Practical Tips & Tricks

Claude Lifehacks
& Tips

Boost your productivity with Claude Code, MCP servers, skills, plugins, and Claude Desktop.

/compactCLAUDE.mdMCPhooksworktreesartifactsskillspluginsRTKcaveman/initmemoryFigmashortcuts/compactCLAUDE.mdMCPhooksworktreesartifactsskillspluginsRTKcaveman/initmemoryFigmashortcuts

Why This Talk

Small habits, big leverage

The difference between using Claude and mastering it is a handful of tricks most people never discover.

10xProductivityShip features faster with the right workflows
75%Fewer TokensCompress output without losing substance
0Context SwitchesStay in your terminal, IDE, or desktop

Claude Code CLI

/compact — Context Management

Compress your conversation to reclaim context window space without losing important information.

claude
$ /compact
Compacting conversation...
✓ Compressed 47 messages → summary
✓ Context: 128k → 12k tokens
# Also works with a focus topic:
$ /compact focus on the auth module
Before
128k tokens
After
12k tokens

Pro tip: Run /compactproactively before your context fills up — don't wait for the warning.

Claude Code CLI

/init & CLAUDE.md

Bootstrap project context so Claude understands your codebase conventions from the start.

~/my-project
$ claude
> /init
Analyzing project structure...
✓ Detected: TypeScript + React + Next.js
✓ Found: package.json, tsconfig.json
✓ Generated CLAUDE.md

Generated Files

📄 CLAUDE.md
📁 .claude/
⚙️ settings.json
📁 settings.local.json

CLAUDE.md tells Claude your project's build commands, test patterns, code style, and architecture decisions.

Claude Code CLI

Memory System

Claude remembers across sessions — your preferences, project decisions, and feedback persist automatically.

memory/feedback_testing.md
---
name: testing-preferences
type: feedback
---
Integration tests must hit a real
database, not mocks.
# Why: prior incident where
# mock/prod divergence masked a
# broken migration

Project Memory

Stored in .claude/memory/ — project-specific context, decisions, references

User Memory

Stored in ~/.claude/memory/ — your role, preferences, feedback across all projects

Session Context

Current conversation — CLAUDE.md, git state, active files

Claude Code CLI

Git Worktrees

Work on multiple branches simultaneously — each agent gets its own isolated working copy.

feature/auth
~/project-auth $
claude "add OAuth login"
Working on auth module...
✓ Created auth/oauth.ts
✓ Updated auth/index.ts
feature/dashboard
~/project-dash $
claude "build analytics page"
Working on dashboard...
✓ Created pages/analytics.tsx
✓ Added chart components
No conflicts
Parallel work
Clean isolation
Auto-cleanup

Claude Code CLI

Keyboard Shortcuts

Navigate faster without leaving your keyboard.

Esc
Interrupt generation
Tab
Accept autocomplete suggestion
Shift+Tab
Toggle plan mode
Ctrl+C
Cancel current operation
Ctrl+L
Clear terminal
/
Open slash command menu
↑
Previous prompt history
Ctrl+R
Search prompt history

Claude Code CLI

Slash Commands — Workflow

Daily Drivers

/clearStart fresh between tasks, keeps project memory
/compactSummarize with focus line to preserve what matters
/contextCheck what's eating your context window
/resumePick up any prior session instead of re-explaining
/rewindUndo button for both code and conversation

Speed Multipliers

/planForce planning before big changes
/model + /effortDrop to Haiku for grunt work, Opus for hard problems
/branchFork conversation to try risky approach safely
/btwQuick side question without polluting main context

Claude Code CLI

Slash Commands — Parallel & Ship

Parallel Work

/backgroundDetach long-running tasks, free your terminal
/batchCodebase-wide changes → subagents in worktrees, each opens PR
/loopWatch for CI, file changes without babysitting

Before Shipping

/diffReview what Claude actually changed before committing
/simplify3 agents review recent changes in parallel, apply fixes
/security-reviewRun before any PR touching auth, input, or data

Claude Code CLI

Slash Commands — Setup & Gems

Setup Once, Save Forever

/init + /memoryGet CLAUDE.md right early — every future session inherits it
/permissionsAllowlist safe commands, stop hitting approval prompts
/fewer-permission-promptsScans history, auto-builds allowlist for you
.claude/commands/*.mdTurn repeated prompts into one-word shortcuts

Hidden Gems

/copy NGrab Nth-latest response; press w to write to file
/teleportPull web session into terminal for real shell access
/insightsSee where you spend time and what trips you up
/goalLet Claude keep iterating until condition met

Claude Code CLI

Built-in Sub Agents

Spawn specialized agents for parallel work — each gets its own context and toolset, reports back when done.

Explore
Read-only code search — find files, grep symbols, trace definitions
Plan
Architect implementation strategies — step-by-step plans with trade-offs
code-reviewer
Diff/branch reviewer — severity-tagged findings, no fluff
code-explorer
Deep codebase analysis — trace execution paths, map architecture
code-architect
Design feature blueprints — files to create, data flows, build sequences
general-purpose
Catch-all for multi-step research and complex tasks
spawning agent
# Claude auto-spawns agents when needed
Launching Explore agent to search codebase...
Agent found: src/auth/oauth.ts:42

Claude Code CLI

Custom Sub Agents

Define your own agent types via markdown files in .claude/agents/ — same pattern as skills.

.claude/agents/code-reviewer.md
---
name: code-reviewer
description: Review code for bugs,
security, and quality issues
tools: Read, Grep, Bash
---
# Code Reviewer Agent
## Role
You review diffs and files for:
- Logic errors and edge cases
- Security vulnerabilities (OWASP)
- Performance regressions
- Style violations
## Output Format
`path:line: severity: issue`
Severity: 🔴 bug, 🟡 warn, 🔵 nit
## Rules
- Skip formatting nits
- No praise, only findings
- Max 10 findings per review

Usage

# Invoke via /agents command
$ claude /agents code-reviewer
# Or Claude spawns it automatically
$ claude "review my last PR"
Spawning code-reviewer agent...

Agent File Structure

Frontmattername, description, tools
RoleWhat agent does + domain focus
Output FormatStructured response template
RulesConstraints and guardrails

Skills & Plugins

What Are Skills?

Markdown files that give Claude specialized capabilities — invoked with /slash-commands or triggered automatically.

📄
Skill File
.md with frontmatter
→
🤖
Claude Code
Reads instructions
→
⚡
Action
TDD, review, deploy…
/tdd
/review
/caveman
/diagnose
/feature-dev
/compress

Skills & Plugins

RTK — Rust Token Killer

Token-optimized CLI proxy — saves 60–90% on dev operations by filtering noise from command output.

rtk gain
$ rtk gain
📊 Token Savings Report
───────────────────────
Before: 15,240 tokens
After: 2,134 tokens
Saved: 86%
# Transparent hook rewriting:
# git status → rtk git status
# ls → rtk ls

How It Works

1Claude runs git status
2Hook rewrites to rtk git status
3RTK filters noise, returns compact output
4Claude sees 86% fewer tokens

Skills & Plugins

Caveman Mode

Ultra-compressed communication — drop filler, keep substance. ~75% fewer output tokens.

Normal

Sure! I'd be happy to help you with that. The issue you're experiencing is likely caused by the authentication middleware not properly validating the token expiry. Let me take a look at the code and suggest a fix for you.

~48 tokens
→
/caveman

Bug in auth middleware. Token expiry check use < not <=. Fix:

~12 tokens

Full technical substance stays. Only fluff dies. Invoke with /caveman — toggle with /caveman lite or /caveman ultra.

Skills & Plugins

Community Skills

Open-source skill collections — install and extend.
github.com/mattpocock/skills

🧪

TDD

Red-green-refactor loop with integration tests

🔍

Code Review

Confidence-based PR review, severity-tagged findings

🏗️

Feature Dev

Guided feature development with architecture focus

🐛

Diagnose

Disciplined bug hunting — reproduce, minimise, fix

📋

To Issues

Break specs into independently-grabbable issue tickets

✨

Simplify

Review changed code for reuse, quality, and efficiency

Skills & Plugins

CocoIndex Code

Semantic code indexing — helps Claude understand large codebases through embeddings and code graph analysis.

cocoindex
$ cocoindex setup
Indexing codebase...
✓ 1,247 files analyzed
✓ Dependency graph built
✓ Embeddings generated
$ cocoindex query "auth flow"
→ src/auth/oauth.ts (0.94)
→ src/auth/session.ts (0.89)
→ src/middleware/auth.ts (0.85)

Capabilities

🔎
Semantic Search

Find code by meaning, not just text matching

🕸️
Code Graph

Map dependencies, imports, and call chains

📊
Context Retrieval

Feed relevant code to Claude automatically

Skills & Plugins

Creating Your Own Skill

A skill is just a markdown file with frontmatter — drop it in your project and Claude picks it up automatically.

.claude/skills/my-deploy.md
---
name: deploy
description: Deploy to staging or production
trigger: when user says "deploy"
---
# Deploy Skill
## Steps
1. Run tests first
2. Build the project
3. Deploy to environment
4. Verify health check
## Rules
- Never skip tests
- Always check env vars

Skill Structure

Frontmattername, description, trigger conditions
InstructionsStep-by-step workflow in markdown
RulesConstraints and guardrails

Place skills in:

.claude/skills/my-skill.md

Or install community skills via claude install-skill

Ecosystem

Claude Marketplace

A growing ecosystem of skills, plugins, and MCP servers — install community-built extensions or publish your own.

📄

Skills

Markdown-driven capabilities — TDD, code review, deployment workflows

/tdd/review/caveman
🔌

Plugins

Installable packages that extend Claude with new tools and integrations

RTKCocoIndexFigma
🌐

MCP Servers

Protocol-based connectors to external services and data sources

GitHubPostgresSlack

Browse and install from the community registry — or create and share your own.

MCP Servers

What is MCP?

Model Context Protocol — a standard for connecting AI models to external tools, data sources, and services.

🤖
Claude
MCP
🔌
MCP Server
📁 Files
🗄️ Database
🌐 APIs
🎨 Figma

One protocol, many integrations — Claude gains superpowers without custom code.

MCP Servers

Figma MCP

Bridge design and code — read Figma designs, generate production code, sync components bidirectionally.

Design File
Button
Nav Bar
📖
Read Designs

Extract layout, styles, and component structure

⚡
Generate Code

Produce production-ready React/CSS from designs

🔄
Sync Components

Map Figma components to codebase via Code Connect

$ claude "implement the Figma design"

MCP Servers

Useful MCP Servers

A growing ecosystem of integrations — connect Claude to your entire stack.

📁

Filesystem

Read, write, and search files across your project

🐙

GitHub

PRs, issues, reviews, actions — full GitHub integration

🗄️

PostgreSQL

Query databases, inspect schemas, run migrations

🌐

Puppeteer

Browser automation, screenshots, testing

💬

Slack

Read channels, send messages, search conversations

🧠

Memory

Persistent context and knowledge across sessions

MCP Servers

Setting Up MCPs

One command to add, auto-configured in settings.

terminal
# Add Figma MCP
$ claude mcp add figma \
-- npx @anthropic-ai/figma-mcp
# Add GitHub MCP
$ claude mcp add github \
-- npx @anthropic-ai/github-mcp
# List active MCPs
$ claude mcp list
.claude/settings.json
{
"mcpServers": {
"figma": {
"command": "npx",
"args": ["@anthropic-ai/figma-mcp"]
},
"github": {
"command": "npx",
"args": ["@anthropic-ai/github-mcp"]
}
}
}

Chrome Extension

Claude in Chrome

Connect Claude Code to your browser — automate web tasks, test UIs, and interact with any web page.

🖥️
Browser Automation

Click, type, scroll, screenshot — Claude controls Chrome directly

📸
Visual Testing

Take screenshots, compare layouts, verify UI changes in real time

🔍
Page Inspection

Read page content, accessibility trees, console logs, network requests

📊
Form Filling & Testing

Automate form interactions, end-to-end test flows, data extraction

Claude Code
→
Chrome Extension
→
Any Web Page

Claude Desktop

Projects & Custom Instructions

Organize conversations into projects with persistent system prompts and shared context.

Projects
Work — Backend API
Personal — Blog
Research — ML Paper
How should I structure the API routes?
Based on your project conventions...
Custom Instructions
# Project: Backend API
You are a senior backend engineer.
Use TypeScript and Express.
Follow REST conventions.
Always include:
- Input validation with zod
- Error handling middleware
- Request/response types
- Integration tests

Claude Desktop

Artifacts

Live, interactive outputs — code, documents, diagrams rendered alongside your conversation.

Conversation
Create a React dashboard component with a line chart showing monthly revenue data.
Here's a dashboard component with revenue visualization...
Add a dark theme toggle.
Updated with dark/light theme support ↗
React ComponentLive Preview
Monthly Revenue
JanFebMarAprMayJun

Practical Hacks

Workflow Tips

Five habits that compound over time.

01

Compact early

Run /compact before context fills — don't wait for the warning

02

CLAUDE.md first

Set up project conventions so every session starts with full context

03

Chain MCP tools

Combine Figma + GitHub + filesystem for end-to-end workflows

04

Worktrees for parallelism

Run multiple Claude agents on different branches simultaneously

05

Custom skills for repetition

Turn any recurring workflow into a reusable /slash-command

Thank You

Start Building Smarter

Small changes in how you use Claude compound into massive productivity gains.

↗
Claude Code Docs
docs.anthropic.com/claude-code
↗
MCP Registry
github.com/modelcontextprotocol/servers
↗
Matt Pocock Skills
github.com/mattpocock/skills
↗
RTK (Rust Token Killer)
github.com/cablehead/rtk
↗
Caveman Mode
github.com/dougbutner/caveman
↗
CocoIndex
github.com/cocoindex/cocoindex
↗
Figma MCP
github.com/anthropics/figma-mcp
↗
Claude in Chrome
chromewebstore.google.com
↗
Claude Desktop
claude.ai/download
Opening01 / 28