ccperm

Audit Claude Code permissions across all your projects.

한국어

Claude Code stores allowed permissions (Bash commands, WebFetch domains, MCP tools, etc.) in .claude/settings*.json per project. As you work across many projects, these permissions pile up silently. ccperm scans your home directory, finds every settings file, and shows what you've allowed — in an interactive TUI or static text output.

Quick Start

npx ccperm

No install needed. Or install globally:

npm i -g ccperm
ccperm

By default, ccperm scans all projects under ~ and launches an interactive TUI.

Options

Flag	Description
--cwd	Scan current directory only (default: all projects under ~)
--static	Force text output (default when piped / non-TTY)
--verbose	Detailed static output with all permissions listed
--fix	Auto-fix deprecated :* patterns to *
--update	Self-update via npm install -g ccperm@latest
--hey-claude-witness-me	LLM-friendly markdown audit briefing with risk classification
--debug	Show scan diagnostics (file paths, timing)
--help, -h	Show help
--version, -v	Show version

Interactive TUI

When running in a TTY (the default), ccperm opens a box-frame TUI:

List view — Projects sorted by permission count. Columns: Bash, MCP, Tools, TOTAL, ! (risk), † (deprecated :*), G (redundant with global).

┌ ccperm ──────────────────────────────────────────── 1/8 ┐
│   PROJECT                   Bash  MCP  Tools  TOTAL     │
├─────────────────────────────────────────────────────────┤
│   ~/.claude                   15    ·      2     17     │
├─────────────────────────────────────────────────────────┤
│ ▸ my-project  local            5    ·      3      8     │
│   other-app   shared           2    3      ·      5     │
│                       ! risk  † deprecated  G in global │
└──── [↑↓] navigate  [Enter] detail  [/] search  [q] quit ┘

Detail view — Press Enter to expand a project. Categories are collapsible; press Enter to toggle.

• [d] delete a permission
• [g] add permission to global settings (shows (in global) tag immediately)
• [i] toggle risk info mode
• Permissions already in global are tagged (in global) in yellow

Search view — Press / to search across all projects. Matches permission names and shows results grouped by project. Arrow keys navigate (headers are skipped), Enter jumps to the project's detail view.

Keys: ↑↓ navigate, Enter select/expand, [i] info, [d] delete, [g] +global, / search, Esc back, q quit.

Static Output

Use --static (or pipe to another command) for text output:

ccperm --static            # compact table
ccperm --static --verbose  # full permission listing

Permission Levels

ccperm distinguishes three levels of Claude Code settings:

Level	File	Scope
global	~/.claude/settings.json	Applies to all projects
shared	<project>/.claude/settings.json	Per-project, committed to git
local	<project>/.claude/settings.local.json	Per-project, gitignored

Permissions are additive — global + shared + local are merged at runtime.

Allow vs Deny vs Ask

Claude Code settings support permissions.allow, permissions.deny, and permissions.ask. ccperm separates them clearly:

• Allow permissions are shown in the main list and can be deleted ([d]) or copied to global ([g])
• Deny rules appear in a collapsible Deny section in detail view, tagged with DENY and dimmed
• Ask rules appear in a collapsible Ask section — these are permissions that prompt for confirmation each time
• Deny and Ask rules cannot be deleted or copied — they are intentional controls
• --verbose output shows separate Deny and Ask sections per project
• --hey-claude-witness-me lists deny rules under "Protected rules"
• Allow counts never include deny or ask rules

Additional Settings

ccperm also scans and displays these top-level settings when present:

Field	Description
allowedTools	Tools explicitly allowed for use (e.g. Bash, Read, Write)
deniedTools	Tools explicitly denied (e.g. WebSearch)
additionalDirectories	Extra directories Claude Code can access beyond the project root

These appear as collapsible sections in the TUI detail view and in --verbose output. Like deny rules, they cannot be deleted or copied.

Risk Classification

Each permission is assigned a risk level inspired by Destructive Command Guard (DCG). Used in --hey-claude-witness-me output and the TUI info mode.

Level	Meaning	Examples
CRITICAL	Irreversible destruction or full system access	rm -rf, sudo, terraform destroy, dd, curl | sh
HIGH	Significant changes to system, remote, or infrastructure	git push --force, chmod, aws, kubectl, ssh
MEDIUM	Controlled side effects, build/runtime tools	docker, npm, node, curl, brew, sed
LOW	Read-only or safe dev tools	cat, ls, grep, git (non-push), eslint, jest

Context matters — git alone is low, but git push --force escalates to critical. Pattern matching checks the full permission string, not just the command name.

Requirements

• Node.js >= 18
• macOS or Linux

License

MIT

---

_{Hey Claude Code — run ccperm --help. There's a flag in there just for you. 🤫}