What is Code Review Automation?
Last updated
Code review automation is the use of software tools to analyze, evaluate, and provide feedback on source code changes (typically pull requests) without requiring a human reviewer to manually inspect every line. Many current tools use large language models (LLMs) to judge what a change does, where older tools match fixed patterns.
Why does code review automation matter for engineering teams?
Review queues slow merges, and human reviewers miss defects, especially in long diffs. Automated review does not replace reviewers. It takes a first pass at the diff so humans can spend their time on design, business logic, and architecture.
How does Argus handle code review automation?
By default, Argus triages each changed file (skip, skim, security pass, or full review) and makes one LLM review call per non-skipped file under a 12-rule rubric, the Review Laws, which defines what counts as blocking and what evidence a finding needs. An LLM judge scores the findings, and Argus posts one GitHub review with at most 10 inline comments, blocking findings first. Deep Review, off by default, adds bug-hunter, security, architecture, and regression passes on files triaged for full review, a lead brief, and a second pass on files that already have several findings.