Warp builds self-improving agents on Claude

Warp created a self-improving agent framework on the Claude platform that splits functionality into two file-based skills: a base skill holding domain knowledge and an outer improver skill that periodically updates the base using human feedback. Reviewers give thumbs-up or detailed critiques, which the improver aggregates, generates focused edits, and merges through standard PR workflows, allowing agents such as code reviewers, spec writers, and triage bots to continuously refine their output in Warp’s open-source repository.