MAGIC Agent Skills

30 agent skills for LLM training data preparation, data science, and computational linguistics.

📖 Documentation: docs.votee.ai/magic-agent-skills

MAGIC (Multi-Agent Generic Intelligence Capabilities) turns any AI coding assistant into a specialist for the data work behind LLM development — from raw data ingestion and cleaning through synthesis, annotation, tokenizer auditing, and evaluation. Works with Claude Code, Cursor, Windsurf, Gemini CLI, and 30 AI tools in total.

How It Works

Each skill is a self-contained knowledge package. When an agent receives a task:

1. Agent reads SKILL.md → gets domain knowledge, code patterns, and procedures
2. Agent reads references/*.md → gets detailed patterns on demand
3. Agent reads scripts/*.py → sees reference implementations (not executed directly)
4. Agent writes its own code adapted to the specific task

Skills provide knowledge and patterns. The agent decides how to act — it may follow the reference scripts closely, adapt them, or write entirely custom code.

Skills

Data Science (12 skills)

Skills for the core data pipeline — loading, profiling, cleaning, transforming, validating, and delivering datasets for LLM training and fine-tuning.

Skill	Description
magic-workspace-init	Workspace scaffolding, environment verification, dependency installation
magic-data-lifecycle	Multi-skill orchestration, routing, quality gating
magic-data-loading	Multi-format file detection, auto-encoding, CJK support, databases, HuggingFace
magic-data-profiling	Quality scoring, distribution analysis, outlier detection, correlation
magic-data-cleaning	Missing values, normalization, sentinel replacement, cleaning plans
magic-data-validation	Schema inference, constraint checking, fitness-for-use assessment
magic-data-exploration	Pattern discovery, segment analysis, relationship exploration
magic-data-transformation	Reshape, aggregate, merge, derive columns, deliver to DB/HuggingFace
magic-data-synthesis	LLM-based generation via DataDesigner — fill missing, translate, enrich
magic-statistical-analysis	Descriptive stats, hypothesis testing, correlation analysis
magic-data-visualization	Chart selection, generation (static + interactive), validation
magic-report-generation	Structured report assembly, table formatting

Computational Linguistics (18 skills)

Skills for low-resource language NLP — tokenizer auditing, corpus building, morphological analysis, annotation, cross-lingual transfer, and evaluation. Essential for extending LLMs to new languages.

Skill	Description
magic-linguistic-orchestrator	Routes tasks to the appropriate linguistic skill
magic-linguistic-scope	Language assessment and resource classification
magic-linguistic-tokenize	Tokenizer fertility audit, vocab extension strategy
magic-linguistic-corpus	Corpus collection and curation
magic-linguistic-morph	Morphological analysis and generation
magic-linguistic-syntax	Syntactic parsing and treebanks
magic-linguistic-semantics	Word embeddings and semantic similarity
magic-linguistic-lexicon	Lexicon building and management
magic-linguistic-annotate	Annotation project design, IAA metrics
magic-linguistic-bitext	Parallel corpus alignment
magic-linguistic-codeswitch	Code-switching detection and handling
magic-linguistic-discourse	Discourse analysis and coherence
magic-linguistic-ethics	Ethical considerations for language technology
magic-linguistic-eval	Evaluation methodology and benchmarks
magic-linguistic-historical	Historical linguistics and language change
magic-linguistic-scripts	Writing system analysis and conversion
magic-linguistic-speech	Speech processing and phonology
magic-linguistic-transfer	Cross-lingual transfer and adaptation

Installation

Prerequisites

• Node.js 20+ — required for the CLI installer
• Python 3.12+ — required for skill scripts and reference implementations
• After installing, run pip install -r requirements.txt for Python dependencies

Option 1: skills.sh (Recommended)

Install all 30 skills at once:

npx skills add Votee-AI/magic-agent-skills

Or install a specific skill:

npx skills add Votee-AI/magic-agent-skills --skill magic-data-cleaning

Option 2: CLI Installer

Granular suite/skill selection with tool detection for 30 AI coding tools:

# Interactive — auto-detects tools in your project
npx @votee-ai/magic-agent-skills init

# Non-interactive — specify tools directly
npx @votee-ai/magic-agent-skills init --tools claude,cursor,windsurf

# Install only data skills
npx @votee-ai/magic-agent-skills init --suites data

# Install only linguistic skills
npx @votee-ai/magic-agent-skills init --suites linguistic

Option 3: Claude Plugin Marketplace

/plugin marketplace add Votee-AI/magic-agent-skills

Option 4: Manual

Clone the repo and copy the skills you need:

git clone https://github.com/Votee-AI/magic-agent-skills.git
cp -r magic-agent-skills/skills/magic-data-cleaning .claude/skills/

Supported Tools

The CLI installer supports 30 AI coding tools including:

Claude Code, Cursor, Windsurf, Gemini CLI, Cline, Aider, Continue, Copilot, Amazon Q, Tabnine, Sourcegraph Cody, JetBrains AI, Zed AI, Replit AI, and more.

Run npx @votee-ai/magic-agent-skills init to auto-detect which tools are in your project.

Project Structure

magic-agent-skills/
├── skills/                     # 30 skill packages + shared utilities
│   ├── magic-data-*/           # 12 data science skills
│   │   ├── SKILL.md            # Knowledge document + frontmatter
│   │   ├── scripts/            # Reference Python implementations
│   │   ├── references/         # Additional reference material
│   │   └── tests/              # Per-skill unit tests (co-located)
│   ├── magic-linguistic-*/     # 18 linguistics skills
│   │   ├── SKILL.md            # Knowledge document + frontmatter
│   │   ├── scripts/            # Reference implementations (where applicable)
│   │   ├── references/         # Linguistic references
│   │   ├── evals/              # Skill-specific evaluation data
│   │   ├── _linguistic_shared/ # Generated bundle (synced from below; do not hand-edit)
│   │   └── tests/              # Per-skill tests (where applicable)
│   └── _linguistic_shared/     # Shared Python utilities — SOURCE OF TRUTH (not a skill);
│                               # synced into each skill by scripts/sync-linguistic-shared.py
├── tests/                      # Test suites by category
│   ├── shared/                 # Cross-cutting tests (all 30 skills)
│   ├── data-agent/             # Data-agent specific tests
│   │   ├── unit/               # Script-level tests
│   │   ├── integration/        # Multi-skill workflow tests
│   │   └── e2e/                # End-to-end pipeline scenarios
│   └── linguistic/             # Linguistic tests (future)
├── commands/
│   ├── data/                   # 13 data slash commands
│   └── linguistic/             # 10 linguistic slash commands
├── cli/                        # npm CLI installer
├── schema/
│   └── SKILL.schema.json       # Frontmatter validation schema
├── docs/images/                # Logo and architecture diagrams
├── .claude-plugin/
│   └── marketplace.json        # Claude plugin manifest
├── skills.sh.json              # skills.sh registry grouping
└── RELEASING.md                # Versioning and release policy

Contributing

See CONTRIBUTING.md for branch strategy, PR process, and development setup.

License

Apache-2.0

---

Built by Votee AI