compaction-explainer

Context compaction in OpenClaw is an official session mechanism that summarizes and compresses conversation history to manage context length. It shrinks the active transcript by condensing previous interactions into a shorter, formalized summary.

Preprocessing (assemble) prepares and structures context before model inference, while formal session compaction (/compact) actively summarizes the conversation transcript to reduce its overall length during the session.

A transcript may not shrink immediately after compression because post-processing (afterTurn) or structural formatting overhead can offset the immediate size reduction achieved by the underlying compaction mechanism.

The compaction-debugger plugin handles context manipulation across preprocessing, formal session compaction, and post-processing stages, whereas OpenClaw's internal compaction focuses solely on the built-in formal session summarization mechanism.

Yes, you can use this approach to get clear, non-technical summaries of context management boundaries, explaining the functional differences between preprocessing, session compaction, and post-processing without diving into implementation details.

You should not rely solely on internal compaction when you need granular control over preprocessing or post-processing stages, as its functional boundary is limited to formal session summarization and it may not shrink transcripts immediately.