improvement-loop

Persistent memory scaffolding for AI agents - Trail, Destination, Improve, Retrospect, Intent, Probe. Self-built across 221 iterations; declared complete only when GPT, Claude, and Gemini independently found nothing left to change.

Autonomous skill improvement loop for Claude Code plugins — inspired by Karpathy's autoresearch. Modify → evaluate → keep/discard → repeat until convergence. Zero-touch quality iteration at scale.

The improvement loop for production AI agents. Failures become eval cases; fixes are regression-tested; approved in plain language.

Autonomous code improvement loop - harness-captured LLM evidence, operator-held Commander's Intent, CI-enforced. Reference implementation of Principles of Earned Autonomy.

Real work, fully auditable, in one file: a standalone, target-agnostic improvement-reasoning skill for LLM agents - examines and improves anything the model can reason about (code, documents, plans, letters) while recording every reasoning step in an auditable trail. Single markdown file, no installer, built on the Principles of Earned Autonomy.