mining-session-patterns

You can mine Claude Code session transcripts to surface recurring error→fix sequences, tool failure rates, and cost signals. This Skill samples the most recent transcript files and extracts actionable learning patterns to support process improvements.

Session mining is the process of analyzing Claude Code transcripts to surface recurring tool errors and fixes. It identifies sequences where tool failures occur, quantifies per-tool failure rates, and estimates token cost signals to drive reliability improvements and learning.

You quantify tool failure rates by applying a sampling strategy to the most recent transcript files, then calculating the frequency of per-tool failures. This Skill processes the latest session logs to generate reliability metrics and writes the findings to a user-specified path.

Yes, you can analyze session transcripts without external dependencies. This Skill operates independently with no required dependencies, directly parsing transcript files to extract tool failure rates and cost signals before writing the results to your specified output path.

You estimate token usage cost signals by mining session transcripts for token consumption data. This Skill analyzes recent Claude Code logs to assess session complexity and calculate cost signals, writing the extracted patterns to a user-specified output path like a markdown file.

Limitations of transcript pattern analysis include its enforced sampling strategy, which restricts mining to only the most recent transcript files. This ensures manageable processing but means older historical sessions outside the sample window are excluded from the pattern extraction.