Track Product Events and Analyze User Retention Funnels

Analytics-Product is a reference and code-generation skill for product analytics work built around PostHog and Mixpanel: event tracking design, funnels, cohort/retention analysis, North Star metrics, OKRs, and product dashboards. It fixes an event-naming convention:

[object]_[past_tense_verb]

Correct:   user_signed_up, conversation_started, upgrade_completed
Wrong:     signup, click, conversion

and ships a full illustrative event taxonomy for an assistant product covering acquisition, activation, retention, and revenue events, plus a working PostHog Python integration (Posthog(project_api_key=..., host=...), a track() and identify() wrapper). It provides ready-made code for a cohort weekly-retention matrix (grouping by first-session week and weeks-since-cohort), a North Star metric worked example ("Weekly Active Conversationalists": users with 3+ conversations of 2+ minutes in a week, against year-1/year-2 targets), a SQL-backed North Star dashboard query, a feature-flag check against PostHog's evaluate_flags API, and a statistical-significance calculator for A/B tests using scipy.stats.chi2_contingency, which validates its inputs and refuses to return a result on sparse counts or insufficient variation rather than forcing a verdict. It also defines seven slash-command prompts (/event-taxonomy, /funnel-analysis, /cohort-retention, /north-star, /ab-test, /dashboard-setup, /okr-template) and a six-step funnel-optimization loop (identify the drop-off, understand why via recordings/surveys, form a hypothesis, run a properly powered A/B test, measure against a predefined window and guardrails without stopping early on a favorable p-value, then learn regardless of outcome).

Use it to define an activation event, investigate funnel drop-off, calculate retention with an explicit denominator and time window, set up event tracking or a North Star metric, or run an A/B significance calculation. Before instrumenting anything, the skill requires recording the product decision, data source, applicable consent, timezone, and unit of analysis. Do NOT treat its illustrative funnel percentages, retention ranges, or assistant event taxonomy as real benchmarks - its own limitations section states they are hypothetical and prove nothing about actual user behavior - and do NOT treat an isolated significance p-value as a rollout decision: the skill explicitly says it doesn't measure product value, eliminate bias, or replace confidence intervals and proper experimental design. It also warns that PostHog/Mixpanel SDKs may send data to external services, so properties should be minimized, conversation text avoided, and consent, data residency, and retention rules validated before tracking is activated.

Inputs are a product's tracked events, an analytics question (funnel drop-off, retention, North Star definition, A/B test), or a request to design an event taxonomy. Outputs are Python code: PostHog tracking/identify calls, a cohort-retention calculation function, a North Star SQL query function, a feature-flag check, and an A/B-test significance function that returns control/variant rates, lift, p-value, and a plain-language recommendation rather than an automatic deploy decision.

Built primarily around the PostHog Python SDK (posthog.capture, posthog.identify, posthog.evaluate_flags) and Mixpanel as an alternative platform, with pandas for cohort analysis and scipy.stats for significance testing; runs as a skill inside Claude Code, Antigravity, Cursor, Gemini CLI, and Codex CLI.

Product managers and engineers setting up or reasoning about product analytics - event taxonomies, funnels, retention cohorts, North Star metrics, or A/B tests - who want working reference code and a disciplined measurement process rather than starting from a blank PostHog dashboard.