agent-skills

Least-privilege permissions

• Create separate API keys for each agent skill with only the minimum access needed (e.g., read-only keys for data extraction, not admin access)
• Store all credentials as encrypted environment variables in your agent platform, never in plain text or code
• Regularly audit which skills have access to sensitive systems and remove permissions that are no longer in use
• Use role-based access controls to ensure agent operators can only trigger skills appropriate for their job responsibilities

Network isolation

• Run agents in a restricted network environment that limits outbound connections to only approved domains and services
• Block agents from accessing internal systems or private networks unless absolutely necessary for the specific task
• Use a firewall or network policy to prevent agents from communicating with unknown or untrusted external services
• Monitor and log all data the agent sends out; disable any skill that attempts unexpected external connections

Monitoring

• Enable audit logging for all agent activities, capturing what data was extracted, when, and by whom
• Set up real-time alerts if an agent processes unusually large amounts of data or makes requests to unexpected websites
• Review agent activity logs weekly to spot suspicious patterns, repeated failures, or unusual data extraction requests
• Track Apify's service status and security announcements through their official channels and apply patches promptly

Red flags to watch for

• An agent skill suddenly requesting new permissions or credentials it didn't need before
• User inputs to extraction tasks containing commands or code (e.g., suspicious quotes, brackets, or syntax) that could change how the agent behaves
• Large data transfers or downloads occurring without a clear business purpose, especially to external cloud storage
• Skills failing repeatedly or the Apify service experiencing outages that leave your automated processes stalled