paper-code-finderlisted

# Paper-Code-Finder: Code Discovery for Academic Papers This skill locates the official or unofficial source code implementations for academic papers, utilizing highly efficient search strategies specifically optimized for the AI/ML ecosystem. ## 1. Input Processing & Entity Extraction First, analyze the user's input to extract key metadata: - **Title, Authors, Affiliations**: Extract these from the prompt, PDF, arXiv link, or DOI. - **ML Framework Semantic Extraction**: If a PDF or abstract is provided, scan the "Experiments" or "Implementation Details" section for keywords like `PyTorch`, `JAX`, `Flax`, `TensorFlow`, or `Diffusers`. Use this to narrow down your search queries. ## 2. Search Strategy (The Waterfall) Execute a structured, sequential search to find the code. Progress through these steps sequentially until you find a match. **For highest precision, explicitly use the `exa-search` or `tavily-web` skills/tools if they are available to you**. ### Phase 1: The Fast-Path (PapersWithCode) 1. Search `"[Title] paperswithcode"`. This is the most reliable database. ### Phase 2: The Hugging Face & Mega-Repo Hunter 1. Search `"[Title] site:huggingface.co/papers"` and `"[Title] site:huggingface.co"`. 2. If the paper is about foundational models, search for pull requests in mega-repos: `"[Title] huggingface/transformers github"` or `"[Title] huggingface/diffusers github"`. ### Phase 3: The Deep-Path (Author Profile Hunting) If the above fail, the repo likely has an obsc