computer-use

Automate desktop and web GUI interactions lacking APIs or CLIs.

4|1|Updated Jun 21, 2026
One-click install
npx skills add https://github.com/meiiie/neko-core --skill computer-use-meiiie
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: computer-use
Source: https://github.com/meiiie/neko-core/tree/main/skills/computer-use
Command: npx skills add https://github.com/meiiie/neko-core --skill computer-use-meiiie

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve?

This Skill eliminates the need for manual, repetitive GUI interactions when automating tasks on desktop or web applications that have no public API, CLI, or scriptable interface, saving hours of tedious clicking and typing.

Core Features & Use Cases

  • Code-first automation: Prefers shell commands, scripts, and APIs for reliability before falling back to GUI control.
  • Accessibility tree-driven control: Uses Windows UI Automation for desktop apps and web DOM for browsers to interact with elements via text, no vision or GUI-trained model required for most use cases.
  • Vision fallback for custom UIs: Supports pixel-based grounding for custom-drawn interfaces with no accessibility tree, with built-in safety guardrails and audit trails.
  • Shared autonomy: Lets users take over control at any time, with automatic resume and re-perception of the screen state when the user is idle again.
  • Use Case: Automate filling out a legacy internal desktop form that has no API, or navigate a complex web app with dynamic GUI elements that can't be accessed via standard selectors.

Quick Start

Use the computer-use skill to open your company's internal web portal, navigate to the expense report page, fill in the required fields with the provided receipt data, and submit the form without manual clicking.

Frequently Asked Questions about computer-use

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I automate GUI tasks on legacy desktop apps that have no API or CLI?

To automate GUI tasks on legacy desktop apps without APIs, this Skill uses a code-first control hierarchy that prefers shell commands before falling back to Windows UI Automation for text-driven element interaction. It reliably fills forms and scripts workflows without manual clicking.

Can I automate web form filling on pages with dynamic DOM elements?

Yes, you can automate web form filling using accessibility tree-driven control that interacts with web DOM elements via text. It grounds dynamic GUI elements without requiring vision or trained models, enabling reliable navigation and submission for complex web applications.

How does vision-based fallback work for custom-drawn interfaces with no accessibility tree?

Vision-based fallback uses pixel-based grounding to interact with custom-drawn interfaces lacking an accessibility tree. It identifies UI elements visually and includes built-in safety guardrails and audit trails to ensure low-risk, reliable operation when standard accessibility methods fail.

What is the best way to control custom-drawn desktop interfaces without standard UI selectors?

The best way to control custom-drawn interfaces is using pixel-based vision grounding as a fallback after standard accessibility tree methods fail. This approach ensures automation coverage for non-standard UIs while maintaining safety guardrails and audit trails for reliable operation.

Does this GUI automation approach allow shared autonomy and manual takeover?

Yes, this GUI automation approach features shared autonomy that lets users take over control at any time. When the user is idle, it automatically resumes and re-perceives the screen state to continue the workflow seamlessly without losing progress.

When should I not use GUI automation for desktop control?

You should not use GUI automation when a programmatic API, CLI, or scriptable interface is available, as this Skill prioritizes code-first control methods for maximum reliability. GUI control is reserved strictly for applications lacking any programmatic interface.