Unverified50% confidenceDecision RuleExact time
A skill-enabled vs. skill-disabled comparison Eval can be used to measure whether a skill is worth occupying context window space; if the baseline accuracy already exceeds 90%, adding a skill may yield minimal gains or even mislead the model.
1
Sources
50%
Confidence
Long-term
Relevance
9/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedSkill分为能力型Skill和偏好型Skill，能力型随模型变强会被淘汰，偏好型更持久需要用评估保护防止性能倒退80% similarUnverified当技能描述过于详尽包含了完成任务所需的全部信息时，模型会跳过技能调用环节76% similarUnverifiedThe article proposes that Skills should be designed according to test workflows, and that the value of a Skill lies not in quantity but in whether it covers a complete workflow closed loop.76% similarUnverifiedThe quality of a Skill is highly dependent on the rigor of its initial creation; if the initial settings are not clear enough, subsequent reuse may amplify errors74% similarUnverifiedSkill的正确使用方式是渐进式优化：使用→发现问题→加规则→再测试，而非追求一次完美70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/925614API
curl https://kongchang.com/api/v1/knowledge/claims/925614MCP
get_claim(id=925614)