Nanoresearch Experiment

Takes your experiment blueprint JSON and scaffolds out a complete Python research codebase with data loaders, model stubs for your proposed method and baselines, training loops with checkpointing, evaluation harnesses, and config files for ablations. Saves you the tedious hour of setting up boilerplate when you want to go from research plan to runnable code. The generated skeleton is opinionated about structure (separate directories for data, models, training, evaluation) but that's the point. You'll still need to fill in the actual model logic and dataset specifics, but you get a working entry point and all the plumbing connected. Good for moving fast on new experiments without reinventing project structure every time.

GitHub stars: 1.5K
last updated: May 26, 2026

View on GitHub

Install

$ npx -y skills add openraiser/nanoresearch --skill nanoresearch-experiment --agent claude-code

Installs into .claude/skills of the current project.

Files

Experiment Skill

Purpose

Take the experiment blueprint and produce a runnable Python code skeleton that implements the proposed method, baselines, training loops, evaluation harness, and ablation configurations.

Tools Required

None. This skill operates entirely through LLM code generation based on the experiment blueprint.

Input

experiment_blueprint: Path to papers/experiment_blueprint.json produced by the planning skill

Process

Parse the experiment blueprint for datasets, baselines, metrics, and ablation groups
Generate the project directory structure (data loaders, models, training, evaluation, configs)
Produce data loading and preprocessing code for each specified dataset
Implement model architecture stubs for the proposed method and each baseline
Generate training loop with logging, checkpointing, and early stopping
Implement the evaluation harness computing all specified metrics
Create configuration files for each ablation group
Add a main entry point that accepts a config and runs the full train-evaluate pipeline

SKILL.md on GitHub