evolution-steward.default

Agent evolution decisions classify root causes of persistent failures into instruction-level, code-level, or systemic categories. It evaluates agent state and evidence to determine if a revision is necessary, then outputs an evolved flag, new revision ID, and detailed rationale.

Root causes of recurring agent failures are classified by reading agent state and evidence, categorizing them as instruction-level, code-level, or systemic. This classification determines whether to trigger an agent revision and delegate the task to the agent factory.

Yes, orchestrating agent evaluation and revision creation without manual triage is possible in the Autonoetic runtime. It uses evidence from the memory curator, maintains safety gates, and delegates revision creation directly to the agent factory.

The best way to route agent improvements after evaluating persistent failures is to delegate the revision task to the agent factory. The evaluation outputs a decision with an evolved flag and new revision ID to trigger the factory-driven revision.

Automated agent evolution should not be used when evidence from the memory curator is insufficient or safety gates are violated. The process requires valid agent state and evidence to accurately classify root causes and decide whether to trigger a factory revision.