Best AI Tools

Permission decision tree · Anthropic docs checked Nov 3, 2026

Choose Claude Code permission rules by user and project

Choose the Claude Code settings file for your user, project, or organization, then configure and verify narrow allow, ask, and deny rules.

Which Claude Code settings file should I use?

Use managed settings for organization policy, `.claude/settings.json` for reviewed rules shared with a project, `.claude/settings.local.json` for your own project-only settings or a trial, and `~/.claude/settings.json` for your user settings across projects. Managed policy and settings precedence can affect the result; check `/status` and the current Anthropic settings guide.

What is the difference between allow, ask, and deny?

Allow lets a matching tool action run without manual approval, ask requires confirmation, and deny blocks a matching action. Claude Code evaluates deny before ask before allow, so an allow rule cannot override a matching deny. Verify the exact matching behavior in the official permission guide.

Should I change the permission mode?

Start with the default mode unless a documented mode matches your broader workflow. Modes such as Plan, `acceptEdits`, `dontAsk`, Auto, and `bypassPermissions` change behavior beyond one rule; review their current descriptions and cautions before using one.

How do I check that a permission rule works?

Try a narrow local rule, inspect `/status` and `/permissions`, then test both the intended action and a nearby action that should not match. Check the diff before moving a tested rule into the shared project settings file. This validates your setup; it does not constitute a security audit.

When were the Claude Code permission details checked?

Anthropic’s settings and permissions documentation was checked on November 3, 2026. Product behavior and documentation can change, so use the linked official pages to confirm current behavior for your installed version.

Sources

Methodology and limitations

The decision path first maps intended audience to Anthropic’s documented settings-file scopes, then maps an individual tool action to the documented allow, ask, and deny outcomes. It starts from the default permission mode and recommends testing a local rule before sharing it. This is a configuration aid, not a security audit; permission rule matching has documented limits and must not be treated as a substitute for sandbox boundaries.

Affiliate placeholder {{AFF_TOOLNAME}} — referral link not configured.