Skip to content
TakoAPI
Back to skills

screen-vision

by ls18166407597-design

AI & LLMs
—Not yet rated
Your rating

Brief

macOS Local OCR & Automation Tool using Vision Framework.

Description

What This Skill Does

A macOS tool that runs OCR on screen content using Apple's built-in Vision framework. It extracts text and [X, Y] coordinates locally, sending only the recognized text to the AI rather than full screenshots. Supports mixed Chinese and English recognition.

Running OCR locally via the Vision framework avoids sending screenshots to an AI API, which eliminates token costs and keeps screen content private.

When to Use It

  • Reading text from a non-standard desktop app that has no API
  • Monitoring a balance or progress indicator on screen without manual checking
  • Clicking a button in a GUI app by locating its label text
  • Automating repetitive UI interactions across any macOS application
  • Extracting text from a screen region that can't be copy-pasted

Example Workflow

INPUT

User asks: help me click the Confirm button in this app

AGENT
  1. 1

    Runs vision_ocr.swift to scan the current screen

  2. 2

    Receives recognized text with [X, Y] coordinates for each element

  3. 3

    Locates the text matching 'Confirm' and retrieves its coordinates

  4. 4

    Passes coordinates to click.swift to perform a physical mouse click

OUTPUT

The Confirm button is clicked at the detected screen coordinates

Requirements

  • macOS Screen Recording permission enabled for your terminal app (System Settings > Privacy & Security > Screen Recording)
  • macOS Accessibility permission enabled for your terminal app (System Settings > Privacy & Security > Accessibility)

Installation

clawhub install ls18166407597-design/screen-vision

Run the CLI command, or paste the prompt into Claude Code, Cursor, Codex or whichever coding agent you use.

Links