Key Features
Survey of reliable self-evolving agents.
549 manuscript-used research works.
Five self-evolution levels.
Eight catalog sections.
Output-level evolution analysis.
Model, scaffold, improver, and criterion levels.
Evidence and oversight focus.
Paper, code, and OpenReview links.
About Diving into Reliable Self-Evolving Agents
The collection organizes research by the deepest component that changes and takes effect: output-level, model-level, scaffold-level, improver-level, and criterion-level self-evolution. It also asks what evidence can support a claim that an update actually improves performance.
With 549 manuscript-used works, five self-evolution levels, and eight catalog sections, the project is useful for researchers, engineers, and students mapping the field. It links to the paper, OpenReview materials, code, and a structured reading collection.