SkillRL Model Card
这是可索引项目报告证据页：它保留 SkillRL 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。
SkillRL Model Card
| Field | Value |
|---|---|
| Repository | aiming-lab/SkillRL |
| Category | Recursive Skill-Augmented RL |
| Stars / forks snapshot | 765 / 59 |
| Language | Python |
| License | MIT |
| Raw capture | raw-github/aiming-lab_skillrl.md |
| Updated by | hourly public metadata update, 2026-05-25 |
1. Role in Self Evolve
SkillRL 是通过自动技能发现连接经验轨迹和策略改进的论文代码，把成功/失败轨迹压缩成层级技能库，并在 RL 中让技能库与 agent policy 递归共进化。
2. Working Principle
experience trajectories -> skill distillation -> hierarchical SKILLBANK -> validation-failure analysis -> recursive skill/policy co-evolution
3. Evidence Path
web GitHub page observed 21 commits, MIT license, 765 stars and 59 forks; README says SkillRL bridges raw experience and policy improvement through automatic skill discovery, distills trajectories into a hierarchical SKILLBANK, recursively evolves skills from validation failures, reports 10-20% token compression, and released code/model/data artifacts during Feb-May 2026. Shell GitHub API access remained blocked by DNS and local gh auth was invalid in this run, so this card treats the current snapshot as web-observed rather than API-verified.
4. Teaching Use
Use this card to explain Recursive Skill-Augmented RL in the raw -> classification -> project card -> site/report pipeline. The reading path is: raw capture -> classification row -> public site card -> project report -> aggregate GitHub analysis.
5. Limits
当前未克隆源码，未运行 benchmark、skill install flows、memory experiments、MCP servers、agent evolution loops、security scanners 或 production deployments；star/fork/commit/release 快照来自公开 GitHub 页面文本或可见页面片段。