From prompt hell to SKILL.md

We keep trying to make agents smarter by cramming more into the system prompt, until it becomes a slow, costly, and confused mess that struggles to apply the right context at the right time. SKILL.md flips the approach: small, on-demand playbooks with optional reference docs, even bundled scripts that the model loads only when a task actually calls for them.

In this workshop we'll author a Skill from scratch, refactor a bloated prompt into something lean and reusable, and learn to write descriptions specific enough that the agent triggers them reliably. And this goes well beyond code: the same pattern can capture how you scaffold a feature, run your tests, or review a pull request just as easily as non-technical work like drafting a document, preparing a report, or following a team's process. Because it's an open format, the very same Skill travels across the agents you already use instead of locking you to one vendor.

You'll leave this talk able to ship your first SKILL.md file!