self_reflect

To extract actionable lessons from completed tasks, you run a post-task reflection flow with the task goal and final status. This process summarizes outcomes and optionally persists the reusable experiences into a memory backend for knowledge retention.

You do not need a memory backend to run post-task reflections, but it is optional for persisting reusable experiences. The core extraction of actionable lessons operates in a stateless invocation model without requiring persistent storage.

The best way to turn runtime task outcomes into reusable experiences is through post-task reflection. By reviewing both successes and failures, you summarize outcomes into lessons that inform future decisions and guide continuous improvement in runtime agents.

Yes, you can use runtime reflection to review failed tasks. It supports runtime-driven reviews after both successes and failures, allowing you to extract actionable lessons that directly inform future decisions and improve subsequent task execution.

The self_reflect Skill operates in a stateless invocation model across task reviews. It requires a runtime reflection backend to process each review independently, while an optional memory backend handles writing the extracted lessons for long-term retention.

A limitation of stateless post-task reflection is that it cannot inherently retain context between invocations. Continuous improvement relies on an optional memory backend to store the extracted lessons, without which the reusable experiences are lost after the review.