Ai Incident Response
Use when an AI system behaves unexpectedly, produces harmful or biased outputs, degrades in performance, is subject to a complaint, or when a potential AI-related incident is identified.
Alignment Review
Use when about to mark any AI feature, model, or system complete, or when a PR is ready to merge. This is the final gate — run before every AI feature ships.
Governance Documentation
Use when preparing an AI system for production deployment, or when asked to document an AI system. Run before go-live — never after.
Bias Assessment
Use when loading datasets, training ML models, evaluating model performance, or preparing data for AI systems. Do NOT skip this for "small" or "simple" models.
Responsible Data Handling
Use when accessing, loading, processing, or storing any dataset — especially one containing personal, sensitive, or demographic data. Run before any data pipeline code.
Fairness Testing
Use when writing tests for any ML model, classifier, or AI system that produces outputs affecting people. Fairness tests must be written before the model is used in production.
Human Oversight Design
Use when designing autonomous AI agents, agentic pipelines, or any system that takes actions without direct user instruction for each step. Must be read before writing any agent orchestration code.
Explainability By Default
Use when building any prediction, classification, recommendation, or scoring system. Explainability is not optional — it is a design requirement.
Getting Started
Use when starting any AI/ML task, or when asked about responsible AI skills. Explains available skills and mandatory workflow.