Birfy / agentdescent

Gradient descent, but the parameters are agents — a parallel, asynchronous framework for self-evolving agents (skills, prompts, harnesses). Diffs are the gradients; the aggregator is the optimizer.

AMAP-ML / SkillClaw

Let Skills Evolve Collectively with Agentic Evolver