Claude Code Tips & Tricks

1. Keyboard Shortcuts

Shortcut	Function
Esc + Esc	Undo menu — revert code changes
Shift+Tab	Toggle permission mode (Normal / Auto / Plan)
Ctrl+C	Interrupt current execution
Ctrl+R	Search prompt history
Ctrl+O	Show detailed conversation log (thinking process)

2. Slash Commands

Command	Purpose
/memory	View loaded memories
/resume	Resume a previous conversation
/clear	Clear conversation, start fresh
/cost	Check session token usage and cost
/context	View remaining context window space
/compact	Compress history to free up space
/init	Auto-generate project CLAUDE.md
/diff	View code changes
/rename name	Name the session for easy resume later
/plan	Enter/exit Plan mode (discuss without executing)
/model sonnet	Switch model (cheaper)

3. Startup Options

claude                        # Normal start
claude --continue             # Continue last conversation
claude --resume               # Choose which session to resume
claude -n "my-project"        # Named session for easy resume
claude -p "check SPY price"   # One-shot question, no interactive mode
claude --dangerously-skip-permissions  # Skip permission prompts

4. Memory System

Memory is the only bridge between conversations. Each conversation is completely independent — anything discussed but not saved to memory is lost next time.

Four Layers of Memory

~/.claude/CLAUDE.md                    → User-level, loaded in ALL conversations
~/project/CLAUDE.md                    → Project-level, only in that directory
~/.claude/projects/.../memory/         → Global memory
~/.claude/projects/...-project/memory/ → Project memory

Key Differences

Type	Auto-loaded	Best For
CLAUDE.md	Every conversation	Rules, workflows, coding standards
MEMORY.md	First 200 lines	Memory index (pointers to files)
Memory files	Must be read manually	Specific info (server configs, API keys)

Best Practices

• Cross-project info (servers, API keys) → global memory
• Project-specific info (portfolio, strategy) → project memory + global summary
• Stable rules → CLAUDE.md
• Keep CLAUDE.md concise (< 200 lines) — it's loaded every time
• Tell Claude "remember: xxx" to save something explicitly

5. Prompt Tips

Be Specific, Not Vague

# Bad
Analyze NVDA

# Good
Check NVDA latest price, compare 3-month trend, give buy/sell
recommendation considering my moderate-conservative risk profile,
save results to log.md

Include Verification Steps

# Deployment
Deploy and test if Telegram messages send correctly

# Development
Run tests after modifying the code

Plan Before Complex Tasks

/plan          # Enter Plan mode, discuss approach
(confirm the plan)
/plan          # Exit Plan mode, start executing

More Tips

• Use @filepath to reference files instead of copy-pasting
• When unsure, ask Claude for options first, then decide
• Describe everything in one message — more efficient than multiple turns

6. Cost Saving

Method	Savings
Use /clear between different tasks	30-50% tokens
Switch to /model sonnet for simple tasks	Much cheaper & faster
Keep CLAUDE.md concise	Loaded every time, shorter = cheaper
/compact to compress long conversations	Free up context space
Use -p for one-shot questions	Most efficient for non-interactive use

7. Multi-Project Management

Named Sessions + Resume (Recommended)

claude -n "website"          # Project 1
claude -n "trading-bot"      # Project 2
claude -n "stock-analysis"   # Project 3

# Later
claude --resume              # List all sessions, pick one

Parallel Terminals

Open different terminal windows for different project directories, each running its own claude instance independently.

8. Hooks & Automation

Configure hooks in .claude/settings.json to auto-run scripts before/after tool calls.

Example: macOS Notification on Completion

{
  "hooks": {
    "Stop": [{
      "hooks": [{
        "type": "command",
        "command": "osascript -e 'display notification \"Claude finished\" with title \"Claude Code\"'"
      }]
    }]
  }
}

Common Use Cases

• Pre-deployment validation
• Block edits to sensitive files (.env)
• Send notifications after tool calls

9. FAQ

Claude doesn't remember what I said before

Each conversation is independent. Make sure important info is saved to memory files. Use /memory to check.

Conversation getting slow

/context to check usage → /compact to compress → or /clear to start fresh.

Claude keeps asking for info I already provided

Check if it was saved to memory. Cross-project info must be in global memory.

Want Claude to always follow a rule

Write it in CLAUDE.md, not just in conversation. CLAUDE.md is loaded every time.