reflection-injection

To prevent an agent from repeating past mistakes, you inject relevant past reflections into its context. This automatically loads filtered lessons learned from previous failures and prepends them to the current iteration for improved decision-making.

Episodic memory injection works by loading reflection history from specified directories and filtering it based on task or error similarity. Relevant lessons are formatted using a provided template and injected into the agent context before new iterations.

The best way to apply lessons learned from previous agent failures is to automatically filter past reflections by error type and inject them into the current task context. This ensures the agent considers why it failed before attempting the new task.

You can track the success rate of injected reflections using built-in usage tracking. It monitors the effectiveness of injected reflections to refine future learning and ensure only the most relevant past lessons are applied.

You do not need to manually format past reflections before adding them to agent context. The system automatically formats relevant reflections using a provided template and tracks their usage when starting new iterations or retrying after failures.