CRCT的最小化上下文加载方案
针对海量代码产生的内存压力,CRCT采用智能的按需加载策略:
core technology::
- 层级式上下文隔离:将项目分解为策略层、模块层、文件层三级结构
- 动态加载算法:运行时仅载入
activeContext.md
标记的当前焦点区域 - LRU缓存机制:自动保留高频访问的上下文片段
Configuration steps::
- exist
.clinerules
set up inContextDepth=2
(默认加载两层关联) - 为关键模块在
dependency_tracker.md
increase[priority]
marking - utilization
#pragma context_keep
注释强制保留特定文件上下文
performance optimization::
- align
cline_docs/config.ini
hit the nail on the headmax_context_size
parameters - 对测试代码目录添加
[nocache]
标签避免占用资源 - Execution on a regular basis
Context GC
命令清理碎片化上下文
This answer comes from the articleCRCT: Recursively Decompose Large Code Tasks and Persistently Save Project StateThe