HAL Harness Model Card
这是可索引项目报告证据页：它保留 HAL Harness 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。
HAL Harness Model Card
| Field | Value |
|---|---|
| Repository | princeton-pli/hal-harness |
| Category | Holistic Agent Evaluation Harness |
| Stars / forks snapshot | 289 / 55 |
| Language | Python |
| License | Unknown |
| Raw capture | raw-github/princeton-pli_hal-harness.md |
| Updated by | hourly public metadata update, 2026-05-25 |
1. Role in Self Evolve
HAL Harness 是 Princeton PLI 的 holistic agent evaluation harness，用统一 CLI 在 SWE-bench、USACO、AppWorld、CORE-bench、tau-bench 等任务上比较 agent，同时记录成本、trace 和 leaderboard 提交流程。
2. Working Principle
agent implementation -> hal-eval benchmark adapter -> parallel local/cloud runner -> traces/cost logs -> encrypted leaderboard upload
3. Evidence Path
web GitHub page observed 701 commits, Python primary language, no license field visible in captured page, unified hal-eval CLI, Weave cost logging, Docker/Azure runners, HAL leaderboard integration, 289 stars and 55 forks. Shell GitHub API access remained blocked by DNS and local gh auth was invalid, so this card treats the current snapshot as web-observed rather than API-verified.
4. Teaching Use
Use this card to explain Holistic Agent Evaluation Harness 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、SDK examples、skill install flows、memory experiments 或 production deployments；star/fork/commit 快照来自公开 GitHub 页面文本或可见页面片段。