Can I restrict Claude to read-only mode?

Yes, by setting the defaultMode to 'plan', Claude is restricted to reading files and cannot modify the filesystem or execute shell commands, making it safe for exploration and auditing.

How do permission rules evaluate in Claude Code?

Rules follow a strict priority order where 'deny' rules are checked first, followed by 'ask' prompts, and finally 'allow' rules. If no match is found, the system falls back to the default behavior for that specific tool type.

What is the difference between permissions and sandboxing?

Permissions are application-level rules that control which tools Claude can request to use. Sandboxing provides OS-level enforcement that physically restricts the resources (like network or specific directories) that Bash commands can access once executed.

How do I apply global permissions for an entire team?

Administrators can deploy a managed-settings.json file to system-wide directories (such as /etc/claude-code/ on Linux or /Library/Application Support/ on macOS). These settings take precedence over user and project-level configurations.

Claude Code Permissions Manager

Name: Claude Code Permissions Manager
Author: Jamie-BitFlight

byJamie-BitFlight

•

보안 및 테스팅

Manages and configures granular access controls, tool approval rules, and security sandboxing for Claude Code environments.

This skill provides a comprehensive framework for governing how Claude Code interacts with your system, offering precise control over tool execution, file access, and network requests. It enables developers and administrators to define tiered permission rules using allow/deny/ask policies, configure specific operating modes like 'plan' or 'acceptEdits', and implement organizational managed settings. By integrating OS-level sandboxing with pattern-based path restrictions, it ensures a secure, defense-in-depth approach to AI-driven coding and command execution.

주요 기능

0126 GitHub stars

02Tiered Evaluation: Implements a strict 'deny → ask → allow' logic to prevent unauthorized actions.

03Security Sandboxing: Combine permission rules with OS-level enforcement to restrict filesystem and network access.

04Flexible Execution Modes: Toggle between 'plan' (read-only), 'acceptEdits', and 'bypassPermissions' for different environments.

05Organization Management: Support for system-wide managed-settings.json to enforce policies across teams.

06Granular Tool Control: Define specific allow/deny/ask rules for Bash, Read, Edit, WebFetch, and MCP tools.

사용 사례

01Automating command approvals for specific safe workflows like 'npm run test' or 'git status'.

02Restricting Claude to read-only access in sensitive production environments or legacy codebases.

03Enforcing organization-wide security policies using managed-settings.json across developer workstations.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add jamie-bitflight/claude_skills permissions

For use in Claude.ai and ChatGPT

Download Skill