claude-permissions

Configure permissions through the /permissions command or by editing settings.json directly. You can set tiered access levels—read-only, Bash with approval, and file edits with session prompts—to balance security with workflow needs and prevent uncontrolled tool execution.

Sandboxing creates safer execution boundaries for Claude Code operations. Enable it in settings.json to isolate tool access and reduce risk when running untrusted commands or scripts across skills, agents, and commands.

Review permissions by viewing settings.json or using the /permissions command to inspect allowed and denied tools across all skills, commands, and agents. This audit helps identify overly permissive rules and enforce consistent security policies.

Yes, Claude Code supports tool-pattern and wildcard matching in permissions rules. This lets you grant or deny access to entire tool categories or specific command patterns without listing each tool individually.

Without configured permissions, Claude Code tools run with default access levels, which may grant broader capabilities than necessary. Configuring permissions ensures explicit control and reduces exposure to unintended or malicious tool use.