A collection of modular AI agent skills for automating academic research workflows — from paper discovery to reproduction.
Each skill is a self-contained tool that can be used:
| Skill | Description | Status |
|---|---|---|
| paper-finder | Search for papers & source code across arXiv, Semantic Scholar, HuggingFace, GitHub | ✅ Ready |
| paper-downloader | Download paper PDFs and clone source code repos | ✅ Ready |
| paper-parser | Parse PDF papers into structured JSON (via MinerU) | ✅ Ready |
| code-analyzer | Deep AST analysis, training loop dissection, reproducibility scoring | ✅ Ready |
| formula2code | Convert LaTeX formulas to PyTorch/NumPy code (15 ML patterns + SymPy pipeline) | ✅ Ready |
| code-writer | Generate project scaffolding from paper descriptions (no source code needed) | ✅ Ready |
| paper-presenter | Generate Beamer presentations summarizing papers | ✅ Ready |
| beamer-skill | Academic Beamer LaTeX presentation lifecycle tool | ✅ Ready (bundled) |
| code-reproducer | Execute reproduction on remote GPU servers via mcp-ssh | ✅ Ready |
| result-analyzer | Compare reproduced results against paper figures/metrics. Reports in MD/JSON/Beamer | ✅ Ready |
git clone https://github.com/orange4664/research-skills.git
cd research-skills
# Use paper-finder
cd paper-finder
pip install -r requirements.txt
python search_paper.py "Attention Is All You Need"
┌──────────────┐ ┌──────────────────┐ ┌──────────────┐
│ paper-finder │───▶│ paper-downloader │───▶│ paper-parser │
│ (search) │ │ (download) │ │ (PDF → JSON) │
└──────────────┘ └──────────────────┘ └──────┬───────┘
│
┌──────────────────┐ ┌───────▼───────┐ ┌──────────────┐
│ code-analyzer │◀───│paper-presenter│───▶│ beamer-skill │
│ (AST/ML scoring) │ │ (summarize) │ │ (LaTeX PPT) │
└────────┬─────────┘ └───────────────┘ └──────────────┘
│
┌──────────────────┼──────────────────┐
│ │ │
┌────────▼─────────┐ ┌──────▼───────┐ ┌────────▼─────────┐
│ formula2code │ │ code-writer │ │ code-reproducer │
│ (LaTeX → code) │ │ (scaffold) │ │ (train via SSH) │
└────────┬─────────┘ └──────┬───────┘ └────────┬─────────┘
│ │ │
└──────────────────┼──────────────────┘
┌────────▼─────────┐
│ result-analyzer │
│ (compare) │
└──────────────────┘
MIT License — see LICENSE
| Component | Author | License | Link |
|---|---|---|---|
| beamer-skill | Noi1r | MIT | GitHub |
| latex2sympy2 | HuggingFace | MIT | GitHub |
| sympytorch | patrick-kidger | Apache-2.0 | GitHub |
| Reproducibility scoring | Papers With Code | — | ML Code Completeness |
| AST analysis approach | PyCG (ICSE'21) | Apache-2.0 | Paper |
| LaTeX paper writing | latex-paper-skills | MIT | GitHub |
14 commits
Python
72.8%
TeX
25.5%
Jinja
1.6%
A collection of modular AI agent skills for automating academic research workflows — from paper discovery to reproduction.
Each skill is a self-contained tool that can be used:
| Skill | Description | Status |
|---|---|---|
| paper-finder | Search for papers & source code across arXiv, Semantic Scholar, HuggingFace, GitHub | ✅ Ready |
| paper-downloader | Download paper PDFs and clone source code repos | ✅ Ready |
| paper-parser | Parse PDF papers into structured JSON (via MinerU) | ✅ Ready |
| code-analyzer | Deep AST analysis, training loop dissection, reproducibility scoring | ✅ Ready |
| formula2code | Convert LaTeX formulas to PyTorch/NumPy code (15 ML patterns + SymPy pipeline) | ✅ Ready |
| code-writer | Generate project scaffolding from paper descriptions (no source code needed) | ✅ Ready |
| paper-presenter | Generate Beamer presentations summarizing papers | ✅ Ready |
| beamer-skill | Academic Beamer LaTeX presentation lifecycle tool | ✅ Ready (bundled) |
| code-reproducer | Execute reproduction on remote GPU servers via mcp-ssh | ✅ Ready |
| result-analyzer | Compare reproduced results against paper figures/metrics. Reports in MD/JSON/Beamer | ✅ Ready |
git clone https://github.com/orange4664/research-skills.git
cd research-skills
# Use paper-finder
cd paper-finder
pip install -r requirements.txt
python search_paper.py "Attention Is All You Need"
┌──────────────┐ ┌──────────────────┐ ┌──────────────┐
│ paper-finder │───▶│ paper-downloader │───▶│ paper-parser │
│ (search) │ │ (download) │ │ (PDF → JSON) │
└──────────────┘ └──────────────────┘ └──────┬───────┘
│
┌──────────────────┐ ┌───────▼───────┐ ┌──────────────┐
│ code-analyzer │◀───│paper-presenter│───▶│ beamer-skill │
│ (AST/ML scoring) │ │ (summarize) │ │ (LaTeX PPT) │
└────────┬─────────┘ └───────────────┘ └──────────────┘
│
┌──────────────────┼──────────────────┐
│ │ │
┌────────▼─────────┐ ┌──────▼───────┐ ┌────────▼─────────┐
│ formula2code │ │ code-writer │ │ code-reproducer │
│ (LaTeX → code) │ │ (scaffold) │ │ (train via SSH) │
└────────┬─────────┘ └──────┬───────┘ └────────┬─────────┘
│ │ │
└──────────────────┼──────────────────┘
┌────────▼─────────┐
│ result-analyzer │
│ (compare) │
└──────────────────┘
MIT License — see LICENSE
| Component | Author | License | Link |
|---|---|---|---|
| beamer-skill | Noi1r | MIT | GitHub |
| latex2sympy2 | HuggingFace | MIT | GitHub |
| sympytorch | patrick-kidger | Apache-2.0 | GitHub |
| Reproducibility scoring | Papers With Code | — | ML Code Completeness |
| AST analysis approach | PyCG (ICSE'21) | Apache-2.0 | Paper |
| LaTeX paper writing | latex-paper-skills | MIT | GitHub |
14 commits
Python
72.8%
TeX
25.5%
Jinja
1.6%