Relyra sits on the SAML trust boundary. The library assumes the IdP is untrusted input and treats all inbound XML, signatures, RelayState values, and response metadata as potentially hostile.

• SHA-256+ for signatures.
• SHA-256+ for digest methods.
• SHA-1 is rejected unless a time-boxed legacy override is explicitly configured.

• DTDs and external entities stay disabled before parse.
• Signatures are verified against configured certificates only.
• Raw RelayState URLs are rejected.
• Replay protection is required.
• Raw assertions/responses must not be logged.

Before tagging or publishing a release:

• Confirm the Security review packet and findings ledger still match the current release candidate.
• Confirm the public domain / namespace values still match the deployed EntityID, ACS URL, and metadata URL surface.
• Confirm any Keycloak example uses a pinned image tag and refresh that pin only after checking the upstream release notes.
• Run the release parity lane (mix ci.release) before publish.
• Run the security review remediation lane (mix ci.security) after any security-sensitive change.

Use a private GitHub Security Advisory or contact the maintainers privately. Please do not open a public issue for a potential security bug.

The canonical reviewer handoff lives in SECURITY_REVIEW.md, with current dispositions in docs/security_findings.md. The remediation policy is exploitability-first: High and Critical findings block release, while Medium, Low, and Informational findings require explicit written disposition.

Include:

• Affected version.
• Reproduction steps.
• Impact assessment.
• Any sample payloads needed to verify the issue.