This project is a defensive runtime guardrail for the Odysseus agent. It is research/educational security tooling. It does not include offensive capabilities (those live in a separate red-team project and are reused only as an evaluation oracle).

If you find a security issue — especially a guardrail bypass (a rail that can be made to fail open, an encoding/obfuscation that slips past a detector, untrusted data reaching a tool without a taint check, or a policy-DSL parser edge case) — please report it privately:

• Use GitHub Security Advisories ("Report a vulnerability" on the Security tab), or
• email the maintainer.

Please do not open a public issue for a sensitive report. Include reproduction steps and, if possible, the affected rail/layer.

• We aim to acknowledge reports promptly and to fix confirmed guardrail bypasses on priority, since a bypass defeats the purpose of the project.
• Fixes ship through the normal PR flow with an adversarial regression test added for the bypass.

• Skill/MCP supply-chain attacks (handled separately; see CLAUDE.md).
• Vulnerabilities in the upstream Odysseus agent, the offensive Security_module, or the Agent eval pipeline — report those to their respective projects.