The settings screen doesn't need its own launcher icon — LSPosed Manager
already provides the entry point for opening a module's settings, via
PackageManager.getLaunchIntentForPackage(), which resolves CATEGORY_INFO
before falling back to CATEGORY_LAUNCHER. Switching to INFO keeps it
launchable that way while dropping it from the home-screen app drawer.
Also: targeting SDK 35 enforces edge-to-edge by default, and the screen
never handled system-bar insets, so content drew straight under the status
bar. fitsSystemWindows on the content root fixes the overlap.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
A static versionName only matches a release tag if someone remembers to bump
it first — nothing enforces that, so they drift. versionName now comes from
`git describe --tags` (the tag itself when built exactly on a tagged commit,
otherwise the usual git-describe dev form), and versionCode from commit
count. Cutting a release is now just tagging a commit; there's nothing left
to hand-edit.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The target app crashed on launch after the root/manager-app and Settings
hooks landed. The actual bug: the pass-through branch of both
hookSettingsClass and hookPackageManager used XC_MethodReplacement and
manually called XposedBridge.invokeOriginalMethod() when a call wasn't one
we wanted to intercept. That routes through Method.invoke(), which wraps
*any* exception the real method throws in InvocationTargetException — including
completely normal ones like NameNotFoundException for a not-installed
package, or SettingNotFoundException for an unrecognized key, unrelated to
anything we hide. The target app's own root-detection code checks a list of
known manager packages and expects to catch NameNotFoundException directly;
the wrapped exception type broke that on the very first miss.
Fixed by switching both to XC_MethodHook.beforeHookedMethod, setting
param.result/param.throwable only for the keys/packages we actually want to
affect and leaving param untouched otherwise. That lets the real method run
through its normal, non-reflective path for everything else, so its
exceptions propagate correctly.
Confirmed on-device: the target app launches and stays in the foreground
with all six features enabled.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
- Real release build type: R8 shrinking + shrunk resources, with a keep
rule for VeilHook specifically — LSPosed finds it by exact class name via
the plaintext xposed_init asset, which R8 can't see, so obfuscating it
would silently break module loading with no build error to point at why.
- Signing config that reads keystore.properties if present (gitignored,
generated via scripts/generate-keystore.sh), falling back to debug
signing otherwise so assembleRelease always produces something
installable.
- scripts/release.sh builds and copies a version-stamped APK into dist/.
- GitHub Actions workflow building and publishing a release on any tag
matching *.*.* (bare semver, e.g. 1.2.0 — no v prefix required), optionally
using repo secrets for real release signing in CI.
- README covering setup, the feature table, and the DenyList gotcha.
No native code in this module, so a single APK already covers every
architecture — no per-ABI splitting needed.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
If an app is genuinely scoped in LSPosed, Veil already picks it up on its
own. If it isn't scoped, typing its package name here wouldn't do anything
anyway — the hook never runs in that process. The manual entry field was
redundant either way.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The screen was built entirely from raw android.widget.* views with no
theme applied at all, so it never had a chance to follow system dark/light
mode and looked dated regardless. Adds AppCompat + Material Components,
switches SettingsActivity to AppCompatActivity with a Theme.Material3.
DayNight theme, and rebuilds the screen with Material widgets (cards,
chips for the hidden-packages list, proper text fields).
Also adds a launcher icon (previously none): a vector adaptive icon, a
moon-like disc partially eclipsed by a diagonal veil in the background
color — no image tool involved, just XML.
Also renames the per-app "Remove" button to "Apply defaults (hide
everything)" — "Remove" read as removing the app from the list, not
resetting it to defaults, which is all it actually does.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Manually re-typing a package name Veil already got injected into was
redundant work. The hook now reports "I got loaded into package X" to a
small ContentProvider in Veil's own app the moment a scoped app's process
starts (found via hooking Application#attach for a reliable early Context),
which records it to SharedPreferences. Veil's settings screen lists the
union of explicitly configured apps and auto-detected ones, tagging the
latter so it's clear they haven't been customized yet.
No persistent service: the OS starts Veil's process on demand for each
call() and can kill it again right after, same footprint as everything
else in this module.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Every scoped app's cold start pays whatever cost the hook installation
itself takes, so it's the one place worth trimming here — there's no
standing service or background work anywhere else in the module.
Previously any enabled feature caused both Settings.Secure and
Settings.Global to get hooked with the full combined key set, even though
each key only ever lives on one of them (accessibility/mock-location keys
are Secure-only, adb/dev-options keys are Global-only). Splitting the key
maps by actual canonical class means an app that only needs e.g. ADB+dev-
options hidden skips hooking Settings.Secure entirely, roughly halving the
reflective hook-install calls for that case.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Collapsible panel on the settings screen covering what came up debugging
this against a real target app: the KernelSU "Umount modules" / DenyList
conflict that silently blocks Zygisk injection, checking a separate Zygisk
implementation's own DenyList, needing a force-stop + relaunch (not a full
reboot) after config changes, and using LSPosed's verbose logging to confirm
injection actually happened.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Checkbox labels like "Accessibility services" read as status, not action —
unclear whether checking it meant "hide this" or something else. Replaces
the flat label map with a label + one-line description per feature, and
adds a short caption on each app card spelling out what checking a box
actually does.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Renamed since the module hides more than accessibility state now — hook class
and app display name become VeilHook/Veil (the module already lived under the
space.bdeshi.veil namespace).
Two new features, same per-app toggle model as everything else:
- Root/manager apps: hides packages in a configurable shared list (default
KernelSU + Magisk) from PackageManager's getPackageInfo/getApplicationInfo/
getInstalledPackages/getInstalledApplications.
- Build tags/type: overwrites Build.TAGS/Build.TYPE to release-keys/user for
the hooked process, hiding the test-keys/userdebug markers apps sometimes
check for.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Replaces "hide everything for every scoped app" with per-package, per-feature
toggles, backed by SharedPreferences the hook reads via XSharedPreferences.
A package with no explicit entry still gets every feature hidden, keeping
prior behavior as the default.
Adds a launcher activity so the settings screen is directly reachable.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
LSPosed module that hooks AccessibilityManager and Settings.Secure/Global
(getString/getStringForUser/getInt/getIntForUser) inside whatever app
process LSPosed injects it into, so that process sees no accessibility
services and no adb/wifi-debugging flags enabled. Target apps are chosen
via the module's scope list in LSPosed Manager, not in-app config.
Confirmed working against a real target app on-device; it previously
refused to launch with an accessibility service active and now starts
normally with the module enabled and scoped to it. Required disabling
KernelSU's "umount modules" option for that app, since it otherwise blocks
Zygisk (and therefore LSPosed) from injecting into the process at all.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude Sonnet 5