I just want code reviews to be useful. They should be about trade-offs, sharing knowledge, and holding each other accountable, not on policing style, naming quirks, or catching trivial regressions and security issues. Tools similar to Cursor Bugbot flood PRs with noise, run in a black box, and bill you per seat for the privilege... they try to solve problems too late, in the wrong place, and without your developer environment tools.

That turned into AgentCheck: an open-source subagent that runs locally with five focused reviewers - logic, security, style, guidelines, and product. It bakes in a lot of internal experience in building enterprise products, working in agentic-native SDLC, and solving Claude Code shenanigans. If you’re curious about the internals or have ideas for new reviewers, feel free to ask questions or contribute.