From one person’s setup to a team-shareable skill

A single SKILL.md on your laptop is L1 — one person has a working setup. The jump to L2 is making that skill shareable so anyone on the team can use it.

A plugin is the packaging format. It is a directory with a manifest:

my-plugin/
  .claude-plugin/
    plugin.json          # manifest: name, version, keywords
  skills/
    my-skill/
      SKILL.md           # skill definition
  scripts/
    my-tool.ts           # backing scripts (if any)

No build step, no compilation, no package manager. Drop it in and Claude picks it up.

Credentials need to be handled so the same skill works across machines. The cascading lookup solves this:

This means you never hardcode secrets, and the same skill works across machines and teams. At L1, one person has a working setup on their laptop. At L2, the skill is shareable — anyone on the team can install it and their own credentials cascade in without touching the skill code.

Read it back

You have a SKILL.md. Read it one more time.

Could a new hire follow it? Could they categorize a transaction correctly using only what is in this document — without calling you to ask?

Could Claude follow it? If you gave this document to an AI system along with an invoice, would the system know what to do?

If the answer to either question is no, the document is not done yet. Add what is missing.

Sarah read her Cornwall Market skill back and immediately spotted three gaps:

1. Chen’s Produce sometimes sends invoices under the name “CP Farms” — same vendor, different name on the paperwork. A new hire would not know that.
2. The bakery suppliers sometimes include delivery surcharges that go under 6100 (Operations), not the ingredient account — she forgot to mention that.
3. When Pacific Foods does a substitution (ships a different brand than ordered), you call them before paying the invoice. That rule existed nowhere except in Sarah’s head.

She added all three. The skill got better. It will keep getting better — every time someone encounters a case the skill does not cover, they add it. This is the beginning of the feedback loop that becomes central in Chapter 09.

Look at what you have done in terms of the distillation pipeline:

1. Conversation (Chapter 02) — you dictated knowledge to Claude
2. Project (this chapter) — the knowledge became standing context for every chat
3. SKILL.md (this chapter) — you formalized it into a shareable skill

The next steps in the pipeline happen in later chapters. In Chapter 06, the skill becomes operational in Claude Code — Claude uses it automatically. In Chapter 07, you will expose the knowledge as an MCP server so it is available on every surface. In Chapter 09, the agent will update the skill itself when you make corrections — the feedback loop closes.

That is the first transition. Not a technology milestone — an organizational one. The knowledge that lived in one person’s head now lives in a document that both humans and machines can follow. Everything else in this course builds on that foundation.

Check your understanding

Answer in your own words — write it down before opening the key. Your tutor grades against the criteria and generates fresh variants on retries.

Q1. A teammate’s skill works perfectly on their laptop and fails on yours with an authentication error. What design principle was violated, and how does the cascading credential pattern fix it?

Q2. State the new-hire test for a knowledge document, and explain why a document that fails it for humans will also fail for Claude — what do the two readers have in common?

Next: The Trust Inversion