Mapping the explosion of self-improving AI loops

What it does This repository is a curated awesome-list that tracks the sprawling ecosystem of projects inspired by karpathy/autoresearch. It collects autonomous improvement loops, research agents, platform ports, and academic papers into categorized sections with one-line descriptions and star counts. Think of it as a field guide to a niche that is expanding faster than most people can track.

The interesting bit The list treats “autoresearch” as a genre rather than a single tool, capturing everything from Claude Code skills and Gemini CLI plugins to ICLR 2026 Oral papers on genetic-pareto prompt evolution. It is a snapshot of a community trying to formalize the loop between an LLM, a metric, and an unwillingness to stop iterating.

Key highlights

• General-purpose descendants include recursive self-improvement frameworks, dashboard-first runtimes like Thoth, and cross-platform glue such as autoresearch-everywhere.
• Research-agent systems cover end-to-end pipelines that generate literature reviews, run SLURM jobs, and draft LaTeX papers with iterative peer review.
• Academic entries are mixed in with hobbyist skills: ICLR 2025/2026 work on meta-agents, self-improving coding agents, and genetic-pareto optimization appear alongside Claude Code plugins.
• Sections for hardware forks, domain-specific adaptations, benchmarks, and writeups attempt to impose order on a rapidly fragmenting space.
• Curated under CC0 by “Boring Dystopia Development,” which feels appropriately on the nose.

Verdict Worth a bookmark if you are building agentic loops and need prior art, or if you want to see how fast a single demo can colonize an entire sub-field. Skip it if you need a unified framework to run; this is strictly a map, not a motor.