hook-preemptive-compaction

Preemptive compaction monitors context window usage and triggers a prompt to run /compact before hitting the context limit, preventing emergency compactions that waste tokens in long-running AI agent sessions.

Preemptive compaction is a context management technique that periodically injects a short reminder message to trigger context summarization before the context window fills up, reducing token waste from emergency compactions.

You can configure the preemptive compaction threshold by setting a custom value (the default is 0.7) and an optional prompt in your harness, enabling the hook to trigger compaction reminders at your preferred context window capacity.

Yes, preemptive compaction is specifically applicable to high-context workflows and long-running sessions where token waste from emergency compactions is a concern, injecting periodic reminders to manage the context window proactively.

The preemptive compaction reminder message consumes approximately 50 tokens, periodically injecting a short prompt into the context window to suggest running the /compact command without adding significant token overhead.