feat(snapshot): add configurable image capture limits

README.md

@@ -73,6 +73,21 @@ Or add it to your own client manually:
 }
 ```
 
+### macOS image capture
+
+`get_app_state` and action tools return a PNG screenshot together with the accessibility tree. On macOS, screenshot size can be tuned with optional environment variables. Set them before starting the `open-computer-use` runtime process; changing shell environment values afterward does not affect an already running process.
+
+| Variable | Default | Meaning |
+| --- | --- | --- |
+| `OPEN_COMPUTER_USE_IMAGE_CAPTURE_TIMEOUT` | `5` | Seconds to wait for ScreenCaptureKit before omitting the screenshot image content from the result. |
+| `OPEN_COMPUTER_USE_IMAGE_MAX_DIMENSION` | `1280` | Integer long-edge pixel cap for the returned PNG. |
+| `OPEN_COMPUTER_USE_IMAGE_MAX_BYTES` | `900000` | Best-effort byte budget for the encoded PNG. |
+| `OPEN_COMPUTER_USE_IMAGE_MIN_SCALE` | `0.25` | Byte-budget retry multiplier applied after the dimension cap. For example, a 500 px capped long edge with `0.25` may retry down to 125 px; it never enlarges the PNG chosen by the dimension cap. |
+
+Invalid, non-finite, or non-positive values fall back to the defaults. `OPEN_COMPUTER_USE_IMAGE_MIN_SCALE` values above `1` also fall back.
+
+Coordinate tools keep using the actual returned PNG dimensions, so downsampling does not change click or drag mapping.
+
 ### Skill
 
 Install the skill directly:

README.zh-CN.md

@@ -72,6 +72,21 @@ ocu install-codex-mcp
 }
 ```
 
+### macOS 截图尺寸
+
+`get_app_state` 和 action 类工具会和 accessibility tree 一起返回 PNG 截图。macOS 上可以通过可选环境变量调整截图尺寸。请在启动 `open-computer-use` runtime 进程前设置这些变量；之后再修改 shell 环境不会影响已经运行的进程。
+
+| 变量 | 默认值 | 作用 |
+| --- | --- | --- |
+| `OPEN_COMPUTER_USE_IMAGE_CAPTURE_TIMEOUT` | `5` | ScreenCaptureKit 等待秒数；超时会从结果里省略截图 image 内容。 |
+| `OPEN_COMPUTER_USE_IMAGE_MAX_DIMENSION` | `1280` | 返回 PNG 的整数长边像素上限。 |
+| `OPEN_COMPUTER_USE_IMAGE_MAX_BYTES` | `900000` | 编码后 PNG 的 best-effort 字节预算。 |
+| `OPEN_COMPUTER_USE_IMAGE_MIN_SCALE` | `0.25` | 作为长边上限之后的字节预算重试倍率。例如长边已限制到 500 px，`0.25` 允许继续重试缩小到 125 px；它不会放大长边上限已经选出的 PNG。 |
+
+非法值、非有限数或非正数会回退到默认值。`OPEN_COMPUTER_USE_IMAGE_MIN_SCALE` 大于 `1` 时也会回退。
+
+坐标类工具会读取实际返回 PNG 的尺寸做映射，因此降采样不会改变 click 或 drag 的坐标换算。
+
 ### Skill
 
 一键安装skill：

apps/OpenComputerUseFixture/Sources/OpenComputerUseFixture/main.swift

@@ -88,6 +88,7 @@ final class FixtureAppDelegate: NSObject, NSApplicationDelegate, NSTextFieldDele
     private let dragPadView = DragPadView(frame: NSRect(x: 0, y: 0, width: 320, height: 120))
     private var scrollView: NSScrollView!
     private var counter = 0
+    private var selectedText: String?
     private weak var observedScrollView: NSScrollView?
     private var commandObserver: NSObjectProtocol?
     private let headless: Bool
@@ -286,15 +287,18 @@ final class FixtureAppDelegate: NSObject, NSApplicationDelegate, NSTextFieldDele
         switch (command.kind, command.identifier) {
         case ("set_value", "fixture-input"):
             inputField.stringValue = command.value ?? ""
+            selectedText = nil
             updateExportedState()
         case ("click", "fixture-increment"):
             handleIncrement()
         case ("click", "fixture-input"):
             window.makeFirstResponder(inputField)
+            selectedText = nil
             updateExportedState()
         case ("click", "fixture-key-capture"):
             window.makeFirstResponder(keyCaptureView)
             keyCaptureView.needsDisplay = true
+            selectedText = nil
             updateExportedState()
         case ("scroll", "fixture-scroll-view"):
             let delta = CGFloat(120 * (command.pages ?? 1))
@@ -321,17 +325,83 @@ final class FixtureAppDelegate: NSObject, NSApplicationDelegate, NSTextFieldDele
         case ("type_text", "fixture-input"):
             inputField.stringValue += command.value ?? ""
             window.makeFirstResponder(inputField)
+            selectedText = nil
+            updateExportedState()
+        case ("select_text", "fixture-input"):
+            selectTextInFixtureInput(command)
             updateExportedState()
         case ("press_key", "fixture-key-capture"):
             keyLabel.stringValue = "Last key: \(command.value ?? "unknown")"
             window.makeFirstResponder(keyCaptureView)
             keyCaptureView.needsDisplay = true
+            selectedText = nil
             updateExportedState()
         default:
             break
         }
     }
 
+    private func selectTextInFixtureInput(_ command: FixtureCommand) {
+        let value = inputField.stringValue
+        guard let target = command.value,
+              let range = fixtureTextSelectionMatch(in: value, text: target, prefix: command.prefix, suffix: command.suffix)
+        else {
+            selectedText = nil
+            return
+        }
+
+        let mode = command.selection ?? "text"
+        let selectedRange = switch mode {
+        case "cursor_before":
+            NSRange(location: range.location, length: 0)
+        case "cursor_after":
+            NSRange(location: range.location + range.length, length: 0)
+        default:
+            range
+        }
+
+        window.makeFirstResponder(inputField)
+        if let editor = inputField.currentEditor() {
+            editor.selectedRange = selectedRange
+        }
+        selectedText = mode == "text" ? (value as NSString).substring(with: range) : nil
+    }
+
+    private func fixtureTextSelectionMatch(in value: String, text: String, prefix: String?, suffix: String?) -> NSRange? {
+        let haystack = value as NSString
+        let target = text as NSString
+        guard target.length > 0 else {
+            return nil
+        }
+
+        let prefix = prefix ?? ""
+        let suffix = suffix ?? ""
+        var searchRange = NSRange(location: 0, length: haystack.length)
+        var matchedRange: NSRange?
+
+        while searchRange.length >= target.length {
+            let found = haystack.range(of: text, options: [], range: searchRange)
+            if found.location == NSNotFound {
+                break
+            }
+
+            let afterStart = found.location + found.length
+            let before = haystack.substring(with: NSRange(location: 0, length: found.location))
+            let after = haystack.substring(with: NSRange(location: afterStart, length: haystack.length - afterStart))
+            if (prefix.isEmpty || before.hasSuffix(prefix)) && (suffix.isEmpty || after.hasPrefix(suffix)) {
+                if matchedRange != nil {
+                    return nil
+                }
+                matchedRange = found
+            }
+
+            let nextLocation = found.location + max(found.length, 1)
+            searchRange = NSRange(location: nextLocation, length: haystack.length - nextLocation)
+        }
+
+        return matchedRange
+    }
+
     private func updateExportedState() {
         guard let contentView = window.contentView else {
             return
@@ -341,6 +411,7 @@ final class FixtureAppDelegate: NSObject, NSApplicationDelegate, NSTextFieldDele
             windowTitle: window.title,
             windowBounds: FixtureRect(rect: windowBoundsInQuartzCoordinates()),
             focusedIdentifier: focusedIdentifier(),
+            selectedText: selectedText,
             elements: [
                 element(identifier: "fixture-window", index: 0, role: "standard window", title: window.title, value: nil, actions: ["Raise"], rect: CGRect(x: 0, y: 0, width: window.frame.width, height: window.frame.height)),
                 element(identifier: "fixture-increment", index: 1, role: "button", title: incrementButton.title, value: nil, actions: [], rect: localRect(for: incrementButton, in: contentView)),