How to use the Live2D Model CLI Toolkit agent skill

What the skill lets the agent do

A skill is a folder of instructions that an AI agent loads when a task matches it. This one is built around a command-line tool called cli-anything-live2d, which offers 42 commands for Live2D Cubism models stored as .model3.json files.

The commands fall into groups: reading (inspect, validate, motions, expressions, textures, diff, manifest and others), editing (edit-motion, add-expr, find-replace and others), generation (init, gen-motion, gen-expr), analysis (lint, runtime-check, atlas, orphan), workflow (undo, batch, snapshot, watch, flatten) and packaging (migrate, pack, yoyo-check). Most of this works without opening the Live2D Editor.

• Inspect a model: version, counts and features
• Validate that all referenced files exist
• Lint a model against best practices
• Diff two model versions or compare them quickly
• Edit motion fade times, expressions and texture paths, with automatic backups
• Batch-edit many models at once, with a dry-run option
• Package a model and its dependencies into a zip for deployment

Install the skill and the command-line tool

There are two parts. First, install the skill itself: download the ZIP file from this page and unzip it into ~/.claude/skills/ for all your projects, or into a project's .claude/skills/ folder for that project only. Alternatively, if you use the Claude apps, upload the skill under Settings, then Capabilities, then Skills.

Second, the skill needs the cli-anything-live2d command-line tool. The source says it is a Python package from the CLI-Anything repository, so follow the installation instructions in that repository. This page does not list a specific install command, so check the repository for the current method.

• Skill files: SKILL.md (the only file listed for this skill)
• Tool: cli-anything-live2d, from the CLI-Anything repository
• Model files: Live2D Cubism .model3.json models and their related assets

How to trigger the skill with natural requests

You do not need to type commands yourself. Describe the task in plain language and mention your model file, and the agent should pick the skill when the request matches its purpose. Name the action you want, such as checking, linting, comparing, batch-changing or packaging.

For edits, say which motion group or file you mean. The skill's edit commands make automatic backups, but you should still review the result.

• Check whether every file referenced by hair.model3.json exists.
• Lint my model and list the problems, with info-level detail.
• Compare the old and new versions of this rig and tell me what changed.
• Set the fade-in time to 0.3 for every model in this folder, but only show me a dry run first.
• Package this model and its dependencies for deployment in a web app.

Examples of commands the agent can run

These are examples from the skill's own documentation. You can run them yourself in a terminal, or ask the agent to run them for you. Each command takes a model file or folder as its input.

• inspect and lint: check a model and its best practices
• find-replace: rename file references in a model
• orphan: find files not referenced by the model
• runtime-check with the web-sdk target: check compatibility with the Web SDK
• flatten with an output folder: prepare a folder for deployment
• batch with a dry run: preview changes across a folder
• snapshot with embed: generate an HTML preview
• stats on a project folder: get a project overview

What you need before you start

You need the cli-anything-live2d tool installed and Live2D Cubism model files, meaning .model3.json files and their related assets. The skill only covers .model3.json-based Cubism models. Other 2D or 3D formats are outside its scope.

The skill runs scripts. It is listed as not using the network. Read commands support a --json option, which is useful when you want machine-readable output.

• cli-anything-live2d installed as a Python package from the CLI-Anything repository
• Live2D Cubism model files (.model3.json and related assets)
• A backup of your models before any batch or edit command

Safety notes before editing

Edit commands create automatic backups, and the undo command can restore from them. However, the source notes that the backups depend on the files the tool can access. Do not treat them as your only protection.

For batch operations, preview with --dry-run first. Keep a separate copy of any model you cannot afford to lose, especially if it is your only copy.

• Keep a separate backup of models before batch edits
• Always preview batch changes with --dry-run
• Scripts run on your machine, so only install skills you trust
• Use backup-clean and restore-file only when you understand what they will change

Tips for smoother results

Run lint and validate before and after editing so you can see what changed. Use diff to compare versions. If you manage a large project, start with stats or scan to see what is there before making changes.

Remember that this is a command-line tool, not a design application. It edits data and files; it does not draw or rig characters visually.

• Validate first, then edit, then validate again
• Use --json when you want output the agent or a script can parse
• Use orphan to find unused files before you package a model