cli-anything

CLI-Anything is a project from HKUDS aimed at making existing software usable by AI agents through command-line interfaces. Its premise is that CLIs are a natural fit for both humans and agents: they are structured, composable, lightweight, self-describing via --help, and can emit structured JSON output. What it provides 1. CLI-Hub: a package manager and web registry for discovering, installing, updating and launching CLI harnesses. Install with `pip install cli-anything-hub`, then use commands such as `cli-hub list`, `cli-hub search <query>`, `cli-hub info <name>`, `cli-hub install <name>`, `cli-hub update <name>`, `cli-hub uninstall <name>` and `cli-hub launch <name> [args...]`. The registry includes both CLI-Anything harnesses and public third-party CLIs from pip, npm, brew or bundled/system tools. Some harnesses wrap desktop or backend software, so the upstream application may also need to be installed. 2. A generator workflow for building new agent-native CLIs. It is distributed as plugins, extensions or skills for several AI coding agents, including Claude Code, Cursor, Pi Coding Agent, OpenClaw, OpenCode, Codex, Hermes, Reasonix, Qodercli and GitHub Copilot CLI. The main entrypoint is `/cli-anything <software-path-or-repo>`, which runs a seven-phase pipeline: analyze source code and map GUI actions to APIs; design command groups, state model and output formats; implement a Click-based CLI with REPL, JSON output and undo/redo; plan unit and end-to-end tests; write the test suite; document results; and publish a setup.py installable to PATH. A `/cli-anything:refine` command performs gap analysis and incrementally adds commands, tests and documentation. 3. SKILL.md generation: generated CLIs ship with AI-discoverable skill definitions, and a CLI-Hub meta-skill lets compatible agents find and install the right CLI for a task. Skills are unified under a top-level `skills/` directory and can be installed with `npx skills add HKUDS/CLI-Anything --skill <skill-name> -g -y`. The README lists many community-contributed harnesses across domains, such as GIMP, Blender, FreeCAD, QGIS, LibreOffice, Inkscape, Krita, MuseScore, Zotero, Obsidian, Joplin, Calibre, n8n, Dify, Zoom, Godot, RenderDoc, Unreal Insights, WireMock, Exa, Safari, Sketch, Rekordbox and others. It also reports a test suite with 2,461 passing tests and links to a technical report on arXiv. The project is licensed under Apache 2.0 and requires Python 3.10+. Typical use cases include letting an agent drive desktop creative tools, GIS or CAD software, knowledge bases, workflow automation platforms and game engines through a consistent command surface, and generating a new harness when no existing CLI covers the target software.