Gardener - Garden-AI Publishing Assistant for Claude Code

A Claude Code skill that helps researchers publish scientific machine learning models on Garden-AI.

What is this?

Gardener is a specialized skill for Claude Code that guides you from a research paper and code repository to a fully deployed, citable function on Garden-AI. Think of it as a pair programmer that:

• Analyzes your research paper to understand your model's purpose
• Explores your code repository to find the inference logic
• Designs domain-appropriate Python APIs for your model
• Generates deployment code for Modal (cloud) or HPC systems
• Tests the generated code and helps you debug it
• Guides you through publishing to Garden-AI

What is Garden-AI?

Garden-AI is a FAIR AI/ML model publishing platform where:

• Published models get DOIs for academic citation
• Users run your models remotely via simple Python APIs
• Models can execute on serverless cloud (Modal) or HPC clusters (via Globus Compute)
• No installation or environment setup required for users

Installation

Prerequisites

• Claude Code CLI installed and authenticated

Install Gardener

# Start Claude Code
claude

# In Claude Code:
/plugin marketplace add Garden-AI/gardener
/plugin install gardener

Usage

Once installed, use the /gardener command to start the workflow, or simply mention "help me publish on Garden-AI" in your conversation.

Example Session

You: /gardener

     I want to publish my protein structure prediction model on Garden-AI.
     Here's the paper PDF: paper.pdf
     And the GitHub repo: https://github.com/user/protein-model

Claude: I'll help you publish this model on Garden-AI. Let me start by
        analyzing your paper to understand the model's purpose...

        [Gardener guides you through 9 phases:]
        1. Gather paper + code
        2. Analyze paper (CHECKPOINT: validates understanding)
        3. Explore repository
        4. Synthesize understanding (CHECKPOINT: confirms model details)
        5. Design API (CHECKPOINT: approves function signatures)
        6. Choose deployment platform (CHECKPOINT: Modal vs HPC)
        7. Generate deployment code (CHECKPOINT: reviews generated code)
        8. Test & debug iteratively
        9. Guide publication to Garden-AI

You: Looks good, let's test it!

Claude: I'll run the code now and we can fix any issues together...

What Makes a Good Use Case?

Great for Gardener:

• You have a paper + code repository for an ML model
• The model performs inference/prediction (not training)
• You want domain scientists to easily use your model
• You want to publish on Garden-AI with a DOI

Not ideal:

• Pure data processing scripts (not ML models)
• Training code (Garden-AI is for inference)
• Code already in Garden-AI format

Key Features

Scientifically-Aware API Design

Gardener designs APIs based on your domain, not ML implementation details:

# ✅ Gardener designs this
def predict_stability(protein_sequences: list[str]) -> list[float]:
    """Predict thermodynamic stability for protein sequences."""

# ❌ Not this
def run_model(inputs: np.ndarray, batch_size: int = 32, hidden_dim: int = 512):
    """Run the neural network."""

Platform-Appropriate Code Generation

Gardener generates correct code for your deployment target:

• Modal - For fast inference, single GPU, standard packages
• groundhog_hpc - For long computations, multi-GPU, HPC libraries

Each platform has different requirements (imports, decorators, calling conventions), and Gardener handles these correctly.

Checkpoint-Driven Workflow

Gardener validates its understanding with you at 5 critical checkpoints before generating code. This ensures the final API matches your scientific intent.

Iterative Testing

Gardener doesn't just write code and declare victory. It runs the code, debugs errors with you, and iterates until it works.

Documentation

• Garden-AI Documentation: garden-ai.readthedocs.io
• Skill Technical Docs: See skills/gardener/README.md for developer documentation
• Claude Code Help: Run /help in Claude Code

Contributing

For developers working on the skill itself:

• CLAUDE.md - Development guide for this repository
• skills/gardener/README.md - Technical documentation for the skill
• skills/gardener/SKILL.md - Skill entry point and overview

Support

• Issues: Report bugs or request features at github.com/Garden-AI/gardener/issues
• Garden-AI Platform: thegardens.ai

License

This project is licensed under the MIT License - see the LICENSE file for details.