Security Best Practices

Keep your Green Monkey Dashboard and AI agent secure.

Sensitive Files

Never commit these files to version control:

The .gitignore file should already exclude these. Verify before pushing.

Security Features

The dashboard includes optional safety guardrails you can configure:

Session Isolation

Separate sessions per contact to prevent context leakage between different conversations.

External Action Confirmations

Require approval before your agent sends emails, posts to social media, or makes API calls.

Tool Restrictions

Sandbox or disable potentially risky operations:

• Web browsing
• File operations
• Code execution

Privacy Protection

• Prevent external logging of conversations
• Block credential exposure in outputs
• Limit data sharing with third parties

Model Safety Warnings

Alert when using small models without sandboxing, which may be more susceptible to prompt injection.

Environment Variables

Always use environment variables for secrets in production:

# Good - secrets in environment
STRIPE_SECRET_KEY=sk_live_...
ANTHROPIC_API_KEY=sk-ant-...
 
# Bad - secrets in code
# Never hardcode API keys in source files

Production Checklist

• Set a strong SECRET_KEY (64+ characters)
• Use environment variables for all secrets
• Enable HTTPS (automatic on Vercel/Heroku)
• Verify Stripe webhook signatures
• Set CORS origins in production
• Use Stripe test mode until fully tested
• Add rate limiting to auth endpoints
• Monitor failed login attempts
• Set up error logging (Sentry, etc.)

Agent Safety

Your AI agent has access to personal data. Follow these guidelines:

Internal vs External Actions

Safe to do freely (no confirmation needed):

• Read files, explore, organize, learn
• Search the web, check calendars
• Work within the workspace

Require confirmation:

• Sending emails, tweets, public posts
• Anything that leaves the machine
• Anything uncertain

Group Chat Safety

In group chats, the agent should:

• Never share private data from USER.md or MEMORY.md
• Only load MEMORY.md in direct (main session) chats
• Participate without dominating the conversation

Memory Security

• MEMORY.md contains personal context — only load in main sessions
• Daily memory files (memory/YYYY-MM-DD.md) are raw logs — review for sensitive content
• Never exfiltrate private data

Reporting Issues

If you discover a security vulnerability:

• Open a private issue on GitHub
• Do not disclose publicly until a fix is available