Slash Commands Mastery

You are three hours into a debugging session. Claude has been reading files, running tests, and narrowing down a race condition in your websocket handler. Then you accidentally close the terminal tab. Without knowing about /resume, that entire session — the context, the debugging thread, the file reads — is gone. You start over from scratch.

Most developers use maybe five slash commands. This guide covers all of them, because the ones you do not know are often the ones that save the most time.

What You Will Walk Away With

• A complete reference for every slash command with real usage patterns
• Keyboard shortcuts that eliminate mouse usage entirely
• Session management workflows for multi-day debugging and feature work
• Techniques for controlling context, costs, and model behavior mid-session

The Essential Commands

Session Control

These commands manage the lifecycle of your conversations with Claude.

Command	What It Does	When to Use It
/clear	Wipes current context, starts fresh	Switching to unrelated work
/compact	Summarizes conversation to free context	Long sessions approaching token limits
/resume	Opens session picker to resume old sessions	Returning to yesterday’s debugging thread
/rename	Names the current session	Before /clear so you can find it later
/cost	Shows token usage for current session	Checking spend before a big operation
/stats	Shows usage patterns (subscribers)	Reviewing your weekly usage trends

Tip

Copy-paste prompt for session hygiene before switching tasks:

/rename auth-refactor-feb
/clear

Name it before you clear it. Otherwise you are scrolling through timestamps trying to find “that session from Tuesday where I was fixing the auth middleware.”

Context and Memory

Command	What It Does	When to Use It
/memory	Opens your auto-memory file in an editor	Reviewing what Claude has learned
/context	Shows what is consuming your context window	Debugging why context is full
/model	Switches the model mid-session	Downgrading to Sonnet for simple tasks
/config	Opens the settings interface	Adjusting permissions, themes, notifications

Tool and Integration Commands

Command	What It Does	When to Use It
/mcp	Shows MCP server status and management	Checking which servers are connected
/hooks	Displays configured hooks	Debugging hook behavior
/agents	Creates or manages subagents	Setting up specialized workers
/install-github-app	Sets up Claude Code GitHub Actions	First-time CI/CD integration

Keyboard Shortcuts

These shortcuts work inside the interactive REPL and eliminate context switches to your mouse.

Shortcut	Action
Tab	Accept Claude’s suggestion
Shift+Tab	Toggle plan mode (think before acting)
Ctrl+C	Cancel current generation
Ctrl+D	Exit Claude Code
Esc	Cancel current input / back out of multi-line
Up/Down	Navigate input history

Tip

Copy-paste prompt for activating plan mode on complex tasks:

Press Shift+Tab to enter plan mode, then type:

I need to refactor the authentication middleware to support both JWT and session-based auth.
Before writing any code, analyze the current auth flow across all routes, identify every
file that imports from auth/, and create a migration plan that avoids breaking existing tests.

Plan mode forces Claude to outline its approach before touching any files. Use it for any change that spans more than three files.

Session Management Patterns

The Multi-Day Feature Branch

When you are building a feature over several days, session management becomes critical.

1. Start a named session on day one

   claude -r "payments-v2"

   If the session does not exist, this creates it. If it does, it resumes it.

2. Work normally throughout the day Claude builds context about your codebase, test patterns, and the specific feature you are implementing.

3. Before ending your day, rename and clear if context is full

   /rename payments-v2-day1
   /clear

4. Resume the next morning

   claude -c

   This continues the most recent session in the current directory. Or use /resume to pick a specific session.

5. Fork when you need to explore a tangent

   claude --resume payments-v2 --fork-session

   This creates a new session with the same context, leaving the original intact.

The Debugging Session

Debugging sessions deserve special handling because you need to preserve the diagnostic context.

# Start with the error context piped in
cat error-log.txt | claude -p "What is causing this TypeError in the payment processor?"

# If you need to go interactive for deeper investigation
claude -c

Caution

Do not use /clear in the middle of a debugging session. The accumulated context — file reads, test results, error traces — is exactly what makes Claude effective at finding root causes. Use /compact instead if you are running low on context space. You can even give it guidance:

/compact Keep all error traces, test output, and file paths. Summarize everything else.

Non-Interactive Mode

The -p flag (print mode) is how you script Claude Code into pipelines, cron jobs, and one-off commands.

# One-shot question, returns answer and exits
claude -p "What does the processPayment function in src/payments.ts do?"

# Pipe file contents for analysis
git diff HEAD~5 | claude -p "Summarize these changes for a changelog entry"

# JSON output for script consumption
claude -p "List all API endpoints in this project" --output-format json

# Budget-limited operation
claude -p "Refactor src/utils/helpers.ts to use modern ES6 patterns" --max-turns 5 --max-budget-usd 2.00

Tip

Copy-paste prompt for a daily standup summary script:

#!/bin/bash
# Save as standup.sh, run each morning
git log --oneline --since="yesterday" --author="$(git config user.email)" | \
  claude -p "Generate a standup update from these commits. Format: What I did yesterday, what I will do today, any blockers." \
  --output-format text

Advanced Flags for Power Users

These flags unlock patterns that most developers never discover.

System Prompt Control

# Add instructions without replacing defaults
claude --append-system-prompt "Always use TypeScript strict mode. Prefer functional patterns."

# Complete prompt replacement for specialized tasks
claude --system-prompt "You are a security auditor. Only analyze code for vulnerabilities."

Tool Restrictions

# Read-only analysis mode -- Claude cannot modify files
claude --tools "Read,Grep,Glob,Bash"
claude --disallowedTools "Edit,Write"

# Allow specific bash commands without prompting
claude --allowedTools "Bash(npm run test *)" "Bash(npm run lint)"

Dynamic Subagents from the CLI

claude --agents '{
  "reviewer": {
    "description": "Reviews code changes for quality and security issues",
    "prompt": "You are a senior code reviewer. Focus on correctness, security, and maintainability.",
    "tools": ["Read", "Grep", "Glob", "Bash"],
    "model": "sonnet"
  }
}'

When This Breaks

Session resume shows “session not found”: Sessions are stored per-directory. If you moved your project or are in a different working directory, Claude Code cannot find the session. Use claude --resume without arguments to see all available sessions.

/compact loses important context: Compaction is lossy by design. Always give it guidance about what to preserve. If you compacted and lost critical debugging context, check whether the original session is still available via /resume.

Non-interactive mode hangs on permission prompts: In -p mode, Claude Code will stop if it needs permission for a dangerous operation. Use --dangerously-skip-permissions for trusted CI environments, or pre-allow specific tools with --allowedTools.

Model switch mid-session resets behavior: When you /model to a different model, the conversation history stays but the model’s interpretation of that history may shift. For critical work, start a new session with the target model instead.

What is Next

• Hooks and Automation — Build deterministic guardrails around the commands you have just learned
• Memory System — Control what Claude remembers between sessions
• Prompt Engineering — Write prompts that get better results from every command