Pinned Loading

1. when-rubrics-fail when-rubrics-fail Public

   When Rubrics Fail: clinician-validated study of hallucination blind spots in HealthBench, HealthBench Professional, and LiveMedBench rubrics

   Python 2

2. experiment-hidden-agenda experiment-hidden-agenda Public

   An experiment to observe a backdoored coding agent's behaviour in a simulated environment

   Jupyter Notebook

3. model-security-scanner model-security-scanner Public

   A lightweight CLI tool to assess NLP model robustness via fast, low-resource adversarial attacks

   Python

4. prompt-shield-distilguard prompt-shield-distilguard Public

   A lightweight DistilBERT + LoRA model that detects and filters prompt injection attempts before forwarding input to large language models (e.g. DeepSeek).

   Jupyter Notebook

5. claude-token-saver claude-token-saver Public

   A Claude Code hook that uses a local Ollama model to refine chosen prompts before sending them to Claude to make them token-efficient and provides token saving tips.

   JavaScript

6. sanityml sanityml Public

   sanityML — A quick sanity check for ML projects. Scans Python code, Jupyter notebooks, dependencies, and serialized models for vulnerabilities.

   Python 1