strategic-compact

Context preservation during long AI-assisted sessions requires strategic compaction at logical task boundaries. By installing a PreToolUse hook, the system tracks tool usage and prompts manual compaction at safe handoff points instead of relying on error-prone auto-compaction.

Strategic context compaction is the process of manually condensing conversation history at natural task phase transitions. It is needed during multi-phase projects, long-running sessions, and frequent task-switching to sustain workflow coherence across research, planning, and implementation steps.

Setting up compaction reminders involves installing the suggest-compact.js hook in your Claude configuration and defining a COMPACT_THRESHOLD. Once configured, the hook monitors your session and triggers prompts to compact context when the threshold is reached.

Manual context compaction works better than auto-compaction because it aligns with logical task boundaries. By marking natural handoff points between phases like research and implementation, it prevents the lost or muddled context that auto-compaction often causes.

You can configure the compaction prompt threshold by adjusting the COMPACT_THRESHOLD variable. This setting controls how frequently the suggest-compact.js hook evaluates tool usage before triggering a reminder to compact your session context.

AI sessions lose context after multiple task switches because auto-compaction drops crucial information at arbitrary limits. Strategic compaction solves this by identifying logical task boundaries and preserving essential context through a guided manual compaction process.