principle-guard-the-context-window

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

Use subagents whenever a payload is large or verbose, such as logs, screenshots, or long documents. The subagent processes the raw data and returns a condensed summary, preserving main-thread context for reasoning.

Content used on every invocation belongs inline in the skill file, since separate files cost a read each time. Only move material to separate reference files when it is needed infrequently.

Context overflow creates compression artifacts and crowds out relevant information, since context spent within a session cannot be reclaimed. Limiting scope per phase and isolating bulk data keeps the window focused on the current task.

These guidelines reduce context pressure but cannot expand the fixed window size. Tasks with inherently large irreducible state may still require session restarts or external persistence of intermediate results.