Can I customize the frequency of compaction suggestions?

Yes, you can set the COMPACT_THRESHOLD environment variable (default 50) to adjust how many tool calls occur before the first suggestion appears.

How does Strategic Compact differ from Claude's built-in auto-compaction?

While auto-compaction triggers at arbitrary points, Strategic Compact suggests manual intervention at logical task boundaries, ensuring you don't lose critical context mid-implementation.

When is the best time to manually compact my session?

The best time is during phase transitions, such as moving from research to planning, or after completing a major milestone and starting a new task.

What information survives a context compaction?

CLAUDE.md instructions, TodoWrite task lists, memory files (~/.claude/memory/), and your git state all survive, while intermediate reasoning and previous tool history are cleared.

Strategic Context Compaction

Name: Strategic Context Compaction
Author: unju-ai

byunju-ai

0•

生産性とワークフロー

Optimizes Claude's memory by suggesting manual context compaction at logical task boundaries to prevent data loss.

Strategic Context Compaction provides a smarter way to manage Claude's context limits by replacing arbitrary auto-compaction with controlled, manual interventions. It monitors tool call frequency and alerts users at critical project milestones—such as moving from research to implementation—ensuring that essential task plans and goals are preserved while clearing out temporary reasoning and bulky intermediate data. By strategically managing when context is cleared, developers can maintain high performance and coherence during long-running sessions without losing focus on their primary objectives.

主な機能

01Suggests manual compaction at logical milestones rather than arbitrary points

02Monitors tool call frequency to track session context pressure

030 GitHub stars

04Includes a Phase Transition Guide for optimal context management

05Prevents accidental mid-task context loss from built-in auto-compaction

06Customizable alert thresholds via environment variables

ユースケース

01Preserving finalized plans in TodoWrite while flushing bulky research logs

02Clearing messy debugging traces before starting a new feature implementation

03Managing high-token sessions (200K+) during complex multi-phase software development

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add unju-ai/ecc strategic-compact

For use in Claude.ai and ChatGPT