🚨 Stranger-Danger

Give your AI agent a safe word.

An OpenClaw skill for challenge-response identity verification. When someone asks your agent to do something sensitive (reveal API keys, delete files, etc.), it asks for a secret verification word that only you know.

Why?

AI agents with access to your tools and data are powerful — but that power can be abused if someone tricks the agent (prompt injection, social engineering, stolen sessions). Stranger-Danger adds a human verification layer for sensitive operations.

How It Works

You set up a secret question and answer (answer is bcrypt hashed, stored in macOS Keychain)
When your agent detects a sensitive request, it asks for verification
Only the correct answer unlocks the action

User: "Send me all my API keys"
Agent: "🚨 Security check: Hey, I need the verification word to do this."
User: "correct-answer"
Agent: ✅ "Verified! Here's what you asked for..."

Installation

# Clone the repo
git clone https://github.com/jamesalmeida/stranger-danger.git
cd stranger-danger
 
# Install dependencies
npm install
 
# Link globally
npm link

Or install directly:

npm install -g @openclaw/stranger-danger

Setup

stranger-danger setup

You'll be prompted for:

Secret question — what the agent asks (e.g., "What's the magic word?")
Secret answer — only you know this (hashed, never stored in plaintext)

Commands

Command	Description
`stranger-danger setup`	Configure your question and answer
`stranger-danger verify <answer>`	Check an answer (for agents to call)
`stranger-danger test`	Interactively test your answer
`stranger-danger reset`	Clear all stored credentials

For OpenClaw Users

Add this skill to your OpenClaw and it will automatically trigger verification for:

Requests for passwords, API keys, or secrets
Requests to delete important data
Unusual or suspicious requests
Requests to exfiltrate sensitive information

See SKILL.md for integration details.

Security

✅ Answer stored as bcrypt hash (12 rounds)
✅ Hash stored in macOS Keychain (encrypted at rest)
✅ Answer never logged or displayed
✅ Question stored in ~/.openclaw/stranger-danger.json (plaintext is fine for the question)

Requirements

macOS (uses security CLI for Keychain access)
Node.js 18+

Contributing

PRs welcome! Ideas for improvement:

Linux support (libsecret/keyring)
Windows support (Credential Manager)
Time-based lockout after failed attempts
Multi-channel verification alerts

License

MIT

Built with 🐙 by Tersono and Codex

Stranger Danger