We use AI every day, but let’s be real: many of us are tired of seeing the same patterns repackaged as innovation. Summarize this. Draft that. Add an agent to a workflow. For security practitioners, the harder question is how to build agent workflows that are useful, predictable, and secure enough to trust.

 

This session focuses on six practical best practices for building precise, secure AI skills. Skills package instructions, domain expertise, decision logic, references, scripts, and expected outputs into reusable workflows. Skill design directly affects how much discretion an agent has, which resources it can access, and how it responds when the expected path breaks down. 

 

We’ll put the practices into action through a security review demo, comparing a loosely defined request with a skill built around explicit scope, an approved methodology, constrained access, evidence-based output, and clear approval points for consequential actions. 

 

The session closes by looking at the administrator side of securing AI at enterprise scale. Attendees leave with best practices for designing and reviewing secure AI skills, plus a model they can explore further for observing, governing, and securing agents across the enterprise.