Show HN: Compactly – Compact Claude Code sessions between tasks (github.com)

The newly launched Compactly plugin for Claude Code offers an innovative solution for managing computational costs during AI/ML task execution. By compacting sessions at natural task boundaries, Compactly creates seamless handoff summaries for subsequent tasks while providing users with real-time insights into session costs and prompt cache expiration. This approach significantly reduces overhead, as large contexts do not need to be re-read in their entirety for unrelated tasks, ultimately optimizing the efficiency of the Claude Code environment. This plugin holds particular significance for developers and researchers in the AI/ML community by enhancing productivity and cost-effectiveness during model training or execution. By learning from user feedback on compaction decisions, it fine-tunes the process to minimize errors while maintaining essential context, making it particularly advantageous for long-running sessions. Compactly requires Claude Code version 2.1.295 or later and adds a user-friendly interface for monitoring session costs and cache utilization, making it a valuable tool for those aiming to maximize their AI/ML workflow efficiency.