Stop repeating the same mistake
Failure clustering is now generally available. Recurring agent failures surface as a pattern — and one line in your config makes them stop.
By The cyql team
The most expensive kind of mistake is the one you make twice. When an autonomous agent misses a convention — the wrong error-handling style, a missing null check, a test file in the wrong directory — it will keep missing it on every subsequent task unless something changes.
Failure clustering groups recurring failures automatically. When cyql sees the same class of error across multiple runs, it surfaces the pattern in your dashboard with a description of what went wrong and a suggested guardrail. You review it, approve it, and it becomes a rule the agents follow on every task going forward.
The guardrail mechanism is deliberately simple. Rules live in a plain-text config file in your repository, versioned alongside your code. They read like conventions — because that is what they are. The goal is not to patch a model; it is to encode what your team already knows.
Failure clustering is now generally available on all plans. If you have been running tasks for a while, open the Failures tab — you will likely find clusters already waiting.