Follow-up to #944: the fixes made after the review that did not make the squash. ## What changes - **Signing note**: delivered only to a call whose device argument resolves to a physical iPhone, including `flow-execute`'s `device`, and held (not dropped) across other calls. The label is now `Apple Development (KEYID)`; the certificate CN's Apple ID no longer reaches agent context or logs. - **Flow device resolution**: the no-device-matched refusal names the cabled phone and the exact `device <udid>` / `--device` spelling. A physical iPhone is never picked automatically. - **`gesture-swipe`**: the description states the hardware semantics (edge gesture needs `fromX` 0, `durationMs` sets drag speed, `momentum:false` does not damp). - **Skills**: `argent-ios-simulator-setup` skips `kind: "device"` entries; `argent-ios-device-setup` and `-interact` state the runner build timing, the full list of tools that do not exist on hardware, the backgrounded-target behaviour of each read tool, `actionButton` only on models that have one, and the swipe and `open-url` rules; the flow, Metro and profiler skills each carry one line on physical-iPhone limits. - **Docs**: `physical-ios-devices`, `flows`, `flow-yaml` and `tools` corrected to match the behaviour above; README platform row; runner `PROTOCOL.md`. ## Verification - tool-server suite 400 files / 5173 tests, `tsc`, lint, prettier and the docs build clean. - Live re-check against 0.23.0 on three simulators and a physical iPhone 15: no regression; the note lands once on the first phone call with no e-mail address, and flow auto-detection is byte-identical to 0.23.0 with 0, 1 and 3 simulators booted.
@argent/ios-device-runner
Argent's on-device automation runner for physical iOS devices: an XCUITest
bundle that, instead of running a scripted test, starts an HTTP command server
and parks in an XCTWaiter wait for 24 hours. XCUITest is the only
Apple-supported process allowed to drive arbitrary apps on real hardware, so
Argent hosts its automation server inside one.
Layout
ArgentRunner/
ArgentRunner.xcodeproj Two targets: ArgentRunner (placeholder host app)
and ArgentRunnerUITests (the runner itself).
ArgentRunner/ Host app: a single black "Argent Runner" screen.
ArgentRunnerUITests/
ArgentRunnerSession.swift Entry test + dispatch, coalescing, busy gate
ArgentRunnerSession+Commands.swift Target resolution, exception guard,
recorded-failure detection,
button / shutdown
ArgentRunnerSession+Gestures.swift tap / longPress / drag / viewport
ArgentRunnerSession+TextEntry.swift type / keyboardReturn / keyboardDelete
ArgentRunnerSession+Snapshot.swift one-shot AX tree capture + flattening
ArgentRunnerSession+Screenshot.swift inline base64 PNG
RunnerProtocol.swift wire models (see PROTOCOL.md)
RunnerHTTPServer.swift NWListener HTTP/1.1 framing
CommandJournal.swift send-once bookkeeping for lost replies
MainThreadGate.swift watchdog + busy/wedged reporting
ArgentExceptionGuard.{h,m} @try/@catch shim for XCTest NSExceptions
The TypeScript side (build orchestration, transport, command client) lives in
packages/tool-server/src/utils/ios-device/.
How it runs
- The tool-server builds this project lazily with
xcodebuild build-for-testing(signed with the user's team; see the environment variables below) into~/.argent/ios-device-runner/derived, stamped with a fingerprint of sources + Xcode + signing; a stamp mismatch (an Argent or Xcode update, a team change) rebuilds in place. - Per session it launches
xcodebuild test-without-buildingdetached; testmanagerd installs and starts the runner on the device. The runner binds a system-assigned loopback port and logs it asARGENT_RUNNER_LISTENING port=N; the tool-server reads that line from the launch log before it sends the first command. - Commands travel as one HTTP POST per command over usbmux (USB cable
only).
PROTOCOL.mddocuments the contract.
Build-time configuration
The signing team resolves in two steps, and the environment variable wins:
ARGENT_IOS_TEAM_ID(optional): your Apple Developer Team ID, a 10-character code. Find it in Xcode > Settings > Accounts (select your Apple ID and team) or at developer.apple.com/account under Membership.- When it is not set, the tool-server detects the team from the Mac's keychain: the Apple Development certificates name their team (and the legacy iPhone Developer name is checked too), and with several teams the one with the newest certificate wins. The detection result is memoized for the tool-server process. With no certificate at all, the build fails with a prompt to sign into Xcode and mint one (Manage Certificates > + > Apple Development).
Everything else is derived or fixed. The build always uses automatic signing
(an Apple ID signed into Xcode is needed so it can mint the provisioning
profile), and the bundle ids are derived from the team
(com.argent.runner.t<teamid> plus its .uitests sibling), which makes them
unique per team by construction. The project intentionally hardcodes no team
and only placeholder bundle ids; the tool-server injects the real values on
the xcodebuild command line. The device needs Developer Mode enabled
(Settings > Privacy & Security > Developer Mode), and on first install iOS
asks to trust the developer on the phone: Settings > General > VPN & Device
Management.
The Xcode project itself needs no locating either: it always sits next to the
bundled tool-server. An npm install of @swmansion/argent ships these Swift
sources there (the build step copies ArgentRunner/ next to the bundle), and
in the argent repo npm run build in packages/argent produces the same
layout.
ARGENT_IOS_RUNNER_PROJECT overrides the location for tool-server runs
outside the bundle (ts-node, tests) or unusual layouts.
Reliability model
- Issue suppression:
record(_:)mutes two benign "Failed to get matching snapshot" accessibility issue shapes that would otherwise tear the long-lived test down; hung-query timeouts still record. - Watchdog: command work runs on the main thread under a hard budget;
overruns are abandoned (XCTest work cannot be cancelled) and the runner
answers
RUNNER_BUSY, escalating toRUNNER_WEDGEDwhen stuck. - Send-once: commands are journaled by
commandId(responses are retained for mutating commands only); a client that lost a reply asksstatusfor the command's fate instead of replaying it. Duplicate in-flight sends coalesce onto the single execution. - Explicit targeting: app commands require
appBundleId; the runner never redirects them to its own host app. - Observation without side effects: the read-only
snapshotnever re-fronts a backgrounded target; it answersAPP_BACKGROUNDEDinstead. Mutating commands, and theviewportread that precedes a gesture, re-front a backgrounded target and stamp the replyreactivated: true.