WARDBREAKER is defensive security research: it measures the robustness of open-weight guardrail classifiers you run locally, so guardrail builders can see and close the gaps.

• Local, open-weight targets only. Do not point WARDBREAKER at a third-party or hosted API you are not authorised to test. The tool refuses unknown/remote targets by design.
• Benign-proxy seeds only. The bundled seeds are everyday-crime proxies and bare injection patterns. Do not add turnkey dangerous payloads (CBRN, mass-harm, CSAM-adjacent). Demonstrate the mechanism, not a weapon.

If you discover a genuinely novel guardrail weakness (not a reproduction of published work), please practise coordinated disclosure: contact the guardrail vendor's security team and agree an embargo before any public write-up. Lead with the defensive takeaway.

Open a GitHub issue for ordinary bugs. For anything you consider sensitive, contact the maintainer through the GitHub profile linked on the repository rather than a public issue.

It is not a turnkey attack CLI. It refuses unknown/remote targets, ships no working exploits for dangerous categories, and frames every finding as what defenders should fix. Guardrails are one layer of defence in depth — never the whole wall.