Awesome Wizard 🧙‍♂️

A curated list of awesome playbooks, sample datasets, community skills, model recipes, and integrations for Wizard — the local-first autonomous data analysis agent.

---

📑 Contents

• 🌟 Core Ecosystem
• 📊 Datasets & Analysis Playbooks
• 🧠 Local Model Recipes
• 🧩 Community Skills Cookbook
• 🛠️ Connectors & Data Tools
• 📚 Research & Architecture
• 🤝 Contributing

---

🌟 Core Ecosystem

• Wizard Core — The autonomous analytical engine, FastAPI backend, Next.js UI, and Go CLI daemon.
• Documentation Site — Official guides, installation walkthroughs, and architecture specifications.
• Skills Registry — Centralized community registry for SKILL.md analytical instructions.
• Homebrew Tap — Official macOS & Linux package installer (brew install Wizard-AIA/wizard/wizard).

---

📊 Datasets & Analysis Playbooks

Step-by-step analytical investigations you can run with Wizard out of the box:

1. SaaS Customer Churn & Cohort Retention

• Dataset: workspace/dataset.csv (10,000 subscriber accounts, MRR, usage metrics, support tickets).
• Prompt: "Analyze which subscriber tiers had the highest churn in Q4, decompose the root drivers by ticket volume, and plot a retention matrix."
• Techniques Used: Kaplan-Meier survival curves, cohort monthly indexing, logistic driver regression.

2. Housing Valuation & Non-Linear Price Modeling

• Dataset: workspace/housing.csv (20,640 census block records, median income, geographical coordinates).
• Prompt: "Train a predictive model for median house value, test for spatial autocorrelation, and identify the top 3 non-linear feature interactions."
• Techniques Used: Random Forest regression, SHAP interaction values, Moran's I spatial diagnostics.

3. Financial Transaction Anomaly Detection

• Techniques Used: Isolation Forest, Mahalanobis distance for multivariate outliers, IQR thresholding.
• Skill Applied: outlier-detection.

---

🧠 Local Model Recipes

Wizard runs with zero cloud API keys required. Recommended local pairings:

Hardware Tier	Memory	Manager (Reasoning)	Worker (Python Code)	Embedding Model
Compact / Laptop	8 GB – 16 GB	qwen2.5:3b / llama3.2:3b	qwen2.5-coder:1.5b / 3b	nomic-embed-text
Standard (Recommended)	16 GB – 32 GB	qwen2.5:7b / mistral:7b	qwen2.5-coder:7b	embeddinggemma
Power User / Studio	32 GB – 64 GB+	qwen2.5:14b / deepseek-r1:14b	qwen2.5-coder:14b	bge-large-en-v1.5

# Setup standard local stack in one command:
ollama pull qwen2.5:7b
ollama pull qwen2.5-coder:7b
ollama pull embeddinggemma

---

🧩 Community Skills Cookbook

Curated analytical domain instructions you can install directly into Wizard:

• cohort-analysis — Uniform cohort assignment, period age indexing, and retention decay modeling.
• data-quality-triage — Pre-aggregate data hygiene, dirty join key detection, and null variance bounds.
• outlier-detection — Distribution-aware anomaly detection (IQR, Z-score, Isolation Forest).
• time-series-forecasting — Seasonal decomposition, naive benchmarking, and out-of-sample backtesting.

# Install all skills into your local Wizard:
./cli/wizard skills add Wizard-AIA/skills --yes

---

🛠️ Connectors & Data Tools

• DuckDB — Embedded analytical SQL engine executed across memory DataFrames.
• Polars — Multi-threaded high-throughput DataFrame library in Rust/Python.
• Apache Arrow — Zero-copy binary IPC streaming between backend and browser.
• Plotly — Interactive client-rendered visualization suite.

---

📚 Research & Architecture

Key concepts underlying Wizard's autonomous design:

• Evidence-Backed Control Plane: Synthesizing answers exclusively from real execution stdout rather than LLM token hallucinations.
• ReAct with Self-Correction: Feeding Python tracebacks and static analyzer failures back into the manager model for automatic recovery.
• OS-Native Subprocess Sandboxing: Secure execution via Linux Landlock/seccomp, macOS sandbox-exec, and Windows Job Objects.

---

🤝 Contributing

Contributions are welcome! Please read CONTRIBUTING.md for submission guidelines.