krzysztofdudek/ResearcherSkill

Implements an autonomous research loop that designs, executes, and evaluates experiments across any measurable domain—code performance, test accuracy, prompt reliability—by branching git commits, reverting failures, and detecting convergence through multi-signal heuristics. Generalizes beyond ML optimization to any codebase or configuration problem where agents can instrument metrics and iterate. Maintains full experiment lineage in a `.lab/` directory while git tracks code changes, enabling session resume and strategy pivoting when plateaus or repeated failures occur.

130 stars.