context-compression

Context overload in long AI sessions is managed by proactively compressing and summarizing relevant information. This Skill detects completed phases, archives old information, and preserves key decisions and file references to maintain clarity.

Context compression is needed when conversations exceed 20 turns and start degrading focus. It identifies lengthy contexts in ongoing sessions, summarizing completed sections like code reviews to improve productivity and retain critical knowledge.

To summarize long conversation context automatically, run the 'context-compression summarize' command. This identifies completed phases in the current session and generates summaries while retaining key decisions and file references.

Yes, context compression retains file references when archiving old session information. It specifically preserves key decisions and file references while summarizing completed phases to ensure important details are not lost.

A limitation of summarizing context for long session optimization is that it targets sessions over 20 turns. It focuses on archiving completed phases and summarizing them, meaning active or ongoing discussion phases are not compressed until completed.

The best way to maintain context in extensive code reviews is using context compression to summarize completed code sections. This preserves key decisions and file references, allowing the session to maintain focus without losing critical information.