status: authoritative
semantics_version: 1.0.0

Epoch-0 governance security policy. Kernel implementation security (credentials, file policy) remains in docs/SECURITY.md.

Full disclosure playbooks and key-compromise runbooks are required before milestone 150 public release (DESIGN_NORTH_STAR.md scorecard row).

Report vulnerabilities to security@clanos.dev (placeholder inbox until public release). Include reproduction steps, affected commit, and threat-node id if known.

Squash and epoch gate commits are GPG-signed. CI verifies signatures against the known-good key registry in keys/reviewer-registry.toml (epoch 0 stub).

Forged epoch_signoffs/ manifests are rejected at gate.

Named ceremony for reviewer key expiry, compromise, or departure:

1. New key added to registry with effective date
2. Historical commits remain verifiable via retired keys section
3. Charter acknowledgment for compromise-driven rotation

Procedure details deferred to milestone 150; attacker class acknowledged in THREAT_MODEL.md.

Re-evaluate signing algorithms against NIST PQC at milestone 150. Migration path noted now; no PQC requirement before 1.0.

None before milestone 150. Policy defined before public release.

Before epoch 0 gate:

• Each domain (kernel, evidence, process) has primary + backup reviewer
• New reviewers co-sign the first 2 epoch gates
• Qualification criteria: demonstrated review of one foundational doc cross-ref cycle

Reviewers must participate in ≥1 epoch gate review per 2 epochs (or 6 months, whichever is shorter).

Stale reviewers re-qualify via co-sign with an active reviewer.

Solo / MV team: same person may satisfy currency for multiple domains with documented multi_domain_roles in sign-offs (CHARTER.md).

Formal proofs and tests verify code as written, not author intent.

Position: malicious or coerced contributors are a residual risk mitigated by:

• Second-reviewer requirement on TCB and unsafe changes (UNSAFE_AUDIT.md)
• GPG-signed gate commits and sign-off manifests
• Split domain review at epoch gates

Full insider-threat modeling in THREAT_MODEL.md.

Rotation, artifact re-sign, revocation of prior images, and runtime response to revoked keys — procedure deferred to BUILD_INTEGRITY epoch 2; attacker class in THREAT_MODEL.md.

Deferred post-150 or epoch-6 hardware measurement chain per ARCHITECTURE_TARGETS.md.

AI-generated or AI-assisted code in kernel TCB paths requires the same second-reviewer treatment as human-authored unsafe code, with explicit reviewer attestation that the output was understood and verified against the relevant KERNEL_OBJECT_MODEL invariants — not merely that it compiles and passes tests.

See also docs/UNSAFE_AUDIT.md.