| name | github-codebase-search |
| description | Use when searching for open-source implementations, code repositories, or reference codebases on GitHub. Triggers when user asks to find code, implementations, libraries, or projects related to a technical topic. |
GitHub Codebase Search
Search GitHub for relevant code repositories and implementations.
Search Strategy
1. GitHub Search URL
https://github.com/search?q=QUERY&type=repositories&sort=stars&order=desc
Search by topic tag:
https://github.com/topics/TOPIC-NAME
Search with filters:
https://github.com/search?q=QUERY+language:python+stars:>100&type=repositories
2. Query Construction
Build 3 query types:
| Type | Example |
|---|
| Exact name | "diffusion model" |
| Task + framework | transformer pytorch implementation |
| Paper title | "Attention Is All You Need" code |
3. Repository Evaluation Criteria
For each repo, assess:
| Signal | Threshold |
|---|
| Stars | >100 = notable, >1000 = popular |
| Last commit | <1 year = active |
| README quality | Clear install + usage |
| License | Check for commercial use |
| Issues/PRs | Active community |
4. Cloning the Repository
When you need the full codebase (not just a few files), clone it:
git clone https://github.com/owner/repo.git
For large repos, use shallow clone to save time:
git clone --depth 1 https://github.com/owner/repo.git
5. Key Files to Inspect
After cloning or fetching a repo, check these paths:
README.md — overview, installation, usagerequirements.txt / setup.py / pyproject.toml — dependencies- Key source files matching the topic
Output Format
## GitHub Codebases: [Topic]
### Repo 1: [owner/repo-name](https://github.com/owner/repo)
- **Stars:** 3.2k | **Language:** Python | **Updated:** 2024-11
- **Description:** [What it implements]
- **Key files:** `src/model.py`, `train.py`
- **Usage:** `pip install X` then `python train.py`
- **Relevance:** [Why useful for the task]
### Repo 2: ...
Search Tips
- Use
Papers With Code links — papers often link to official implementations
- Search
awesome-TOPIC repos for curated lists
- Check org accounts (e.g., huggingface, facebookresearch, google-research) directly
- Look at forks of popular repos for improvements/variants
Common Mistakes
- Only checking star count — a 50-star specialized repo may be more relevant
- Skipping README — always verify the repo solves the right problem
- Missing official implementations — check paper authors' GitHub profiles
