We've been using the community version of Claw-Code for about six months, mostly as a learning aid for our junior devs. The move to "Pro" this week, and the new "Smart Rules" they've enabled by default, has fundamentally changed the experience—and not for the better.
Previously, we could tune it to flag a few key patterns (e.g., potential NPEs in our Java services, or missing error handling in new API endpoints). Now, the Pro version comes with a sprawling set of "context-aware" rules that seem to generate a comment on nearly every diff. The signal-to-noise ratio has plummeted. For example:
* It now insists on commenting on every single `@Autowired` constructor in our Spring Boot apps, suggesting we refactor to use Lombok—a style choice we deliberately decided against as a team.
* It's adding "consider extracting this method" suggestions on blocks of code that are already simple, well-named private methods.
* The worst part is the "security" rules: they're flagging theoretical issues in test fixtures and generated code, which just trains the team to ignore all of its comments.
It feels less like a focused reviewer and more like a noisy linter that's been bolted onto our PRs. We're spending more time dismissing its comments than considering meaningful feedback.
Has anyone else made this transition? I'm trying to see if this is just a configuration hurdle. Specifically:
* Did you manage to dial it back to something useful, or did you have to disable entire rule categories?
* Are the new rules actually providing high-value catches in your codebase that make the noise worth it?
I'm starting to think we might need to re-evaluate our tooling choice if this is the intended direction.
gh2
ship early, test often