Skill
Internal procedural knowledge that tells Claude how to perform a specific task. Skills contain step-by-step instructions, examples of good and bad output, edge cases, and expected result formats — all written in plain Markdown, not code. Skills handle the "how"; Connectors handle the "where from" and "where to."
Exam context: The exam tests whether you can distinguish Skills from Connectors. Skills cannot make network calls or access external APIs. If the workflow needs live data from an external service, you need a Connector.
See also: 3.1 Understanding the Ecosystem
Connector
An integration that reaches out to external APIs (Salesforce, Google Drive, Slack, Gmail) to fetch or push data using the Model Context Protocol (MCP). Connectors handle data retrieval and delivery — they are the delivery service that brings ingredients from outside Claude's kitchen.
Exam context: Know that Connectors use OAuth tokens (never passwords), that permissions are granular and revocable, and that each connector requires a one-time browser-based authentication flow.
See also: 3.2 Connectors: Integrating Productivity Tools
Plugin
A unified distribution bundle that packages one or more Skills, Connectors, and slash commands into a single installable unit. When you install a "Salesforce Plugin" from the marketplace, you get the connector for API access, the skills for data analysis, and the slash commands for quick actions — all in one package.
Exam context: Questions test whether you understand the relationship between Plugins, Skills, and Connectors. A Plugin is the container; Skills and Connectors are the components inside it. "Install a skill" when you mean "install a plugin" is a common confusion the exam exploits.
See also: 3.1 Understanding the Ecosystem
Model Context Protocol (MCP)
An open standard for AI-to-tool communication, governed by the Linux Foundation's Agentic AI Foundation — not proprietary to Anthropic. MCP provides a universal interface so any tool can plug into Claude (or any other AI client) without custom integration code. Think of it as the USB-C port for AI.
Exam context: A classic trap claims MCP is proprietary to Anthropic. It is an open standard. Also know that MCP Apps can render interactive UI elements directly inside the conversation.
See also: 3.1 Understanding the Ecosystem
MCP App
An interactive UI component rendered directly inside the Claude conversation via the MCP protocol. Instead of Claude describing your Jira board in text, an MCP App shows you the actual board with drag-and-drop task cards, editable fields, and live status indicators — all within the chat window.
Exam context: Know that MCP Apps are not Claude-exclusive — they work across multiple MCP-compatible clients including VS Code and other AI tools.
See also: 3.2 Connectors: Integrating Productivity Tools
OAuth Authentication
The secure, one-time browser-based flow used to connect external services. You authenticate directly with Google, Slack, or Salesforce — never through Claude. The resulting access token gives Claude scoped permission until you revoke it. Claude never sees or stores your password.
Exam context: Questions test the OAuth flow mechanics. Know that installation and authentication are separate steps, that tokens are scoped (not universal), and that permissions are revocable.
See also: 3.2 Connectors: Integrating Productivity Tools
Progressive Disclosure
The token-efficient mechanism Claude uses to discover and load skills. Rather than reading your entire skill library at once (which would waste thousands of tokens), Claude scans metadata labels first (skill name, trigger conditions) and only loads the full instructions when a specific skill is needed. Like scanning book spines rather than reading every book cover to cover.
Exam context: Know that progressive disclosure makes skills more efficient than pasting detailed instructions into the system prompt. This efficiency advantage is a key justification for using skills over long manual prompts.
See also: 3.1 Understanding the Ecosystem
Built-in Agent Skills
Pre-installed capabilities that Claude has out of the box — including reading, writing, and manipulating Excel spreadsheets, Word documents, PowerPoint presentations, and PDFs. These require no plugin installation or MCP server configuration. They run locally within the sandboxed environment.
Exam context: A trap answer may claim you need a plugin to process spreadsheets in Cowork. You do not. Excel and CSV processing is a built-in Agent Skill requiring no installation.
See also: 3.3 The Skills System
SKILL.md
The core Markdown file that defines a custom skill. It contains YAML frontmatter (name, version, trigger conditions) followed by plain-English instructions describing how to perform the task, examples of good and bad output, and the expected result format. No coding knowledge is required.
Exam context: Know that the YAML frontmatter is mandatory (without it, the skill is invisible to Claude), that the file must be stored in ~/.claude/skills/ to be discoverable, and that skills are written in Markdown, not code.
See also: 3.4 Creating Custom Skills
YAML Frontmatter
The metadata block at the top of a SKILL.md file, placed between triple-dash markers (---). It contains the skill's name, version number, description, and trigger conditions. Without this frontmatter, Claude cannot discover, trigger, or load the skill — it is completely invisible.
Exam context: A skill without YAML frontmatter does not work with "reduced functionality" — it does not work at all. The frontmatter is mandatory, not optional.
See also: 3.4 Creating Custom Skills
Skills Directory
The specific filesystem path (~/.claude/skills/) where Claude scans for custom skills. Storing a skill file in any other location — your Desktop, Downloads, or a project folder — means Claude will never find it. The directory path is non-negotiable.
Exam context: Questions test whether you know the correct storage location. The path must be exact: ~/.claude/skills/.
See also: 3.4 Creating Custom Skills
Workflow Chaining
Linking multiple tools in sequence to handle complex, multi-step workflows. The standard pattern is Connector (fetch data) to Claude's reasoning engine (process) to Skill (apply formatting or logic). Each component does one job well, and they connect in sequence to handle end-to-end automation.
Exam context: Know the correct component for each role: Connectors for external data access, Skills for internal procedural logic. The exam tests whether you can correctly assign each step of a workflow chain.
See also: 3.5 Designing Connected Workflows
Tool Priority Hierarchy
Claude's natural selection order when choosing tools: Connectors first (structured API data, fastest), browser second (web navigation, slower), Skills for processing logic. Connected workflows should follow this hierarchy to maximise efficiency and minimise token consumption.
Exam context: Questions ask you to design the most efficient workflow. The answer follows the hierarchy — use the fastest available tool for each step.
See also: 3.5 Designing Connected Workflows
Departmental Bundles
Curated collections of plugins pre-configured for a specific role or team. A "Sales Analytics" bundle might include the Apollo connector for lead data, a lead-scoring skill, and a pipeline reporting template. Teams install a single bundle tailored to their workflow rather than assembling components individually.
Exam context: Know that bundles demonstrate the Plugin concept in practice: multiple components packaged as one installable unit.
See also: 3.3 The Skills System
FindSkill Directory
The marketplace and community catalogue for discovering, installing, and managing skills and plugins — accessible via Customise > Browse Plugins inside Cowork. It is the authoritative source for verified, installable components. Community-built skills extend Claude with domain-specific procedures and logic.
Exam context: Know that community skills are not vetted by Anthropic for accuracy. Always install from trusted authors and test with non-sensitive data first.
See also: 3.3 The Skills System
Token Budget Management
The practice of designing connected workflows with context window efficiency in mind. Every tool definition, loaded skill, and piece of data consumes tokens. Overloading a workflow with too many simultaneous connectors and bloated skill definitions degrades response quality. Disable connectors you are not actively using.
Exam context: Questions test whether you understand that even unused connectors consume context tokens when their tool definitions are loaded.
See also: 3.5 Designing Connected Workflows