Week 12 · Lecture outline

Week 12 — Lecture Outline · Claude Cowork II: Skills, Connectors & Artifacts

Using Artificial Intelligence · AI 101 Fall 2026 · Prof. Quinn Fictional sample

Course: Using Artificial Intelligence (AI 101) · Silver Oak University (fictional sample) · Prof. Quinn
Objective covered: Objective 5 — Use Claude Cowork to build agentic workflows — projects, connected folders and file read/write, tasks, skills, connectors (MCP), and live artifacts.
SLOs touched: A (produce high-quality results through agentic workflows) · B (use AI safely with least-privilege thinking and accurate feature knowledge)
Meeting pattern: 2 sessions × 75 min = 150 min. Segment minutes below total ~150; scale to your own pattern.

Week at a Glance


The week's big question	"When AI can reach your email, calendar, files, and apps — what's the right way to set that up, and where do you draw the line?"
By the end of the week, students can…	(1) define and distinguish skill, connector, artifact, and plugin; (2) explain MCP (open standard, Anthropic); (3) use or describe a built-in skill, connect a low-risk connector, or build a live artifact; (4) apply least-privilege thinking; (5) catch and correct an over-claim about Cowork features.
Key vocabulary	skill · SKILL.md · connector · MCP (Model Context Protocol) · connectors directory · artifact (live) · plugin · least privilege · built-in skill · custom skill · permission scope
Materials	slides (Deck 12); Claude Cowork desktop (for the live demo); official Anthropic docs: Use connectors · Use plugins · Connectors Directory FAQ · Extend Claude with skills
Timing note	8 segments, ~150 min total. Session 1 = Segments 1–4 (~75). Session 2 = Segments 5–8 (~75).

Segment 1 — Hook & the Promise (8 min) · Session 1 opens

Hook. Put one question on the slide: "Claude just helped a student by pulling her three most-recent unread emails and her next two calendar events, then drafted a morning briefing. She didn't type or paste anything — Claude retrieved the data automatically. How?" Pause. Take guesses. Most students will say "it's connected to Gmail" — which is exactly right, and that's the opening.

"When an AI can reach your inbox, your calendar, your files, and your Slack messages, two things become true at once: it becomes dramatically more useful — and your privacy decisions become dramatically more consequential. This week we learn exactly how those connections work, what they're called, and how to set them up safely."

The promise: "By Friday you'll be able to build (or design) a real skill, connector, or live artifact in Cowork — and you'll be able to catch a claim about these features that isn't quite right."

Memory hook: "Four things: skill = instruction set. Connector = app link. Artifact = live view. Plugin = bundle of all three."

Segment 2 — Skills: Teaching Claude a Reusable Task (22 min)

Plain language first. A skill is a reusable instruction set — a file called SKILL.md with YAML frontmatter and Markdown instructions — that teaches Claude how to do a specific kind of task. When you or Claude invoke a skill, the contents of that file enter the conversation and Claude follows the instructions. Source: code.claude.com/docs/en/skills.

Built-in skills (demonstrate live):
- Cowork ships with built-in document-format skills: docx (Word), pptx (PowerPoint), xlsx (Excel), pdf. Ask Claude in a Cowork task to "create a Word document summarizing these notes" — and a real .docx file appears in your connected folder.
- Show the result live: open the folder, open the file.

Custom skills:
- You (or your organization) can create a custom skill: write a SKILL.md with frontmatter (name, description) and instructions; drop it in a skills location; Claude adds it to its toolkit.
- Example: a "weekly-meeting-notes" skill that tells Claude exactly what format, what headers, and what summary style to use every time — so you never have to re-explain.

Key misconception to kill: "A skill connects Claude to an external app." No — a skill is a local instruction file. It teaches Claude how to do a task; it doesn't pull external data. (That's what a connector does — next.)

Memory hook: "A skill is a recipe. The connector is the grocery store. Don't mix them up."

Segment 3 — Connectors (MCP): Linking Claude to External Apps (22 min)

Plain language first. A connector links Claude to an external app or data source. Connectors use the Model Context Protocol (MCP) — an open standard created by Anthropic to let AI systems connect to tools and data. Connectors are vetted MCP servers, each running only with the permissions you grant. Source: support.claude.com — Use connectors.

The connectors directory:
- Cowork includes a connectors directory with a large, growing list of available connectors. Examples confirmed in the official docs: Gmail, Google Calendar, Google Drive, Slack, Notion, GitHub, Microsoft 365/Outlook, Linear.
- Each connector's page in the directory shows what it can read and write, and what permissions it requests.
- You can also build custom connectors using remote MCP if you have a service with an MCP endpoint.

Key permission principle (teach this carefully):
- Claude inherits your permissions from the connected service. If you can't access a file in Google Drive without those connector permissions, Claude can't either.
- Least privilege: grant only what a specific task needs. You don't have to connect Gmail and Calendar and Drive all at once — connect one, use it, then decide if you need more.

Live demo (do it on the projector):
- Open Cowork → navigate to the connectors directory → show a connector's detail page (read/write capabilities, permission scope). Do not connect a personal account live — explain what you'd see; students connect their own in the Studio.

Key misconception to kill: "A connector is the same thing as a skill." No — a connector pulls data from an external app; a skill is a local instruction set. Different purposes; different mechanisms.

Note on connector count: the number of available connectors grows frequently. Never quote a specific count — say "a large, growing directory."

Segment 4 — Misconceptions + Quick Interaction (23 min) · Session 1 closes (~75)

Name the misconceptions out loud, then cure each:

❌ "A connector and a skill are the same thing."
✅ Cure: a connector links Claude to an external app (Gmail, GitHub, Notion) via MCP; a skill is a local instruction file (SKILL.md) that teaches Claude how to do a task. Totally different mechanism, totally different purpose.
❌ "An artifact is a static file."
✅ Cure: a live artifact is an interactive view that refreshes with current data from your connectors each time you open it. It persists across tasks and keeps version history. Not a one-off export — it's a living view.
❌ "Plugins are individual skills."
✅ Cure: a plugin bundles skills + connectors + sub-agents into one installable package — a ready-to-go setup. One skill is not a plugin; a plugin assembles several capabilities together.
❌ "MCP is something Claude invented."
✅ Cure: MCP (Model Context Protocol) is an open standard created by Anthropic — it's designed so any AI system, not just Claude, can connect to tools and data through compatible servers.

Interaction — Classification sprint (rapid-fire, ~10 min):
Put 5 scenarios on a slide; students vote A (skill) / B (connector) / C (artifact) / D (plugin):
1. "I installed a 'Legal Research' package that gave me Westlaw access and 3 custom instruction sets." → D (plugin)
2. "I created a SKILL.md file that tells Claude how to format my weekly meeting notes." → A (skill)
3. "I authorized Claude to read my GitHub issues." → B (connector)
4. "Claude built me a dashboard that updates my open project tasks from Linear every time I open it." → C (artifact)
5. "I asked Claude to write a Word document from my outline." → A (built-in skill)

Go through each with the correct answer and the one-sentence reason why.

Segment 5 — Live Artifacts: Persistent, Refreshing Views (15 min) · Session 2 opens

Hook back in: "We said an artifact is not a static file. Let's be precise about what that means — because it's the feature most students initially misunderstand."

What a live artifact is (teach from the official doc):
- You ask Claude (within a Cowork project that has connectors set up) to "build me a live view of my open GitHub issues."
- Claude creates an artifact — a persisted, interactive view that refreshes with current data from your connectors each time you open it. It lives in its own place, not buried in a chat thread. It keeps version history. Source: support.claude.com — Use live artifacts in Claude Cowork.
- Contrast: a one-off chat response that lists your issues is just text in a thread. A live artifact is a separate, saved, refreshing view.

When would you use one?
- A dashboard of your open tasks, refreshing from Linear or GitHub.
- A weekly sales summary, refreshing from Salesforce or Notion.
- A morning briefing that pulls from your calendar and Gmail.

The limit to name explicitly: a live artifact can only use the connectors you approved when you created or updated it — it doesn't gain new data sources on its own. This is a safety feature. Source: support.claude.com — Use live artifacts.

Segment 6 — Plugins & the Marketplace (15 min)

Plain language first. A plugin bundles skills + connectors + sub-agents into a single installable package for a particular role or workflow. You install one plugin and get a ready-to-go setup — pre-configured skills, pre-wired connectors, and (sometimes) sub-agents — instead of configuring each piece yourself. Source: support.claude.com — Use plugins in Claude.

Where to find plugins:
- Claude includes built-in marketplaces (e.g., a Knowledge Work marketplace) added by default.
- You can add plugins from other sources (Anthropic-built or from a GitHub repo).
- The Customize menu in the sidebar is where plugins, skills, and connectors are all gathered.
- There is also a unified directory where skills, connectors, and plugins can be browsed together. Source: support.claude.com — Browse skills, connectors, and plugins.

The key analogy: a connector is a single app link; a plugin is a pre-assembled toolkit. "Just as you might install a productivity suite rather than sourcing each app individually, a plugin sets up the whole environment in one go."

What a plugin is NOT: a plugin is not just a skill, and not just a connector. It's a bundle. Confusing "plugin" with "skill" is this week's classic distractor on the quiz.

Segment 7 — Technology Workflow & the Verify-the-AI Moment (12 min)

Technology workflow — safe connector setup, step by step:
1. Open the connectors directory in Cowork.
2. Read the connector's detail page — what can it read? what can it write? what permissions does it request?
3. Grant the minimum permissions the task needs (least privilege).
4. Run a task that uses the connector; review the output.
5. Verify: did Claude actually use the connector data, or did it fill in something from its training? (Ask it to show you the source.)
6. Revoke or narrow permissions if you no longer need them.

AI-critique / verify-the-AI moment (students do this, not watch):

Ask an approved assistant: "What connectors are available in Claude Cowork? Name them all and describe what each can do." The assistant will very likely produce a list — some accurate, some invented, some with overstated capabilities. Compare that list against the official connectors directory at support.claude.com. Which claims were accurate? Which were fabricated? What would a user who trusted the AI uncritically have gotten wrong?

This is a load-bearing exercise: the course's product-accuracy discipline applied live.

Segment 8 — Callback, Tease & Hand-off (13 min) · Session 2 closes (~75)

Callback:
- Skill = instruction set (SKILL.md). Connector = MCP link to an external app, with your granted permissions. Artifact = live, refreshing view built from connector data. Plugin = bundle (skills + connectors + sub-agents).
- MCP is an open standard, created by Anthropic. Connectors only run with the permissions you grant. Least privilege always.

Tease next week:
"We know how to connect apps and build live views. Next week — scheduled tasks and dispatch — we set those connections to work automatically, on a schedule, in the background. And we'll be very specific about one constraint you need to know."

Hand-off (the week's graded work):
- Lecture Tutorial 12 (AI tutor, share-link submission).
- Quiz 12 (no AI) — the four-way distinction.
- Discussion 12 — "Where's the line?" on connectors, privacy, and trust.
- Assignment 12 — four problems distinguishing skill / connector / artifact / plugin.
- AI Build Studio 12 — "Skill, Connector, or Artifact?" — build one (or design it carefully) and catch an over-claim.

Instructor FAQ — Common Stumbles

Student says / does	Quick cure
Confuses skill and connector.	Skill = local instruction file (SKILL.md, teaches a task). Connector = external app link (MCP, pulls data).
Says "artifact" when they mean the output of a task.	A live artifact persists, refreshes from connectors, and has version history. A one-off chat output is not an artifact.
Thinks a plugin is one skill.	A plugin bundles skills + connectors + sub-agents.
Claims connectors give Claude unlimited access.	Connectors run only with the permissions you grant; Claude inherits your own access level in the service.
Wonders "what's MCP exactly?"	MCP = Model Context Protocol — an open standard created by Anthropic for AI systems to connect to tools and data. Not proprietary to Claude alone.
Wants to know "how many connectors are there?"	Say "a large, growing directory" — the count changes frequently; naming a specific number risks being wrong.
Thinks custom skills require coding.	Custom skills are written in Markdown (a SKILL.md file with YAML frontmatter). No code required for basic skills.

Scope flag

This outline covers Objective 5 (Cowork skills, connectors, artifacts, plugins — W12 half). Scheduled tasks, dispatch, and automation are Week 13. Computer use, Claude in Chrome, and cross-app workflows are Week 14. Every Cowork feature claim here traces to verified official Anthropic documentation (links above). The instructor and institution are fictional; real products (Claude Cowork, MCP) are named factually.

~ Prof. Quinn's edition · Fall 2026 · built with thecoursemaker.com