Claude Code CLI Commands: Full Reference (2026)

Claude Code is primarily a conversational tool, but it has a rich set of CLI flags, in-session slash commands, and keyboard shortcuts that change how it behaves. This is the complete reference.

CLI flags (when launching claude)

These flags are passed when you start Claude Code from the terminal.

Basic flags

Model and configuration

Context and files

Permissions and safety

Warning: --dangerously-skip-permissions and bypassPermissions disable all confirmation prompts. Use only in trusted automated environments where you have reviewed the task in advance.

Session control

Non-interactive (pipe) mode

Claude Code can be used in scripts and CI pipelines without an interactive terminal. Use -p with stdin:

# Read a file and ask a question about it
cat main.ts | claude -p "What does this file do? List the exported functions."

# Pipe multiple files
cat src/auth.ts src/middleware.ts | claude -p "Find any security issues in these files."

# Save output to a variable
REVIEW=$(git diff HEAD~1 | claude -p "Review this diff for bugs.")
echo "$REVIEW"

# JSON output for programmatic use
git diff | claude -p "List any breaking changes as JSON" --output-format json

In-session slash commands

Once Claude Code is running, these commands are typed in the conversation input.

Session management

Model switching

Example use case: start a session with Sonnet for planning, switch to Opus with /model claude-opus-4-7 for the complex implementation, then switch back.

Permissions

Memory and context

Custom slash commands

If you have .md files in .claude/commands/ (project-level) or ~/.claude/commands/ (user-level), they appear as slash commands. Example:

~/.claude/commands/ship.md  →  /ship
.claude/commands/qa.md     →  /qa

Run them by typing the command name. Arguments are passed as $ARGUMENTS in the command template.

Keyboard shortcuts (interactive mode)

Output format options (--output-format)

Used with -p (non-interactive) mode.

JSON output example:

claude -p "List all TODO comments in this file" --output-format json < main.ts

Returns:

{
  "result": "Found 3 TODO comments:\n1. Line 42: TODO: handle edge case\n2. ...",
  "cost_usd": 0.0023,
  "session_id": "abc123",
  "is_error": false
}

stream-json example:

claude -p "Explain this module" --output-format stream-json < module.ts

Each line is a JSON event:

{"type": "system", "subtype": "init", "session_id": "..."}
{"type": "assistant", "message": {"content": [{"type": "text", "text": "..."}]}}
{"type": "result", "cost_usd": 0.004}

Environment variables

These environment variables control Claude Code's behavior without CLI flags.

Common usage patterns

Start a session in a project with context

cd my-project
claude  # Claude Code reads CLAUDE.md and .claude/settings.json automatically

Run a quick task without entering interactive mode

# Summarize the last 10 commits
git log --oneline -10 | claude -p "Summarize these commits as release notes."

# Find security issues in a directory
claude -p "Scan for SQL injection vulnerabilities" --allowed-tools Read,Grep

Resume a specific session

# Get the session ID from /status or --output-format json
claude --resume abc123def456

CI/CD integration

#!/bin/bash
# In your CI pipeline — review PR diff for bugs
DIFF=$(git diff origin/main...HEAD)
REVIEW=$(echo "$DIFF" | claude -p "Review this PR diff. List any bugs or security issues as JSON." --output-format json --dangerously-skip-permissions)
echo "$REVIEW" | jq '.result'

Restrict Claude to read-only mode

# Claude can only read and search — no writes, no bash
claude --allowed-tools Read,Grep,Glob

FAQ

Can I use Claude Code without a terminal? Yes — via the VS Code extension (adds a sidebar panel) or the JetBrains plugin. The slash commands and keyboard shortcuts work the same way. The CLI flags apply when launching via the terminal.

What is the difference between /clear and /compact? /clear starts a completely fresh context — Claude forgets everything from the current session. /compact summarizes the conversation into a shorter form, reducing token usage while preserving key context. Use /compact for long sessions where you want to continue working without losing the thread.

How do I see what tools Claude is using? All tool calls are shown in the conversation output. Use claude --debug to see raw tool input/output for debugging.

Can I disable specific tools permanently? Yes — add disabledTools to your ~/.claude/settings.json:

{
  "disabledTools": ["WebFetch", "WebSearch"]
}

What is the max-turns flag for? --max-turns N limits how many agentic back-and-forth turns Claude takes before stopping. Use it to prevent runaway loops in automated scripts.

Sources

1. Claude Code CLI documentation — Anthropic, April 2026
2. Claude Code configuration — April 2026