claude-code-tutor

Your Claude Code companion. Fresh docs, built-in tutor (Pulse), always in sync.

npm: claude-code-tutor | GitHub: Erikole21/claude-code-tutor

Installation

Recommended (global install)

npm install -g claude-code-tutor

Then verify:

pulse --version

Global install makes the pulse command available system-wide, but you still need to initialize each repository where you want to use it:

pulse init

Update

npm update -g claude-code-tutor

Then re-initialize each repo to apply hook and skill fixes:

pulse init --force

Use --yes to skip the confirmation prompt:

pulse init --force --yes

Quick run (no global install)

npx pulse init

This command installs pulse skills in .claude/skills, runs an initial sync, and adds a safe SessionStart hook.

How it works across repositories and new Claude sessions

• Global npm install adds the CLI binary only.
• pulse init configures the current repository only.
• Each repository must be initialized once.
• New Claude Code sessions in an initialized repository use that repo's .claude configuration and hooks automatically.
• In a brand-new repository, run pulse init first.

How Pulse activates

Pulse is your built-in tutor — it has a name, a personality, and remembers your progress. It activates automatically inside Claude Code — no slash command needed. It triggers when you:

• Mention Pulse anywhere in your message: "pulse", "hey pulse, help me", "en que vamos pulse?"
• Are new to Claude Code: "it's my first time", "how do I get started?"
• Ask for help or guidance: "how can you help me?", "what can I do here?"
• Ask about any Claude Code feature: hooks, MCP, skills, agents, settings, permissions, channels, plugins, workflows, and more — anything covered by the installed skills

Pulse does not activate for regular coding tasks like "fix the bug in app.ts" or "add a login page". It only responds when the conversation is about Claude Code itself.

Tip: You can also invoke Pulse explicitly with /pulse if you prefer.

What can you do with pulse?

Beginner — Your first steps

Never used Claude Code? Pulse guides you from zero:

You:   "It's my first time using Claude Code, help me"
Pulse: "Welcome! I'm Pulse, your Claude Code tutor. What's your name?
        ...Great! Let's start with the basics. Open any project folder
        and type 'claude'. That's it — you're in a session.
        Try asking Claude to explain a file in your project."

Ask anything, Pulse adapts to your language automatically:

You:   "No entiendo qué es CLAUDE.md"
Pulse: "Es un archivo que le da contexto a Claude sobre tu proyecto.
        Piensa en él como un README pero para Claude. Creemos uno juntos:
        crea un archivo CLAUDE.md en la raíz de tu proyecto..."

Learn what commands are available:

You:   "What commands can I use?"
Pulse: "Here are the essentials to get started:
        /help     — see all available commands
        /compact  — compress conversation when it gets long
        /cost     — check how much you've spent this session
        /memory   — see what Claude remembers about your project
        Try /help now and tell me what catches your eye!"

Intermediate — Level up your workflow

Already using Claude Code? Pulse helps you unlock powerful features:

You:   "I want to automate something every time I start a session"
Pulse: "That's what hooks are for! A SessionStart hook runs a shell command
        every time Claude opens. For example, to auto-format your code:

        In .claude/settings.json:
        { 'hooks': { 'SessionStart': [{ 'command': 'npm run lint:fix' }] } }

        Want me to walk you through setting one up for your project?"

Connect external tools:

You:   "Can Claude access my database?"
Pulse: "Yes! Through MCP servers. You configure them in settings.json
        and Claude gets read/write access. Here's how to connect PostgreSQL:
        ..."

Create custom skills for your team:

You:   "How do I create a skill for code review?"
Pulse: "Create a file at .claude/skills/code-review/SKILL.md with
        instructions for how Claude should review code in your stack.
        Let me show you the structure..."

Advanced — Scale with your team

For power users building automation and team workflows:

You:   "I need multiple agents working in parallel on different tasks"
Pulse: "Agent teams let you coordinate multiple Claude instances.
        You define agents with roles (reviewer, implementer, tester)
        and they work on isolated worktrees. Here's the pattern:
        ..."

Integrate with CI/CD:

You:   "How do I get Claude to review PRs automatically?"
Pulse: "With GitHub Actions. Add @claude mentions in your workflow
        and Claude reviews PRs, suggests fixes, and can even push
        commits. Here's a working workflow file:
        ..."

Run Claude in scripts:

You:   "I want to use Claude in a bash script"
Pulse: "Headless mode! Use 'claude --print' for single-shot tasks:

        claude --print 'summarize the changes in the last 5 commits'

        It returns the output directly to stdout. Perfect for
        CI pipelines, cron jobs, or chaining with other tools."

Pulse remembers you across sessions

Your first session:

Pulse: "Hey! I'm Pulse, your Claude Code tutor. What's your name?"
You:   "Ana"
Pulse: "Nice to meet you, Ana! Let's figure out your level..."

Your next session, days later:

Pulse: "Hey Ana! Last time we set up your first hook. Ready to
        learn about MCP servers? That was your next step."

Pulse saves progress incrementally after each topic it explains, not just at the end of the session. If you close the terminal mid-conversation, your progress is still there.

Check your progress anytime:

$ pulse memory
  Name:     Ana
  Level:    intermediate
  Language: es
  Joined:   4/1/2026

  Topics: 5 completed, 2 in progress, 3 pending
    ✓ basic-commands
    ✓ permissions
    ✓ claude-md
    ✓ git-workflow
    ✓ hooks
    … mcp-servers
    … sub-agents

  Next steps:
    → mcp-servers: Ready after mastering hooks

Set it up once per repo, works every session

npm install -g claude-code-tutor
cd my-project
pulse init
# Done. Every Claude Code session in this repo now has fresh docs and Pulse as your tutor.

What pulse does

• Auto-discovery: reads llms.txt to discover every documentation page published by Anthropic. New pages are detected automatically — no package update needed.
• 120+ skills: ships with 23 curated skills and ~100 auto-discovered skills covering the entire Claude Code documentation.
• Doc sync: keeps all skills fresh via daily ETag-based sync. Only changed docs are re-downloaded.
• Bundled fallback: if the official docs are unreachable, pulse installs pre-built skills so you are never left without reference material.
• Built-in tutor: includes pulse for guided help in-session. Reads your memory first, greets you by name, saves progress incrementally, and notifies you about new features relevant to your level.
• Learning path: includes cc-learning-path for structured onboarding from beginner to advanced, with exercises and checkpoints.
• Skill index: dynamically generates a reference index of all installed skills and appends it to both pulse and cc-learning-path, so they always know what's available.

How it works (context architecture)

Having 120+ skills installed does not bloat Claude's context window. Here's why:

1. At rest: All 120+ skills sit as files in .claude/skills/. Nothing is loaded into context.
2. Every session: Only the description field of pulse (~250 characters) enters Claude's context. It's the only skill with auto-activation enabled (disableModelInvocation: false). All other skills have this disabled — not even their descriptions are loaded.
3. When you say "pulse": Claude activates the pulse skill. Its full content loads into context, including a dynamic index of all available skills. Pulse now knows what exists and can recommend the right one.
4. When a specific skill is needed: Claude invokes it (e.g., /cc-hooks-guide), and only that skill's content loads into context.

Session start     → ~250 chars (pulse description only)
"hey pulse"       → +pulse SKILL.md (tutor + skill index)
"how do hooks work?" → +cc-hooks-guide SKILL.md (loaded on demand)

The result: zero context overhead from having 120 skills vs 23. Skills load on demand, one at a time, only when relevant.

Skills

Curated skills (23)

Hand-configured with custom split strategies, token budgets, and descriptions:

| ID | Description | Priority | |---|---|---| | pulse | Claude Code companion: tutor + motivator with practical guidance and ideas. | critical | | cc-learning-path | Structured multi-level learning path. | high | | cc-changelog | Recent updates and changes in Claude Code. | critical | | cc-hooks-events | Hook events table and when they trigger. | critical | | cc-hooks-guide | Practical guide for implementing hooks. | critical | | cc-mcp | MCP server setup and usage. | critical | | cc-settings | settings.json reference and scopes. | critical | | cc-permissions | Permission system and per-tool rules. | critical | | cc-sub-agents | Sub-agent setup and usage. | high | | cc-agent-teams | Coordinating multiple agents in teams. | high | | cc-skills-guide | How to create and maintain skills. | high | | cc-memory | Using CLAUDE.md and project memory. | high | | cc-cli-reference | CLI commands and flags reference. | high | | cc-commands | Built-in Claude Code commands. | high | | cc-model-config | Model configuration and aliases. | high | | cc-plugins | Plugin architecture and authoring. | medium | | cc-channels | Channels and webhook workflows. | medium | | cc-scheduled-tasks | Scheduled tasks with /loop and cron. | medium | | cc-headless | Headless execution and SDK usage. | medium | | cc-sandboxing | Sandboxing and isolation. | medium | | cc-common-workflows | Common development workflows. | medium | | cc-best-practices | Recommended practices and patterns. | medium | | cc-github-actions | Claude Code integration with GitHub Actions. | medium |

Auto-discovered skills (~100)

Automatically discovered from the official docs index. These cover every documentation page Anthropic publishes, including: Chrome integration, VS Code extension, JetBrains, voice dictation, computer use, code review, desktop app, Slack, remote control, ultraplan, enterprise deployments (Bedrock, Vertex AI, Foundry), and more.

Discovered skills are installed alongside curated skills during init and updated daily via the sync hook. When Anthropic adds a new doc page, it appears automatically on the next sync — no package update required.

Run pulse list to see all installed skills.

Team usage

If your team wants the same skills, commit the generated directories:

.claude/skills/cc-*/

Pulse only manages skills marked with _pulse: true, so user-owned custom skills are preserved.

CLI commands

| Command | Description | Key flags | |---|---|---| | pulse init | Initializes pulse in current project. | --force, --yes, --skills <ids...> | | pulse sync | Fetches, transforms, and installs skills. | --force, --if-stale <seconds>, --silent, --skills <ids...> | | pulse greet | Emits session context for the SessionStart hook. | (none) | | pulse list | Lists installed skills and freshness state. | (none) | | pulse status | Shows global pulse status and diagnostics. | (none) | | pulse uninstall | Removes pulse hook and pulse-managed skills. | --keep-hook, --keep-skills, --purge-memory |

Your tutor progress

Pulse remembers your learning progress across sessions in ~/.claude/pulse/memory.json. This is the only source of truth for tutor memory — Pulse does not use Claude's native auto-memory system for session continuity.

| Command | Description | |---|---| | pulse memory | View your tutor progress summary | | pulse memory --reset | Reset memory (keeps session logs) | | pulse memory --export | Export memory as JSON | | pulse memory --import <file> | Import memory from a JSON file | | pulse memory --update '<JSON>' | Update specific memory fields | | pulse memory --exercise <id> --status <status> | Update exercise status | | pulse memory --next-step "<desc>" --reason "<reason>" | Add a next step |

Memory is stored locally and never leaves your machine. Use pulse uninstall --purge-memory to delete all memory data.

Sync mechanism

• Auto-discovery: fetches llms.txt from the official docs, detects new pages not in the curated registry, and installs them as skills automatically.
• ETag caching: unchanged docs (including llms.txt itself) return 304 and are skipped — zero overhead on most syncs.
• Static transformer: condenses docs to fit token budgets. Truncated skills include a note directing Claude to fetch the full official documentation instead of guessing.
• Bundled fallback: init installs 120+ pre-built skills instantly from the package — no network needed.
• SessionStart hook: runs daily stale-aware sync (--if-stale 86400) and injects tutor activation context every session.

Configuration

Create .pulserc.json in your project:

{
  "skills": ["critical", "high"],
  "maxAge": 86400,
  "transformer": "auto",
  "silent": false
}

| Option | Default | Valid values | Description | |---|---|---|---| | skills | ["critical","high"] | ["critical"], ["critical","high"], ["all"], etc. | Skill priorities to sync. | | maxAge | 86400 | number (seconds) | Staleness threshold for auto-sync decisions. | | transformer | "static" | "static" | Transformation mode. Static transformer condenses docs locally. | | silent | false | true, false | Suppress non-error output. |

Troubleshooting

• Hook does not run: check .claude/settings.json has a SessionStart entry with _pulse: true.
• Windows behavior: pulse uses npx.cmd on native Windows and Unix format on WSL.
• Skills look stale: run pulse sync --force to bypass ETag cache.
• New docs not showing up: run pulse sync --force to re-run discovery and fetch all docs.

Contributing

• Improve static skill sources under src/static-skills/.
• Improve tutor and learning content in pulse.md and cc-learning-path.md.
• Adjust the dynamic skill index generator in src/core/skill-index.ts.
• Add/adjust tests in tests/ when changing fetch, transform, split, or hook behavior.
• After changes, run npm run build:skills to regenerate skills-fallback/.