/tutorial

allowed-tools: Read, Glob, Grep
description: Interactive tutorial for learning the plugin. Walks through agents, skills, hooks, and commands with hands-on examples.

Plugin Tutorial

Welcome the user and guide them through the plugin's features interactively. This is a read-only walkthrough -- no files are modified.

Step 0: Detect Context

Before starting, silently gather context:

1. Use Glob to find `package.json`, `pyproject.toml`, `Cargo.toml`, `go.mod`, or other project markers in the working directory 2. Identify the project's primary language and framework 3. Use Glob to list the plugin's available components. They live in the plugin's install directory, `${CLAUDE_PLUGIN_ROOT}` (if that placeholder still appears literally here, the workflow was installed with npx: use `.claude/` instead):

• `${CLAUDE_PLUGIN_ROOT}/agents/*.md` for agents
• `${CLAUDE_PLUGIN_ROOT}/skills/*/SKILL.md` for skills
• `${CLAUDE_PLUGIN_ROOT}/hooks/hooks.json` for hooks
• `${CLAUDE_PLUGIN_ROOT}/commands/*.md` for commands

Then begin the tutorial.

Step 1: Meet the Agents

Greet the user and introduce the agent system.

Read each file in `${CLAUDE_PLUGIN_ROOT}/agents/*.md` and present a summary table:

Welcome to the Claude Workflow Plugin tutorial.

Let's explore what this plugin gives you. First up: Agents.

Agents are specialized assistants that activate automatically based on
your request. Here's what's available:

| Agent           | Triggers When You...                    | Model  |
|-----------------|----------------------------------------|--------|
| orchestrator    | Need multi-file coordinated changes    | opus   |
| code-reviewer   | Ask for a code review                  | sonnet |
| debugger        | Report a bug or error                  | sonnet |
| ...             | ...                                    | ...    |

Agents marked PROACTIVELY in their description will activate
automatically when your prompt matches their trigger keywords.
You don't need to invoke them explicitly.

Explain that the orchestrator can spawn subagents for parallel work, and how agent selection works based on prompt keywords.

**Try it**: Suggest the user try a prompt like "review the last commit for issues" to see the code-reviewer agent activate.

Step 2: Commands at Your Fingertips

Read the list of command files from `${CLAUDE_PLUGIN_ROOT}/commands/*.md` and present them:

Commands are slash-invocable workflows. Think of them as recipes
that combine multiple steps into one action.

Available commands:

| Command              | What It Does                              |
|---------------------|-------------------------------------------|
| /project-starter:commit         | Auto-generate conventional commit  |
| /project-starter:verify-changes | Multi-agent verification suite     |
| /project-starter:review         | Code review with structured output |
| ...                             | ...                               |

Commands can be simple (commit) or complex (verify-changes spawns
5+ parallel subagents).

Read `${CLAUDE_PLUGIN_ROOT}/commands/commit.md` and walk through how it works:

Let's look at how /project-starter:commit works:

1. It gathers context: current branch, staged changes, recent commits
2. It analyzes the diff to understand what changed
3. It generates a conventional commit message (feat/fix/docs/...)
4. It executes the commit

Notice the frontmatter:
  allowed-tools: Bash(git status:*), Bash(git diff:*), ...

This pre-approves those git commands for the turn, so the
command runs without permission prompts. It does not
restrict Claude to them.

**Try it**: Suggest staging a file and running `/project-starter:commit`.

Step 3: Hooks That Protect Your Code

Read `${CLAUDE_PLUGIN_ROOT}/hooks/hooks.json` and explain each hook:

Hooks run automatically before or after certain actions.
They protect your code without you thinking about it.

Active hooks:

| Hook                | Triggers On       | What It Does              |
|--------------------|-------------------|---------------------------|
| format-on-edit     | File save/edit    | Auto-formats changed files|
| ...                | ...               | ...                       |

Hooks use exit codes to control behavior:
  Exit 0 = allow the action to proceed
  Exit 2 = block the action and show a message

This means a hook can prevent you from committing secrets,
saving malformed config, or other mistakes.

Read one hook script to show the pattern:

Here's a simplified hook structure:

  1. Receive input as JSON from stdin
  2. Inspect the content or action
  3. Exit 0 (allow) or exit 2 (block with message)

Hooks fail silently on errors (exit 0) so they never block
your workflow unexpectedly.

**Try it**: Suggest the user edit a file and observe the format-on-edit hook in action (if applicable to their project).

Step 4: Skills That Enhance Responses

Read `${CLAUDE_PLUGIN_ROOT}/skills/*/SKILL.md` files and present the skill domains:

Skills inject domain knowledge into responses. When a skill's
trigger keywords match your prompt, its instructions are loaded
automatically.

Available skill domains:

| Skill Domain     | Enhances Responses About...             |
|-----------------|----------------------------------------|
| [skill-name]    | [description from SKILL.md]            |
| ...             | ...                                    |

Skills are passive -- they don't run commands. They shape how
the assistant thinks about your problem by providing specialized
knowledge, patterns, and best practices.

**Try it**: Suggest a prompt that would activate one of the skills relevant to their detected project type.

Step 5: Your Personalized Next Steps

Based on the project context detected in Step 0, provide tailored recommendations:

Based on your project ([language/framework]), here's what I'd
recommend trying first:

1. **[Most relevant command]