Unverified85% confidenceFactTime unknown
Context Mode的快照机制在压缩发生前将任务状态（包括规划树、文件修改记录、决策上下文）序列化保存，压缩后再注入新的上下文
1
Sources
85%
Confidence
Long-term
Relevance
6/3/2026
First Seen
Sources
两个MCP插件组合，解决Claude Code大项目失忆难题
bilibiliAi学习的老章
Related Entities
Related Claims
UnverifiedContext Layer会在每次发消息时自动准备相关背景知识，属于确定性触发而非等LLM决定，包含Ontology Context、Document Context和Function Backed Context三种方式73% similarUnverified在上下文即将超载前，将关键信息压缩总结转移到新对话实例可规避长上下文成本爆炸，该模式在Cursor、Windsurf等AI编程工具中被称为Context Condensation73% similarUnverified上下文编译（Context Compilation）是在调用大模型之前对原始资料进行结构化处理的过程，包括扫描证据目录、生成证据清单、抽取哨兵字段、登记冲突风险四个步骤73% similarUnverifiedReAct框架因三元组编码在文本序列中，上下文窗口消耗随任务步骤数线性增长，在长任务场景下会面临上下文截断问题72% similarUnverifiedCommon context trimming strategies include sliding window (keeping the most recent N messages), summary compression, and importance filtering.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/39426API
curl https://kongchang.com/api/v1/knowledge/claims/39426MCP
get_claim(id=39426)