CLI-Anything: Making ALL Software Agent-Native with Automated CLI Generation

CLI-Anything by HKUDS is a groundbreaking framework that transforms any software with a codebase into agent-native tools through automated CLI generation. With 38.2K stars and growing at 1,038+ per day, it solves a fundamental problem: AI agents are great at reasoning but terrible at using real professional software. CLI-Anything bridges this gap with a fully automated 7-phase pipeline that generates production-ready CLIs for 18+ major applications.

The Agent-Software Gap

Today’s software serves humans through GUIs. Tomorrow’s users will be AI agents. But current solutions for agent-software interaction are fragile:

Current Pain Point	CLI-Anything’s Fix
AI can’t use real tools	Direct integration with actual software backends (Blender, LibreOffice, FFmpeg)
UI automation breaks constantly	No screenshots, no clicking, no RPA fragility. Pure command-line reliability
Agents need structured data	Built-in JSON output for seamless agent consumption
Custom integrations are expensive	One command auto-generates CLIs for ANY codebase
Prototype vs Production gap	2,280+ tests with real software validation

The 7-Phase Automated Pipeline

CLI-Anything generates complete, production-ready CLIs through a fully automated 7-phase pipeline:

1. Analyze - Scans source code, maps GUI actions to APIs
2. Design - Architects command groups, state model, output formats
3. Implement - Builds Click CLI with REPL, JSON output, undo/redo
4. Plan Tests - Creates TEST.md with unit + E2E test plans
5. Write Tests - Implements comprehensive test suite
6. Document - Generates SKILL.md for agent discovery
7. Publish - Creates setup.py, installs to PATH

One command triggers the entire pipeline:

# Generate a complete CLI for GIMP (all 7 phases)
/cli-anything ./gimp

# Build from a GitHub repo
/cli-anything https://github.com/blender/blender

Multi-Agent Platform Support

CLI-Anything works across all major AI coding agents:

Platform	Installation	Status
Claude Code	/plugin marketplace add HKUDS/CLI-Anything then /plugin install cli-anything	Production
Pi Coding Agent	bash .pi-extension/cli-anything/install.sh	Production
OpenClaw	Copy SKILL.md to ~/.openclaw/skills/cli-anything/	Community
Codex	bash codex-skill/scripts/install.sh	Experimental
OpenCode	Copy commands to ~/.config/opencode/commands/	Experimental
GitHub Copilot CLI	copilot plugin install ./cli-anything-plugin	Community

Key Features

JSON Output for Agent Consumption

Every command supports --json flag for structured data:

$ cli-anything-libreoffice --json document info --project report.json
{
  "name": "Q1 Report",
  "type": "writer",
  "pages": 1,
  "elements": 2,
  "modified": true
}

Interactive REPL Mode

Every generated CLI includes a branded REPL with command history, progress indicators, and undo/redo:

$ cli-anything-blender
+==========================================+
|       cli-anything-blender v1.0.0       |
|     Blender CLI for AI Agents           |
+==========================================+

blender> scene new --name ProductShot
Created scene: ProductShot

blender[ProductShot]> object add-mesh --type cube --location 0 0 1
Added mesh: Cube at (0, 0, 1)

blender[ProductShot]*> render execute --output render.png --engine CYCLES
Rendered: render.png (1920x1080, 2.3 MB) via blender --background

Iterative Refinement

After the initial build, refine the CLI to expand coverage:

# Broad refinement - agent analyzes gaps across all capabilities
/cli-anything:refine ./gimp

# Focused refinement - target a specific functionality area
/cli-anything:refine ./gimp "I want more CLIs on image batch processing and filters"

Each refinement run is incremental and non-destructive.

SKILL.md Generation

Every generated CLI ships with a SKILL.md file for AI agent discovery:

• YAML frontmatter with name and description
• Command groups with all available subcommands
• Usage examples for common workflows
• Agent-specific guidance for JSON output, error handling, and programmatic use

18+ Production-Ready Harnesses

Software	Domain	CLI Command	Tests
GIMP	Image Editing	cli-anything-gimp	107
Blender	3D Modeling & Rendering	cli-anything-blender	208
Inkscape	Vector Graphics	cli-anything-inkscape	202
Audacity	Audio Production	cli-anything-audacity	161
LibreOffice	Office Suite	cli-anything-libreoffice	158
OBS Studio	Live Streaming	cli-anything-obs-studio	153
Kdenlive	Video Editing	cli-anything-kdenlive	155
Shotcut	Video Editing	cli-anything-shotcut	154
Draw.io	Diagramming	cli-anything-drawio	138
FreeCAD	CAD	cli-anything-freecad	258 cmds
ComfyUI	AI Image Generation	cli-anything-comfyui	70
Ollama	Local LLM Inference	cli-anything-ollama	98
Godot Engine	Game Development	cli-anything-godot	24
Zoom	Video Conferencing	cli-anything-zoom	22
MuseScore	Music Notation	cli-anything-musescore	56
QGIS	GIS & Mapping	cli-anything-qgis	22
Exa	AI Web Search	cli-anything-exa	40
Mailchimp	Email Marketing	cli-anything-mailchimp	303 cmds

Total: 2,280+ tests with 100% pass rate across 1,682 unit tests + 579 end-to-end tests + 19 Node.js tests.

CLI-Hub: Package Manager for Agent CLIs

CLI-Hub lets agents autonomously discover and install the CLIs they need:

# Install CLI-Hub
pip install cli-anything-hub

# Browse and install CLIs
cli-hub install <name>

The meta-skill enables agents to browse the full catalog, pick the right CLI for the task, and use it - all autonomously:

Find appropriate CLI software in CLI-Hub and complete the task: <your task here>

Core Design Principles

1. Authentic Software Integration - The CLI generates valid project files and delegates to real applications for rendering. No Pillow replacements for GIMP, no custom renderers for Blender.

2. Flexible Interaction Models - Every CLI operates in dual modes: stateful REPL for interactive agent sessions + subcommand interface for scripting/pipelines.

3. Agent-Native Design - Built-in --json flag on every command delivers structured data for machine consumption. Agents discover capabilities via standard --help and which commands.

4. Zero Compromise Dependencies - Real software is a hard requirement. Tests fail (not skip) when backends are missing, ensuring authentic functionality.

5. Consistent User Experience - All generated CLIs share unified REPL interface (repl_skin.py) with branded banners, styled prompts, and command history.

HARNESS.md: The Methodology SOP

HARNESS.md is the definitive standard operating procedure for making any software agent-accessible. It encodes critical lessons from building 18+ production harnesses:

Lesson	Description
Use the real software	The CLI MUST call the actual application for rendering. Generate valid project files, then invoke the real backend.
The Rendering Gap	GUI apps apply effects at render time. If your CLI uses a naive export tool, effects get silently dropped.
Filter Translation	When mapping effects between formats, watch for duplicate filter merging, parameter space differences, and unmappable effects.
Timecode Precision	Non-integer frame rates (29.97fps) cause cumulative rounding. Use round() not int().
Output Verification	Never trust that export worked because it exited 0. Verify magic bytes, ZIP structure, pixel analysis, audio levels.

Real-World Demos

AI agents using generated CLIs to produce complete artifacts without any GUI:

• FreeCAD - Agent-built Curiosity-style rover with preview, live preview, and trajectory loops
• Blender - Orbital relay drone assembled through CLI commands with real Blender rendering
• Draw.io - Complete HTTPS handshake sequence diagram built entirely through CLI
• Slay the Spire II - Automated gameplay with strategic decision-making
• VideoCaptioner - AI-powered video captioning with bilingual text rendering

Getting Started

# Step 1: Add the marketplace (Claude Code)
/plugin marketplace add HKUDS/CLI-Anything

# Step 2: Install the plugin
/plugin install cli-anything

# Step 3: Build a CLI for any software
/cli-anything ./path/to/software

# Step 4: Install the generated CLI
cd software/agent-harness && pip install -e .

# Step 5: Use it
cli-anything-software --help
cli-anything-software          # enters REPL
cli-anything-software --json <command>   # JSON output for agents

Key Takeaways

• 7-phase automated pipeline transforms any codebase into a production-ready CLI
• 18+ real software harnesses with 2,280+ passing tests at 100% pass rate
• Multi-agent support for Claude Code, Pi, OpenClaw, Codex, OpenCode, and GitHub Copilot CLI
• JSON output on every command enables seamless agent consumption
• Interactive REPL with undo/redo, command history, and branded experience
• CLI-Hub package manager lets agents autonomously discover and install CLIs
• SKILL.md generation makes every CLI discoverable by AI agents
• HARNESS.md methodology encodes proven patterns from 18+ production harnesses
• Authentic software integration - no toy implementations, no compromises

CLI-Anything represents a paradigm shift: from software that serves humans through GUIs to software that serves both humans and AI agents through structured, reliable command-line interfaces. One command makes any software agent-native.

Links

• Repository: HKUDS/CLI-Anything
• CLI-Hub: hkuds.github.io/CLI-Anything
• Documentation: HARNESS.md
• Contributing: CONTRIBUTING.md

Enjoyed this post? Never miss out on future posts by following us