Live site: https://leililab.github.io/susvibes-leaderboard/
- Node.js (version 16 or higher)
- npm (comes with Node.js)
-
Navigate to the leaderboard directory
cd susvibes-leaderboard -
Install dependencies
npm install
-
Start the development server
npm run dev
-
Open your browser
- Navigate to
http://localhost:5173(or the URL shown in your terminal) - The application will automatically reload when you make changes
- Navigate to
We welcome community submissions. Results are added through pull requests. At a high level:
- Evaluate your model with SusVibes.
- Assemble a submission directory — a
submission.json(metadata + scores) plus atrajectories/folder with your run's trajectory and summary files. - Open a pull request adding that directory under
public/submissions/and listing it inmanifest.json.
📋 Follow docs/SUBMISSION_GUIDE.md for the full step-by-step process (standard vs custom submissions, naming, the review checklist).
| Doc | What it covers |
|---|---|
| docs/SUBMISSION_GUIDE.md | How to submit results via pull request (standard & custom), step by step. |
| docs/TRAJECTORY_FORMAT.md | The exact trajectories/ file format — trials.json + summary.json (the single source of truth). |
src/
├── components/ # React components
│ ├── Leaderboard.jsx # model performance leaderboard
│ ├── TrajectoryVisualizer.jsx # trajectory / dataset explorer
│ ├── BlogContent.jsx # interactive paper/blog page
│ └── ... # Carousel, ExploitViewer, VulnerabilityGame, etc.
├── assets/ # logos, css
├── App.jsx # app shell + hash routing (home/blog/leaderboard/visualizer)
└── main.jsx # entry point
public/
├── submissions/ # all submissions (loaded via manifest.json)
│ ├── manifest.json # list of displayed submission directories
│ ├── schema.json # submission.json JSON schema
│ └── <DIR>/ # submission.json + trajectories/
└── datasets/ # susvibes_dataset_<version>.jsonl (per dataset version)
docs/ # SUBMISSION_GUIDE.md, TRAJECTORY_FORMAT.md
This project is licensed under the MIT License - see the LICENSE file for details.
This is a modified version of tau2-bench by Sierra Research.
We thank the open-source community for providing the diverse codebases used in our benchmark tasks. We also thank the Sierra Research for their work on tau2-bench.