Please report vulnerabilities privately first.

1. Do not open a public GitHub issue.
2. Email: jonathansantilli@gmail.com with:
   • affected version
   • reproduction steps
   • impact assessment
   • proof-of-concept (if available)
3. We will acknowledge receipt within 5 business days and assign a tracking status.

• We validate and triage the report.
• We coordinate a fix and release timeline.
• We publish an advisory after a fix is available (or mitigation guidance if no fix is immediately possible).

Dependabot reports ip-address and undici advisories against this repository. Both come from copies bundled inside npm, which is vendored by @semantic-release/npm, and both are dev-only: the published package contains dist, README.md and LICENSE, so none of it reaches anyone who installs codegate-ai.

There is no upstream fix to take. semantic-release and @semantic-release/npm are already at their latest versions, @semantic-release/npm requires npm ^11.6.2, and npm 12.0.2 — the newest release — still bundles the same ip-address@10.2.0 and undici@6.27.0. Upgrading changes nothing. Re-check when npm refreshes its bundled dependencies.

• Use --format sarif in CI to keep security findings visible in code-scanning workflows.
• Treat exit code 2 as a deployment/blocking condition in CI/CD.
• Use codegate run <tool> as a local pre-flight guard before launching AI coding tools; it blocks dangerous findings, can require confirmation for warning-only findings, and rechecks the scanned config surface before launch.
• Use --deep only when you explicitly want Layer 3 remote metadata analysis.
• Layer 3 requests are consent-gated per resource; skipped consent is reported for auditability.
• CodeGate does not execute untrusted MCP stdio command arrays during tool-description scanning.
• Use codegate scan --reset-state only when you intentionally want to clear MCP config change history stored at the resolved scan_state_path location.