Indexion Segment

This splits text into meaningful chunks for RAG pipelines and embeddings, which is harder than it sounds if you want segments that actually respect context boundaries. It offers three strategies: window-based divergence detection (the default), TF-IDF for topic shifts, and simple punctuation breaks. The hybrid mode combines divergence and TF-IDF with configurable weights, which is probably what you want for mixed content documents. You get sensible defaults (500 char target, adaptive thresholds) but can tune min/max sizes and divergence sensitivity. Honestly, if you're building a RAG system and currently just chunking every N tokens, this will likely improve your retrieval quality with minimal setup.

Install to Claude Code

npx -y skills add trkbt10/indexion-skills --skill indexion-segment --agent claude-code