Unverified50% confidenceBenchmarkExact time
SkillScaffold was evaluated on four benchmarks—ALFWorld, WebShop, ScienceWorld, and τ²-Bench—and surpassed strong agentic RL baselines and skill-driven RL baselines on both short-horizon and long-horizon tasks.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/14/2026
First Seen
Valid until: 1/12/2027
Sources
Related Entities
Related Claims
UnverifiedSkillBench focuses on measuring the incremental performance of a model+Skill combined system on specific tasks, unlike traditional benchmarks such as MMLU and HumanEval that measure general model capabilities.70% similarUnverifiedA total of 6 skills were loaded in the Dataverse plugin demonstration, with the top-level DV Overview skill serving a routing role to determine which skills should be invoked.68% similarUnverifiedThe key skills corresponding to each stage are: Prompt Engineering for the Prompt stage, RAG/vector retrieval for the Context stage, tool integration/MCP protocol for the Harness stage, and Loop Engineering for the Loop stage.67% similarUnverifiedSkills Store采用策展型（Curated）平台模式，审核机制包含功能可用性测试、安全性审查和用户体验标准67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1028036API
curl https://kongchang.com/api/v1/knowledge/claims/1028036MCP
get_claim(id=1028036)