autoimprove

To improve inconsistent agent prompts automatically, this skill dispatches an auto-improver agent on a separate thread to gather test inputs and run an iterative hill-climbing evaluation loop. It outputs improved prompt versions along with a detailed changelog.

Iterative hill-climbing for prompt optimization works by applying targeted improvements to underperforming prompts using guardrails. The auto-improver agent runs a configurable evaluation loop to iteratively tighten prompts until the desired performance and consistency are achieved.

The best way to optimize a skill without blocking your workflow is to use a separate-thread optimization process. This skill dispatches the auto-improver agent on a dedicated thread, allowing autonomous prompt optimization to run while your main workflow remains active.

Yes, you need to provide test inputs for autonomous prompt optimization. When you supply the target skill or agent name, the system prompts you to approve the test inputs and checklist required to run its configurable evaluation loop.

You should use automated prompt optimization guardrails when a skill or agent produces inconsistent results and requires iterative tightening. Guardrails ensure the hill-climbing mechanism safely targets underperforming prompts without degrading existing functionality.