Mining Hidden Skills from Claude Code Session Logs with Semantic Knowledge Graphs
The article introduces crune, a tool that analyzes Claude Code session logs to extract and surface reusable skills without requiring manual documentation.
Why it matters
Automating the extraction of reusable skills from interaction logs can significantly improve the productivity and efficiency of LLM-based coding assistants.
Key Points
- 1Session logs contain implicit knowledge about user workflows and decision-making processes
- 2crune builds a semantic knowledge graph from session logs, detecting recurring patterns
- 3The tool uses multi-signal feature extraction (text, tool usage, structural) to cluster sessions and identify skill candidates
Details
The article discusses the challenge of turning tacit knowledge into reusable skills when using LLM-based coding agents like Claude Code. The author presents crune, a tool that analyzes session logs to extract this implicit knowledge. crune builds a semantic knowledge graph from the logs, using a combination of text-based (TF-IDF), tool-based (Tool-IDF), and structural features to cluster sessions and identify recurring workflow patterns. This allows the tool to surface potential skill candidates without requiring users to first document their knowledge. The knowledge graph visualization shows how sessions are grouped into topics with multi-signal edges, providing insights into the underlying decision-making processes.
No comments yet
Be the first to comment