Skip to content

ricet logo

ricet

Automate scientific research using Claude Code with structured skills, persistent knowledge, and overnight autonomous execution.

Created by Luca Fusar Bassini.

ricet is a CLI tool and framework that manages the full lifecycle of scientific research projects. It pairs Claude Code with eight research skills (slash commands), a persistent knowledge system that compounds insights across sessions, reproducibility enforcement, and a complete paper pipeline -- so you can focus on the science while automation handles the scaffolding.

Why ricet?

Running a research project involves dozens of repetitive tasks: environment setup, literature searches, experiment tracking, figure generation, paper writing, and more. ricet provides a single ricet command that orchestrates all of these through structured AI skills operating inside a safe, containerized environment.

Problem Solution
Ad-hoc experiment tracking Lab/stable bipartition with promotion gates and provenance tracking
Scattered knowledge Persistent encyclopedia, rules, and decision log that grow automatically
Tedious boilerplate Project templates with skills, hooks, LaTeX, and CI/CD out of the box
Unsafe autonomous runs Docker sandbox with permission model and auto-backup
Manual paper formatting Integrated LaTeX pipeline with colorblind-safe figures and citation management
Presentation preparation AI-generated slide decks with Nano Banana Pro schematics
Lost insights between sessions Meta-learn hook auto-captures rules, insights, and decisions from every prompt
No remote monitoring Mobile access via Tailscale or Cloudflare Tunnel

Quick Start

# Install Claude Code (requires Node.js 20+)
npm install -g @anthropic-ai/claude-code

# Authenticate with Claude subscription (Pro or Team required, no API key needed)
claude auth login

# Clone the repository
git clone https://github.com/lucafusarbassini/research-automation
cd research-automation

# Install the CLI
pip install -e .

# Create your first project (interactive wizard)
ricet init my-project

# Edit your research goal
cd my-project
$EDITOR knowledge/GOAL.md

# Start an interactive session
ricet start

# Or run overnight
ricet overnight --iterations 20

The ricet init wizard auto-detects your system (GPU, conda, Docker), installs the full toolchain (uv, tectonic, biber, screen, Tailscale), walks you through notification and credential setup, deploys 8 research skills, registers the meta-learn hook, and optionally creates a GitHub repository. See the full Quickstart Tutorial for a step-by-step walkthrough.

Feature Highlights

  • 8 Research Skills -- Slash commands (/lit-review, /experiment-review, /paper-draft, /falsify, /reproduce, /research-retro, /overnight, /slides) give Claude structured workflows for every research task.
  • Persistent Knowledge System -- RULES.md (behavioral), ENCYCLOPEDIA.md (domain), DECISION_LOG.md (architectural) -- all auto-populated by the meta-learn hook from your interactions.
  • Lab/Stable Bipartition -- Experimental code in lab/, promoted to stable/ via ricet promote with provenance tracking (git hash, timestamp, metrics).
  • Code Indexing & Search -- ricet index-code extracts function/class signatures; ricet search-code does semantic search over the index.
  • Feature Request Pipeline -- ricet feature-request logs ideas; ricet implement-features builds selected ones in parallel worktrees.
  • Collaborative Research -- Multiple researchers on the same repo with personal branches, ricet morning-sync merges, and user attribution.
  • Adopt Existing Repos -- ricet adopt transforms any GitHub repo into a ricet project with fork, scaffold, and personal branch.
  • Sandbox Infrastructure -- ricet sandbox manages a fully isolated Docker sandbox with auto-backup and work extraction.
  • Slide Deck Generation -- ricet slides creates .pptx decks with AI-generated schematics via Nano Banana Pro (Gemini 3 Pro).
  • Mobile Access -- Tailscale (default) or Cloudflare Tunnel for secure access from any device. Screen sessions for persistence.
  • Overnight Mode -- Autonomous execution loop with auto-debug, resource monitoring, and recovery.
  • Paper Pipeline -- LaTeX template, BibTeX management, tectonic compilation, style transfer.
  • gstack Integration -- ricet gstack install adds Garry Tan's startup workflow skills alongside ricet's research skills.
  • Voice Prompting -- ricet voice transcribes audio in 30+ languages and executes the prompt.
  • Cascading Self-Update -- When ricet is updated, existing projects get refreshed skills and defaults automatically.
  • Cross-Repository Code Search -- Link external repos with ricet link so Claude can search across all your code.
  • context-hub -- Versioned API documentation via ricet chub.
  • MCP Discovery -- ricet mcp-search searches 1300+ MCP servers and installs on demand.
  • Auto-Commit & Push -- Every state-modifying command automatically commits and pushes.
  • Global Credential Store -- ~/.ricet/credentials.env stores API keys once across all projects.
  • Interactive Dashboard -- ricet dashboard Rich TUI with live progress and resource monitoring.
  • Zenodo Publishing -- ricet zenodo publishes software, datasets, and papers with permanent DOIs.

Explore all features in the Features page.

Project Philosophy

ricet is built on six core principles:

  1. Never please the user -- Be objective, challenge assumptions, report flaws.
  2. Popperian falsification -- Try to break results, not validate them.
  3. Never guess -- Search or ask when uncertain.
  4. Test small, then scale -- Downsample first, run one epoch, then scale up.
  5. Commit aggressively -- Meaningful commits after every subtask.
  6. Accumulate knowledge -- The encyclopedia grows with every task.

Project Status

ricet is under active development. The core modules, CLI, templates, and skill system are implemented. Contributions and feedback are welcome.

Component Status
CLI (ricet command, 55+ commands) Implemented
Core modules (50+ modules) Implemented
Research skills (8 slash commands) Implemented
Persistent knowledge system Implemented
Meta-learn hook (auto-capture) Implemented
Lab/stable bipartition Implemented
Paper pipeline Implemented
Slide deck generation Implemented
Docker sandbox infrastructure Implemented
claude-flow integration Implemented (optional)
gstack integration Implemented
Mobile access (Tailscale/Cloudflare) Implemented
Collaborative research workflow Implemented
Cascading self-update Implemented
Code indexing & search Implemented
Feature request pipeline Implemented
Reproducibility pipeline (/reproduce) Under development
Manual curation of MD skills Under development
Documentation site You are here

About

ricet is designed and maintained by Luca Fusar Bassini.

License

MIT