fix: use NSApplication.shared for proper WindowServer connection

[Korkyzer]

Improve OCR text extraction quality (dark UIs, resolution, extraction logic)

Problem

The current OCR pipeline misses most text on dark-themed applications (WhatsApp, Slack, Discord, etc.). On a WhatsApp conversation with dozens of visible messages, agent-watch only captured ~80 characters (menu bar text: "File Edit Chat Call View Window Help").

Root causes identified

1. NativeTextExtractor short-circuits OCR: The accessibility extractor runs first. If it returns ≥ minimumAccessibilityChars (even just menu bar items), OCR is skipped entirely. For WhatsApp, accessibility returns ~100 chars of sidebar/menu text, satisfying the minimum — so the Vision framework OCR never runs on the actual message content.

2. Frame buffer resolution too low: FrameBufferStore downscales captures to maxDimension = 1280, which halves Retina resolution (2560 → 1280). Text becomes too small for reliable OCR, especially in dense UIs.

3. Apple Vision framework struggles with dark themes: VNRecognizeTextRequest performs poorly on light-on-dark text. The Vision framework was designed primarily for document scanning (dark text on light backgrounds).

Changes

1. NativeTextExtractor.swift — Always run both extractors, keep the best

Before: Accessibility runs first; if it returns enough chars, OCR is skipped.
After: Both accessibility AND OCR always run. The result with more text wins.

// Before
if let accessibilityText = accessibilityExtractor.extractText(),
   accessibilityText.count >= minimumAccessibilityChars {
    return ExtractedText(text: accessibilityText, source: .accessibility, metadata: metadata)
}
// OCR only runs as fallback

// After
let accessibilityText = accessibilityExtractor.extractText()
var ocrText: String? = nil
if ocrEnabled {
    ocrText = try ocrExtractor.extractText()
}
// Return whichever extracted more text
if ocrLen > accLen { return ocr } else { return accessibility }

2. FrameBufferStore.swift — Increase resolution to full Retina

// Before
maxDimension: Int = 1280

// After
maxDimension: Int = 2560

Disk impact: frames go from ~250KB to ~400-800KB. With the existing retention/pruning policy this remains well under control.

3. OCRTextExtractor.swift — Color inversion for dark themes

Runs OCR twice: once on the original image, once on a color-inverted version (using CoreImage CIColorInvert). Keeps whichever result contains more text. Also lowered minimumTextHeight from 0.005 to 0.002 to catch smaller text.

Results

Metric	Before	After
WhatsApp text captured	~80 chars (menu bar only)	1567 chars (all messages, contacts, timestamps, links)
Frame resolution	1280×831	2560×1662
text_source for WhatsApp	accessibility (short-circuited)	ocr (full Vision + inversion)

Environment

• macOS 26.2 (Tahoe)
• MacBook Pro M-series (Retina display)
• WhatsApp desktop, Slack, Discord (dark theme)

4. LaunchAgent — Fix TCC accessibility permission persistence

Problem: accessibilityGranted always showed false after reboot, even with AgentWatch toggled ON in System Settings.

Root cause: The LaunchAgent ran a bash script (launch) as intermediary. macOS TCC grants permissions per-process based on code signing identity. The launch bash script had its own identity (launch-3c34cc36...), so when the user granted accessibility to "AgentWatch.app", the TCC entry didn't match the actual process calling AXIsProcessTrusted().

Fix: Split into two separate LaunchAgents that launch the agent-watch binary directly:

• com.differentai.agentwatch.plist → agent-watch daemon
• com.differentai.agentwatch.serve.plist → agent-watch serve --host 127.0.0.1 --port 41733

Both have RunAtLoad: true. The TCC grant now correctly matches the process identity and persists across reboots.

Results

Metric	Before	After
WhatsApp text captured	~80 chars (menu bar only)	1567 chars (all messages, contacts, timestamps, links)
Frame resolution	1280×831	2560×1662
text_source for WhatsApp	accessibility (short-circuited)	ocr (full Vision + inversion)
accessibilityGranted after reboot	false (TCC mismatch)	true (direct binary launch)

Environment

• macOS 26.2 (Tahoe)
• MacBook Pro M-series (Retina display)
• WhatsApp desktop, Slack, Discord (dark theme)

Related

• Issue fix: CGDisplayCreateImage returns desktop wallpaper instead of screen content #1: CGDisplayCreateImage returns wallpaper on macOS Sequoia+ (the prerequisite fix that made screen capture work at all)

[benjaminshafii]

Thanks. But why the app icon though? Is it necessary for this to be successfully registered?

[benjaminshafii]

lemme know @Korkyzer

[Korkyzer]

@benjaminshafii Honestly it's not strictly necessary, just added it to make the bundle feel more complete/clean. Happy to remove it if you'd prefer to keep the PR focused on the fix only.

[Korkyzer]

TCC permissions breaking after codesign (fix)

If accessibilityGranted or screenRecordingGranted show false in /status even though they're toggled ON in System Settings, the issue is the launch bash script in the LaunchAgent.

Root cause: The LaunchAgent runs Contents/MacOS/launch (a bash script) which spawns agent-watch serve and execs agent-watch daemon. macOS TCC can't properly associate processes spawned through a bash intermediary with the app bundle's code signature — so grants don't match.

Fix: Split into two LaunchAgents that launch agent-watch directly:

<!-- com.differentai.agentwatch.plist -->
<key>ProgramArguments</key>
<array>
    <string>/path/to/AgentWatch.app/Contents/MacOS/agent-watch</string>
    <string>daemon</string>
</array>

<!-- com.differentai.agentwatch.serve.plist -->
<key>ProgramArguments</key>
<array>
    <string>/path/to/AgentWatch.app/Contents/MacOS/agent-watch</string>
    <string>serve</string>
    <string>--host</string>
    <string>127.0.0.1</string>
    <string>--port</string>
    <string>41733</string>
</array>

After any codesign --force, you need to:

1. tccutil reset Accessibility com.differentai.agentwatch.app
2. tccutil reset ScreenCapture com.differentai.agentwatch.app
3. Re-add ~/Applications/AgentWatch.app in both Accessibility and Screen Recording settings (Cmd+Shift+G → ~/Applications)
4. Restart both LaunchAgents

Also pushed OCR improvements to my fork (fix/fullscreen-capture) — captures went from ~18-78 chars (menu bars only) to 1000-2600 chars (full page content).