web-quality-skills

Web Quality Skills Security Hardening Guide

Least-privilege permissions

• Restrict API access: Grant the agent read-only access to analytics platforms (e.g., Google Analytics, monitoring tools). Never allow it to modify settings, delete data, or change configurations.
• Scope credentials tightly: If the agent needs authentication keys or passwords, create dedicated accounts with minimal permissions—separate from production admin accounts.
• Rotate access regularly: Change any API keys or credentials at least monthly, and immediately if the skill is deactivated or transferred to different teams.

Network isolation

• Block outbound data leaks: Configure firewalls to restrict where the agent sends performance reports and metrics; only allow connections to approved internal systems and trusted Lighthouse services.
• Encrypt communications: Ensure all data traveling between the agent and external services (APIs, analytics platforms) uses HTTPS encryption to prevent interception of sensitive domain or performance information.
• Separate test environments: Run quality audits against test versions of websites, not live production sites, to avoid exposing real user data or domain vulnerability patterns.

Monitoring

• Log all activities: Record which websites were audited, what data was retrieved, and where results were sent. Review logs weekly for unexpected activity or suspicious patterns.
• Alert on anomalies: Set up automatic notifications if the agent suddenly makes unusual API calls, connects to new services, or reports impossible performance metrics (e.g., 0ms load times).
• Track dependency updates: Monitor when Lighthouse and related audit tools release security patches, and apply updates promptly—vulnerabilities in these underlying tools directly affect the agent's security.

Red flags to watch for

• Unusual metric changes: If performance reports suddenly show extreme improvements or degradation without corresponding code changes, suspect prompt injection (someone tricking the agent into false reporting).
• New external connections: If the agent begins communicating with unknown APIs or services not in the original configuration, this may indicate compromise or misconfiguration.
• High-risk setup signals: Limited community adoption and low public testing of this skill means fewer eyes on potential security issues—treat it as higher-risk than mainstream alternatives and conduct extra internal security reviews.