rag-knowledge-capture-cli

To save Claude Code CLI sessions as RAG knowledge chunks, run this skill at session end to transform problem-solution pairs into self-contained 150–400 word Markdown documents. These standalone chunks are then saved to .claude/summaries using a date-slug filename for local persistence and easy retrieval.

Generating embeddings from CLI sessions with Qdrant involves creating self-contained Markdown chunks that are structurally optimized for embedding. The skill outputs a push-to-qdrant.sh command, providing the exact shell instruction needed to index the locally saved knowledge chunks into your Qdrant collection.

The Session End Protocol supports automatic knowledge capture by integrating directly with this skill. It triggers the summarization workflow before your Claude Code CLI session ends, ensuring that problem-solution pairs are automatically converted into embeddable chunks without manual intervention.

You can use this skill to create self-contained chunks for retrieval systems by enforcing a strict frontmatter schema. It processes Claude Code sessions into standalone 150–400 word Markdown documents that carry full context independently, making them immediately ready for embedding and retrieval.

The best way to format Claude Code session summaries for Qdrant is to use a strict frontmatter schema and chunk them into 150–400 word self-contained documents. This skill applies that formatting automatically, saving the output to .claude/summaries to ensure optimal embedding quality and retrieval accuracy.

You do not need Qdrant installed to generate knowledge chunks, as the skill saves all output locally to .claude/summaries. However, to complete the pipeline and index the generated chunks, you will eventually need a Qdrant collection to execute the provided push-to-qdrant.sh command.