Do you want the AI to control your computer?
This workflow grants the AI agent full control over your desktop environment — web browsers, native applications, mouse, and keyboard — using a strict 3-pillar hierarchy (DOM control → native app automation → physical mouse/keyboard) for maximum speed and safety.
Under the Hood
This is the actual text of the workflow
Computer Access Cookbook
I. Authorization Rule & Prime Directive
CRITICAL: You may NEVER grant yourself computer access. The user must explicitly say the words "computer access" to authorize you. This is defined in
gemini.mdunder the Computer Access rule. If the user has not said those words in the current conversation, you do NOT have permission to use any of the techniques below.
THE PRIME DIRECTIVE: Once the user grants you Computer Access, you have total power to automate EVERYTHING yourself. Do NOT ask the user to manually click buttons, navigate menus, or change settings that you are fully capable of doing via the methods below. The only time you should ask the user to intervene is when entering sensitive credentials.
II. PART 1: The Domain of the Computer Access workflow
[!WARNING] THIS WORKFLOW DOES NOT EXPLAIN HOW TO BROWSE THE WEB: If your task involves interacting with a website, read
gemini.mdfor instructions and use the relevant skill file.
When you have computer access for native apps, you must follow this strict hierarchy:
III. Pillar 1: Native Windows Apps (WinApp-MCP)
Rule: Read the windows fast automation mcp skill file.
IV. Pillar 2: The Nuclear Fallback (Physical Clicks)
[!CAUTION] DO NOT GUESS PIXEL COORDINATES FROM A RAW SCREENSHOT.
Rule: Read the take picture and click skill file.
V. PART 2: Visual Verification
Liberally take pictures of the screen to make sure everything is going as expected! If you click a button or type text into a native app, take a screenshot immediately afterward to visually confirm that the app responded and progressed as you intended. To get the PowerShell script for taking screenshots, refer to the Universal Vision Utilities section in take picture and click skill.
VI. PART 3: Utilities
VII. Utility 1: Data Extraction Integrity
If a programmatic extraction step (WinApp-MCP or Vision) returns unexpected output (wrong format, wrong content, an error, or nothing), you must stop and tell the user. Never substitute fabricated data to fill a gap. If you cannot extract a value, set it to "UNVERIFIED" in the dataset and report what went wrong.
The Output
None
How to Set This Up
Option 1: The Easy Way (Automated)
Just point Antigravity to this webpage and ask it to figure it out for you. Antigravity can read this documentation, copy the workflow script, and automatically generate all the required skill files in the correct directories on your machine.
Option 2: The Hard Way (Manual Copy & Paste)
If Antigravity fails to set this up automatically, you will need to manually copy the scripts into your local directories:
- Copy the raw workflow script from the "Under the Hood" section above.
- Save it as
C:\Users\[Your Name]\.gemini\antigravity\global_workflows\computer-access.md. - You must also click every hyperlinked skill file and save its contents into your skills directory. You must do this for every single skill file linked in the workflow. For example, the
legal_analysis_partnerskill must be saved toC:\Users\[Your Name]\.gemini\antigravity\skills\legal_analysis_partner\SKILL.md. - Once all files are saved, open Antigravity and type
/computer-accessin the chat to run it.
New to Antigravity? Read the Master Installation Guide first.