ralph

To reduce ambiguity and drift in long-running AI coding tasks, use a specification-first workflow that surfaces hidden assumptions through Socratic interviewing before execution begins. This crystallizes requirements iteratively.

The best way to expose hidden assumptions in an ambiguous prompt is applying Socratic interviewing. This technique iteratively questions the task requirements to crystallize specifications before any code is generated.

You can set up a specification-first loop by initiating a persistent cycle that moves through interview, seed, execute, evaluate, and evolve phases. This loop repeats until verification passes and ontology converges.

Yes, you can configure max iterations and set completion promises for long-running AI agent tasks. The persistent loop mode supports these configurations alongside multiple agent minds and safe drift monitoring.

Yes, the specification-first approach supports multiple agent minds and includes safe drift monitoring. It maintains a persistent evolutionary loop that manages these minds until ontology convergence is achieved.

You should not use a Socratic interviewing loop for simple, well-defined tasks that lack ambiguity. The loop is designed for long-running tasks and ambiguous prompts where hidden assumptions need to be exposed and requirements crystallized.