vermapragya
UserSkills library for product data scientists working with Claude
Categories
Indexed Skills (14)
ab-test-analysis
Analyzes A/B test results with significance testing, confidence intervals, sample ratio mismatch check, guardrail evaluation, and a stakeholder-ready readout. Use when the user mentions A/B test results, experiment readout, test analysis, lift, significance, p-value, treatment vs control, or asks "did the experiment work."
ab-test-design
Designs A/B tests with power analysis, minimum detectable effect (MDE), sample size estimation, randomization unit selection, guardrail definition, and a pre-registration checklist. Use when the user mentions experiment design, A/B test setup, power analysis, sample size, MDE, pre-registration, randomization, or asks "how should I run this experiment."
causal-inference
Estimates causal effects when RCTs aren't feasible — using difference-in-differences, propensity score matching, or instrumental variables — with a decision framework for which method fits. Use when the user mentions causal inference, DiD, difference-in-differences, propensity matching, IV, instrumental variable, "estimate the impact of X without an A/B test", quasi-experiment, or natural experiment.
cohort-analysis
Builds cohort retention tables and retention curves with a single consistent denominator policy. Use when the user mentions cohort, retention, retention curve, "by signup month", "by acquisition channel", N-day retention, churn over time, or lifecycle analysis.
data-quality-audit
Runs a structured audit on a table covering nulls, duplicates, freshness, schema drift, primary key uniqueness, value distributions, and referential integrity. Use when the user mentions data quality, DQ check, audit this table, "can I trust this data", null check, dedup, freshness, or before relying on a new source.
funnel-analysis
Builds step-by-step funnel analyses with conversion rates, drop-off diagnosis, and segmentation. Use when the user mentions funnel, conversion rate, drop-off, signup-to-activation, step-by-step conversion, onboarding flow, or "where are users falling off."
linear-regression
Fits, evaluates, and interprets linear regression for continuous outcomes (revenue, session time, NPS scores) with residual diagnostics and assumption checks. Use when the user mentions linear regression, OLS, continuous outcome, "predict <numeric KPI>", coefficient interpretation, R-squared, or regression diagnostics.
logistic-regression
Fits, evaluates, and interprets logistic regression for binary outcomes (churn, conversion, fraud, adoption). Use when the user mentions logistic regression, binary outcome, churn modeling, conversion prediction, propensity model, odds ratio, classification, AUC, or asks "what predicts X" where X is yes/no.
metric-definition
Writes precise metric specs with grain, owner, source, formula, guardrails, and known caveats. Use when the user mentions metric definition, metric spec, KPI definition, "what is a session", "define X", North Star metric, or needs to disambiguate a metric across teams.
modular-sql-ctes
Refactors SQL into staging, intermediate, and fact CTE layers with explicit grain and naming conventions. Use when the user asks to refactor a SQL query, clean up a model, build a dbt model, modularize a query, or mentions CTE structure, query readability, or "this SQL is hard to follow."
stakeholder-readout
Structures a stakeholder-ready analysis writeup with TL;DR, evidence, decision, and next steps. Use when the user asks for a readout, analysis writeup, insight doc, exec summary, "share this with the team", "make this stakeholder-friendly", or "summarize the findings."
survival-analysis
Runs censoring-aware time-to-event analysis using Kaplan-Meier curves and Cox proportional hazards models. Use when the user mentions survival analysis, time-to-event, time-to-churn, time-to-conversion, hazard ratio, Kaplan-Meier, Cox model, censored data, or duration modeling.
sql-correctness-review
Audits a SQL query's logic for wrong-results bugs — duplicate rows, join fanout, wrong join types, NULL handling traps, and CASE branch issues — with evidence queries that prove or clear each suspicion. Use when the user says "numbers look wrong", "double counting", "rows are duplicated", "this join is exploding", "check my query logic", or when totals don't reconcile.
sql-query-review
Static review of a SQL query — finds anti-patterns, structural problems, and performance issues from the query text alone, then proposes an optimized rewrite with a verification plan. Use when the user says "review this query", "check my SQL", "is this query okay", "clean up this query", or "optimize this" without runtime profile data.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.