Compress natural language memory files to reduce token usage.
npx skills add https://github.com/Ikki-Group/erp --skill caveman-compress-ikki-group
Please help me install this Agent Skill. Skill: caveman-compress Source: https://github.com/Ikki-Group/erp/tree/main/.agents/skills/caveman-compress Command: npx skills add https://github.com/Ikki-Group/erp --skill caveman-compress-ikki-group
Compress natural language memory files (CLAUDE.md, todos, preferences) to reduce the token load on Claude sessions, preserving content and structure while enabling faster session starts.
Run caveman:compress <filepath> to compress a memory file and create a .original.md backup.
High-intent search queries and answers about installing and using this skill.
To compress memory files and save tokens, run caveman:compress <filepath>. This activates a deterministic workflow that reduces natural language content and writes the compressed version back to the original file.
Compressing markdown memory files creates a safe backup of the original content automatically. Your uncompressed file is preserved as <filename>.original.md to prevent data loss during the token reduction process.
If compression loses important content, the workflow runs automatic validation with targeted fixups. It attempts up to two retries to ensure content and structure are preserved while minimizing token costs.
You can use this to compress personal todos and preferences for Claude sessions. The tool detects natural-language memory files and applies a Claude-based reduction to enable faster session starts.
Safety checks when reducing token usage include file-size checks and sensitive path heuristics. These features minimize data leakage risk while backing up your original markdown files before compression.
selfagencyCompress markdown memory files into shorter phrasing while preserving code blocks and URLs.
JeffallanAutomate fine-tuning workflows for large language models using LoRA/QLoRA and PEFT methods.
catlog22Capture session memory entries into Markdown files and a structured JSON index.
junechen7414Compress prose-heavy markdown memory files into a compact caveman format.
