Agents that review agents

A single model call produces a draft. The gap between a draft and a mergeable PR is review — and that's where most autonomous coding falls down.

cyql closes the gap with a review loop. After an agent writes code, it runs your tests and linters, then a separate reviewer agent checks the diff against your conventions and the original plan. Disagreements trigger a revision, not a merge.

When the loop can't converge, the task replans instead of shipping something wrong. The result is fewer, smaller, better-justified PRs — and a trace that shows exactly how each one got there.