Security model overview
The layered claim in one page, with a link to the page that proves each layer.
Identity & the platform envelope
The Kamiwaza forwarded envelope as the only identity source Kaizen accepts.
Permissions & ACL enforcement
Enforcement inside the query, the trusted Postgres boundary, and viewer projections.
Guardrails & policy
Budgets, citation discipline, redaction, and loop control on every turn.
Approvals & the write guard
Why an agent can propose a side effect but never authorize one.
Isolation boundaries
Sandbox, package runner, egress proxy, and per-participant workspaces.
Audit trail
What is recorded, what is deliberately excluded, and how to read a denial.
Audit event reference
Every audit action, its actor, and its safe fields — as a lookup table.
Data handling & retention
Where content lives, how long it stays, and what deletion does and does not promise.
Task automation security
Why scheduling work creates no new identity, credential, or approval authority.
Threat model summary
The trust boundaries, the assets behind each, and the controls on the edge.
Compliance & evidence
What Kaizen can produce for a reviewer, and what it will not retain.
Licensing & notices
The platform's licensing boundary, EULA acceptance, and the bundle in nine release images.