strategic-compact

To prevent AI context loss during long Claude sessions, use milestone-based compaction to prompt manual context preservation at logical phase transitions instead of relying on automated truncation. This preserves essential information before context risk increases.

You should run /compact for AI context management after exploration phases, after reaching milestones, and before major context shifts. A PreToolUse hook tracks tool usage and emits timely prompts when a configurable threshold is exceeded.

To configure a PreToolUse hook for manual compaction, load the strategic-compact script into Claude, set the COMPACT_THRESHOLD to your preferred value (default is 50), and rely on the hook's prompts to guide your /compact execution.

The best way to manage context during exploration-to-implementation work is tracking tool usage to identify when context should be compacted, prompting for manual compaction at logical milestones to avoid arbitrary truncation of essential task information.

Yes, you can adjust the threshold for context compaction prompts. The skill provides a configurable COMPACT_THRESHOLD variable with a default value of 50, allowing you to control when the hook triggers guidance to run /compact.

Automated context truncation causes problems during milestone reviews because it arbitrarily removes information without recognizing logical phase transitions. Manual compaction at milestones ensures essential exploration and implementation context is preserved intentionally.