What problem does it solve? Operators running many scheduled automation skills lack a single ranked, fleet-wide view: per-run heartbeats show only binary ok/not-ok, and per-skill health checks audit one skill at a time, so silent skills, failure streaks, and skip-heavy behavior go unnoticed. ## Core Features & Use Cases - Fleet-wide ranking: Pulls GitHub Actions run history via a skill-runs script and ranks every skill by run count and success rate over a configurable window (default 7 days). - Anomaly detection: Flags silent scheduled skills, all-fail runs, consecutive failure streaks, low success rates, all-skip behavior, and duplicate runs using a priority-ordered taxonomy. - Exit taxonomy mining: Greps daily logs for markers like SKIP_UNCHANGED and NEW_INFO to distinguish healthy-quiet skills from silently broken ones. - Use Case: An operator with 80 scheduled skills runs this weekly to produce a Markdown report, a dashboard JSON spec, and a significance-gated notification only when anomalies exist. ## Quick Start Run the skill-analytics skill to generate a 7-day fleet report ranking all scheduled skills by runs, success rate, and anomaly flags.
