Анализ эффективности скиллов по сессиям

data

skill-monitor is a Claude Code skill that analyzes per-skill effectiveness across sessions by reading `.claude/session-metrics/metrics.jsonl` produced by `/session-scan` and computing aggregated signals for every skill in use. Metrics include total invocations, trigger distribution (user-slash, claude-proactive, nested-skill), action rate, average post-errors, average post-corrections, outcome distribution, and an effectiveness score. Thresholds are type-adjusted — execution skills are held to a higher action-rate standard than analysis or check skills, preventing false flags on tools like verify or triage. The skill also detects auto-load gaps: auto-loadable skills that have never fired as claude-proactive despite five or more invocations are flagged as a description-routing problem. Run with `--improve` to spawn a recommendations agent, or `--window 30d` to change the comparison period from the default seven days.