How to review AI-generated code before the PR

Use a local security scan loop, a finding-first prompt, and a CI gate to keep agent output reviewable.

Apply this guide locally Inspect product evidence

radar scan . --quick

The working loop

Scan locally, fix the first risky lane, ask your agent for a scoped change, then rescan before pushing.

Guide

Evidence to inspect

Use “How to review AI-generated code before the PR” as the scope for this decision: verify the input, finding detail, workflow handoff, and product boundary before you install or buy.

CriterionEvidence to inspectBoundary

Input scopeSelected files, configuration, scan mode, and enabled rules.Only included paths and configured checks are evaluated.

Finding detailFile, line, rule ID, severity, explanation, and repair direction.Illustrative output is not a result from your repository.

Workflow handoffLocal result, report format, agent context, and optional CI signal.Enable exports or CI only when the workflow needs them.

Decision fitUse the same criteria on a real repository before choosing a plan or tool.No universal winner or guaranteed outcome is claimed.

Guide

Apply this guide locally

Official source

Primary sources

NISTSecure Software Development Framework (SSDF) 1.1
OASIS OpenSARIF 2.1.0 standard

Validate the workflow on your own code.

Apply this page’s evidence to one real repository. For “How to review AI-generated code before the PR”, confirm which finding is produced, whether the proposed next step is reproducible, and where local scanning, reports, agents, or CI should stop or expand.