Awesome-Self-Improving-Agents: Explore Self-Improving Agent Research

Overview

Awesome-Self-Improving-Agents is a curated research index accompanying a survey on self-improvement in modern agentic systems. It helps readers navigate a broad literature by distinguishing updates to a foundation model from changes to an agent's surrounding scaffold, such as its prompts, memory, tools, or workflows.

This is a reference resource, not an agent framework or runnable implementation. It is useful when surveying the field, finding related work, or locating papers, code links, benchmarks, and other learning materials. The project also has a website and welcomes contributions through pull requests.

Key Features

• Groups literature into foundation-model improvement and scaffolding improvement.
• Breaks down model improvement by self-generated demonstrations, evaluative feedback, and exploratory experience.
• Organizes scaffold research around prompts, memory, tools, and full-agent systems.
• Includes paper tables with publication years, venues, paper links, and code links where listed.
• Covers evaluation and benchmarking, including metric-based and judge-based measurement.
• Collects related resources such as blogs, podcasts, interviews, videos, workshops, and courses.
• Provides an accompanying survey paper and an agent-oriented guide.

Use Cases

• Researchers starting a literature review can use the taxonomy to identify relevant strands and representative papers.
• Agent developers comparing approaches can browse examples of model updates, memory changes, prompt optimization, and tool adaptation.
• Students learning the field can use the survey and categorized reading lists to build a structured study path.
• Benchmark designers can consult the evaluation section to find examples of measurement approaches and benchmarks discussed in the literature.

Getting Started

No software installation or run command is provided. Browse the README or the project website to explore the survey and categorized resources.

Alternatives

• Awesome-Self-Evolving-Agents: Covers agent self-evolution across model, environment, and combined evolution, while Awesome-Self-Improving-Agents focuses on model and scaffold improvements.

Considerations

• This repository is a curated bibliography and resource hub, not a software package for building or running agents.
• The resource lists are broad and evolving. Check linked papers and repositories for current details and availability.
• The listed implementation language is TeX, reflecting the survey materials rather than an agent runtime.