Hermes Agent Self-Evolution Model Card

这是可索引项目报告证据页：它保留 Hermes Agent Self-Evolution Model Card 的源材料入口、机制线索和限制提醒；正文仍需 reader/editor 与 academic public-copy review 后才能当作最终结论引用。

Hermes Agent Self-Evolution Model Card

One Sentence

Hermes Agent Self-Evolution turns Hermes skills, prompts, tool descriptions, and code into explicit evolutionary search objects.

Three Sentences

The repository uses DSPy and GEPA to generate candidate variants, evaluate them, and select better versions. It is valuable because it makes the improvement surface concrete: the agent does not vaguely “learn”; a skill or prompt changes under evaluator pressure. It also helps connect personal-agent products with benchmarkable self-improvement loops.

Model Card

Teaching Use

Use this project to teach the difference between “agent has skills” and “agent evolves skills.” The first is a library problem; the second is an evaluator and lineage problem. Hermes Self-Evolution is a clean example of a no-GPU, API-driven optimization loop over agent artifacts.

Evidence And Limits

The raw capture reports roughly 3.4k stars, 370 forks, Python implementation, and a design focused on DSPy + GEPA. The main missing evidence is whether reported gains are reproducible across tasks and whether failed variants are archived for regression analysis.