Vibe Code Guardian guides AI-assisted review and remediation. It is not a security boundary, trusted authorization service or certification. Host permissions and human authority still apply.

Do not post credentials, private code, personal data, raw account/session records or exploit-ready details in a public issue or pull request.

If this repository offers GitHub private vulnerability reporting, use Security → Report a vulnerability. If that option is unavailable, open a minimal public issue asking for a private reporting channel, without disclosing sensitive details. No private email address or guaranteed response time is advertised.

A useful private report includes:

• Affected release, component and host/tool versions when known.
• The expected boundary, actual behavior and potential impact.
• A minimal synthetic reproducer with prerequisites and a legitimate control.
• The observed evidence and limitations, separated from assumptions.
• Any proposed containment or repair, without running it against other users or systems.

Test only systems and data you own or are explicitly authorized to assess. Do not exfiltrate secrets, probe unrelated accounts or expand access to demonstrate impact.

Relevant issues include unsafe helper input/output handling, evidence or decision inconsistencies, instructions that improperly expand authority, disclosure through report handling and reproducible failures to preserve the skill's stated execution boundaries.

The intentionally vulnerable synthetic evaluation fixtures are demonstrations, not application templates. A seeded defect is not a new vulnerability unless the report shows an unintended effect or escape beyond the fixture's declared scope.

Because AI and host behavior vary, distinguish a reproducible boundary failure from an unavailable capability or a model's unverified answer. Neither should be hidden; the evidence needed to act on them differs.

This project is in prerelease. Report issues against the latest public prerelease and identify older affected versions when possible. No long-term support or backport commitment is currently made.

Confirmed fixes should preserve the first failure, add appropriate regression evidence and state remaining limitations. Disclosure timing and mitigations are handled case by case. A report alone does not authorize publishing another party's private details.