Build a self-improvement loop for your coding agents

A poorly written agent Skill can burn through hundreds of unnecessary tool calls, and still produce code that falls short of your team’s standards.

Agents need a feedback loop for continuous improvement. In this hands-on workshop, we’ll show you how to build one: a system that evaluates agent runs, identifies weaknesses, and turns those findings into concrete Skill improvements.

An evaluator scans your team’s conversation logs, delegates each run to scorer agents, and grades the results against criteria you define, such as code quality, efficiency, and task compliance. A self-improvement agent then uses those scores to propose Skill updates for your team to review and merge.

We’ll share real self-improvement results from our own repositories, including the Skill changes we shipped and how they improved the code produced by our agents.

Plus: how to run the full evaluation and improvement loop with Warp Factories, so agent quality improves continuously as your team works.

Who it’s for: Engineers and engineering leaders deploying coding agents who want a repeatable, measurable process for improving code quality over time.