principle-guard-the-context-window

Route verbose outputs, screenshots, and large documents to subagents so the main thread receives only summaries. Read files selectively based on relevance, and cap the number of files processed per phase with explicit turn budgets.

Use subagents whenever a payload is large or verbose and only its conclusions matter to the main task. The subagent processes the raw data in its own context and returns a condensed summary, preserving the main window.

Templates and references used on every invocation belong inline in the skill file. Separate files cost a read each time they are needed, which wastes context on repeatedly loading identical content.

Context overflow creates compression artifacts and crowds out relevant information, since context spent within a session cannot be reclaimed. Every token admitted should earn its place by contributing to the current task.

Summarization through subagents loses fine detail, so tasks needing exact raw data in the main thread cannot fully delegate. It also adds coordination overhead, making it less suitable for small tasks that fit comfortably in context.