self-improve

Autonomous codebase improvement uses an orchestration loop coordinating research, planning, execution, and evaluation to iteratively enhance repositories. It manages setup and stop conditions without user prompts, applying tournament-style evaluation to benchmark parallel experiments and merge winners into an improvement branch.

Tournament-style evaluation runs parallel experiments, benchmarks winners, and archives results for visualization. It maintains detailed histories and plots for progress monitoring, supporting iterative hypothesis evaluation to measure and compare codebase improvements against defined benchmarks.

Provide the path to your target repository and start the self-improvement loop. The tool requires a software project with measurable goals and a defined benchmark, using isolated git worktrees for experiments and merging winners into an improvement branch.

Yes, the orchestration creates isolated git worktrees for parallel experiments, merges winners into an improvement branch, and pushes changes when configured. It requires a defined benchmark to evaluate measurable goals and iteratively test hypotheses for repository enhancement.

Yes, matplotlib is a required dependency for plotting progress. The orchestration maintains detailed histories, reports, and plots for stateful traceability, enabling visualization of benchmark results and iterative improvement progress across the codebase.

Autonomous self-improvement requires a software project with measurable goals and a defined benchmark. It is limited to repositories where iterative hypothesis evaluation can be applied, needing git-centric workflows with isolated worktrees and clear stop conditions for effective orchestration.