strategic-compact

To preserve context across multi-phase AI workflows, apply manual context compaction at logical task boundaries. This summarizes and resets the workspace state, keeping only relevant information for the next phase and preventing context bleed.

Manual context compaction is the process of summarizing workspace state to retain essential information during long-running tasks. Use it during milestone transitions or phase boundaries to prevent context loss before moving to execution.

You can trigger context compaction prompts automatically using hook-based triggers via PreToolUse and a configurable threshold. This emits compact prompts and reminders at logical boundaries to refresh context for the next phase.

Yes, this approach works for research-to-implementation workflows by prompting compaction after exploration and before execution. This clears the mental model and prevents context bleed, ensuring only relevant information carries forward.

Manual compaction is preferred over opaque auto-compaction because it prompts you to summarize and reset state at logical task boundaries. This ensures essential information is preserved across phases without losing critical context.

Yes, you can configure the threshold for context compaction reminders. The implementation uses hook-based triggers via PreToolUse to emit compact prompts when the threshold is reached during long-running sessions.