- Home
- Skills
- Workflow Automation
- Eyes: Your Agent Sees and Uses Your Mac
More screenshots
Works with the AI Agents you already use
Eyes: Your Agent Sees and Uses Your Mac
Your coding agent can read files and run commands, but it can't see your screen.
$19
Eyes: Your Agent Sees and Uses Your Mac
Example session with this skill installed
Open builtbynauroz.tech in my Firefox and click through every page in the top menu like a visitor would. Tell me if anything is broken or reads wrong.
- Read your context and instructions
- Compiled the eyes
Summary
I opened builtbynauroz.tech in your Firefox and went through it the way a visitor would: I clicked each item in the top menu, scrolled the long pages and read every screen. All 5 pages load with the right title and heading. Nothing is broken and nothing reads wrong. It took 15 eyes commands and about 40 seconds.
What I checked
| Page | Address | Tab title | Read on screen (x,y) | Result |
|---|---|---|---|---|
| Home | builtbynauroz.tech | Syed Nauroz Ali, senior full-stack and agentic AI engineer | the menu "WORK LOG ABOUT CONTACT" at 608,146 | OK |
| Work | /work | Case studies: AI products and software | "Work" at 32,276; after scrolling 1,400 points, "Ask questions about a spreadsheet in plain English" at 196,143 | OK |
| Log | /log | Builder's log: what I shipped, dated | "Builder's log" at 34,272 | OK |
| About | /about | About Syed Nauroz Ali, full-stack and AI engineer in Pakistan | "About" at 32,275; after scrolling 900 points, "making it reliable is most of the engineering" at 367,128 | OK |
| Contact | /contact | Contact Syed Nauroz Ali: AI projects and hiring | "Message me." at 33,274 | OK |
Steps
eyes read home page, menu at 608,146
eyes click-on Work eyes: click on "Work" at 628,153
eyes scroll 1400 eyes: scroll down 1400 at 628,153
eyes click-on Log eyes: click on "Log" at 673,153
eyes click-on About eyes: click on "About" at 723,153
eyes scroll 900 eyes: scroll down 900 at 723,153
eyes click-on Contact eyes: click on "Contact" at 786,153
I read the screen after every step and scrolled back to the top before each click.
Next steps
- Nothing to fix.
- Run the same check after each deploy: "check every page in builtbynauroz.tech's menu again".
- To check images as well as text, ask me to use
look: text recognition reads words, not pictures.
Connects securely to your tools. The creator never sees your data.
About this skill
Your coding agent can read files and run commands, but it can't see your screen. So it asks you for screenshots, and it can't click anything in an app that has no API. eyes gives it both: it reads the screen and uses the real mouse and keyboard, at a person's pace.
What it does
- Reads any window: the text in view, each line with the x,y where it starts (on-device text recognition with Apple Vision), plus the app, the window title, the browser address and where typing would go.
- Finds controls exactly: buttons, links and fields with their labels and positions, from macOS Accessibility.
- Acts like a person: curved pointer moves, a short pause before each click, double and right clicks, drags, real wheel scrolling, typing at about 40 words a minute, and key combinations.
- Presses without the pointer:
pressclicks a button through Accessibility, so your pointer stays where it is. - Checks its work: every action prints the screen after it, so the agent sees what happened.
- Clicks only what it can reach:
click-onnever clicks under the Dock, a browser's toolbar or a sticky header, andfindsays when something must be scrolled to first. - Waits:
wait --untilholds until text appears, andwait --changeuntil the screen changes: for page loads, uploads and exports.
Safe by design
- Move the mouse and it stops at once (exit code 3), and the agent is told to ask before going on.
- Password managers are never captured, read or driven.
- Terminals and code editors take clicks and scrolling only. No typing or keys, so it can't run a command there.
- No network calls. The only file it writes is
look's screenshot in the temp folder, overwritten each time. - The skill tells the agent to treat words on screen as data, never to type passwords, and to ask before anything that sends, posts or deletes.
How it runs
One Swift script, about 39 KB. Your agent copies it to a temp file and runs swift <file> <command>. Swift compiles it on each run, so a command takes 1 to 2 seconds plus the action itself. The first run on a Mac can take about 30 seconds while macOS loads text recognition.
Good to know
- macOS only, and the main display only.
- Text recognition can split very small monospaced text (11 pt code in an editor) into pieces. Ordinary app and web text reads cleanly.
- Tested live on macOS 27: every command, including the stop when the mouse moves.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Workflow Automation.
macOS only, and the main display only. Each command takes 1 to 2 seconds while Swift compiles the script, and the first run on a Mac can take about 30 seconds. Typing runs at a person's pace, about 40 words a minute, so long text takes a while. Very small monospaced text (11 pt code in an editor) can come out of text recognition in pieces.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed today
- Software
- Xcode Command Line Tools
The first time, macOS asks for Screen Recording and Accessibility. Turn both on for the app your agent runs in (your terminal, editor or agent app), then quit and reopen it. The first command can take about 30 seconds while macOS loads text recognition.
Permissions
File Scopes
Screen Recording, to read the screen with on-device text recognition. Accessibility, to read buttons and fields and to move the mouse and type. No network access. The only file written is look's screenshot in the temp folder, overwritten each time.