Prompt Engineering
From prompt to skill
When a repeatable prompt should become an Agent Skill — and how to make the jump.
Every technique in this section points the same direction: once a prompt reliably produces the outcome you want, re-typing it every session is waste. That's the moment it should become an Agent Skill.
The signals it's time
- You've re-explained the same workflow more than a couple of times.
- You keep a growing set of examples or a paragraph of standing instructions you paste in.
- The task is repeatable and matters — review, debugging, migrations — where "the agent improvised differently today" is a real cost.
What packaging adds
A skill turns that prose into something the agent loads automatically when it's relevant,
across every client — no paste, no drift. The
description becomes the trigger; the body becomes the
procedure.
Make the jump
- Write the prompt down as a
SKILL.md— anatomy here. - Validate it so it actually loads.
- If it's high-stakes, eval it so "it works" is a number, not a hope.
This is the whole throughline
A repeatable prompt, hardened by evals, distributed as a skill. The free set is three prompts that made exactly this jump — start there.