How to Review PRs Automatically with AI Agents (and Not Die Trying)
In teams with fast cycles, PRs pile up and superficial review becomes the norm. AI agents don't replace human judgment — but they can do the first pass and free the reviewer for architecture decisions.
The three-layer workflow: code-review ultra analyzes the complete diff, detects correctness bugs, and leaves inline comments with --comment. security-review runs on auth and validation changes. verify checks that the fix doesn't introduce regressions by opening the browser. Rule: delegate bug detection to the agent, reserve design and business logic for the human.
Install the skills with: npm install -g @anthropic-ai/claude-code && npx autoskills