context-compactor

Reduce token bloat by triggering context compression to produce condensed, context-aware summaries. This workflow discards verbose content while preserving essential information to extend the life of memory-heavy interactive sessions.

Context compression works by using a private <analysis> scratchpad for chain-of-thought reasoning that is stripped from the final output. It then generates a 9-section summary to maintain key context during iterative dialogue and session handoffs.

You need to compress session context when token limits are approached in long-running conversations. Triggering compression refreshes memory and extends session life without losing the critical insights built up during the dialogue.

Trigger the compression workflow by saying compact or issuing the /compact command. This activates a trigger-based process driven by context_trigger to refresh memory and produce a 9-section summary of the session.

The best way to handle session handoffs is to apply iterative context compression to produce a condensed, context-aware summary. This preserves essential information while discarding verbose content for the next session.

Yes, you can use a private <analysis> scratchpad for chain-of-thought-like reasoning during context compression. This scratchpad reasoning is stripped from the final output, ensuring only the 9-section summary remains in the active context.