[ ABORT TO HUD ]
SEQ. 1
SEQ. 2
SEQ. 3

Permissions, Hooks & Safety

💻 Claude Code CLI⏱ 20 min⭐ 950 BASE XP

Tiered Permission System

Claude Code implements a sophisticated tiered permission model to balance speed and safety:

TierActionsApproval
Read-OnlyFile reads, grep, directory listingAuto-approved
WriteFile edits, new file creationPer-session or per-project approval
Bash/ExecuteShell commands, npm scriptsRequires explicit approval
DestructiveFile deletion, git operationsAlways requires manual approval

Auto Mode

Auto Mode is an AI-powered risk classifier that sits between Claude and your machine. It evaluates each proposed action for risk level and automatically approves low-risk operations while blocking dangerous ones — eliminating "permission fatigue" without sacrificing safety.

Lifecycle Hooks

Hooks are deterministic code that executes automatically during Claude Code's lifecycle. Configure them in .claude/settings.json:

  • Pre-tool hooks: Run before a tool executes — can block dangerous commands, enforce linting rules, or validate file paths.
  • Post-tool hooks: Run after a tool completes — auto-format code, run tests, send notifications.
  • Session hooks: Trigger on session start/end — initialize environments, save state, alert teams.

Permission Hooks

For team-based workflows, the --permission-prompt-tool CLI flag lets you route approval requests to external systems like Slack, email, or custom webhooks. This enables delegated oversight — a senior engineer can approve risky operations from their phone while Claude Code continues working.

SYNAPSE VERIFICATION
QUERY 1 // 3
What does Auto Mode do in Claude Code?
Automatically writes all code
An AI risk classifier that auto-approves safe actions and blocks dangerous ones
Disables all permissions
Runs in the background without output
Watch: 139x Rust Speedup
Permissions, Hooks & Safety | Claude Code CLI — Claude Academy