Skip to content
Claude Code Docs Claude Code Docs

Permission modes

Permission modes determine Claude’s default behavior when a tool has no matching allow/deny/ask rule. The mode applies as a fallback after all explicit rules are checked.

default

Section titled “default”
  • Value in settings.json: "default"
  • Behavior: Standard interactive mode. For tools not covered by a rule, Claude prompts the user for permission on first use. File editing permissions persist until session end. Bash commands are approved per-command per-directory.
  • When to use: General development workflows where you want to approve tool use interactively.
  • Related: --permission-mode CLI flag; permissions.defaultMode setting. See Settings/permissions-security.md.

acceptEdits

Section titled “acceptEdits”
  • Value in settings.json: "acceptEdits"
  • Behavior: Auto-accepts file edits and common filesystem commands (mkdir, touch, mv, cp, rm -specific-file) for paths in the working directory or additionalDirectories. Bash commands are auto-approved if they only modify files in allowed directories. Other tools still prompt.
  • When to use: When you trust Claude to edit and create files without approval, but want control over other operations (Bash exploratory commands, network requests).
  • Related: permissions.additionalDirectories for expanding writable scope. See Settings/permissions-security.md.

plan

Section titled “plan”
  • Value in settings.json: "plan"
  • Behavior: “Plan mode” — Claude can read files, reason about them, and propose changes, but cannot execute tool invocations (Bash, Write, Edit, etc.). Use for analysis-only workflows. When useAutoModeDuringPlan is true (default), plan mode uses auto-mode semantics if auto mode is available, allowing safe reads and analysis.
  • When to use: Brainstorming, code review, architecture analysis. Claude generates plans but does not execute them.
  • Related: useAutoModeDuringPlan (Settings/permissions-security.md); switches plan to auto-mode semantics if available.

auto

Section titled “auto”
  • Value in settings.json: "auto" (feature-gated; may not be available in all deployments)
  • Behavior: Auto-approves tool calls using background safety classifiers that check whether actions align with the user’s stated request and organizational trust boundaries. The classifiers consult autoMode.environment, autoMode.allow, and autoMode.soft_deny prose rules. Actions flagged as risky (exfiltration, force-push, curl | sh, production deploys to untrusted destinations) are denied. Internal, routine actions (pushing to your own repo, writing to staging) are allowed. Auto mode does not bypass deny rules; explicit denies still block.
  • When to use: Development workflows where you trust Claude to make routine decisions (commit, push to your repo, write to ephemeral buckets) without prompting. Requires configuration of trusted infrastructure via autoMode.environment.
  • Related: autoMode object (environment, allow, soft_deny); skipAutoPermissionPrompt; disableAutoMode; useAutoModeDuringPlan. See Settings/permissions-security.md.

dontAsk

Section titled “dontAsk”
  • Value in settings.json: "dontAsk"
  • Behavior: Auto-denies all tools unless explicitly allowed via permissions.allow rules or /permissions command approval. No prompts. Intended for workflows where only a narrow set of tools are permitted.
  • When to use: Restricted environments; CI/CD pipelines where only specific tools (e.g., Read, Bash(specific-test-command)) are needed.
  • Related: Pair with explicit permissions.allow rules.

bypassPermissions

Section titled “bypassPermissions”
  • Value in settings.json: "bypassPermissions"
  • Behavior: Skips permission prompts for all tool invocations except writes to protected directories (.git, .claude, .vscode, .idea, .husky, and tooling symlinks). Writes to .claude/commands, .claude/agents, .claude/skills do not prompt (Claude routinely modifies these). Use only in isolated environments (containers, VMs) where accidental damage is contained.
  • When to use: Automated or sandboxed environments where you want minimal friction and accept the risk.
  • Related: skipDangerousModePermissionPrompt (skip the activation confirmation); disableBypassPermissionsMode to prevent this mode organization-wide. ⚠ Security risk in production; use with caution.

← Back to Permissions/README.md