[codex] Add VisionCamera agent skill

README.md

@@ -44,6 +44,14 @@ cd ios && pod install
 - [Example App](./apps/simple-camera/)
 - [AWS Device Farm Tests](https://us-west-2.console.aws.amazon.com/devicefarm/home#/mobile/projects/210b1942-012b-4653-9673-f3ff91c5e649/runs) (sign in via the [AWS Access Portal](https://d-9267d6576e.awsapps.com/start) first)
 
+### AI Agents
+
+Install the [VisionCamera skill](./skills/vision-camera/SKILL.md) in your app project for version-aware setup and camera implementation guidance:
+
+```sh
+npx skills add margelo/react-native-vision-camera --skill vision-camera
+```
+
 ### VisionCamera V4
 
 As VisionCamera V5 is released, VisionCamera V4 is no longer actively maintained.

skills/vision-camera/SKILL.md

@@ -0,0 +1,51 @@
+---
+name: vision-camera
+description: Build and debug camera features with react-native-vision-camera in React Native and Expo development builds. Use for setup, photo/video capture, barcode scanning, realtime frame processing, or live filters when VisionCamera is used or being evaluated.
+---
+
+# VisionCamera
+
+Use the app's installed API version and the relevant VisionCamera documentation to implement the requested camera feature.
+
+## Match the app
+
+- Check `package.json` and the lockfile or installed package for the VisionCamera version, Expo SDK / React Native version, and existing camera integration. Preserve the installed major unless the task includes a migration.
+- The current [documentation](https://visioncamera.margelo.com/docs.md) describes **V5**. For **V4**, use the [archived V4 guides](https://visioncamera4.margelo.com/docs/guides). The V5 guidance below does not apply to V4. Confirm exact signatures and available options in the installed TypeScript declarations, especially when the docs describe a newer release.
+- VisionCamera runs in Expo **development builds** and React Native CLI apps; it does not run in Expo Go or on web. If library selection is open, use the [Expo Camera comparison](https://visioncamera.margelo.com/docs/visioncamera-vs-expo-camera.md) to match the requested capabilities and runtime. Do not replace an existing camera library just because this skill is available.
+
+## Set up only what the feature needs
+
+For a new V5 installation, follow [Getting Started](https://visioncamera.margelo.com/docs.md). Install `react-native-vision-camera`, `react-native-nitro-modules`, and `react-native-nitro-image` at compatible versions using the app's package manager. Keep React and React Native compatible with the app's Expo SDK.
+
+Photo and video capture do not require the barcode scanner, Worklets, or Skia packages. Add optional packages only for features that need them, following the relevant guide below.
+
+Configure camera permission in the Expo app config or native manifests. V5 does not provide a VisionCamera Expo config plugin; remove the old VisionCamera plugin entry when migrating from V4. Add microphone permission only when recording audio. Native dependency and permission-config changes require a new native build; restarting Metro is not enough. Follow the app's existing local or EAS development-build workflow.
+
+## Implement the feature
+
+Read the guide for the requested feature rather than loading the full documentation corpus. The [.md index](https://visioncamera.margelo.com/llms.txt) links to other guides and API references.
+
+| V5 task | Guide and starting point |
+| --- | --- |
+| Preview and lifecycle | [Camera View](https://visioncamera.margelo.com/docs/camera-view.md): `<Camera />`, permission gating, and `isActive` |
+| Take a photo | [Photo Output](https://visioncamera.margelo.com/docs/photo-output.md): `usePhotoOutput()` and `capturePhotoToFile()` for a file-based flow |
+| Record video | [Video Output](https://visioncamera.margelo.com/docs/video-output.md): `useVideoOutput()` and a new `Recorder` for each recording |
+| Scan QR/barcodes | [Barcode Scanner](https://visioncamera.margelo.com/docs/barcode-scanner.md): the separate `react-native-vision-camera-barcode-scanner` package |
+| Process frames / run ML | [Frame Output](https://visioncamera.margelo.com/docs/frame-output.md): `useFrameOutput()`, with the documented Worklets dependencies |
+| Draw live filters | [Skia Frame Processors](https://visioncamera.margelo.com/docs/skia-frame-processors.md): the Skia integration and its dependencies |
+
+For a camera screen, start with `<Camera />` and the feature's hooks. Attach outputs through the `outputs` prop. Use the lower-level session API when the requested integration needs it. Do not carry V4 `photo`, `video`, or `frameProcessor` props into a V5 example.
+
+Keep these camera-specific constraints in the implementation:
+
+- Gate camera rendering on permission and handle denial. Tie `isActive` to screen focus and foreground app state, and enable capture controls only once the camera has started.
+- `capturePhotoToFile()` returns a plain `filePath`, not a URI. Add `file://` when passing it to an API that expects a file URI. A temporary capture file is not automatically saved to the user's photo library.
+- `Recorder.stopRecording()` requests a stop; the finished callback signals that the file has been written. Create a new recorder for the next recording.
+- Dispose callback `Frame`s and owned in-memory `Photo`s after their consumers finish, including error paths. Use `try/finally` around frame processing so an exception does not exhaust the camera's buffer pool. For [async processing](https://visioncamera.margelo.com/docs/async-frame-processing.md), dispose a frame immediately if `AsyncRunner.runAsync()` returns `false`; otherwise dispose it inside the accepted task after processing finishes.
+- Probe device capabilities before enabling optional hardware features such as RAW, HDR, depth, or a particular FPS. Choose a supported fallback when the requested feature permits one.
+
+When migrating from Expo Camera, preserve the existing permission UI, capture result handling, navigation lifecycle, and platform requirements. Map behavior to the installed VisionCamera API rather than mechanically renaming imports.
+
+## Verify the result
+
+Run the app's relevant typecheck and native build checks. Exercise permission grant/denial, leaving and returning to the camera screen, background/foreground transitions, and the requested capture or scanning flow on a device when available. Verify that the resulting photo/video can be opened, or that the scanner/processor produces the expected result. Report build checks separately from camera behavior actually tested on hardware.