superpowers-zh

Security Hardening Guide: superpowers-zh

Least-privilege permissions

• Restrict API access: Only connect superpowers-zh to APIs and services it absolutely needs (e.g., code repository, build tool). Don't give it blanket access to your entire infrastructure.
• Separate credentials: Use dedicated, limited accounts for each AI tool—never share production passwords or admin tokens with skills.
• Disable unused skills: Remove or deactivate any of the 250k+ skills your team doesn't actively use; fewer active skills means fewer potential weak points.
• Review Chinese-origin skills closely: The 4 Chinese-created skills should undergo extra scrutiny before deployment; verify the creator's reputation and check what data they can access.

Network isolation

• Sandbox execution environment: Run superpowers-zh in a container or isolated virtual machine so compromised skills can't access your internal network.
• Egress filtering: Block outbound network traffic from the AI agent except to explicitly approved domains (e.g., your GitHub, CI/CD platform); this prevents accidental data leaks.
• No direct terminal access: If possible, prevent the skill from running system commands or shell scripts that could modify files or steal data.

Monitoring

• Log all skill activity: Record which skills executed, what code they generated, and what external systems they contacted. Review logs weekly for suspicious patterns.
• Track API calls: Monitor authentication token usage and API request volume; unexpected spikes or failed authentications may signal compromise.
• Audit code output: Spot-check the code snippets generated by superpowers-zh, especially when it touches sensitive systems or repositories.

Red flags to watch for

• Unexpected network connections: If superpowers-zh tries to contact servers outside your approved list, treat it as a potential security incident.
• Skills requesting admin/root permissions: Be extremely cautious if any skill asks for elevated access; this is rarely necessary and often indicates a problem.
• Generated code with hardcoded credentials or secrets: If the AI outputs passwords, API keys, or internal hostnames, immediately audit the skill and regenerate your credentials.
• Vague or undocumented skill behavior: Avoid skills with no clear description of what they do or whose repository has poor maintenance history; unclear intent increases risk.