What problem does it solve? It answers whether your prompting is actually improving over time by auditing every Claude Code transcript on your machine, separating real trends from statistically unsupported patterns like model-release confounds and prompt-form correlations. ## Core Features & Use Cases - Corpus-wide trend analysis: Parses all local sessions to compute rework rates, week-over-week direction, and per-project comparisons with statistical gating. - Incident taxonomy with cost attribution: Counts and categorizes friction incidents, surfacing repeat-prompt and correction exemplars you can learn from. - Honest reporting: Marks unreliable signals, non-comparable model pairs, and caveats explicitly instead of inventing confident conclusions. - Use Case: Run /qtrends after a month of Claude Code usage to see whether your rework rate is dropping, which projects go badly, and which prompt habits correlate with friction — rendered as an interactive HTML report. ## Quick Start Ask the assistant to run /qtrends to analyze all Claude Code sessions on this machine and open the trends report.
