ClawProBench Model Card
这是可索引项目报告证据页：它保留 ClawProBench 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。
ClawProBench Model Card
| Field | Value |
|---|---|
| Repository | suyoumo/ClawProBench |
| Category | Live OpenClaw Benchmark Harness |
| Stars / forks snapshot | 690 / 50 |
| Language | Python |
| License | Apache-2.0 |
| Raw capture | raw-github/suyoumo_clawprobench.md |
| Updated by | hourly public metadata update, 2026-05-26 09:12 +0800 |
1. Role in Self Evolve
ClawProBench is a live-first benchmark harness for evaluating LLM agents in the OpenClaw runtime with deterministic grading, profile selection and repeated-trial reliability. It matters because self-evolving agents need reliable harnesses, memory/skill surfaces, feedback evidence and benchmark loops before any improvement claim can be trusted.
2. Working Principle
OpenClaw runtime task -> live scenario execution -> deterministic grading -> structured report -> leaderboard/profile selection
3. Evidence Path
web-observed GitHub page showed 690 stars, 50 forks, 44 commits, Apache-2.0 license, Python primary language, and README claims for 102 active scenarios, 162 catalog scenarios, deterministic grading, core profile ranking and public leaderboard. Shell GitHub API access remained blocked by DNS and local gh auth was invalid, so this card treats the snapshot as web-observed rather than API-verified.
4. Teaching Use
Use this card to explain Live OpenClaw Benchmark Harness: it shows which part of the agent improvement stack is made operational, and where evaluator, policy, memory, skill, task-state or artifact evidence enters the loop.
5. Limits
The repository was not cloned in this iteration; no benchmark, install flow, workflow run, policy audit, memory experiment, OpenClaw run, or agent loop was executed. Counts and claims are visible public-page/search signals unless independently revalidated later.