AgentLens is a local-first, zero-backend tool. It never sends your page content, brand facts, or NCS constraints to any remote server. The attack surface is limited, but we take security seriously.

In-scope vulnerabilities:

• Prototype pollution via malformed facts.config.json input
• ReDoS (Regular Expression Denial of Service) in robots.txt parser or HTML differ
• Path traversal in the CLI when writing compiled output files
• Arbitrary code execution via malformed JSON-LD in audited pages
• MCP server authentication bypass or command injection
• Chrome extension content-script XSS or privilege escalation

Out of scope:

• Security issues in audited third-party websites (that's what AgentLens finds, not causes)
• Social engineering attacks
• Physical attacks against devices running AgentLens

Please do NOT open a public GitHub issue for security vulnerabilities.

Report vulnerabilities privately via GitHub Security Advisories.

You can expect:

• Acknowledgement within 48 hours
• Status update within 7 days
• Patch or mitigation within 30 days for high/critical severity
• Public disclosure coordinated with you after a fix is available

We follow responsible disclosure and will credit reporters in the release notes (unless you prefer to remain anonymous).

• Pin AgentLens to a specific version in CI: npx agentlens@1.0.0 audit
• The MCP server (@agentlens/mcp) listens only on stdio — never expose it over a network port
• facts.config.json is public by design — do not store secrets in it
• Review NCS constraint reviewBy dates before deploying; expired constraints fail CI by default