Goldie: App Store and Google Play screenshot automation for coding agents

The gap Goldie fills between simulator runs and store listings

Store listings need a fixed set of images: iPhone screenshots at 1320 x 2868, Google Play phone screenshots at 1080 x 1920, and an Apple preview video between 15 and 30 seconds. Producing those by hand means re-running the same app states, cropping, adding bezels and headlines, and then re-checking dimensions after every design change. Goldie takes the opposite route. It assumes the app states are already reproducible because argent replays them on an iOS simulator or an Android emulator, and it treats framing, copy and verification as a rendering step over those captures.

The audience is narrow but real. Goldie is framework agnostic: the README states it works the same for SwiftUI, UIKit, Jetpack Compose, Flutter, React Native and Kotlin Multiplatform, because it drives the app through the simulator or emulator rather than through the app's own test hooks. If your team already writes argent flows, Goldie is an extension of that work. If you have never automated a UI run, Goldie is not a starting point; it is a finishing step.

How the capture, frame, render and verify pipeline is arranged

The data flow has four visible stages. First, argent replays flows and produces raw captures. Second, Goldie frames each capture with a device bezel, a background and a headline. Third, for previews it joins clips into a video. Fourth, it checks the result against the upload rules of the stores. The CLI exposes these as separate commands, which matters when a run fails: you can re-run framing without re-capturing, or re-render a preview without touching the screenshots.

Configuration lives in goldie.config.ts, and the repository ships goldie.config.example.ts to copy. Scenes point at argent flows stored in .argent/flows. The design layer is separate: the studio writes choices to goldie.design.json, and the README states the CLI renders the same result from that file, so a browser session and a CI run converge on the same output. Output paths follow a device and locale pattern: out/screenshots/<device>/<locale>/ and out/previews/<device>/<locale>/.

The templating surface is wider than the four-line summary suggests. theme.template accepts editorial, showcase, magazine, storyboard or dynamic, or a custom layout sequence drawn from classic, copy-below, hero, offset, tilt, tilt-right, duo, duo-tilt, panorama, panorama-duo and minimal. frame accepts 17-pro-blue, 17-pro-silver, 17-pro-orange or a custom bezel image, and theme.screenOnly: true removes the bezel entirely. Six fonts ship bundled, including Noto Sans SC for Simplified Chinese, which implies locale-aware copy is an expected use rather than an afterthought.

Installing Goldie and producing a first screenshot set

The prerequisites are Node 20 or newer and ffmpeg on the PATH. App Store screenshots require macOS with iOS simulators and Xcode. Google Play screenshots require an Android emulator, which the README says works on macOS, Linux and Windows; on Linux, argent needs extra packages for Android emulators. Install the CLI globally with npm.

npm i -g goldie

If you want a coding agent to drive the setup instead, the README gives a skills command for Cursor, Codex and any agent that supports the skills format, plus a plugin marketplace pair for Claude Code.

npx skills add kacperkapusciak/goldie

By hand, copy the example config to goldie.config.ts, point its scenes at argent flows in .argent/flows, then run the doctor command. The README states that goldie doctor checks tools, simulators and flows, and that it warns when an iOS device is configured on a host that cannot run it, which is the failure you hit on Linux or Windows if you leave an iPhone key in the devices list. After that, goldie all captures, frames, renders the preview and verifies in one pass, and goldie studio opens the browser preview for tweaking. Expect the iPhone output at 1320 x 2868 with an 886 x 1920 H.264 preview, and Google Play phone screenshots at 1080 x 1920.

For Android, the pixel-10-pro device key renders Play phone screenshots from the same scenes, and the example config already lists it. The captures come from an emulator whose AVD must use the Pixel 10 Pro or Pixel 9 Pro hardware profile, both of which have a 1280 x 2856 screen. The README says Goldie reuses a running emulator that matches or boots the AVD itself. The config then needs the build path and application id.

devices: ["iphone-6.9", "pixel-10-pro"],
android: {
  appPath: "/path/to/app-release.apk",
  applicationId: "com.example.app",
},

Where Goldie breaks: flow drift, debug builds and the Play promo video

The README is direct about the main failure mode: flows fail when the app changes. That is inherent to selector-driven replay, and the suggested repair is to ask the coding agent to fix them or re-record with argent. There is no documented rollback or versioning for flows, so a rename in your UI can invalidate a screenshot pipeline that previously worked. Budget for that maintenance rather than treating the first green run as permanent.

The second trap is the build type. The README warns to use a Release build because debug builds paint LogBox banners into the captures. This is easy to miss locally and expensive to catch after a submission.

The third is platform asymmetry. Goldie frames Android tiles with the bundled Pixel 10 Pro bezel, but the frame variant is iPhone art and does not apply to them; android.frame swaps in your own art and its geometry. And the Play Store promo video is a YouTube link, so preview and all record the preview scene on the emulator and render a portrait video for you to post on YouTube yourself. Apple's 15 to 30 second window does not apply to it. If your workflow assumes Goldie uploads the Play video, it does not; it hands you a file.

Goldie against scripted simulator screenshot tooling

The closest comparison is a hand-written fastlane snapshot setup or a custom XCUITest script that captures screenshots and lets you compose frames in an image editor. The difference is where the state comes from. A snapshot script runs inside the app's own test target, so it is tied to that platform's test framework. Goldie drives the app through the simulator or emulator via argent, which is why the README can claim the same pipeline for SwiftUI, UIKit, Jetpack Compose, Flutter, React Native and Kotlin Multiplatform. One flow set can serve both stores when its selectors match.

What you give up is the maturity of those older tools. Goldie is at version 0.3.1 with no releases retrieved, so the API surface in goldie.config.ts and goldie.design.json should be treated as movable. A fastlane setup has years of CI recipes behind it; Goldie's CI story is not documented in the README beyond the fact that the CLI renders from the same design file the studio writes. That is a reasonable foundation, not a proven one.

Maintenance cost, licence and what the repository actually declares

The last push to the default branch was on 2026-09-02, and the repository is not archived. That is recent, but it is a single data point and the README does not publish a support policy, a release cadence or a deprecation process. The practical upgrade cost sits in two places: the flow files in .argent/flows, which break when the app changes, and the design JSON, which the studio rewrites. Both are plain files in your repository, so a version bump that changes template names or frame keys would show up as a rendering difference rather than a crash.

Licensing has a discrepancy worth resolving before adoption. package.json declares "license": "MIT", while the repository metadata reports NOASSERTION. The LICENSE file exists at the repository root, so the answer is one file read away, but the two signals disagree and that is the kind of thing a legal review will ask about. Nothing here is legal advice; read the LICENSE file and the attribution file shipped in assets before you redistribute the bundled bezels and fonts.

Editorial conclusion

Adopt Goldie if your mobile app already has, or can get, reproducible argent flows and you want store screenshots and previews generated from those flows instead of assembled by hand. Skip it if you ship on iOS only from a Linux or Windows host with no Mac available, or if you have no appetite for maintaining flow selectors against a moving app. Before committing, run goldie doctor on the target machine, confirm that the bundled iphone-6.9 and pixel-10-pro device keys match hardware profiles you can actually boot, and check the LICENSE file in the repository root, since package.json declares MIT while the repository metadata reports NOASSERTION.

Frequently asked questions

Official sources

1. Issues
2. kacperkapusciak/goldie on GitHub
3. README