SenseNova-Skills

Least-privilege permissions

• Limit what the skill can access: Configure credentials so skills can only read or modify specific calendars, email folders, or document libraries—not entire accounts. Think of it like giving a spare key to just one room, not the whole building.
• Separate accounts for testing: Use dedicated low-permission test accounts during development and trials; never use production admin or shared accounts to evaluate the skill.
• Audit permission requests: Before deploying, review exactly what data and actions each skill requests (e.g., "read emails" vs. "delete emails"). Reject any requests broader than the actual task needs.
• Rotate and revoke regularly: Change passwords or tokens every 90 days, and immediately revoke access if a skill is no longer in use.

Network isolation

• Use private networks for skill execution: Run skills on internal servers or isolated cloud environments, not on public-facing internet-connected machines.
• Block outbound data exfiltration: Configure firewalls to prevent skills from sending data to external services, cloud storage, or analytics platforms unless explicitly required and approved.
• Encrypt data in transit: Ensure all communication between the skill and email, calendar, or document systems uses encrypted connections (HTTPS/TLS).

Monitoring

• Log all actions with timestamps: Track what each skill does (which emails were read, which events were created) and who triggered it, but do not log sensitive content like email body text or file contents.
• Set alerts for suspicious activity: Flag unusual patterns—e.g., mass deletions, access outside business hours, or access to calendars/folders the skill should not touch.
• Review logs weekly: Assign someone to spot-check logs for unauthorized or unexpected actions before they cause harm.

Red flags to watch for

• No recent security updates: Check the skill's GitHub or vendor site for updates in the last 3 months. If inactive for 6+ months, pause deployment and request maintenance status.
• Unexpected prompts or requests: If the skill suddenly asks for new permissions, asks to "update" without explanation, or behaves differently than trained, stop using it immediately and investigate.
• Vague or missing documentation: Avoid skills where the vendor cannot clearly explain what data they access, how they store it, or who can see it.
• Unexplained credential errors or retries: If the skill repeatedly fails to authenticate or requests credentials multiple times, it may indicate a compromise or misconfiguration—disable and contact support.