Level 3 — Customisation

Slash commands, hooks, MCP servers, skills, and plugins — making Claude Code truly yours

Node.js / Express / Lambda Vue 3 React Native QA

Use to reveal tips • to go back • next slide • previous slide

Level 3 — Customisation

Custom Slash Commands

Reusable prompt templates that your whole team shares.

How they work: Create a markdown file at .claude/commands/name.md (project-shared) or ~/.claude/commands/name.md (personal). The filename becomes the /name command.
Example — /review-security:
# .claude/commands/review-security.md
Perform a security review of the current changes:
1. Check for SQL injection, XSS, CSRF vulnerabilities
2. Verify all user inputs are validated with Zod
3. Ensure no secrets are hardcoded
4. Check auth middleware is applied to protected routes
Focus on: $ARGUMENTS
Example — /new-endpoint:
# .claude/commands/new-endpoint.md
Create a new API endpoint following our conventions:
1. Add the OpenAPI spec to api/openapi.yaml
2. Generate the route in src/routes/
3. Create the service in src/services/
4. Write tests covering 200, 400, 401, 404
Endpoint: $ARGUMENTS
$ARGUMENTS placeholder: Whatever you type after the command becomes $ARGUMENTS. E.g., /review-security the auth middleware.
Frontmatter for control:
---
allowed-tools: Read, Grep, Glob
model: opus
---
Deep analysis prompt here...
Team sharing: Commit .claude/commands/ to your repo. Everyone gets the same commands. Personal variants go in ~/.claude/commands/.
Level 3 — Customisation

Hooks — Deterministic Automation

Unlike CLAUDE.md (advisory), hooks are guaranteed to run. Use them for things that must happen every time.

What are hooks? Shell commands triggered by Claude Code lifecycle events. Configured in .claude/settings.json. They run your scripts at specific points.
Hook events: SessionStart (session begins), PreToolUse (before a tool runs), PostToolUse (after a tool completes), Stop (agent finishes), Notification, PostCompact.
Example — Auto-lint on save:
// .claude/settings.json
{ "hooks": { "PostToolUse": [
  { "matcher": "Write|Edit",
    "hooks": [{ "type": "command",
      "command": "npx eslint --fix $CLAUDE_FILE_PATH",
      "timeout": 15 }] }
] } }
Example — Block dangerous commands:
// PreToolUse hook: exit code 2 = BLOCK the action
{ "matcher": "Bash",
  "hooks": [{ "type": "command",
    "command": "echo $CLAUDE_TOOL_INPUT | grep -qE 'rm -rf|drop table' && exit 2 || exit 0" }] }
Example — Auto-test after code changes:
{ "matcher": "Write|Edit",
  "hooks": [{ "type": "command",
    "command": "npm test -- --related $CLAUDE_FILE_PATH",
    "timeout": 60 }] }
PreToolUse exit codes: 0 = allow, 2 = block the action (Claude gets notified why). Any other code = error.
Rule of thumb: If it must happen 100% of the time → hook. If it's guidance Claude should follow → CLAUDE.md. Hooks are deterministic; CLAUDE.md is advisory.
Level 3 — Customisation

MCP Servers — Extending Claude's Reach

Model Context Protocol servers give Claude access to external tools and data sources.

What are MCP servers? External processes that expose tools to Claude via a standard protocol. Think "plugins for Claude's tool system." Each server provides new capabilities.
Configuration: Add to .claude/settings.json or ~/.claude/settings.json:
{
  "mcpServers": {
    "browser": {
      "command": "npx",
      "args": ["@anthropic-ai/mcp-server-browser"]
    },
    "postgres": {
      "command": "npx",
      "args": ["@anthropic-ai/mcp-server-postgres", "postgresql://..."]
    }
  }
}
Popular MCP servers: Database access (Postgres, MySQL), browser automation, Slack integration, Jira/Linear, filesystem watchers, Docker management, AWS services.
Useful for your stack: AWS MCP server for Lambda logs & metrics. Postgres MCP for direct database queries during debugging. Browser MCP for E2E verification.
Each server exposes tools: After configuring, Claude sees new tools. E.g., with a Postgres MCP: "Query the users table for accounts created this week" — Claude calls the database tool directly.
Build your own: MCP servers are simple JSON-RPC over stdio. Build one for your internal APIs: deploy status, feature flags, monitoring dashboards. The SDK has TypeScript and Python templates.
Level 3 — Customisation

Skills — Reusable Knowledge Packs

More powerful than slash commands: skills bundle prompts, scripts, and resources together.

Skills vs Commands: Commands = simple prompt templates. Skills = rich packages with a SKILL.md, helper scripts, reference files, and configuration. Use skills when the task needs more than a prompt.
Structure:
.claude/skills/
  deploy-checker/
    SKILL.md         # Main instructions
    scripts/
      check-deploy.sh  # Helper script
    reference/
      deploy-checklist.md  # Reference doc
SKILL.md frontmatter:
---
description: "Verify deployment readiness for Lambda services"
allowed-tools: Bash, Read, Grep
model: sonnet
---
# Deploy Checker
Before deploying, verify:
1. All tests pass (npm test)
2. No TODO/FIXME in changed files
3. OpenAPI spec is up to date
...
Invoke: Skills appear as slash commands: /deploy-checker. They can also be triggered by Claude automatically when their description matches the task.
Sharing skills: Commit .claude/skills/ to your repo for team skills. Use ~/.claude/skills/ for personal ones. Skills can also be distributed via plugins.
Idea for your team: Create skills for: api-scaffold (new endpoint from OpenAPI), lambda-debug (analyse CloudWatch logs), rn-screen (new React Native screen), vue-component (new Vue component with tests).
Level 3 — Customisation

Plugins — All-in-One Packages

Plugins bundle skills, commands, hooks, and MCP configs into a single installable package.

What's in a plugin? A plugin can include: slash commands, skills, hooks, MCP server configurations, and settings — all bundled together. Install once, get everything.
Install a plugin:
/plugin install anthropic/code-review
/plugin install anthropic/memory
Built-in plugins worth knowing: code-review — automated PR review with specialised agents. memory — persistent facts across sessions. todo — task tracking.
Create a team plugin: Bundle your team's commands, skills, and hooks into a plugin. Distribute via a GitHub repo. New team members run one install command and get the full setup.
Plugin structure:
my-team-plugin/
  plugin.json           # Plugin manifest
  commands/             # Slash commands
  skills/               # Skills with SKILL.md
  hooks/                # Hook scripts
  settings.json         # MCP servers + permissions
RESOURCES

References & Learning Resources

Official docs, free courses, video tutorials, and community guides to deepen your Claude Code skills.

Official Docs — Start Here: code.claude.com/docs — The complete official documentation: overview, quickstart, how it works, configuration, hooks, skills, agents, CLI reference, and more.
Quickstart Guide: code.claude.com/docs/en/quickstart — Up and running in minutes. The official first-steps walkthrough.
Best Practices (Official): code.claude.com/docs/en/best-practices — Anthropic's own guide on effective prompting, context management, and scaling workflows.
Common Workflows (Tutorials): code.claude.com/docs/en/tutorials — Step-by-step walkthroughs: exploring codebases, fixing bugs, refactoring, writing tests, creating PRs, and managing sessions.
Hooks Reference: code.claude.com/docs/en/hooks — Full reference for SessionStart, PreToolUse, PostToolUse, Stop, Notification, PostCompact events and handler types.
CLI Reference: code.claude.com/docs/en/cli-reference — Every flag, environment variable, and invocation mode documented.
Free Course — Claude Code 101: anthropic.skilljar.com/claude-code-101 — Official free Anthropic Academy course. From installation to advanced customisation, with certificate on completion.
Free Course — Claude Code in Action: anthropic.skilljar.com/claude-code-in-action — Deeper official course: building, configuring, and sharing Skills. Also free with certificate.
Video — Anthropic's YouTube Channel: youtube.com/@anthropic-ai — Official Anthropic channel with product demos, workflow walkthroughs, and feature deep-dives.
Video — Anthropic Workflows: Even Anthropic Engineers Use This Claude Code Workflow — Real-world workflow patterns from the team that built Claude Code.
GitHub — Source & Community: github.com/anthropics/claude-code — Source, issue tracker, plugins, and community discussions. Also see awesome-claude-code for curated skills, hooks, and plugins.
Community — Best Practices Repo: shanraisshan/claude-code-best-practice — 20k+ stars. 84 compiled best practices covering prompting, CLAUDE.md, skills, and debugging. A great second opinion alongside official docs.

Next: Level 4 — Agent Architect

Sub-agents, multi-agent orchestration, the Claude Code SDK, CI/CD automation, and scheduled tasks — building autonomous systems.

Master the tools. Now orchestrate them.

reveal tip   hide tip   next slide   prev slide   F fullscreen