api-analytics-posthog-analytics（API 分析 posthog 分析）

PostHog Analytics Patterns

Quick Guide: Use PostHog for product analytics with structured event naming (category:object_action), server-side tracking for reliability, and proper user identification integrated with your authentication flow. Client-side for UI interactions, server-side for business events. Always call reset() on logout, never store PII in event properties, and use captureImmediate() or await shutdown() in serverless environments.

Detailed Resources:

• examples/core.md - Event naming, user identification, property conventions
• examples/client-tracking.md - React hooks, provider setup, component tracking
• examples/server-tracking.md - posthog-node, serverless patterns, auth events
• examples/group-analytics.md - B2B organization tracking
• examples/privacy-gdpr.md - GDPR consent, cookieless mode, PII filtering
• reference.md - Decision frameworks, anti-patterns, event taxonomy

<critical_requirements>

CRITICAL: Before Using This Skill

All code must follow project conventions in CLAUDE.md (kebab-case, named exports, import ordering, import type, named constants)

(You MUST call posthog.identify() ONLY when a user signs up or logs in - never on every page load)

(You MUST include the user's database ID as distinct_id in ALL server-side events)

(You MUST call posthog.reset() when a user logs out to unlink future events)

(You MUST use the category:object_action naming convention for all custom events)

(You MUST NEVER include PII (email, name, phone) in event properties - use user IDs only)

</critical_requirements>

Auto-detection: PostHog, posthog-js, posthog-node, usePostHog, PostHogProvider, capture, identify, group analytics, product analytics, event tracking, funnel analysis

When to use:

• Tracking user behavior and product analytics
• Setting up conversion funnels and retention analysis
• Implementing group analytics for B2B multi-tenant apps
• Understanding feature adoption and user journeys
• A/B testing analysis (in conjunction with feature flags)

When NOT to use:

• Feature flag implementation (separate concern)
• Error tracking and logging (use dedicated error tracking tools)
• Infrastructure monitoring (use observability tools)

Key patterns covered:

• Event naming conventions (category:object_action)
• Property naming patterns (object_adjective, is_/has_ booleans)
• User identification with authentication flow integration
• Client-side tracking with React hooks
• Server-side tracking with posthog-node
• Group analytics for B2B organizations
• Privacy and GDPR consent patterns
• TypeScript patterns for type-safe events

Philosophy

PostHog analytics follows a structured taxonomy approach: consistent naming conventions, meaningful properties, and strategic placement (client vs server). Track what matters for product decisions, not everything.

Core principles:

1. Server-side for business events - User signups, purchases, subscriptions (reliable, not blocked)
2. Client-side for UI interactions - Button clicks, page views, form interactions
3. Identify once per session - Not on every page load
4. Structured naming - Makes querying and analysis possible at scale

Core Patterns

Pattern 1: Event Naming Conventions

Use the category:object_action framework for consistent, queryable event names.

// category: Context (signup_flow, settings, dashboard)
// object: Component/location (password_button, pricing_page)
// action: Present-tense verb (click, submit, view)

"signup_flow:email_form_submit";
"dashboard:project_create";
"settings:billing_plan_upgrade";

// Simpler alternative: object_verb
"project_created";
"user_signed_up";

Why good: Category prefix groups related events in PostHog UI, enables wildcard queries like signup_flow:*, consistent naming makes analysis possible at scale.

Property naming rules:

• object_adjective: project_id, plan_name, item_count
• is_ / has_ for booleans: is_first_purchase, has_completed_onboarding
• _date / _timestamp suffix: trial_end_date, last_login_timestamp

See examples/core.md for complete naming examples.

Pattern 2: User Identification with Authentication

Call identify() only on auth state change (not every render). Use database user ID as distinct_id. Call reset() on logout.

// Check _isIdentified() to prevent duplicate calls
useEffect(() => {
  if (session?.user && !posthog._isIdentified()) {
    posthog.identify(session.user.id, {
      plan: session.user.plan ?? "free",
      created_at: session.user.createdAt,
      is_verified: session.user.emailVerified ?? false,
    });
  }
}, [session?.user]);

// Always reset on logout
posthog?.capture("user_logged_out");
posthog?.reset(); // Unlink future events from this user

See examples/core.md for full identification hook and logout handler.

Pattern 3: Server-Side Tracking

Track business events reliably from your backend with posthog-node.

// Serverless: use captureImmediate (guarantees HTTP completion)
await posthogServer.captureImmediate({
  distinctId: user.id,
  event: "subscription_created",
  properties: { plan: "pro", is_annual: true },
});

// Always call shutdown before returning in serverless
await posthogServer.shutdown();

Key rules:

1. Always include distinctId (user's database ID)
2. Use captureImmediate() for serverless (guarantees HTTP completion)
3. Always call shutdown() before returning in serverless
4. Configure flushAt: 1 and flushInterval: 0 for serverless

See examples/server-tracking.md for complete server setup and route examples.

Pattern 4: Group Analytics (B2B)

Associate events with organizations using PostHog groups for B2B metrics.

// Client-side: identify organization
posthog.group("company", org.id, {
  name: org.name,
  plan: org.plan ?? "free",
  member_count: org.memberCount,
});

// Server-side: include groups in event
posthogServer.capture({
  distinctId: user.id,
  event: "organization:member_invited",
  properties: { role: data.role },
  groups: { company: data.organizationId },
});

Limitations: Maximum 5 group types per project. One group per type per event.

See examples/group-analytics.md for complete group patterns.

Pattern 5: Privacy and GDPR Consent

PostHog supports cookieless tracking and consent management.

// Cookieless mode: "always" (no consent needed) or "on_reject" (with banner)
posthog.init(POSTHOG_KEY, {
  cookieless_mode: "on_reject",
  person_profiles: "identified_only",
});

// Consent methods
posthog.opt_in_capturing(); // User accepts
posthog.opt_out_capturing(); // User rejects

Key rule: Never store PII (email, name, phone, IP, address) in event properties. Use pseudonymized IDs only.

See examples/privacy-gdpr.md for consent banner integration and before_send filtering.

Performance Optimization

Web Apps (default batching): Use default settings -- PostHog batches efficiently out of the box.

Serverless (immediate delivery):

const posthogServer = new PostHog(POSTHOG_KEY, {
  flushAt: 1, // Flush after 1 event
  flushInterval: 0, // No interval batching
});
// Use captureImmediate() or capture() + await shutdown()

Reducing Costs:

posthog.init(POSTHOG_KEY, {
  person_profiles: "identified_only", // Anonymous events 4x cheaper
  autocapture: false, // Disable for high-traffic sites
});

<red_flags>

RED FLAGS

High Priority Issues:

• Using email as distinct_id -- PII should not be the identifier
• Missing posthog.reset() on logout -- users get mixed together
• No await shutdown() in serverless -- events are lost
• PII in event properties -- GDPR violation risk
• Calling identify() on every render -- performance degradation

Common Mistakes:

• Importing posthog directly instead of using usePostHog hook in React
• Not setting up reverse proxy (api_host: "/ingest") -- events blocked by ad blockers
• Different event names for same action on frontend vs backend
• Not using person_profiles: "identified_only" -- 4x higher costs on anonymous events
• Using capture() instead of captureImmediate() in serverless -- events may not complete

Gotchas & Edge Cases:

• distinct_id is required for ALL server-side events (unlike client-side which auto-generates one)
• group() must include group ID with every event (not persisted like identify())
• Maximum 5 group types per project
• cookieless_mode: "always" disables identify() entirely -- privacy trade-off
• PostHog web SDK is client-side only -- will not work in server components
• Session IDs must be manually passed to server-side events for session linking

</red_flags>

<critical_reminders>

CRITICAL REMINDERS

All code must follow project conventions in CLAUDE.md (kebab-case, named exports, import ordering, import type, named constants)

(You MUST call posthog.identify() ONLY when a user signs up or logs in - never on every page load)

(You MUST include the user's database ID as distinct_id in ALL server-side events)

(You MUST call posthog.reset() when a user logs out to unlink future events)

(You MUST use the category:object_action naming convention for all custom events)

(You MUST NEVER include PII (email, name, phone) in event properties - use user IDs only)

Failure to follow these rules will cause analytics data quality issues, privacy violations, or lost events.

</critical_reminders>