Security Policy

Threat Model

MirrorOS is a governance layer — it gates agent actions, not a network perimeter. The relevant threat surface:

Threat	Mitigation
Agent bypasses MRS gate	All actions must pass `MRSBridge.check_authorization()` — never call backend directly
Malicious Prolog assertion	`assert_fact()` validates against Codex before asserting — contradictions are rejected
Tampered audit trail	immudb `verifiedSet` — every write returns a cryptographic proof; `result.verified == True`
Leaked API keys	Keys are loaded from environment variables only — never committed to the repo
Concordance drift (Z3 ↔ Prolog desync)	Boot check in `MRSBridge._verify_concordance_coverage()` raises `ConcordanceError` on mismatch

The following are intentionally absent from the public codebase:

Please do not open a public GitHub issue for security vulnerabilities.

Include:

You will receive a response within 72 hours. We will coordinate a fix and disclosure timeline with you.