CLI-Anything: Making ALL Software Agent-Native

Today's Software Serves Humans👨‍💻. Tomorrow's Users will be Agents🤖.
CLI-Anything: Bridging the Gap Between AI Agents and the World's Software

🌐 CLI-Hub: Explore all community-built CLIs and install with one command at the CLI-Hub. Want to add your own? Open a PR — the hub updates instantly.

One Command Line: Make any software agent-ready for OpenClaw, nanobot, Cursor, Claude Code, etc.  Chinese Documentation | Japanese Documentation

📰 News

Thanks to all invaluable efforts from the community! More updates continuously on the way everyday..

• 2026-03-23 🤖 Launched CLI-Hub meta-skill — agents can now discover and install CLIs autonomously via cli-hub-meta-skill/SKILL.md. The catalog auto-updates from registry.json via GitHub Actions, making the entire marketplace agent-native.
• 2026-03-22 🎵 MuseScore CLI merged with transpose, export, and instrument management. Community contributions continue expanding domain coverage.
• 2026-03-21 🔧 Infrastructure improvements — refined test harnesses and documentation across multiple CLIs. Enhanced Windows compatibility for several backends.
• 2026-03-20 🌐 Novita AI CLI added for OpenAI-compatible API access. Registry metadata improvements for better hub discovery.
• 2026-03-19 📦 Package structure refinements across harnesses. Improved SKILL.md generation with better command documentation.
• 2026-03-18 🧪 Test coverage expansion — additional E2E scenarios and edge case validation across multiple CLIs.
• 2026-03-17 🌐 Launched the CLI-Hub — a central registry where you can browse, search, and install any CLI with a single pip command. Contributors can add new CLIs or update existing ones by simply opening a PR with a registry.json entry. The hub updates automatically on merge.

🤔 Why CLI?

CLI is the universal interface for both humans and AI agents:

• Structured & Composable - Text commands match LLM format and chain for complex workflows

• Lightweight & Universal - Minimal overhead, works across all systems without dependencies

• Self-Describing - --help flags provide automatic documentation agents can discover

• Proven Success - Claude Code runs thousands of real workflows through CLI daily

• Agent-First Design - Structured JSON output eliminates parsing complexity

• Deterministic & Reliable - Consistent results enable predictable agent behavior

🚀 Quick Start

Prerequisites

• Python 3.10+
• Target software installed (e.g., GIMP, Blender, LibreOffice, or your own application)
• A supported AI coding agent: Claude Code | OpenClaw | OpenCode | Codex | Qodercli | GitHub Copilot CLI | More Platforms

Pick Your Platform

# Add the CLI-Anything marketplace
/plugin marketplace add HKUDS/CLI-Anything

# Install the cli-anything plugin from the marketplace
/plugin install cli-anything

# /cli-anything:cli-anything 
Generate a complete CLI for GIMP (all 7 phases)
/cli-anything:cli-anything ./gimp
Note: If your Claude Code is under 2.x, use "/cli-anything" instead.

# 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"

# Clone the repo
git clone https://github.com/HKUDS/CLI-Anything.git
Copy plugin to Claude Code plugins directory
cp -r CLI-Anything/cli-anything-plugin ~/.claude/plugins/cli-anything
Reload plugins
/reload-plugins

# Clone the repo
git clone https://github.com/HKUDS/CLI-Anything.git
Global install (available in all projects)
cp CLI-Anything/opencode-commands/*.md ~/.config/opencode/commands/
cp CLI-Anything/cli-anything-plugin/HARNESS.md ~/.config/opencode/commands/
Or project-level install
cp CLI-Anything/opencode-commands/*.md .opencode/commands/
cp CLI-Anything/cli-anything-plugin/HARNESS.md .opencode/commands/

# Generate a complete CLI for GIMP (all 7 phases)
/cli-anything ./gimp
Build from a GitHub repo
/cli-anything https://github.com/blender/blender

# Broad refinement — agent analyzes gaps across all capabilities
/cli-anything-refine ./gimp
Focused refinement — target a specific functionality area
/cli-anything-refine ./gimp "batch processing and filters"

/cli-anything:cli-anything ./gimp
/cli-anything:refine ./gimp "batch processing and filters"

git clone https://github.com/HKUDS/CLI-Anything.git
bash CLI-Anything/qoder-plugin/setup-qodercli.sh

/cli-anything:cli-anything ./gimp
/cli-anything:refine ./gimp "batch processing and filters"
/cli-anything:validate ./gimp

# Clone the repo
git clone https://github.com/HKUDS/CLI-Anything.git
Install to the global skills folder
mkdir -p ~/.openclaw/skills/cli-anything
cp CLI-Anything/openclaw-skill/SKILL.md ~/.openclaw/skills/cli-anything/SKILL.md

# Clone the repo
git clone https://github.com/HKUDS/CLI-Anything.git
Install the skill
bash CLI-Anything/codex-skill/scripts/install.sh

.\CLI-Anything\codex-skill\scripts\install.ps1

Use CLI-Anything to build a harness for ./gimp
Use CLI-Anything to refine ./shotcut for picture-in-picture workflows
Use CLI-Anything to validate ./libreoffice

git clone https://github.com/HKUDS/CLI-Anything.git
cd CLI-Anything
copilot plugin install ./cli-anything-plugin

/cli-anything:cli-anything ./gimp
/cli-anything:refine ./gimp "batch processing and filters"
/cli-anything:validate ./gimp

Use the Generated CLI

Regardless of which platform you used to build it, the generated CLI works the same way:

# Install to PATH
cd gimp/agent-harness && pip install -e .
Use from anywhere
cli-anything-gimp --help
cli-anything-gimp project new --width 1920 --height 1080 -o poster.json
cli-anything-gimp --json layer add -n "Background" --type solid --color "#1a1a2e"
Enter interactive REPL
cli-anything-gimp

Each installed CLI ships with a SKILL.md inside the Python package (cli_anything//skills/SKILL.md). The REPL banner automatically displays the absolute path to this file so AI agents know exactly where to read the skill definition. No extra configuration needed — pip install makes the skill discoverable.

🤖 Empower Your Agents with CLI-Hub

CLI-Hub lets agents autonomously discover and install the CLIs they need — zero human intervention required.

We published a meta-skill that lets any AI agent freely explore the full catalog of community CLIs and pick the right one for the task.

Install in one command:

# OpenClaw
openclaw skills install cli-anything-hub
nanobot
nanobot skills install cli-anything-hub

Then just prompt your agent:

Find appropriate CLI software in CLI-Hub and complete the task: 

How it works under the hood:

• The meta-skill points to the live catalog at https://hkuds.github.io/CLI-Anything/SKILL.txt
• The agent reads 20+ CLIs organized by category with one-line pip install commands
• The agent installs whichever CLI fits the task, then reads that CLI's own SKILL.md for detailed usage

For Claude Code users: Copy cli-hub-meta-skill/SKILL.md into your project or skills directory for the same automatic CLI discovery.

💡 CLI-Anything's Vision: Building Agent-Native Software

• 🌐 Universal Access - Every software becomes instantly agent-controllable through structured CLI.

• 🔗 Seamless Integration - Agents control any application without APIs, GUI, rebuilding or complex wrappers.

• 🚀 Future-Ready Ecosystem - Transform human-designed software into agent-native tools with one command.

🔧 When to Use CLI-Anything

| Category | How to be Agent-native | Notable Examples | |----------|----------------------|----------| | 📂 GitHub Repositories | Transform any open-source project into agent-controllable tools through automatic CLI generation | VSCodium, WordPress, Calibre, Zotero, Joplin, Logseq, Penpot, Super Productivity | | 🤖 AI/ML Platforms | Automate model training, inference pipelines, and hyperparameter tuning through structured commands | Stable Diffusion WebUI, ComfyUI, Ollama, InvokeAI, Text-generation-webui, Open WebUI, Fooocus, Kohya_ss, AnythingLLM, SillyTavern | | 📊 Data & Analytics | Enable programmatic data processing, visualization, and statistical analysis workflows | JupyterLab, Apache Superset, Metabase, Redash, DBeaver, KNIME, Orange, OpenSearch Dashboards, Lightdash | | 💻 Development Tools | Streamline code editing, building, testing, and deployment processes via command interfaces | Jenkins, Gitea, Hoppscotch, Portainer, pgAdmin, SonarQube, ArgoCD, OpenLens, Insomnia, Beekeeper Studio, iTerm2 | | 🎨 Creative & Media | Control content creation, editing, and rendering workflows programmatically | Blender, GIMP, OBS Studio, Audacity, Krita, Kdenlive, Shotcut, Inkscape, Darktable, LMMS, Ardour | | 🔬 Scientific Computing | Automate research workflows, simulations, and complex calculations | ImageJ, FreeCAD, QGIS, ParaView, Gephi, LibreCAD, Stellarium, KiCad, JASP, Jamovi | | 🏢 Enterprise & Office | Convert business applications and productivity tools into agent-accessible systems | NextCloud, GitLab, Grafana, Mattermost, LibreOffice, AppFlowy, NocoDB, Odoo (Community), Plane, ERPNext | | 📞 Communication & Collaboration | Automate meeting scheduling, participant management, recording retrieval, and reporting through structured CLI | Zoom, Jitsi Meet, BigBlueButton, Mattermost | | 📐 Diagramming & Visualization | Create and manipulate diagrams, flowcharts, architecture diagrams, and visual documentation programmatically | Draw.io (diagrams.net), Mermaid, PlantUML, Excalidraw, yEd | | 🌐 Network & Infrastructure | Manage network services, DNS, ad-blocking, and infrastructure through structured CLI commands | AdGuardHome | | 🔬 Graphics & GPU Debugging | Analyze GPU frame captures, inspect pipeline state, export shaders, and diff rendering state | RenderDoc |

| ✨ AI Content Generation | Generate professional deliverables (slides, docs, diagrams, websites, research reports) through AI-powered cloud APIs | AnyGen, Gamma, Beautiful.ai, Tome |

CLI-Anything's Key Features

The Agent-Software Gap

CLI-Anything's Solution: Transform any professional software into agent-native tools without losing capabilities.

| Current Pain Point | CLI-Anything's Fix | |----------|----------------------| | 🤖 "AI can't use real tools" | Direct integration with actual software backends (Blender, LibreOffice, FFmpeg) — full professional capabilities, zero compromises | | 💸 "UI automation breaks constantly" | No screenshots, no clicking, no RPA fragility. Pure command-line reliability with structured interfaces | | 📊 "Agents need structured data" | Built-in JSON output for seamless agent consumption + human-readable formats for debugging | | 🔧 "Custom integrations are expensive" | One Claude plugin auto-generates CLIs for ANY codebase through proven 7-phase pipeline | | ⚡ "Prototype vs Production gap" | 1,839+ tests with real software validation. Battle-tested across 16 major applications |

🎯 What Can You Do with CLI-Anything?

✨ ⚙️ How CLI-Anything Works

🤖 SKILL.md Generation

Each generated CLI includes a SKILL.md file inside the Python package at cli_anything//skills/SKILL.md. This self-contained skill definition enables AI agents to discover and use the CLI through Claude Code's skill system or other agent frameworks.

What SKILL.md provides:

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

🎬 Demonstrations

🎯 General-Purpose

🏭 Professional-Grade Testing

🎨 Diverse Domain Coverage

✅ Full CLI Generation

100% pass rate across all 1,858 tests — 1,355 unit tests + 484 end-to-end tests + 19 Node.js tests.

📊 Test Results

Each CLI harness undergoes rigorous multi-layered testing to ensure production reliability:

| Layer | What it tests | Example | |-------|---------------|---------| | Unit tests | Every core function in isolation with synthetic data | test_core.py — project creation, layer ops, filter params | | E2E tests (native) | Project file generation pipeline | Valid ODF ZIP structure, correct MLT XML, SVG well-formedness | | E2E tests (true backend) | Real software invocation + output verification | LibreOffice → PDF with %PDF- magic bytes, Blender → rendered PNG | | CLI subprocess tests | Installed command via subprocess.run | cli-anything-gimp --json project new → valid JSON output |

================================ Test Summary ================================
gimp          107 passed  ✅   (64 unit + 43 e2e)
blender       208 passed  ✅   (150 unit + 58 e2e)
inkscape      202 passed  ✅   (148 unit + 54 e2e)
audacity      161 passed  ✅   (107 unit + 54 e2e)
libreoffice   158 passed  ✅   (89 unit + 69 e2e)
mubu           96 passed  ✅   (85 unit + 11 e2e)
obs-studio    153 passed  ✅   (116 unit + 37 e2e)
kdenlive      155 passed  ✅   (111 unit + 44 e2e)
shotcut       154 passed  ✅   (110 unit + 44 e2e)
zoom           22 passed  ✅   (22 unit + 0 e2e)
drawio        138 passed  ✅   (116 unit + 22 e2e)
mermaid        10 passed  ✅   (5 unit + 5 e2e)
anygen         50 passed  ✅   (40 unit + 10 e2e)
notebooklm     21 passed  ✅   (21 unit + 0 e2e)
comfyui        70 passed  ✅   (60 unit + 10 e2e)
adguardhome    36 passed  ✅   (24 unit + 12 e2e)
ollama         98 passed  ✅   (87 unit + 11 e2e)
sketch         19 passed  ✅   (19 jest, Node.js)
renderdoc      59 passed  ✅   (45 unit + 14 e2e)
──────────────────────────────────────────────────────────────────────────────
TOTAL        1,917 passed  ✅   100% pass rate

🏗️ CLI-Anything's Architecture

🎯 Core Design Principles

• Authentic Software Integration — The CLI generates valid project files (ODF, MLT XML, SVG) and delegates to real applications for rendering. We build structured interfaces TO software, not replacements.
• Flexible Interaction Models — Every CLI operates in dual modes: stateful REPL for interactive agent sessions + subcommand interface for scripting/pipelines. Run bare command → enter REPL mode.
• Consistent User Experience — All generated CLIs share unified REPL interface (repl_skin.py) with branded banners, styled prompts, command history, progress indicators, and standardized formatting.
• Agent-Native Design — Built-in --json flag on every command delivers structured data for machine consumption, while human-readable tables serve interactive use. Agents discover capabilities via standard --help and which commands.
• Zero Compromise Dependencies — Real software is a hard requirement — no fallbacks, no graceful degradation. Tests fail (not skip) when backends are missing, ensuring authentic functionality.

📂 Project Structure

cli-anything/
├── 📄 README.md                          # You are here
├── 📁 assets/                            # Images and media
│   ├── icon.png                          # Project icon
│   └── teaser.png                        # Teaser figure
│
├── 🔌 cli-anything-plugin/               # The Claude Code plugin
│   ├── HARNESS.md                        # Methodology SOP (source of truth)
│   ├── README.md                         # Plugin documentation
│   ├── QUICKSTART.md                     # 5-minute getting started
│   ├── PUBLISHING.md                     # Distribution guide
│   ├── repl_skin.py                      # Unified REPL interface
│   ├── commands/                         # Plugin command definitions
│   │   ├── cli-anything.md               # Main build command
│   │   ├── refine.md                     # Expand existing harness coverage
│   │   ├── test.md                       # Test runner
│   │   └── validate.md                   # Standards validation
│   └── scripts/
│       └── setup-cli-anything.sh         # Setup script
│
├── 🤖 codex-skill/                      # Codex skill entry point
├── 🎨 gimp/agent-harness/               # GIMP CLI (107 tests)
├── 🧊 blender/agent-harness/            # Blender CLI (208 tests)
├── ✏️ inkscape/agent-harness/            # Inkscape CLI (202 tests)
├── 🎵 audacity/agent-harness/           # Audacity CLI (161 tests)
├── 🌐 browser/agent-harness/            # Browser CLI (DOMShell MCP, new)
├── 📄 libreoffice/agent-harness/        # LibreOffice CLI (158 tests)
├── 📝 mubu/agent-harness/               # Mubu CLI (96 tests)
├── 📹 obs-studio/agent-harness/         # OBS Studio CLI (153 tests)
├── 🎞️ kdenlive/agent-harness/           # Kdenlive CLI (155 tests)
├── 🎬 shotcut/agent-harness/            # Shotcut CLI (154 tests)
├── 📞 zoom/agent-harness/               # Zoom CLI (22 tests)
├── 🎵 musescore/agent-harness/          # MuseScore CLI (56 tests)
├── 📐 drawio/agent-harness/             # Draw.io CLI (138 tests)
├── 🧜 mermaid/agent-harness/            # Mermaid Live Editor CLI (10 tests)
├── ✨ anygen/agent-harness/             # AnyGen CLI (50 tests)
├── 🖼️ comfyui/agent-harness/            # ComfyUI CLI (70 tests)
├── 🧠 notebooklm/agent-harness/         # NotebookLM CLI (experimental, 21 tests)
├── 🛡️ adguardhome/agent-harness/       # AdGuard Home CLI (36 tests)
├── 🦙 ollama/agent-harness/             # Ollama CLI (98 tests)
├── 🎨 sketch/agent-harness/             # Sketch CLI (19 tests, Node.js)
└── 🔬 renderdoc/agent-harness/          # RenderDoc CLI (59 tests)

Each agent-harness/ contains an installable Python package under cli_anything./ with Click CLI, core modules, utils (including repl_skin.py and backend wrapper), and comprehensive tests.

🎯 Plugin Commands

| Command | Description | |---------|-------------| | /cli-anything | Build complete CLI harness — all 7 phases | | /cli-anything:refine [focus] | Refine an existing harness — expand coverage with gap analysis | | /cli-anything:test | Run tests and update TEST.md with results | | /cli-anything:validate | Validate against HARNESS.md standards |

Examples

# Build a complete CLI for GIMP from local source
/cli-anything /home/user/gimp
Build from a GitHub repo
/cli-anything https://github.com/blender/blender
Refine an existing harness — broad gap analysis
/cli-anything:refine /home/user/gimp
Refine with a specific focus area
/cli-anything:refine /home/user/shotcut "vid-in-vid and picture-in-picture compositing"
Run tests and update TEST.md
/cli-anything:test /home/user/inkscape
Validate against HARNESS.md standards
/cli-anything:validate /home/user/audacity

🎮 Demo: Using a Generated CLI

Here's what an agent can do with cli-anything-libreoffice:

# Create a new Writer document
$ cli-anything-libreoffice document new -o report.json --type writer
✓ Created Writer document: report.json
Add content
$ cli-anything-libreoffice --project report.json writer add-heading -t "Q1 Report" --level 1
✓ Added heading: "Q1 Report"
$ cli-anything-libreoffice --project report.json writer add-table --rows 4 --cols 3
✓ Added 4×3 table
Export to real PDF via LibreOffice headless
$ cli-anything-libreoffice --project report.json export render output.pdf -p pdf --overwrite
✓ Exported: output.pdf (42,831 bytes) via libreoffice-headless
JSON mode for agent consumption
$ cli-anything-libreoffice --json document info --project report.json
{
  "name": "Q1 Report",
  "type": "writer",
  "pages": 1,
  "elements": 2,
  "modified": true
}

REPL Mode

$ 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 (1920×1080, 2.3 MB) via blender --background
blender[ProductShot]> exit
Goodbye! 👋

📖 The Standard Playbook: HARNESS.md

HARNESS.md is our definitive SOP for making any software agent-accessible via automated CLI generation.

It encodes proven patterns and methodologies refined through automated generation processes.

The playbook distills key insights from successfully building all 16 diverse, production-ready harnesses.

Critical Lessons

| Lesson | Description | |--------|-------------| | Use the real software | The CLI MUST call the actual application for rendering. No Pillow replacements for GIMP, no custom renderers for Blender. Generate valid project files → invoke the real backend. | | The Rendering Gap | GUI apps apply effects at render time. If your CLI manipulates project files but uses a naive export tool, effects get silently dropped. Solution: native renderer → filter translation → render script. | | Filter Translation | When mapping effects between formats (MLT → ffmpeg), watch for duplicate filter merging, interleaved stream ordering, parameter space differences, and unmappable effects. | | Timecode Precision | Non-integer frame rates (29.97fps) cause cumulative rounding. Use round() not int(), integer arithmetic for display, and ±1 frame tolerance in tests. | | Output Verification | Never trust that export worked because it exited 0. Verify: magic bytes, ZIP/OOXML structure, pixel analysis, audio RMS levels, duration checks. |

See the full methodology: cli-anything-plugin/HARNESS.md

📦 Installation & Usage

For Plugin Users (Claude Code)

# Add marketplace & install (recommended)
/plugin marketplace add HKUDS/CLI-Anything
/plugin install cli-anything
Build a CLI for any software with a codebase
/cli-anything 

For Generated CLIs

# Install any generated CLI
cd /agent-harness
pip install -e .
Verify
which cli-anything-
Use
cli-anything- --help
cli-anything-                    # enters REPL
cli-anything- --json    # JSON output for agents

Running Tests

# Run tests for a specific CLI
cd /agent-harness
python3 -m pytest cli_anything//tests/ -v
Force-installed mode (recommended for validation)
CLI_ANYTHING_FORCE_INSTALLED=1 python3 -m pytest cli_anything//tests/ -v -s

🤝 Contributing

We welcome contributions! CLI-Anything is designed to be extensible:

• New software targets — Use the plugin to generate a CLI for any software with a codebase, then submit your harness via cli-anything-plugin/PUBLISHING.md.
• Methodology improvements — PRs to HARNESS.md that encode new lessons learned
• Plugin enhancements — New commands, phase improvements, better validation
• Test coverage — More E2E scenarios, edge cases, workflow tests

Limitations

• Requires strong foundation models — CLI-Anything relies on frontier-class models (e.g., Claude Opus 4.6, Claude Sonnet 4.6, GPT-5.4) for reliable harness generation. Weaker or smaller models may produce incomplete or incorrect CLIs that require significant manual correction.
• Relies on available source code — The 7-phase pipeline analyzes and generates from source code. When the target software only provides compiled binaries that require decompilation, harness quality and coverage will degrade substantially.
• May require iterative refinement — A single /cli-anything run may not fully cover all capabilities. Running /refine one or more times is often needed to push the CLI's performance and coverage to production quality.

Roadmap

• [ ] Support for more application categories (CAD, DAW, IDE, EDA, scientific tools)
• [ ] Benchmark suite for agent task completion rates
• [ ] Community-contributed CLI harnesses for internal/custom software
• [ ] Integration with additional agent frameworks beyond Claude Code
• [ ] Support packaging APIs for closed-source software and web services into CLIs
• [x] Produce SKILL.md alongside the CLI for agent skill discovery and orchestration

📖 Documentation

| Document | Description | |----------|-------------| | cli-anything-plugin/HARNESS.md | The methodology SOP — single source of truth | | cli-anything-plugin/README.md | Plugin documentation — commands, options, phases | | cli-anything-plugin/QUICKSTART.md | 5-minute getting started guide | | cli-anything-plugin/PUBLISHING.md | Distribution and publishing guide |

Each generated harness also includes:

• .md — Architecture SOP specific to that application
• tests/TEST.md — Test plan and results documentation

⭐ Star History

If CLI-Anything helps make your software Agent-native, give us a star! ⭐

📄 License

MIT License — free to use, modify, and distribute.

Thanks for visiting ✨ CLI-Anything!

--- Tranlated By Open Ai Tx | Last indexed: 2026-03-29 ---