Key Takeaways
- AI Research Skills by Orchestra Research provides 86 skills across 22 categories for autonomous AI research
- Compatible with Claude Code, Codex, Gemini, and any agent supporting the
SKILL.mdstandard - 5K+ GitHub stars — the go-to skill library for ML researchers
- The autoresearch meta-skill orchestrates the full loop: ideation → literature review → experiments → paper writing
- One-command install:
npx @orchestra-research/ai-research-skills
What Are AI Research Skills?
AI Research Skills is a library of 86 skills across 22 categories that enables autonomous AI research — from initial idea to published paper.
Core attributes:
- Developer: Orchestra Research
- Skills count: 86 across 22 categories
- Compatibility: Claude Code, Codex, Gemini, any
SKILL.md-compatible agent - GitHub stars: 5K+
- Install:
npx @orchestra-research/ai-research-skills
The 22 Categories
The skills cover the entire AI/ML research lifecycle:
| Stage | Categories |
|---|---|
| Discovery | Autoresearch, Ideation, ML Paper Reading |
| Design | Model Architecture, Experiment Design |
| Build | Fine-Tuning, Post-Training, Distributed Training |
| Optimize | Optimization, Inference, Quantization |
| Evaluate | Benchmarking, Evaluation, Ablation Studies |
| Apply | Agents, RAG, Multimodal, Tool Use |
| Understand | Mechanistic Interpretability, Emerging Techniques |
| Govern | Safety & Alignment, Ethics |
| Write | ML Paper Writing, Scientific Communication |
The Autoresearch Skill
The standout feature is the autoresearch skill — a meta-skill that orchestrates the full research loop:
- Idea Generation — brainstorm research questions from a topic
- Literature Review — search and summarize related work
- Experiment Design — plan the methodology
- Implementation — write and run code
- Analysis — interpret results, generate plots
- Paper Writing — draft the manuscript with proper structure
Trigger with a single prompt:
Use the autoresearch skill to investigate whether sparse attention patterns
in vision transformers can match dense attention on ImageNet while reducing
FLOPs by 50%.The agent works through the full loop, checking in with you at each stage.
Quick Start
One-Command Install
npx @orchestra-research/ai-research-skillsThis copies all 86 skills to your local .claude/skills/ directory (or equivalent for your agent).
Manual Install
git clone https://github.com/Orchestra-Research/AI-Research-SKILLs.git
cp -r AI-Research-SKILLs/skills/* ~/.claude/skills/Creating Your Own Skills
Orchestra provides a skill creation guide and template:
cat skills/docs/SKILL_CREATION_GUIDE.md
cat skills/docs/SKILL_TEMPLATE.mdEach skill is a Markdown file with:
- Description — what it does
- Instructions — step-by-step agent behavior
- Tools — required APIs/libraries
- Examples — sample interactions
Who Should Use This?
| User Type | Use Case |
|---|---|
| ML researchers | Automate literature reviews and experiment iterations |
| PhD students | Structure papers and run ablation studies |
| Research engineers | Build reproducible training pipelines |
| Any developer | Give coding agents deep ML research understanding |
FAQ
Q1: Does this work with agents other than Claude Code?
Yes. Any agent that supports the SKILL.md standard can use these skills — including Codex, Gemini, and custom agents. The skills are plain Markdown files.
Q2: Can I use individual skills without the full library?
Yes. Each skill is a standalone Markdown file. Copy only the ones you need to your agent's skills directory.
Q3: How do I contribute a new skill?
Follow the SKILL_CREATION_GUIDE.md template. Submit a PR to the GitHub repository with your skill in the appropriate category folder.
Q4: What's the difference between this and K-Dense Scientific Skills?
Orchestra focuses on ML/AI research (model training, benchmarking, paper writing). K-Dense focuses on domain-specific science (bioinformatics, genomics, clinical research). They're complementary — many researchers install both.
Q5: Does autoresearch actually produce publishable papers?
It produces well-structured drafts with literature review, methodology, and results sections. Human review and refinement are still needed for publication quality, but it handles 60–80% of the mechanical work.
Summary
AI Research Skills by Orchestra Research is the most comprehensive skill library for ML research automation. With 86 skills across 22 categories, it covers the full lifecycle from ideation to paper writing. The autoresearch meta-skill is its standout feature — turning a single research question into a structured paper draft.