关于
This skill provides a comprehensive framework for managing LLM context windows in complex, long-running agent tasks. It moves beyond simple truncation to implement Anchored Iterative Summarization, a method that preserves critical information like file modifications, architectural decisions, and session intent through structured metadata. By focusing on the tokens-per-task metric, it ensures that Claude remains technically accurate and maintains an unbroken artifact trail even when dealing with codebases exceeding millions of tokens, preventing the common issue of agents forgetting previous file changes or technical constraints.
