Claude Code is Anthropic's official CLI for Claude, used as the primary coding assistant across this project.

Key Features

• Agentic task execution with tool use (read, edit, bash, search)
• Subagent delegation via the Agent tool
• Hook system for pre/post tool events
• MCP server integration
• Persistent memory via file-based memory system

Hooks

Claude Code hooks are shell commands wired to lifecycle events in ~/.claude/settings.json. Four hooks are active globally:

Hook	Event	Purpose
block-destructive	PreToolUse	Blocks destructive shell commands (e.g. rm -rf, DROP TABLE)
protect-sensitive	PreToolUse	Prevents reads/writes to credential files and .env paths
audit-log	PostToolUse	Appends every completed tool call to an audit log
permission-requests	PermissionRequest	Logs every permission prompt for allowlist tuning

PermissionRequest hook — allowlist tuning

The PermissionRequest hook fires whenever Claude Code is about to show a permission prompt to the user. It logs the event to ~/.claude/permission-requests.log as a tab-separated line:

<ISO-8601 timestamp>\t<tool_name>\t<tool_input JSON>

Example entry:

2026-03-17T14:22:05Z	Bash	{"command":"docker build -t myimage ."}

Why: Over time the log accumulates the full set of commands that hit permission prompts. Review it periodically to identify patterns and promote frequently-approved commands into the allow list in settings.json, reducing interruptions without sacrificing auditability.