Localize metric changes through decomposition, drift classification, and causality checks.
npx skills add https://github.com/hero-engine/hero --skill metric-rca-hero-engine
Please help me install this Agent Skill. Skill: metric-rca Source: https://github.com/hero-engine/hero/tree/main/domains/pm/skills/metric-rca Command: npx skills add https://github.com/hero-engine/hero --skill metric-rca-hero-engine
High-intent search queries and answers about installing and using this skill.
Decompose the metric into its components (e.g., Revenue = Users × ARPU) and build a table showing which component carries the move. Then classify the drift type and rank candidate causes, each with a confirming cut that would validate or kill it.
Metric-tree decomposition breaks a top-line metric into the components that multiply or add up to it, such as Conversions = Sessions × Conversion-rate. It localizes a move to a specific leaf so the investigation targets one component instead of the headline number.
Check the influence drift class: segment the metric and weight each segment by prior-period shares to hold the mix constant. If every segment is flat while the blended number falls, the move is a composition shift, not a real decline.
No. Temporal coincidence is a hypothesis, not a cause. Run a confirming cut such as comparing the metric for users who saw the change versus those who did not in the same window before asserting causality.
When observational cuts cannot resolve the cause, the answer is a pre-registered experiment rather than a stronger assertion. RCA produces ranked, disconfirmable hypotheses; experiments settle causation that cuts cannot.
oldwinterDiagnose two-sided marketplace liquidity bottlenecks and generate management packs.
AndurilCodeDesign experiments with pre-registered metrics, sample sizing, and validity mitigation.
mverabAnalyze incident logs to identify error patterns and root causes.
huiferMonitor business metric streams and alert on abnormal patterns.
