Waza

How to Stay Secure with Waza Skills

Least-privilege permissions

• Only enable skills that you actively use—disable everything else. Each skill is like giving someone a key to part of your house; don't hand out keys you don't need.
• Review what folders and files each skill can access before installing it. Skills should only touch the code they're meant to work with, not your entire computer or repository.
• Run skills in isolated projects first, not directly in your main production codebase. Test them like you'd test a stranger's advice before following it completely.

Network isolation

• Check if Waza skills need internet access—they shouldn't unless clearly necessary. If a skill claims to work offline but connects online, that's suspicious.
• Use a firewall or network tool to block any unexpected outbound connections from your AI agent or development environment running Waza.
• Keep skills and your codebase on an internal network when possible, away from public internet exposure.

Monitoring

• Log what each skill does—what files it changes, what commands it runs. You should be able to review a record afterward, like checking your security camera footage.
• Watch for unusual behavior: skills taking much longer than expected, accessing files outside their intended scope, or making network calls you didn't authorize.
• Set up alerts that notify you if a skill modifies critical files (like configuration files, authentication keys, or deployment scripts).

Red flags to watch for

• Skills from unknown maintainers or those with no activity history. Check if the skill creator's GitHub account looks legitimate and active.
• Installation prompts that ask for vague permissions ("access to your system") instead of specific ones ("read-only access to src/ folder").
• Skills that require you to input passwords, API keys, or secrets directly into their setup—legitimate tools shouldn't ask for these.