getting-started

Responsible AI Skills — Getting Started

You have a suite of Responsible AI Skills. These are mandatory workflows, not suggestions.

What These Skills Do

They ensure every AI system you build is:

• Explainable — decisions can be understood and justified
• Fair — performance is equitable across population groups
• Governed — documentation exists for audit and accountability
• Aligned — behaviour matches stated intent and human values
• Safe — incidents have a response plan before they happen

Available Skills

Skill	When It Activates
bias-assessment	Loading datasets, training models, evaluating model performance
fairness-testing	Writing tests for any ML model or classifier
explainability-by-default	Building prediction, classification, or recommendation systems
governance-documentation	Before deploying any AI system to production
responsible-data-handling	Accessing, loading, or processing datasets
human-oversight-design	Designing autonomous or agentic AI systems
ai-incident-response	AI system behaves unexpectedly, produces harmful output, or underperforms
alignment-review	Before marking any AI feature complete

Mandatory Rules

1. Before building any AI feature — check if a skill applies. If it does, read and follow it.
2. Before calling any AI task "done" — run alignment-review.
3. Before deploying to production — run governance-documentation.
4. You cannot skip these steps by claiming time pressure or simplicity.

How Skills Work

Skills load progressively — this file uses ~100 tokens. Full skill content loads only when relevant. This means you can have all skills available without context cost.

Quick Reference

Building a classifier?        → bias-assessment + fairness-testing + explainability-by-default
Loading training data?        → responsible-data-handling
Deploying to production?      → governance-documentation
Designing an agent?           → human-oversight-design
Something went wrong?         → ai-incident-response
Finishing any AI feature?     → alignment-review (ALWAYS)