Top AI Product

If you’ve been letting Claude Code, Cursor, or Cline write code for you — and let’s be honest, who isn’t at this point — you’ve probably had that stomach-dropping moment where you realize the agent just did something… wrong. Maybe it half-implemented a feature and claimed it was done. Maybe it quietly introduced a SQL injection. Maybe it deleted something it shouldn’t have. That’s exactly the problem [Vet](https://imbue.com/product/vet/) is trying to solve.

Built by [Imbue](https://imbue.com/), the AI research company, Vet is an open-source tool that acts as a second pair of eyes on your coding agent’s work. It sits between the agent and your codebase, reviewing changes before they land. What makes it interesting is that it doesn’t just do generic linting — it actually looks at your conversation history with the agent and checks whether what was delivered matches what you asked for. It catches the stuff that’s easy to miss: features that are only partially done, tests that were supposedly written but never actually run, hardcoded credentials, path traversal bugs, and plain old logic errors.

You can run it locally after every change, or plug it into your CI pipeline via a [GitHub Action](https://github.com/imbue-ai/vet). It snapshots your repo, diffs the changes, optionally pulls in your agent conversation for context, runs its checks, then deduplicates findings into a clean list of issues. No noise, no walls of text — just the stuff that actually matters.

The timing couldn’t be better. The [Clinejection incident](https://adnanthekhan.com/posts/clinejection/) in February — where a prompt injection hidden in a GitHub issue title led to a supply chain attack compromising around 4,000 developer machines — made it brutally clear how much trust we’re placing in these AI tools without adequate guardrails. That story blew up on [Hacker News](https://news.ycombinator.com/) with 549 upvotes, and it put agent security front and center for the dev community. Vet showed up on [bestofshowhn.com](https://bestofshowhn.com/) shortly after, which feels like perfect timing.

I’ve been running it alongside Claude Code for a couple of weeks now, and it’s caught a handful of things I would’ve missed — nothing catastrophic, but the kind of subtle issues that compound over time. It’s not going to replace your own code review skills, but as a quick sanity check on agent output, it’s earned a permanent spot in my workflow. The (https://github.com/imbue-ai/vet) if you want to poke around or contribute.