AI Code Review Misses Bugs? Split Finding From Filtering

Why does Claude code review miss obvious bugs even on clean diffs? A confidence gate buried in your prompt trades recall for quiet. In a real run, the reviewer noted a 100x billing bug and still returned "No high-confidence issues." This video shows that run, then the two fixes: a finder and verifier split, and a reviewer that can search and read the repo. Every demo is a real run on a demo repo you can try yourself. Try it yourself (demo repo, every command, recorded output): https://github.com/darindeters/field-... Chapters 0:00 A real review that saw the bug and said no issues 0:18 A confidence gate hides real bugs 1:29 Split finding from filtering 2:39 Cross-file bugs need a reviewer that can search 4:07 Review in a fresh context 4:55 Teach taste: rules, examples, focused passes 6:17 REVIEW.md tunes the managed reviewer Sources (official docs) https://code.claude.com/docs/en/best-... https://code.claude.com/docs/en/code-... https://code.claude.com/docs/en/code-... https://code.claude.com/docs/en/code-... Support the channel: https://buymeacoffee.com/techwithdarin #FieldGuide #ClaudeCode #CodeReview