- Behavior change:
resolveHubAppReference, shipped throughagent-device/plugins/webdriver, now needscwdto resolve a relative app path. Without it the call fails withINVALID_ARGSanddetails.reason: provider_app_cwd_missing; it used to resolve against the process's working directory. Absolute paths, URLs and provider references are unaffected. The daemon fills it from the client's working directory (or its host's) for every provider lease request, soconnectandlease_allocateare unchanged. - Fix
ensureDaemon(), the CLI and MCP killing a daemon they had just asked to leave. After the daemon answers the quiesce request withdraining, the client now waits up to 60 s for it to exit by itself and clean up its registration, so plugin disposal and lease finalization finish (on Windows a signal ends the process outright, and on POSIX a drain longer than 3 s was killed). A daemon still running at the deadline that fails its liveness probes is force-retired; one that still answers is left running and the call fails withdetails.reason: 'daemon-drain-timeout', with a hint namingagent-device daemon stop --state-dir <path>. @agent-device/testmuis now a version 2 provider plugin.import testmu from '@agent-device/testmu'gives a factory,testmu({ credentials: { username, accessKey } }), that reads those credentials instead ofLT_USERNAMEandLT_ACCESS_KEY; without options it reads the variables as before. The package still declares the version 1 plugin entry, soagent-device plugins add @agent-device/testmukeeps working on hosts that predate the version 2 contract.- Provider plugins have a new contract. A package exports a factory at
./agent-devicethat returns{ apiVersion: 2, provider, credentials, connection?, setup(host) }, typed byAgentDevicePluginFactoryfromagent-device/plugins. The store calls the factory with the package'sconfig.jsonoptions, and a JavaScript host can call it directly. Version 1 plugins (agentDevicePlugin.apiVersion: 1) keep loading. - Bundled providers and installed plugins now register through one registry, in that order, with a third slot for embedder plugins (#3757). The first plugin to name a provider id claims it; a later duplicate is recorded as failed, and
cloudandproxystay reserved. - A plugin that fails to load, validate, or set up no longer stops the daemon. It is listed as
failedwith a typederror.reason, and the other providers still register. The newagent-device daemon status [--json]shows the inventory, which the daemon also publishes indaemon.json;plugins list --jsonworks without a daemon and reports a running daemon's inventory underdaemon. - The client reads provider credential variable names from that inventory, narrowed per lease backend, and falls back to version 1 manifests. A Limrun credential fingerprint now hashes the exact variable values, so padding whitespace around a value counts as a change, as it does for BrowserStack.
- A partial Limrun instance set now appears in
daemon statusas a failed provider and asprovider_plugin_failedindaemon.log, replacingprovider_runtime_skipped. - Behavior change: the CLI and MCP now call
ensureDaemon()explicitly once their config is resolved, then attach to the daemon it leaves. A command that finds a daemon from another version or build replaces it only if the daemon reports no request, session, repair session or live lease. A resumable upload still streaming and a host human-control hold that names no lease are not counted, so a replacement can end them. A daemon that answers nothing on any transport is retired after repeated failed liveness probes, which cannot show it idle. A busy one fails the command withdetails.reason: 'daemon-busy'instead of being replaced with its sessions, and a daemon from a release that predates quiescing fails it with'daemon-incompatible'; stop either withagent-device daemon stop --state-dir <path>. The takeover notice moves fromReplacing daemon … : <reason>toReplaced idle daemon …(an incompatible daemon) orReplaced unreachable daemon …(one that answered nothing). A request that times out no longer retires an unresponsive daemon from the CLI or MCP; the error namesagent-device daemon stop --state-dir <path>. A one-shotreplayortestwith an explicit--state-dirleaves the daemon it started running until its idle reap instead of stopping it. Remote--daemon-base-urlconnections and replay on its private ephemeral daemon are unchanged. Node.js clients keepdaemonLifecycle: 'managed'until 0.22. - Add
ensureDaemon({ stateDir })and the client configdaemonLifecycle: 'attach' | 'managed'. Anattachclient connects to the daemon registered in the state directory and never starts, replaces or kills one, including after a request times out. It fails withdetails.reason: 'daemon-not-running'(the hint points atensureDaemon()) or'daemon-incompatible'(both identities indetails, the hint namesagent-device daemon stop).ensureDaemon()starts a missing daemon and replaces an incompatible one that answers only through an authenticated quiesce request the daemon answers itself (a resumable upload still streaming and a human-control hold that names no lease are not counted, so a replacement can end them): it refuses withdaemon-busywhile any request is in flight or any session, repair session or live lease exists, and a daemon from a release before this one is never replaced. A registered daemon that answers nothing is retired, as is one that times out on the quiesce request and then fails repeated liveness probes; neither can be asked. A remotedaemonBaseUrlconnection never starts or retires a local daemon in either mode.managedstays the default and behaves as before until 0.22. lease_allocate(andclient.leases.allocate()) with aleaseProviderthat no registered runtime claims now refuses before recording a lease, withUNSUPPORTED_OPERATION,error.details.reason: provider-not-registered,details.provider,details.registeredProviders, anddetails.failedPluginsnaming a plugin that failed to register. Previously a plugin provider that the daemon never loaded passed admission and recorded a lease with no device behind it. Leases with no provider (aleaseBackendalone included),proxyandcloudneed no runtime and still allocate. Restart the daemon with the provider configured, then retry.- Add
uninstall <app>andclient.apps.uninstall({ app })to remove an installed app on adb-backed Android targets, iOS simulators, and CoreDevice iOS devices. Both reportwasInstalled. A package name or bundle id that is not installed succeeds withwasInstalled: false. Other targets, Appium and Limrun sessions included, refuse it before anything runs. - Add
settings mobile-data on|offfor adb-backed Android targets. It switches mobile data throughsvc dataand reports themobileDatastate stored in the globalmobile_datarow; multi-SIM read-back is not covered yet. Every other target refuses it before anything runs, and--appis refused because the switch is device-wide. - On iOS, a
fillthat runs under a degraded XCTest snapshot channel now aims at a point, including ref, selector and replay fills. When the runner identifies no text input under that point, it refuses withNO_TEXT_INPUT_AT_POINT(error.details.reason: no_text_input_at_point,dispatched: no) before dispatching its focus tap, and no field changes; when the input lookup itself cannot answer, it refuses withTEXT_INPUT_PROBE_UNAVAILABLE(error.details.reason: text_input_probe_unavailable). Previously such a fill cleared and typed through the app-level first responder, so it could rewrite a field the caller never named, or report a clear failure after emptying it. A ref or selector fill whose rect centre misses the real field (an outer container, a non-input node) now needs a point inside the input's own rect. - Opening a macOS app by bundle identifier now waits briefly when the app is already running but not yet registered with LaunchServices — the case when a supervisor or dev script started it directly — and if registration arrives within that bounded window the open attaches to the running copy, so the session drives the app the caller started, with its arguments. A copy still unlisted when the window closes gets the plain
openas before, which can still duplicate; the wait only bounds the race, it does not close it. Launching an app that is not running costs one extra LaunchServices lookup and one process-table read, plus one bundle-identifier read when app-shaped processes exist. - Limrun iOS snapshots no longer fail when the instance reports a numeric or boolean accessibility value (the system photo picker's tabs report
1/0): the provider seam normalizes variant scalars to text, the way the Apple tree does. Limrun nodes now also publishselected: truefor nodes carrying theSelectedtrait, so snapshot output andselected:selectors can tell a picked photo from an unpicked one. settingsrefuses a setting the target does not serve before anything changes, withUNSUPPORTED_OPERATION,settings <setting> is not supported on this device, and the target'serror.details.reason(such asunsupported-platform-leaf). This replaces the macOSINVALID_ARGSrefusal and the Applesetting-unsupported-on-leafreason, and now also covers settings that used to reach the device first, such as Androidreset-keychainor Appleanimations.settingsrefuses, before touching the device, a state its setting does not accept (Invalid <setting> state: …), a permission action or mode its target does not take (such aspermission deny accessibility), and a stray extra argument (such asclear-app-state <app> <extra>). The CLI and the daemon accept the same state spellings, in any casing or surrounding whitespace.alert wait 0on Android reports an alert that is already visible instead of timing out.settings --helpdescribes each setting on its own line, and its usage folds settings that share a form.- Added (android):
telephony call|accept|cancel <number>andtelephony sms <number> <text>, plusclient.command.telephony({ action, number, text }), simulate an incoming phone call or SMS on an Android emulator through the emulator console (adb emu gsm …,adb emu sms send …), so a test can check how an app behaves when a call or message interrupts it, such as video playback pausing and resuming. The number must be up to 15 digits with an optional leading+; SMS text is sent as one adb argument, with backslashes and line breaks escaped for the console; the emulator drops characters outside the Basic Multilingual Plane, such as emoji. A console refusal (for exampleacceptfor a number with no call) fails withCOMMAND_FAILEDandreason: telephony-console-refused. Physical Android devices, iOS, every other target, and the WebDriver (Appium) and Limrun providers refuse withUNSUPPORTED_OPERATION. foldon Android emulators: the posture and hinge-angle console calls now time out after 10 s instead of waiting on a console that stopped answering, and a failed call names only its first two words (for exampleadb emu sensor get).- Carry provider keyboard facts through snapshots. Appium captures, including wait polls and fill pre-captures, add one sequential keyboard visibility probe with a 500 ms timeout. Probe results preserve tap-time docked-tree checks; unavailable probes leave visibility unknown.