General questions for a security architecture or design review. Not every question applies to every system, and answering them is not a complete security assessment. Record each answer and any follow-up, rather than just ticking it off.
Trust boundaries and communication
Start by agreeing where trust changes, because most of the questions below apply at those points.
Watch for: flows that are described as internal and therefore trusted, without a stated reason.
Identity and authentication
Every boundary crossing should answer the question: who is calling, and how do we know?
Watch for: a single gateway assumed to authenticate every route, without a check that it does.
Authorization
Authentication says who; authorization says what they may do.
Watch for: authorization enforced only in the user interface, not in the API behind it.
Sensitive data handling
Follow sensitive data from where it enters to where it leaves.
Watch for: copies of production data in lower environments or analytics tools.
External dependencies
Each dependency is a trust decision.
Watch for: outbound integrations that accept a user-supplied URL.
Interfaces and input handling
Every entry point is attack surface.
Watch for: undocumented or legacy endpoints that are still deployed.
Logging, monitoring and detection
If something goes wrong, will anyone know, and can they reconstruct what happened?
Watch for: actions that cannot be attributed to a person or service, which creates repudiation risk.
Failure scenarios and resilience
Security includes behaving safely when things fail.
Watch for: a single compromised component that holds credentials to everything else.
Recording the review
For each question, record the answer, the evidence (a diagram, configuration or document), and any follow-up as a specific finding with an owner. Unanswered questions are findings too. Write down the assumptions the design relies on, so they can be checked again when the system changes. The fundamentals guide describes what a useful finding contains.
Astrix gives a review a shared threat model to work from, with attack paths and heat maps to set priorities, comment threads, owner, editor and viewer roles, an audit trail, and PDF or DOCX reports. It does not answer these questions automatically or approve an architecture; reviewers do. Read the design review use case.