yb2460/harness-anything

What it solves

This project provides a suite of CLI-based toolsets that allow AI agents to programmatically control professional productivity and design software. It bridges the gap between LLMs and desktop applications by providing a standardized command-line interface for software that typically requires manual GUI interaction, specifically targeting office suites, academic research tools, and graphic design software.

How it works

The system uses Windows COM (Component Object Model) automation interfaces to communicate with installed applications. It wraps these complex COM bridges into a set of simplified CLI commands (via Python and the click library). This allows an AI agent to execute a command like cli-anything-illustrator shapes rect to trigger a specific action within the Adobe Illustrator engine without needing to interact with the visual interface.

Who it’s for

• AI Agent Developers: Those building agents that need to perform real-world tasks like creating presentations, editing vector graphics, or managing academic literature.
• Researchers: Users who want to automate the process of writing papers, conducting meta-analyses, and generating academic charts.
• Designers/Power Users: Individuals looking to automate repetitive layout and document generation tasks in Office or Adobe Creative Cloud.

Highlights

• Office Automation: 47 CLI commands for WPS/Microsoft Office (Word, Excel, PowerPoint) including a dedicated PPT design system with 14 layouts.
• Academic Intelligence: Integration with Zotero featuring 27 "skills" for literature search, IMRAD manuscript writing, and peer review simulation.
• Design Tooling: Full COM-based control over Adobe Illustrator (vector graphics) and Adobe Photoshop (raster images), including layer management and exporting.
• PPT Generation Engine: A JSON-driven engine (pptengine) that allows for programmatic slide creation using reusable primitives and automatic layout validation via PyMuPDF.