strategic-compact

Strategic Compact is a specialized skill for Claude Code that replaces arbitrary, disruptive auto-compaction with intentional, user-controlled memory management. By monitoring tool usage through PreToolUse hooks, this skill tracks interaction volume and suggests compaction specifically when logical task boundaries occur—such as after finishing research, before starting a new implementation phase, or upon debugging completion. This approach ensures that vital task context remains preserved during complex multi-step operations where standard auto-compaction might otherwise purge necessary historical data prematurely. It provides developers with granular control over the agent's context window, optimizing token usage while maintaining task continuity.

• Tracks tool invocations such as Edit and Write to provide data-driven suggestions.

• Configurable threshold detection (default 50 calls) with periodic 25-call follow-up reminders.

• Enables cleaner session states by allowing users to flush exploration noise before moving to execution phases.

• Integrates directly into ~/.claude/settings.json using the PreToolUse hook architecture.

• Supports custom tuning via environment variables like COMPACT_THRESHOLD for different workflow intensities.

• Recommended for developers managing large, multi-file software projects requiring long-term session context.

• Best used after planning phases or error-resolution cycles to clear temporary context without losing the main project state.

• Avoid usage mid-implementation to ensure the model retains immediate code-change history.

• Provides a manual safety valve; the system prompts the user, but the final decision to compact remains with the developer.

• Ideal for environments where tokens must be optimized to keep core project logic in memory while discarding transient exploration traces.