Merge pull request #61 from instil/copilot/update-skill-lesson-documentation

docs: add walk-then-document pattern to skill lesson

Author: calebwilson706

tutorial/lessons/11-agent-skills.md

@@ -50,6 +50,37 @@ After that, `/docs` is all you need.
 - A teammate asks "how do I get Copilot to do X?" and you have a saved prompt for it
 - You're about to write a long prompt and think "I've done this before"
 
+### Walk the process first, then capture it as a skill
+
+One of the best patterns for building a skill is to **walk the agent through the process interactively first** — and only create the skill once you're happy with the result.
+
+Here's why this works so well: the first time you do something with Copilot, you're exploring. You iterate, correct course, and refine the approach. Once you reach a result you're satisfied with, you have a proven process — and that's exactly what belongs in a skill.
+
+**The workflow:**
+
+1. **Walk through the process step by step** — guide the agent through each stage of the task in a normal conversation. Correct it when it goes wrong, refine the steps, build up to the result you want.
+2. **Confirm it worked** — once you're happy with the outcome, pause and take stock of what you actually did.
+3. **Ask the agent to document the approach as a skill** — say something like:
+
+   > "That worked well. Now document the steps we just followed as a skill called `<name>` so I can reuse this process in future."
+
+   Copilot will look back at the conversation, distil the successful steps, and write a `SKILL.md` that captures them.
+
+**Example:**
+
+Suppose you've just walked the agent through auditing a React component for accessibility issues — checking ARIA labels, contrast ratios, keyboard navigation, and focus order. It found and fixed several issues. Now say:
+
+> "That process worked great. Create a skill called `a11y-audit` that documents how to audit a React component for accessibility so I can run it on any component in future."
+
+Copilot will turn the conversation into a reusable skill that captures exactly the checks it performed.
+
+**Why this matters:**
+
+- You don't have to write skill instructions speculatively — you capture what already worked
+- The skill reflects your actual codebase and conventions, not a generic template
+- The agent is better at describing a process it just performed than one you describe from memory
+- You end up with a skill that's been validated in practice before it's ever written down
+
 ### SKILL.md format
 
 A skill lives in a named folder with a `SKILL.md` file: