes instead: drop a replacement only when its settled inventory is empty. A dead channel still publishes nothing past the grace, so the retire/reopen cycle stays quiet. * refactor(device): fold the replacement filter into one retain Clippy's too_many_lines rejected enumerate_reporting_completeness at 101 lines after the previous commit. The same condition fits in one retain call, which keeps the function at the 100-line limit without an allow. --------- Co-authored-by: AprilNEA <github@sku.moe> * fix(overlay): draw the Windows Actions Ring without a window frame (AprilNEA#1913) * fix(overlay): draw the Windows Actions Ring without a window frame GPUI creates WindowKind::PopUp windows with a zero style, which is WS_OVERLAPPED, and CreateWindowExW gives every overlapped window a caption. Windows 11 then draws that frame — an opaque backdrop, a thin border and rounded corners — around the transparent ring and notice. Make the window a borderless WS_POPUP before placing it, applying the change with SWP_FRAMECHANGED on the bootstrap move, and ask DWM for square corners and no border colour. Fixes AprilNEA#1897 * fix(overlay): keep placing the window when frame removal fails * style(overlay): space the windows-sys feature list --------- Co-authored-by: AprilNEA <github@sku.moe> * fix(windows): post the layout's scan code with keys a layout can move (AprilNEA#2028) Keys a layout can move now carry the scan code of the key that produces their virtual key under the foreground window's layout. Illustrator matches shortcuts by that scan code under a Russian or Ukrainian layout, and ignored the remapped key without it. The reporter confirmed the fix with the PR build. Fixes AprilNEA#2026 * fix(cli): route backlight operations through the agent (AprilNEA#2031) Route `openlogi backlight` through the running agent instead of opening a competing HID connection. The agent bumps the existing settings revision after each backlight attempt, so an open GUI reads the keyboard again. Fixes AprilNEA#2029 Refs AprilNEA#521 Refs AprilNEA#1200 * fix(i18n): translate the receiver chooser, old-pairing cleanup, and update channel in Czech (AprilNEA#2034) Co-authored-by: Imbecile6197 <335541687+Imbecile6197@users.noreply.github.com> * feat(desktop): add a keyboard backlight duration timer (AprilNEA#1879) Keyboards that report HID++ `0x1982` version 3 or later can now read and set how long the keys stay lit after typing stops. - Device tab: a "Turn off after" row with 5-30 s steps; "Enable extended durations" adds 1 min-2 h steps and writes nothing to the keyboard. - CLI: `openlogi backlight duration SECONDS` (a multiple of 5, from 5 through 7200), sent through the agent. - Core/IPC: `BacklightControls` is a bit set with a `DURATION` control; the timer is recorded under protocol v41, and older recorded profiles still load. Closes AprilNEA#1834 * fix(assets): follow the 0.1 asset release line instead of pinning 0.1.0 (AprilNEA#2039) * fix(assets): follow the 0.1 asset release line instead of pinning 0.1.0 * docs(assets): describe the release-line asset mirrors * feat(assets): ask the jsDelivr data API for the newest asset release (AprilNEA#2040) * feat(assets): ask the jsDelivr data API for the newest asset release * docs(assets): describe the jsDelivr data API lookup * fix(assets): keep the range release when the newer jsDelivr release fails * docs(assets): state the measured jsDelivr cache delays * chore: release v0.8.18 (AprilNEA#2007) Co-authored-by: aprilnea[bot] <41898282+github-actions[bot]@users.noreply.github.com> * fix(flow): complete upstream snapshot merge --------- Co-authored-by: Xuan Zhang <github@sku.moe> Co-authored-by: Manus <dev-agent@manus.ai> Co-authored-by: Hima <19400811+HimaAramona@users.noreply.github.com> Co-authored-by: Sergey Afonin <safonin@users.noreply.github.com> Co-authored-by: c1m <2175815055@qq.com> Co-authored-by: Moon <16745613+MoonZhe@users.noreply.github.com> Co-authored-by: Libor <jsem@libormacak.eu> Co-authored-by: Imbecile6197 <335541687+Imbecile6197@users.noreply.github.com> Co-authored-by: Kabir Singh <26483873+ASparkOfFire@users.noreply.github.com> Co-authored-by: aprilnea[bot] <263112868+aprilnea[bot]@users.noreply.github.com> Co-authored-by: aprilnea[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Jay <jayboy137@Jay-MacMini.local>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
[0.8.18] - 2026-10-25
Added
Fixed