vibeguard

---

Overview

VibeGuard sits between Claude Code and your filesystem as a tool hook. Before any shell command runs or any file is written, VibeGuard intercepts the call, scans for secrets and dangerous patterns, and either allows, warns, or blocks the action — giving you an audit trail along the way.

What it catches

Secrets in file content and command arguments

Pattern	Example shape
OpenAI / Anthropic API keys	sk-proj-..., sk-ant-...
AWS access keys	AKIA... followed by 16 alphanumeric characters
Generic API key assignments	api_key = "<value>"
Passwords in connection strings	postgres://user:hunter2&#64;db.example.com/mydb
PEM private key blocks	BEGIN PRIVATE KEY header line and base64 body

Dangerous shell commands

Command	Why it is blocked
rm -rf /	Wipes the root filesystem
rm -rf ~ / rm -rf $HOME	Wipes the home directory
chmod 777 /etc	Removes all permission boundaries
Recursive deletes inside home subdirectories	e.g. rm -rf ~/projects
Pipe-to-shell patterns	e.g. curl ... | bash

---

How it works

You type  →  Claude calls tool  →  VibeGuard hook intercepts  →  Scan  →  Allow / Block / Warn

Internally:

1. Claude Code fires a PreToolUse hook before every tool invocation.
2. VibeGuard receives the tool name and full input payload as JSON on stdin.
3. The scanner checks content against a set of regex rules for secrets and command patterns.
4. VibeGuard writes a structured log entry, then exits 0 (allow), 2 (block), or outputs a warning to stderr.
5. Claude Code respects the exit code and either proceeds or surfaces the block reason to you.

---

Installation

npm install -g @embodot/vibeguard

Verify the install:

vibeguard --version

Register with Claude Code

vibeguard install

This writes the hook entry into your Claude Code settings (~/.claude/settings.json). To confirm:

vibeguard status

---

Commands

Command	Description
vibeguard install	Register hooks in Claude Code settings
vibeguard uninstall	Remove hooks from Claude Code settings
vibeguard status	Show current hook registration and config path
vibeguard logs	Stream the live action log
vibeguard logs --tail 50	Show last 50 log entries
vibeguard dashboard	Open the terminal dashboard (requires a TTY)
vibeguard audit	Print a summary report of blocked/warned events
vibeguard config	Print resolved configuration

---

Dashboard

vibeguard dashboard renders a live terminal UI showing:

• Real-time tool call feed with allow / block / warn status
• Rolling count of events per tool type
• Last blocked command and reason

Requires a true terminal (does not work over a plain pipe). Exit with q or Ctrl-C.

---

Configuration

VibeGuard reads configuration from the first file found in this order:

1. VIBEGUARD_CONFIG environment variable (path to a JSON file)
2. .vibeguard.json in the current working directory
3. ~/.config/vibeguard/config.json

Example .vibeguard.json

{
  "logPath": "~/.vibeguard/actions.log",
  "blockSecrets": true,
  "blockDangerousCommands": true,
  "warnOnly": false,
  "allowlist": [
    "Read",
    "ListDirectory"
  ],
  "extraSecretPatterns": [
    "MY_INTERNAL_TOKEN_[A-Z0-9]+"
  ]
}

Key	Type	Default	Description
logPath	string	~/.vibeguard/actions.log	Where to write the action log
blockSecrets	boolean	true	Block tool calls containing secret patterns
blockDangerousCommands	boolean	true	Block known dangerous shell commands
warnOnly	boolean	false	Downgrade all blocks to warnings (still logs)
allowlist	string[]	[]	Tool names to skip scanning entirely
extraSecretPatterns	string[]	[]	Additional regex patterns to treat as secrets

---

Log format

Each line in the log is newline-delimited JSON:

{
  "ts": "2026-04-05T10:22:01.443Z",
  "tool": "Bash",
  "action": "blocked",
  "reason": "dangerous_command: rm -rf /",
  "input": { "command": "rm -rf /" }
}

action is one of allowed, warned, or blocked.

---