Skip to main content
Back to Webinars
RecordedMay 28, 2026

Building with Liatrio | Agent Skills: Turning AI workflows into reusable team capabilities.

Explore how Agent Skills transform repeatable AI workflows into reusable, installable, and governable team capabilities, and why they are emerging as a more powerful foundation than prompt libraries for scaling AI-native engineering practices across organizations.

Presenters

Damien Storm

Damien Storm

Lead AI Enablement Engineer

Robert Kelly

Robert Kelly

VP of Innovation

Your Team Has Great AI Workflows. They Just Live on One Person's Laptop.

Most teams have at least one person who's figured it out. They've got a multi-agent setup with code review baked in, automated security scans, and they're cranking out quality code at a pace that feels almost unfair. Their teammates know it, too.

But ask them to document it. Ask them to package it so the rest of the team can use it the same way, in a new session, on a different machine. That's where it falls apart. The workflow lives in their head. The prompts are scattered across local files. Someone else on the team is doing the same thing completely differently, and neither approach has made it into any shared system.

This is the problem Robert Kelly, VP of Innovation at Liatrio, and Damien Storm, Lead AI Enablement Engineer, set out to solve in Liatrio's latest AI-native webinar: how do you take the AI workflows that are actually working and make them repeatable at the team level?

Their answer is agent skills.

A skill is just a folder

That's worth repeating, because the name sounds heavier than it is. A skill is a folder. At the center of it is a skill.md file - the name, a description, and the core instructions. That's what the agent reads first to understand what the skill does and when to use it.

From there, you can layer in as much or as little as the workflow needs. A references folder holds deeper documentation, things like CLI tool specs or conventions the agent needs to follow for a specific task. A scripts folder is where you put deterministic code - Python, Node, whatever runs the same way every time so the agent isn't generating the same logic from scratch on every invocation. Templates lock in output shapes: always the same HTML report structure, always the same JSON contract, always the same folder layout for a new repo.

Skills are also portable. Claude Code, Cursor, Codex, Windsurf - they all support them. It's become a standard primitive across the major agent harnesses.

Why this is different from an MCP

MCPs and skills solve different problems, and Damien spent time on this distinction in the webinar because it trips people up.

When you load an MCP, every tool definition it contains gets loaded into the context window. If that MCP has 40 tools, that's 40 descriptions eating into your available context before you've typed a single prompt. Skills don't work that way. Only the name and description get loaded per session. The agent checks those against whatever's happening in the conversation, and if it recognizes a match, it pulls in the rest of the skill on demand. The full instructions, the references, the scripts - none of that loads until it's actually needed.

Damien's phrase for it stuck: skills protect the context window by loading the playbook only when the task calls for it.

Where to start

One skill. Not a full marketplace, not a complete catalog of your team's workflows. One workflow that already works, packaged as a skill, committed to a shared repo, and adopted by the team.

The next step isn't getting better at one-off prompts. Everyone's past that. The next step is writing down how you work so it can be shared, tested, and given to agents and teammates in a way that's actually safe and repeatable.

Related Resources