Unverified50% confidenceFactExact time
Besides timestamps, content in the system prompt that sits near the front and changes on every call—such as usernames, tool lists generated in dictionary order that change, debug request ids, and randomly selected examples—destroys prefix stability and causes cache misses.
1
Sources
50%
Confidence
Long-term
Relevance
10/15/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDynamically adding or removing tool definitions in each request will cause the prefix hash to change, invalidating the cache.74% similarUnverified缓存命中的前提是提示词内容保持稳定，若系统提示词中包含动态插入的时间戳、会话ID或随机内容则会造成缓存失效73% similarUnverifiedOnce any minor change appears in the first half of a prompt — such as a dynamic timestamp, a user's name, or a real-time retrieval result — the entire cache prefix is invalidated.73% similarUnverifiedPrompt 缓存依赖于前缀的精确匹配，一旦技能定义的顺序或内容发生微小变化，缓存即失效70% similarUnverifiedWhen the number of tools increases and tools with similar names and overlapping functions appear (such as get_x, fetch_x, read_x), the LLM's tool selection becomes unreliable.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1036603API
curl https://kongchang.com/api/v1/knowledge/claims/1036603MCP
get_claim(id=1036603)