Unverified50% confidenceFactExact time
在使用Claude Code时，模型在自然语言层面的描述与底层工具调用的实际参数之间可能存在不一致
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/3/2026
First Seen
Valid until: 10/1/2026
Sources
Claude Code会话回滚功能详解：配合Git工作流实战指南
bilibili啥都懂的程序员6/22/2026
Related Claims
VerifiedClaude Code搭载DeepSeek模型时，与Claude原生模型相比在工具调用（Tool Use）和多步推理的稳定性上存在差异，实际体验未达预期77% similarVerifiedClaude Code的Skill description字段依赖大语言模型的语义理解能力进行意图匹配，而非传统的关键词精确匹配77% similarUnverified当新旧两个技能的触发关键词相同时，Claude Code可能因加载顺序或路径优先级的不确定性而调用错误版本74% similarUnverifiedClaude Code的工具调用机制将模型的思考能力与执行能力解耦，模型输出结构化的工具调用指令，由外部运行时负责实际执行74% similarUnverifiedClaude Code differs from traditional code completion tools by handling the entire workflow from planning to implementation based on natural language descriptions73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/61728API
curl https://kongchang.com/api/v1/knowledge/claims/61728MCP
get_claim(id=61728)