What is strategic compaction vs auto-compaction?

Auto-compaction triggers at arbitrary token limits and can lose mid-task context. Strategic compaction is a manual process triggered at logical milestones to ensure important plans and states are preserved while clearing noise.

What data is lost when I run the /compact command?

Compaction clears intermediate reasoning, previous tool call history, and the memory of file contents you've read, though it preserves instructions in CLAUDE.md and your Todo list.

How does the skill know when to suggest a compaction?

The skill tracks the number of tool calls (defaulting to 50) and provides a non-intrusive reminder when you reach high-intensity points in your workflow.

When is the worst time to compact context?

You should avoid compacting mid-implementation, as losing access to local variable names, specific file paths, and partial state reasoning can lead to coding errors.

Strategic Context Compactor

Name: Strategic Context Compactor
Author: flatrick

byflatrick

0•

生产力与工作流

Optimizes Claude Code performance by suggesting manual context compaction at logical task boundaries to prevent information loss.

The Strategic Context Compactor is an essential utility for long-running Claude Code sessions, providing intelligent reminders to use the /compact command at optimal moments. Rather than relying on Claude's arbitrary auto-compaction which can interrupt complex reasoning mid-task, this skill monitors tool usage and alerts developers at logical phase transitions. By clearing out bulky research context or debugging traces before starting new implementation phases, it ensures the AI remains coherent, responsive, and focused on the most relevant information.

主要功能

01Threshold-based monitoring that tracks tool invocations to detect context pressure.

02Customizable tool-call thresholds via environment variables to suit different workflows.

03Decision support logic to help determine if current state is safe to compact.

040 GitHub stars

05Integrated hooks for Edit and Write actions to prompt users at strategic intervals.

06Phase-aware guidance for transitioning between research, planning, and implementation.

使用场景

01Transitioning from an exhaustive research phase to active code implementation.

02Managing high-token sessions (200K+) to maintain AI response speed and coherence.

03Clearing session history after resolving complex bugs to start a new feature fresh.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add flatrick/mdt strategic-compact

For use in Claude.ai and ChatGPT