Yunjue Agent Model Card
这是可索引项目报告证据页：它保留 Yunjue Agent 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。
Yunjue Agent Model Card
| Field | Value |
|---|---|
| Repository | YunjueTech/Yunjue-Agent |
| Category | In-Situ Self-Evolving Agent System |
| Stars / forks snapshot | 426 / 49 |
| Language | Python |
| License | Apache-2.0 |
| Raw capture | raw-github/yunjuetech_yunjue-agent.md |
| Updated by | hourly public metadata update, 2026-05-26 05:44 +0800 |
1. Role in Self Evolve
Yunjue Agent is a high-signal research system because it makes tool synthesis, execution feedback, trace release, and benchmark reproduction part of the self-evolution loop.
2. Working Principle
open-ended task stream -> tool generation and verification -> reusable tool/capability accumulation -> benchmark reproduction and trace audit -> next task starts with expanded capabilities
3. Evidence Path
Web-observed GitHub evidence showed 426 stars, 49 forks, 34 commits, Apache-2.0 license, Python-dominant code, public reproduction branch, and public benchmark trace links. The README names HLE, DeepSearchQA, FinSearchComp, xbench-ScienceQA, and xbench-DeepSearch as evaluation surfaces.
4. Teaching Use
Use this card when explaining in-situ self-evolution: the agent does not only tune prompts, it turns execution experience into tools that become future reusable capability.
5. Limits
The repository was not cloned and no benchmark was rerun in this iteration. Shell GitHub API DNS failed, so metadata is web-observed rather than API-verified.