pm-root-cause
Diagnostic engine for in-flight anomalies - drills to the real root cause, not the first plausible one
/pm-root-cause▊When to use
When something live behaves unexpectedly: a metric dropped, churn spiked, a feature isn't adopted, conversion fell, tickets are rising. Cross-phase - run whenever an anomaly needs structured diagnosis, at any point in the product lifecycle.
Agent mode is not supported. The value is the live investigative dialogue - hypothesis drill-down requires interactive back-and-forth. Invoking with --agent will warn once and proceed interactively.
What it produces
A structured root cause investigation with:
- Reality check - is this a real anomaly or a measurement artifact (data quality, segment definition, seasonality, instrumentation bug)?
- Concentration analysis - where does the anomaly concentrate? Segment / funnel stage / cohort / geography / feature / device?
- Change inventory - what changed (product, marketing, competition, external event) in the relevant time window?
- Candidate cause differential - causes ranked by evidence strength, across categories: product, data/measurement, market, operational, team
- 5-Whys drill - structured iterative drill on the highest-confidence candidate
- Evidence vs. guess separation - what is confirmed vs. inferred vs. assumed
- Cheapest next test - the lowest-cost experiment to confirm or reject the root cause hypothesis
- Testable hypotheses - formatted for input into
pm-hypotheses
Built-in diagnostic methods
- 5-Whys drill
- Fishbone (Ishikawa) categorization
- Anomaly-to-cause differential library
- Measurement trap catalogue (common instrumentation errors)
- Cognitive bias catalogue (availability, confirmation, recency bias checks)
How to invoke
/pm-root-cause # interactive only - --agent not supported
Describe the symptom: what metric, what change, over what time period, compared to what baseline. The skill takes it from there - you do not need a hypothesis before starting.
Dependencies
Recommended (provides context):
pm-kpis- KPI framework helps scope the anomaly and confirm the metric definitionpm-personas- segment breakdown informs where the anomaly concentratespm-hypotheses- testable hypotheses from this session feed directly into a new experiment plan
Produces for:
pm-hypotheses- root cause hypotheses become entries in the Hypothesis Register for validationpm-prioritize- confirmed root cause may change feature priority
Related skills: pm-hypotheses, pm-problem-validation, pm-kpis, pm-personas, pm-prioritize